LATE QUATERNARY SEDIMENT & LANDSCAPE DYNAMICS IN THE DIJLE CATCHMENT

Transcription

1 EXCURSION GUIDE LATE QUATERNARY SEDIMENT & LANDSCAPE DYNAMICS IN THE DIJLE CATCHMENT BELQUA excursion September 22nd, 2011 Dr. Bastiaan Notebaert Nils Broothaerts Prof. Dr. Gert Verstraeten Prof. Dr. Jean Poesen Earth & Environmental Sciences KU Leuven Division of Geography

2 Excursion stops: Meerdaalforest: more than 3000 years of human impact (including erosion processes) Nodebais (Hamme Mille): erosion and colluviation in the Belgian Loam belt since the introduction of agriculture Lunch break near Korbeek Dijle Korbeek Dijle: floodplain development during the Holocene how natural are our river systems? Rotselaar: pleniglacial and Holocene changes in river systems When time permits: Haacht Sint Adriaan (also human vs. natural factors in explaining the development of the fluvial landscape of the lower River Dijle) Papers in this excursion guide: Vanwalleghem et al., Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium). Geomorphology, 56 (1 2), Vanwalleghem et al., Reconstructing rainfall and land use conditions leading to the development of old gullies. The Holocene 15,3. 4. Vanwalleghem et al., Prehistoric and Roman gullying in the European loess belt: a case study from central Belgium. The Holocene, 16 (3). Vanwalleghem et al., Origin and evolution of closed depressions in central Belgium, European loess belt. Earth surface processes and landforms 32, Rommens et al., Soil erosion and sediment deposition in the Belgian loess belt during the Holocene : establishing a sediment budget for a small agricultural catchment. The Holocene 15,7. Rommens et al., Holocene alluvial sediment storage in a small river catchment in the loess area of central Belgium. Geomorphology, 77, pp Rommens et al., Reconstruction of late Holocene slope and dry valley sediment dynamics in a Belgian loess environment. The Holocene, 17(6), Verstraeten et al., A temporarily changing Holocene sediment budget for a loess covered catchment (central Belgium). Geomorphology, 108(1 2), Notebaert et al., Establishing a Holocene sediment budget for the river Dijle. Catena, 77(2), Notebaert et al., Changing hillslope and fluvial Holocene sediment dynamics in a Belgian loess catchment. Journal of Quaternary Science, 26(1), Notebaert et al., Modeling the sensitivity of sediment and water runoff dynamics to Holocene climate and land use changes at the catchment scale. Geomorphology, 126 (1 2), Notebaert et al., Fluvial architecture of Belgian river systems in contrasting environments: implications for reconstructing the sedimentation history. Some figures from: De Smedt, P., Paleogeografie en kwartair geomorfologie van het confluentiegebied Dijle Demer. Acta geographica Lovaniensia, vol. 11.

3 Fig 1. Cross-section of the Dijle floodplain, Korbeek-Dijle Anthropogenic deposits Holocene bed and lateral sediment accretion Holocene floodplain deposits Holocene organic floodplain deposits Height (m.a.s.l.) Holocene peaty deposits Pre-Holocene braided river deposits 25 Top of the peat layer: - diachronic - ranging between 5600 AD and 1400 BC - 15 AMS radiocarbon dates Bottom of the peat layer: 20 - ranging between to 8000 BC - 11 AMS radiocarbon dates m A B n= 11 n= 15 Fig 2. Cumulative probability functions (CPF) for radiocarbon ages in the cross-section at Korbeek-Dijle. A) Bottom of peat layer; B) Top of peat layer

4

5

6 This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier s archiving and manuscript policies are encouraged to visit:

7 Author's personal copy Available online at Catena 77 (2009) Establishing a Holocene sediment budget for the river Dijle Bastiaan Notebaert, Gert Verstraeten, Tom Rommens, Bart Vanmontfort, Gerard Govers, Jean Poesen K.U. Leuven, Department of Earth and Environmental Sciences, Celestijnenlaan 200E 3001, Leuven, Belgium Received 8 November 2007; received in revised form 1 February 2008; accepted 4 February 2008 Abstract A Holocene sediment budget was constructed for the 758 km 2 Dijle catchment in the Belgian loess belt, in order to understand long-term sediment dynamics. Hillslope sediment redistribution was calculated using soil profile information from 809 soil augerings, which was extrapolated to the entire catchment using morphometric classes. As large parts of the forests within the catchment prove to have undergone little or no erosion since medieval times, a correction was applied for the presence of forests. Total Holocene erosion amounts 817±66 Mt for the catchment, of which 327±34 Mt was deposited as colluvium. This corresponds with a net Holocene soil erosion rate of 10.8± Mg ha 1 for the entire Dijle catchment. Alluvial deposits were studied through 187 augerings spread over 17 cross-valley transects. The total alluvial sediment deposition equals 352±11 Mt or 42% of total eroded sediment mass. Results indicate that at the scale of a medium-sized catchment the colluvial sediment sink is as important as the alluvial sediment sink and should not be neglected. As a result the estimation of erosion through alluvial storage and sediment export would yield large errors. Dating of sediment units show an important increase in alluvial deposition from medieval times onwards, indicating the important influence of agricultural activities that developed from that period. Mean sediment export rates from the catchment for the last years range between 0.8 and 1.3 Mg ha 1 a 1 and are consistent with present suspended sediment measurements in the Dijle. Erosion for agricultural land for this period is 9.2±2.2 Mg ha 1 a 1. Sediment budgets for the various tributary catchments provide an insight in the sources and sinks of sediment at different scales within the catchment Elsevier B.V. All rights reserved. Keywords: Soil erosion; Alluvial sediment storage; Sediment budget; Human impact; Holocene; Sediment delivery 1. Introduction Historical soil erosion and sediment deposition have reshaped the landscape, especially in those regions where anthropogenic land cover change has been important. These changes can be very clear, as is demonstrated by traces of ancient gully systems, infilled valleys and cultivation steps (e.g. Bork, 1989; Larue, 2002; Stankoviansky, 2003; Vanwalleghem et al., 2003; Larue, 2005). However in most cases, slope morphology has changed Corresponding author. Fax: addresses: Bastiaan.notebaert@ees.kuleuven.be (B. Notebaert), Gert.verstraeten@ees.kuleuven.be (G. Verstraeten), rommens.tom@gmail.com (T. Rommens), Bart.Vanmontfort@ees.kuleuven.be (B. Vanmontfort), Gerard.govers@ees.kuleuven.be (G. Govers), Jean.poesen@ees.kuleuven.be (J. Poesen). only moderately, with slope profile truncation on steep slopes and sediment deposition at footslopes and in alluvial environments (e.g. Bork, 1983; Trimble, 1999; Knox, 2006; Rommens et al., 2007). A thorough understanding of the present landscape therefore implies the study of past processes. Much information can be gained from the study of contemporary processes and their associated rates. The alluvial and colluvial deposits are also important as an archive of past environments. They can be used to construct temporal frameworks for historical erosion, and in this way they give an indication of the importance of past erosion processes (e.g. Lang, 2003). Using these data, the influence of land use and climate change on the fluvial systems can be studied (Lang and Bork, 2006). An important tool to study the sediment dynamics in catchments and the controlling factors are sediment budgets. Such budgets are the accounting of sources, sinks and pathways of /$ - see front matter 2008 Elsevier B.V. All rights reserved. doi: /j.catena

8 Author's personal copy B. Notebaert et al. / Catena 77 (2009) sediment in a unit region for a time unit (Slaymaker, 2003; Reid and Dunne, 2003). Temporarily changing sediment budgets can indicate how land use changes within the catchment impact sediment transport at various scales (e.g. Trimble, 1999). According to Foulds and Macklin (2006), sediment budgeting is a necessary tool to understand the role of land use change on catchment stability, as it identifies reach-scale zones of sediment transfer and storage. Several studies have constructed catchment-wide sediment budgets for relative short periods ranging from days to a few decades (e.g. Trimble, 1983; Walling and Quine, 1993; Beach, 1994; Page et al., 1994; Trimble, 1999; Fryirs and Brierley, 2001; Walling et al., 2002, 2006). However, few studies have concentrated on long-term sediment budgets (e.g. spanning the entire Holocene), as this is more difficult, mainly because of the larger uncertainties and the lack of precise dates. Most long-term sediment budget studies therefore focus on relatively small catchments (b100 km 2 ) for which sufficiently detailed data can be gathered relatively easily (e.g. Clemens and Stahr, 1994; Rommens et al., 2005; Houben, 2006; Verstraeten et al., in press). For larger river catchments, often only one sediment budget component, for instance floodplain storage, is assessed (e.g. Houben, 2006; Hoffmann et al., 2007). In some case studies, a late- Holocene sediment budget has been reconstructed using a combined field and modelling approach. For the 380 km 2 large Geul catchment, de Moor and Verstraeten (2008) reconstructed a sediment budget for the last 1000 years by estimating the alluvial storage from field data and the hillslope erosion and sediment deposition masses using a spatially distributed model approach. For medium-sized catchments, which provide the link between processes operating at hillslopes and fluvial processes in large drainage basins, balanced sediment budgets spanning the entire Holocene are still rare (Macaire et al., 2002). The objectives of this study are therefore threefold. Firstly, we aim to construct a Holocene sediment budget for the mediumsized Dijle catchment (758 km 2 ) in the Belgian Loess Belt. For this river catchment, information on historic erosion and deposition amounts is available for smaller spatial units (Rommens et al., 2005, 2006). Secondly, the possibility of upscaling of local information to larger spatial units, in order to arrive at a Holocene sediment budget for the entire catchment. In this way an attempt was made to identify the main sediment sources and sinks. Thirdly, in addition to the quantitative approach, a timeframe for depositional processes is established and will be compared with existing information from local sediment archives. 2. Study area This study focuses on the 758 km 2 catchment of the river Dijle upstream the city of Leuven, Belgium (Fig. 1). The topography consists of an undulating plateau in which several rivers are incised. The slopes are in general less than 5%, but slopes steeper than 50% occur regularly along the axes of the major streams. The height of the plateau varies between 165 m a.s.l. in the south to 80 m a.s.l. in the north of the catchment, while the floodplain is situated around 25 m a.s.l. around Leuven. The southern part of the drainage network consists of three important rivers (Dijle, Thyle and Orne) with more or less the Fig. 1. The study area with indication of the main cities (1: Leuven; 2: Wavre; 3: Ottignies; 4: Court-St-Etienne; 5: Genappes) and the main rivers (a: Dijle; b: Thyle; c: Orne; d: Train; e: Laan; f: Nethen; g: IJsse). same dimensions. These three rivers join near Court-Saint- Etienne to form the main River Dijle (Fig. 1). The width of the floodplain downstream from this confluence varies between 200 and 1800 m, whereas in the upstream parts the floodplain width generally does not exceed 150 m. More to the north, several other tributaries join the main branch, of which the Laan, IJse, Nethen and Train are the most important. The plateau mainly consists of Tertiary sands and clays covered with Pleistocene loess. Locally, this loess cover can be absent, resulting in sandy outcrops. In the southern part of the Dijle catchment the rivers have locally cut through the sandy Tertiary deposits, resulting in small outcrops of more resistant Palaeozoic rocks. Soils in the region are mainly loess-derived luvisols according to the FAO (1998). Current land use is rather diverse: the plateau areas are mainly covered with cropland, but several large forest stands occur, especially in the northeastern (Meerdaal forest) and northwestern (Zonien forest) part of the catchment (Fig. 2). The floodplains are used for grassland and forests. Large built up areas cover both the floodplain and the nearby slopes and plateaux between Court-St- Etienne and Bas-Wavre, and some important residential areas were built around forested plateau sites. Palynological information (Mullenders and Gullentops, 1957; Mullenders et al., 1966; De Smedt, 1973) taken from sediment cores in the Dijle floodplain near Leuven shows that the catchment was mainly forested during the first half of the Holocene and that the first agricultural crops occurred in the Subboreal phase (Mullenders et al., 1966) The oldest known settlement dates from around 5000 BP (Diriken, 1989) and numerous finds from other historical periods (Bronze Age, Iron Age, Roman Period

9 Author's personal copy 152 B. Notebaert et al. / Catena 77 (2009) within the floodplain, and the mass of soil eroded and sediment redeposited on the hillslopes Fluvial deposition In total 17 cross-sections across the floodplain were established to characterize and quantify the fluvial sediment storage within the catchment (Fig. 3) and 187 hand augerings were made along these cross-sections. A detailed description for each 5 cm soil depth was made for each augering. The interpretation of the cross-sections and the identification of sedimentological units were based on fluvial architecture concepts, for which the sediment texture was assessed in the field (e.g. Miall, 1985; Houben, 2007). In order to quantify the total volume and mass of sediment stored within the entire catchment using information derived from these cross-sections, a floodplain polygon was digitized using soil maps, topographic maps, field observations and, where available, detailed LIDAR DEM's. The floodplain was consequently divided in homogenous zones, for which it is assumed that the thickness of the Holocene alluvial deposits is effectively constant, and for each zone a representative cross-section was selected. The sediment mass for each zone was calculated using: Fig. 2. Spatial distribution of forests within the Dijle catchment. M zone ¼ A zone d M cs with: ð1þ and Medieval Period) can be found throughout the catchment (Van Hove et al., 2005; Peeters, 2007; Vanwalleghem et al., in press). Although anthropogenic disturbances become more important from the Subatlantic period onwards (Mullenders et al., 1966; De Smedt, 1973), it is only in the Roman period that the human influence on the landscape becomes significant. The forest was extensively cleared to provide agricultural land. Continuity between the Roman and medieval land occupation is not very clear, but it appears that agricultural land remained widespread. However, some large areas remained forested from at least the 14th century until present, such as the Meerdaal forest in the northeastern part of the catchment (Vanwalleghem et al., 2006). In these forests pre-medieval landforms like manmade closed depressions (Vanwalleghem et al., 2007) and gullies dated by 14 C and OSL dating to the Bronze Age and Roman Period (Vanwalleghem et al., 2003, 2005, 2006, in press) were preserved. Comparable gullies were found in the Zonien forest (Arnould-De Bontridder and Paulis, 1966) and were observed within this study in several other forests in the Dijle catchment. This indicates that these forests were not prone to severe erosion for at least the last few centuries. The Zonien forest is reported to have been forested since the 12th century (Langohr, 2001). 3. Materials and methods M zone A zone the mass of mineral sediment (t) deposited in the alluvial zone; the area (m 2 ) of the alluvial zone; In order to reconstruct a Holocene sediment budget, all components of the sediment budget were quantified. A different approach was followed to quantify the mass of sediment deposited Fig. 3. Location of augering sites. Sites with dating results of alluvial deposits: 1: Korbeek-Dijle; 2: St-Joris-Weert. Sites with augering data for slope processes: 3: Loonbeek; 4: Billande; 5: Ottenburg; 6: Nodebais; 7: Hamme-Mille; 8: Beauvechain.

10 Author's personal copy B. Notebaert et al. / Catena 77 (2009) M cs the average mineral sediment mass per unit surface area (Mg m 2 ) for the cross-section representative for the alluvial zone. Values for M cs were calculated as follows: M cs ¼ Xj with: j M unit i i¼1 M unit i ð2þ number of alluvial units present in the cross-section; the average mineral sediment mass per unit surface area (Mg m 2 ) for alluvial unit i in the cross-section representative for the alluvial zone. Values for M unit i are calculated with a correction factor for the occurrence of high percentages of organic matter (Verstraeten and Poesen, 2001): M unit i ¼ d unit i with: d unit i 1 ð3þ 1 kom DBD MS þ ð1 komþdbd OM average thickness of alluvial unit i in the considered cross-section (m); %OM percentage organic matter; DBD OM the dry bulk density of the organic matter (Mg m 3 ); DBD MS dry bulk density of the clastic component (Mg m 3 ). The sum of the sediment masses of the different fluvial units within a zone, gives the total mass of deposited sediments within the alluvial zone. For the Nethen subcatchment, we made use of the data reported by Rommens et al. (2006). In order to derive more information on the chronology of fluvial sediment deposition, AMS-radiocarbon dating was performed on 8 samples. All samples were chosen at the transition from one fluvial unit to another, corresponding to changes in organic material content. Additional radiocarbon data for the Nethen floodplain were obtained from Rommens et al. (2006). Calibration of radiocarbon dating was performed with Oxcal 4.0 ( ac.uk/oxcal, build number 22; Bronk Ramsey, 2008) using the Intcal 04 curve (Reimer et al., 2004). The precision of the calculations can be estimated using Gaussian error propagation. Errors of average sedimentological unit thickness (augering depth), DBD and the area of the alluvial plain can be considered independent (e.g. Rommens et al., 2006). The error on individual sedimentary units in coring observations is 5 cm, while the thickness of these units ranges from 1 m to more than 3 m, resulting in an relative error of 5% or less. Therefore an error of 5% was assumed on average depths of the different sedimentological units when these depths are calculated based on profiles. Where insufficient augerings where undertaken to make a detailed subdivision into different sedimentological units, the average depth is taken to be the arithmetic mean of the individual augerings. Values and errors for %OM, DBD OM and DBD MS are based on Rommens et al. (2006). The error associated with the digitizing of the alluvial plain was estimated for each polygon individually, based on the expected error in the width of the polygon. The error in the polygon width mainly depends on the ability to define of the alluvial plain (e.g. steep or gentle valley slopes). It was assumed that all cross-sections are representative for the given floodplain zones, thus no error was used for the extrapolation of the cross-section data to the entire floodplain zones Hillslope erosion and deposition The loess in which most soils are developed typically contained 10 13% CaCO 3 upon deposition (Goossens, 1987), but gradually decalcification of the upper loess layers occurred. This allowed the mobilization of clay and the development of a typical luvisol with an eluviation horizon (E) and a clay accumulation horizon or argic horizon (B t ). Later on these typical luvisol soil profiles were altered by soil erosion and sediment deposition. Under the assumption that the depth of the different soil horizons is constant within the study area, i.e. that decalcification and soil profile development occurred independently of the slope gradient and slope aspect, observed soil profiles can be compared with the theoretical local undisturbed luvisol profile to estimate the depth of erosion. For a detailed description of the methodology we refer to Rommens et al. (2005). Data from 633 augerings at four sites (Beauvechain, Nodebais, Hamme-Mille, Ottenburg; see Fig. 3) were derived from existing datasets (Vanmontfort et al., 2004; Rommens et al., 2005). In addition, 176 hand augerings were made at two other study sites, namely Loonbeek and Bilande (Fig. 3). For each augering the sedimentological record was described in detail and (where possible) the different soil horizons were determined. Apparently uneroded soil profiles on plateau top sites were used to determine the local (undisturbed) reference soil profiles. All datasets were filtered to exclude soil augerings in man-made closed depressions (Gillijns et al., 2005; Vanwalleghem et al., 2007; Rommens et al., 2007). Finally the depth of the decalcification front was used to determine the local erosion depth for each augering. In order to extrapolate the augering data to the whole Dijle catchment, morphometric classes were used. This Average Per Unit method (APU) was also successfully used by other studies (Lewis and Lepele, 1982; Bork, 1983; Macaire et al., 2002; Rommens et al., 2005). Mapping of the morphometric units for each augering site was performed by digitizing the contour lines with a height interval of 2.5 m from topographic maps at a scale of 1:10,000. These contour lines were used to create a raster DTM with a pixel resolution of 20 m (hereafter called DTM1 ), using ArcGIS. This raster was in turn used to define the morphometric classes. Colluvial (dry) valleys were defined as a buffer around pixels with an upslope area of 4 ha or more draining towards that pixel. However, the width of this buffer depends again on the upslope area, as downstream these colluvial valleys tend to become wider. All other pixels were classified according to the local slope gradient in 4 classes (0% 3%, 3% 5%, 5% 8%, N8%). These classes were chosen arbitrarily, but in accordance with Rommens et al. (2005) to make comparison possible. Average erosion and deposition depths of the morphometric units were calculated for each corresponding soil augering dataset.

11 Author's personal copy 154 B. Notebaert et al. / Catena 77 (2009) Fig. 4. A typical cross-section with indication of the various alluvial sedimentary units in the main valley of the River Dijle near Korbeek-Dijle (see also Fig. 3). Datings are uncalibrated ages BP (see also Table 3). A limited number of augerings (in total 26) were reallocated to another unit, as they were erroneously classified as being in- or outside the colluvial valleys due to small shifts of the thalweg on the DTM. The average erosion and deposition depths for each morphometric unit at the different sites were weighted using the local distribution of the morphometric units, in order to calculate a total average value for each unit. Using this weighting, the influence of unevenly distributed augering densities on the average value was avoided. For the entire Dijle catchment a second DTM (hereafter called DTM2 ) was available based on contour lines of the 1:50,000 topographical map of the Belgian National Geographic Institute (NGI). For a description of the DTM creation we refer to Van Rompaey et al. (2001). This DTM was used to define the same morphometric classes as described above, complemented with the alluvial plains. As the topography is somewhat smoothed on this DTM2 compared to the DTM1, the slope values had to be multiplied by a factor of before defining the slope-units. This factor was calculated by matching the slope-histograms. The DTM2 was not used initially for the average erosion and deposition calculations per unit area as described above, because the position of the geomorphic units is rather poorly defined, resulting in a very large number of wrongly classified soil augerings, mainly around the dry valleys. On the other hand, the general pattern is well represented. Next, the morphometric map derived from DTM2 was combined with the calculated average erosion and Table 1 Overview of soil augering data in alluvial valleys Location Subcatchment Floodplain width (m) Drainage area (km 2 ) Average thickness of alluvial sediment deposits (m) Korbeek-Dijle Main valley ±0.5 Sint-Joris-Weert Main valley ±0.3 Pecrot Main valley ±0.4 Archennes Main valley ±0.5 Thy Upper-Dijle ±0.5 Loupoigne 1 Upper-Dijle ±0.6 Loupoigne 2 Upper-Dijle ±0.6 Terlanen Laan ±0.7 Couture-Saint-Germain Laan ±0.1 Bonlez Train ±0.7 Blanmont Orne ±0.2 Chastres Orne ±0.4 Cortil 2 Orne ±0.3 Cortil 1 Orne ±0.5 Bois Quinaux Orne ±0.3 Nil-St-Vincent-St-Martin Nil ±0.7 Tourinnes-St-Lambert Nil ±0.5 Average mineral sediment mass per unit surface (Mg m 2 )

12 Author's personal copy B. Notebaert et al. / Catena 77 (2009) (colluvial) deposition rates to obtain total eroded and colluvial sediment masses for the entire Dijle catchment. When extrapolating soil erosion and sediment deposition depths derived from the different augering sites to the whole catchment, it is assumed that the erosion and deposition history for these sites is representative of the whole catchment. However, the augering data were all collected at agricultural sites (cropland or pasture), while large parts of the catchment are at present covered by forests. Most of these forests occur on historical maps of the end of the 18th century (Carte de Cabinet des Pays-Bas Autrichiens et de la Principauté de Liège, count de Ferraris, 1777), and for several forests there are strong indications that they were forested since the Middle Ages onwards (see above). This means that total erosion and hillslope sediment deposition depths will be overestimated, if we assume that these areas experienced the same rates of erosion and deposition as those that were continuously used for cropland. According to Langohr and Sanders (1985), there is no indication of substantial colluvial deposits within the Zonien forest. To include the occurrence of these historical forests within the sediment budget, a correction factor for these areas was applied. This correction factor was based on the dating of two colluvial deposits in the Nethen subcatchment that show that about 60% to 85% of the colluviation depth took place during and after the Medieval period (Rommens et al., 2005; Rommens, 2006; Rommens et al., 2007). As demonstrated above, it can be expected that most of these forests were not prone to soil erosion since the Middle Ages, but erosion phases before this period cannot be excluded. Extrapolating the dating results of colluvial deposits, it can be assumed that total erosion under present-day forests is, as a maximum, 25% of total erosion on current agricultural land. Therefore, the estimated erosion and colluvial deposition depths within the forests were reduced by 75% compared to agricultural land. For the colluvial valleys within forests smaller than 30 ha, no correction was applied, as these forests are typically trapping the sediment originating in the surrounding agricultural land. The spatial extent of forest within the Dijle catchment was digitized from topographical maps ( ) and aerial photographs ( ) (Fig. 2). Finally, sediment volumes were converted to sediment masses using a soil and sediment dry bulk density of 1.53 Mg m 3 (see Rommens et al., 2005). The errors estimated for the slope results are based on Gaussian error propagation (see above). It was assumed that there was no error on the morphometric mapping. the top a small but well developed peat layer. The third and upper depositional unit contains no recognizable plant remains or peat, except for the presence of contemporary roots. This layer is often the thickest layer. Alluvial deposits in some of the main tributaries (e.g. Laan, Orne and parts of the Upper-Dijle) generally consist of the same units, although they are less distinguishable, which confirms data for the Nethen (Rommens et al., 2006). In other main tributaries (e.g. Train, parts of the Upper-Dijle) this image is disturbed by the presence of calciferous deposits including travertine (own data; Geurts, 1976). Furthermore, in the smaller tributaries there is a large lateral variation in the presence of peat and organic deposits in units 2 and 3. There is a large variation in average thickness of the alluvial deposits between the different augering sites, and as a consequence also in the average mineral sediment mass per unit surface area (Mg m 2 )(Table 1). The largest deposited masses per unit area of floodplain can be found in the upper valleys of the Dijle and Laan. The lowest values are found in the Lower Orne valley (Blanmont cross-section). The general trend shows that the thickness of alluvial deposits and deposition per unit area within the floodplain is decreasing downstream. Average sediment storage in mass per m valley length is represented in Fig. 5, to give an insight in the geographic distribution. Combining the augering data with the digitized floodplain, yields a total alluvial deposition of 352±11 Mt (megaton or 4. Results 4.1. Fluvial deposits Alluvial deposits in the main floodplain of the Dijle show a typical pattern with three major units (Fig. 4). The oldest depositional unit consists of a peat layer or highly organic clastic sediments containing ca 15% organic matter. This unit is overlain by a second unit, which contains mainly clastic material, albeit mixed with few peaty and organic layers, and often containing plant remains. The upper parts of unit 2 are often less organic and contain almost no peat and plant remains, but frequently there is at Fig. 5. Alluvial sediment storage per unit of downstream valley length within the catchment (Mg m 1 ).

13 Author's personal copy 156 B. Notebaert et al. / Catena 77 (2009) Table 2 Overview of the floodplain area and the alluvial sediment deposition masses within the different subcatchments Catchment Area (km 2 ) Floodplain area (km 2 ) Number of studied cross-sections Floodplain deposits (Mt) Upper tributaries ± % Upper-Dijle ± % Thyle ± % Orne ± % Lower tributaries ± % Train ± % Nethen ± % Laan ± % IJse ± % Small tributaries ± % Main valley ± % Total catchment ± % Fraction of floodplain deposits Table 3 AMS-radiocarbon dating results Sample ID Lab code Location Depth (m) Stratigraphic position Material 14 C age (BP) 1 Sigma calibrated calendar age (BC AD) 2 Sigma calibrated calendar age (BC AD) SJW Beta St-Joris-Weert 2.65 Top of the peat layer Organic C 710± AD 1299 AD 1224 AD 1314 AD at the top of unit 2 residue 1370 AD 1380 AD 1357 AD 1389 AD SJW Beta St-Joris-Weert 3.40 Top of unit 1 Wood 5770± BC 4582 BC 4718 BC BC 4557 BC SJW Beta St-Joris-Weert 4.45 Top of unit 1 Wood 5500± BC 4424 BC 4450 BC 4318 BC 4371 BC 4327 BC 4296 BC 4263 BC 4283 BC 4271 BC KOR_0163 UtC Korbeek 2.90 Top of unit 1 Organic C 2473± BC 686 BC 765 BC 483 BC residue 668 BC 611 BC 467 BC 415 BC 597 BC 523 BC KOR_0164 UtC Korbeek 4.60 Bottom of unit 1 Plant 6646± BC 5543 BC 5639 BC 5488 BC KOR_04C UtC Korbeek 3.30 Peat layer at the top of unit 2 OH_03340 UtC Korbeek cm below top of peat layer (unit 1); valley edge OH_07455 UtC Korbeek cm below top of peat layer (unit 1) Charcoal 1280± AD 724 AD 658 AD 783 AD 739 AD 771 AD 789 AD 813 AD 844 AD 857 AD Organic C residue Organic C residue 973± AD 1049 AD 996 AD 1006 AD 1086 AD 1123 AD 1012 AD 1157 AD 1138 AD 1151 AD 6508± BC 5466 BC 5606 BC 5595 BC 5441 BC 5423 BC 5560 BC 5367 BC 5406 BC 5383 BC Table 4 Average erosion (E, m) and deposition (D, m) depths of the different morphometric units for the used augering datasets (n: number of augerings used for the calculation is given in parenthesis) Nodebais Hamme-Mille Beauvechain Ottenburg Loonbeek Bilande Average Number of augerings Slope 0 3% E (n) 0.34 (39) 0.36 (7) 0.39 (65) 0.77 (72) 0.69 (11) 0.27 (1) 0.46 (195) D (n) 0.10 (46) 0.14 (8) 0.07 (65) 0.16 (113) 0 (11) 0 (4) 0.10 (247) Slope 3 5% E (n) 0.91 (20) 0.78 (5) 0.45 (53) 1.20 (13) 1.88 (2) 0.91 (11) 0.81 (104) D (n) 0.13 (22) 0.18 (9) 0.11 (54) 0.27 (35) 0 (7) 0 (14) 0.13 (141) Slope 5 8% E (n) 1.22 (22) 0.97 (15) 0.91 (28) 0.35 (1) 1.50 (11) 1.31 (27) 1.10 (104) D (n) 0.05 (23) 0.17 (22) 0.09 (29) 0.24 (14) 0.47 (16) 0.03 (31) 0.12 (135) Slope N8% E (n) 1.58 (25) 1.28 (15) 1.93 (17) 1.66 (30) 1.56 (87) D (n) 0.9 (39) 0.17 (25) 0.32 (7) 0.27 (36) 0.30 (41) 0.24 (148) Thalweg E (n) 0.75 (18) 1.17 (6) 0.85 (28) 2.46 (4) 1.55 (2) 0.41 (58) D (n) 1.72 (55) 2.27 (23) 1.33 (39) 2.01 (5) 2.89 (9) 3.88 (7) 2.38 (138)

14 Author's personal copy B. Notebaert et al. / Catena 77 (2009) Table 5 Areas of the different morphometric units within the study area (km2) and the percentage of forest within each hillslope unit Slope 0 3% Slope 3 5% Slope 5 8% Slope 8%+ Thalweg Alluvium Area on morphometric map (km 2 ) Of which is forested 12.8% 13.3% 17.5% 37.0% 16.9% 10 6 Mg) clastic material. About 22% is deposited in upper tributaries (south of Court-St-Etienne), 41% in the lower tributaries (north of Court-St-Etienne) and 36% in the main valley (Table 2). AMS-dating results for the main floodplain are represented in Table 3 (see also Fig. 4). For the top of the first sedimentological unit several dates are available, suggesting that the end of the peat development varies within the same cross-section. Two dates are available for the (top of the) peat layer at the top of the second unit, indicating that the top of this unit dates from the Middle Ages Hillslope erosion and deposition Erosion and deposition rates for the different augering sites are represented in Table 4. Erosion depths could be calculated for only 547 of the 809 augering, as for the other augerings no information about the depth of the decalcification front was present, mainly because the loess layer was completely decalcified, or because no loess layer was present. The morphometric map (summarized in Table 5) was combined with the average erosion depths (Table 4) to calculate the erosion and deposition volumes for the entire catchment (Table 6, Fig. 6). Incorporating the correction factor for the forests in the calculation, the total erosion amounts to m 3 or 817±66 Mt sediment, while the total colluvial deposition amounts to m 3, which is equivalent to 327±34 Mt sediment. This means that 490±75 Mt has moved from the hillslopes towards the fluvial system. The total Holocene soil erosion for the entire catchments is ± Mg ha 1. If the correction factor for forests is not applied, the rates are much higher, i.e m 3 (1004 Mt) erosion, m 3 (383 Mt) deposition and a hillslope sediment export of 621 Mt The sediment budget of the Dijle catchment With the given calculations a sediment budget for the entire Dijle catchment was calculated (Fig. 6, Table 7). Sediment export out of the catchment amounts to 138±75 Mt. In order to compare our results with those of a previous estimate of the sediment budget for the Nethen (Rommens et al., 2006), we extracted data for this subcatchment from the morphometric map and combined it with the erosion and deposition depths. The results of this comparison are given in Table 6, using the morphometric maps and the average erosion and deposition rates reported in Table 4. Similar information for the other tributaries provides more insight in the geographic distribution of erosion and sedimentation (Table 7). 5. Discussion 5.1. Hillslope sediment redistribution A major weakness of the calculations for slope and plateau positions is the assumption that the original soil profile is homogenous throughout the different study sites. However, 184 augerings of undisturbed soil profiles in the Meerdaal forest show an average depth of 2.48 m for the decalcification front, with a standard deviation of 0.63 m (Vanwalleghem, pers. com.). The same database contains 215 cores where no decalcification front was observed. The upper and lower limits of the B t horizon show less variation, respectively 0.13 m (n=233) and 0.24 m (n=216). This shows that these borders would be better indicators for the soil profile truncation calculations. However, in practice the upper B t border is often truncated under agricultural land, while the identification of the lower B t border is difficult or even impossible for large parts of the augering datasets. Only soil textural data could present a solution for a reliable delineation of the B t. Table 6 Erosion and sediment deposition masses (Mt) for the different morphometric units Slope 0 3% Slope 3 5% Slope 5 8% Slope 8%+ Thalweg Total Morphometric map without correction for forests Erosion (Mt) ±76 Erosion (%) 14% 17% 20% 45% 4% Deposition (Mt) ±40 Deposition (%) 8% 7% 6% 17% 62% Morphometric map with forests Erosion (Mt) ±66 Erosion (%) 15% 19% 21% 40% 4% Deposition (Mt) ±34 Deposition (%) 9% 7% 6% 15% 64% Nethen catchment Erosion (Mt) ±4 Erosion (%) 33% 22% 17% 22% 7% Deposition (Mt) ±2 Deposition (%) 13% 6% 4% 6% 68% Both the cases with and without forests considered within the calculations are represented.

15 Author's personal copy 158 B. Notebaert et al. / Catena 77 (2009) Fig. 6. Sediment budget of the Dijle catchment. With the applied method, the calculations for the most severely eroded sites are probably somewhat underestimated: for augerings where the calcareous loess occurs at the surface, the erosion depth was set equal to the reference profile depth of the decalcified layer. For areas where the calcareous loess is outcropping, it is most probable that also a part of this calcareous layer was eroded. Between and within augering sites, there is a large variation in erosion and deposition depths for the different morphometric units (Table 4). When plotting the local erosion depths and slope for the different augerings, within the different subcatchments, it is evident that there is no linear relation between both (Fig. 7). This shows that slope gradient alone is not enough to explain the local variation in erosion depths. Although Rommens et al. (2005) showed the applicability of the APU method for the erosion and deposition calculations, further research is needed for a better extrapolation of total historic soil erosion depths, including factors other than slope. It is clear that these methodological limitations result in a large uncertainty on the calculated total hillslope erosion volumes. However, it remains difficult to quantify this uncertainty. It was attempted to establish this largely by applying a statistical analysis whereby a large number of augering data were used. It is expected that in this way, individual augerings or entire augering sites with somewhat deviating results, will have a limited influence on the overall average erosion and deposition amounts. The error associated with the eroded soil mass equals 8%, while the error associated with the deposited sediment mass is ±10%. These errors are rather low as a large dataset was used. However, it should be stressed that this error only includes extrapolation errors, under the assumption that soil erosion and sediment deposition is related to topography Alluvial storage Dating results show large variations in the age of the top of the first (oldest) sedimentological unit (Table 3, Fig. 4). Moreover, two samples from the top of unit 1 with an absolute height difference of about 1 m and a horizontal distance of about 30 m, yield dates within the same time period (BETA and BETA ). Although we have insufficient data too, this suggests that the assumption of a uniform vertical aggradation of the floodplain is not correct. It is, however, clear that the major peat development of the first unit probably stopped around 6500 BP (5527 cal BC 5383 cal BC) for some parts of the valley, and continued for much longer at the edges of the valley. The second unit indicates a transition towards a phase with more sediment input in the valley. But on top of it often a peat layer is present, indicating a new period with reduced sediment input in the valleys. Dating results show that this peat layer formed somewhere during the Medieval period, yet there are insufficient dates available to better constrain the exact timing of this peat formation. The general pattern indicates that the majority of the floodplain deposition took place after the early Medieval period (Fig. 8). This suggests that the important agricultural activities Table 7 Sediment budget for the Dijle catchment and the different subcatchments Catchment Surface (km 2 ) Erosion (Mt) Colluvial deposition (Mt) Hillslope export (Mt) HSDR (%) Alluvial deposits (Mt) Catchment export (Mt) Nethen Thyle IJse Train Upper-Dijle Orne Laan Entire Dijle catchment SDR (%)

16 Author's personal copy B. Notebaert et al. / Catena 77 (2009) Fig. 7. Relation between slope gradient and hillslope erosion (gross soil profile truncation) for all augerings where the decalcification front is present. which developed from this period onwards are responsible for the large sediment input. This general pattern fits well into the timeframe of sediment deposition developed for the Nethen valley (Rommens et al., 2006), although the recent sediment deposition rates seem to be somewhat lower for the Nethen. Dating results from colluvial deposits within the Nethen catchment (Rommens, 2006; Rommens et al., 2006, 2007) show a start of colluviation around BP ( cal BC) with increasing amounts from BP ( cal BC) onwards. From the Roman period onwards, colluviation was already significant. At first sight this pattern is not evident in the alluvial deposits and there seems to be a time lag between colluvial and alluvial depositions (see also Rommens et al., 2006; Verstraeten et al., in press). More precise dating of alluvial deposits, and more specific dating of the upper parts of the second unit is necessary to confirm this time lag. It would also provide a better understanding of the history and mechanisms of sediment transport within the catchment Sediment budget In the calculation of the sediment budget, we took into account that parts of the catchments were most likely forested for long time periods. If no correction was applied, total eroded and colluvial sediment masses, as well as the sediment export, would be larger. Indeed, sediment export would then increase by 95% to 270 Mt, whereas total erosion would equal 1004 Mt, which corresponds to a 23% increase. However, the relative importance of the various components of the sediment budget changes only slightly (Table 7). Direct comparison between the sediment budget estimated in this study with that of the Nethen subcatchment estimated by Fig. 8. Average accumulated sediment in the floodplain through time: 1) Sint-Joris-Weert, Dijle main floodplain (this study); 2) Nethen floodplain (Rommens et al., 2006); 3) Shaded area: combination of dates for Korbeek, Dijle main floodplain (this study).

17 Author's personal copy 160 B. Notebaert et al. / Catena 77 (2009) Rommens et al. (2006) and Verstraeten et al. (in press) is not straightforward. First of all, they did not take into account the presence of forests. This is not realistic, especially for the Nethen catchment, as the large Meerdaal forest makes up 25% of the catchment. Secondly, in the present study, we made use of a more extensive augering dataset which resulted in average erosion and colluvial sediment deposition volumes for the entire Dijle catchment, which differ slightly from the average volumes used in earlier studies. And finally, a more detailed mapping of the colluvial valleys was carried out in this study, whereby the width of the colluvial deposits varies with upslope drainage area as can be evidenced in the field. Therefore, it can be expected that the calculated sediment budget for the Nethen subcatchment yields more realistic results than former calculations. The sediment budgets that were established in this paper incorporate hillslope erosion, hillslope deposition and net floodplain deposition. It does not mean, however, that this sediment budget can be considered as a fluvial sediment budget. Indeed, we estimated total volumes of erosion and sediment deposition on hillslopes, irrespective of the process that caused the redistribution of soil. For instance, tillage operations can be held responsible for the majority of the current-day soil redistribution volumes within fields (e.g. Govers et al., 1996; Van Oost et al., 2005). From the observed soil profile truncations and colluvial profiles, it cannot be estimated which fraction was eroded or deposited by tillage or by water erosion processes. Van Oost et al. (2005) show that there is an important shift in the relative contribution of tillage and water erosion processes to soil redistribution in recent decades, whereby tillage erosion is nowadays more important than water erosion. In the longer term it is clear that water erosion is the dominant process, and thus, the majority of the erosion and sediment deposition volumes on hillslopes calculated in our sediment budget do reflect water erosion processes. Another important soil degradation process is soil loss due to crop harvesting (Poesen et al., 2001; Ruysschaert et al., in press), which has become important in this study area since the late 19th century. Part of the observed soil profile truncation is caused by this process, and thus it is inherently incorporated in the sediment budget. There is, however, no detailed time frame available to estimate the importance of tillage erosion and soil losses due to crop harvesting on a decade scale for this catchment, and thus it is very difficult to estimate the relative contribution of water erosion in the sediment budget. On the other hand, some fluvial erosion and deposition processes are not incorporated in the sediment budget, namely river channel erosion and deposition. Changes in channel width and depth would possibly have an influence on total sediment export. Available data give insufficient information about historical changes in channel dimensions, but even if these changes would be relatively important, they would still represent a negligible net erosion amount compared to the hillslope erosion amounts. For instance, an average channel widening of 5 m combined with an average deepening of the channel by 1 m over the 39 km long river stretch downstream Court-St-Etienne, would represent a net contribution of 0.28 Mt sediment to the river channel, which is very low compared to the 482 Mt entering the fluvial network by hillslope processes. On the other hand, the gross channel erosion and deposition amounts can have a much stronger influence on the relative importance of the various sediment budget components (see e.g. Trimble, 1997, 1999). There is insufficient information to estimate these values over the long term. Recent data on a 1010 m long stretch show that deposition caused by channel displacement equals roughly 10.3 Mg m 1, or m 3 for the entire stretch, for a 30 year period ( ) (Notebaert et al., submitted for publication). Extrapolating this value over the main stretch over the last 10,000 years implies a total deposition of 134 Mt. Total channel erosion is the same if we assume that channel dimensions remained constant (no net deepening or widening). These values are similar to the total estimated sediment export and represent one third of the total floodplain deposition amount. However, this value does not necessarily correspond with the total mass of bed and pointbar deposits within the catchment. Indeed, augering data show that these deposits often cover small parts of the alluvial deposits, suggesting that the river is mostly eroding and reworking these old bed deposits, instead of eroding floodplain deposits. There are no data available to validate the estimates, nor are there indications that the values for the studied stretch are representative for other stretches or time periods, and thus their use within the budget is not justified Implications of the sediment budget The construction of a long-term sediment budget makes it possible to calculate sediment delivery ratios (SDR) for longer time periods. The SDR of the catchment is defined as the ratio between the sediment export and the gross erosion within the catchment, while the hillslope sediment delivery ratio (HSDR) is defined as the ratio between the sediment exported from hillslopes towards the fluvial system and the gross erosion within the catchment. Although recently it was argued that the concept of SDR is a fallacy (Parsons et al., 2006), we believe that SDR remains an important tool for understanding the sediment dynamics within a catchment, as it represents an essential part of the sediment budget which is closely connected with all the processes occurring in the sediment cycle (erosion, transport and deposition at different scales). For the Dijle catchment, the SDR equals 17%. The SDR values for the different subcatchment show also large variations, which cannot only be attributed to their differences in size (Table 7). It is clear that differences in catchment environment play an important role. If the correction for forests is not incorporated in the overall budget, the SDR increases to 27%. This indicates that the assumptions and definitions that are used for the different terms of a sediment budget have important implications for this sediment budget and in particular the SDR. Therefore it is only possible to compare budgets or delivery ratios when all calculations were made in the same way. As was demonstrated above, the budget for the Dijle catchment is not a pure water erosion budget. Therefore it is not possible to make a comparison with sediment budgets which were, for instance, the erosion component was estimated using a water erosion modelling approach (e.g. Trimble, 1999; de Moor and Verstraeten, 2008). The sediment budget clearly shows that at the scale of the Dijle catchment, colluvial deposition (both on slopes and in dry valleys) is as a sink as important as the floodplain. This implies

18 Author's personal copy B. Notebaert et al. / Catena 77 (2009) that the calculation of erosion amounts from floodplain storage amounts or from sediment export rates (e.g. Hoffmann et al., 2007) is not justified. This will systematically lead to a major underestimation of erosion amounts. The HSDR of the different subcatchments show important variations. From this it is clear that extrapolating alluvial storage and export data to derive colluvial deposition amounts can result in large errors. For the non-forested area (557 km 2 ) in the Dijle catchment, the total erosion mass was estimated at 754 Mt, which corresponds to a mean erosion rate of 1.4±0.1 Mg ha 1 yr 1 for the entire Holocene. If we assume that about 75±15% of soil erosion took place during the last years (see above), the average erosion rate on agricultural land for this period equals 9.2±2.2 Mg ha 1 yr 1. Other studies that applied similar methodologies in the Belgian loess areas report erosion rates ranging from 2.8 to 17.6 Mg ha 1 yr 1 for the last 1000 years, whereas present-day rates of water erosion processes equal 2.6 and 16.7 Mg ha 1 yr 1 (Verstraeten et al., 2006). Likewise, historical sedimentation rates calculated with different but comparable methods (like colluvial infilling of closed depressions) also yield comparable rates (Verstraeten et al., 2006). No long-term independent data on sediment export are available to validate the sediment budget estimated in this study. However, it is possible to compare the estimated sediment export rate with present-day rates of fluvial sediment transport by the River Dijle. From the radiocarbon dates it can be concluded that roughly 60% of alluvial, and 75% of colluvial sediment deposition took place in the last years. If we apply the same ratio to the sediment export, sediment export for the last 1000 years ranges between 0.8 and 1.3 Mg ha 1 yr 1. These data are in correspondence with contemporary suspended sediment measurements for the Dijle near Leuven, which equals 0.9 Mg ha 1 yr 1 (Verstraeten et al., 2006). The above-mentioned values are those that were estimated by applying a correction factor for forests. If we were to establish a sediment budget without considering the role of forests, average sediment export for the last 1000 years would equal 1.6 to 2.4 Mg ha 1 yr 1. Such high values are very unlikely for a time period of 1100 years. It was only during an exceptionally wet winter period that sediment export values of 2.1 Mg ha 1 yr 1 were measured (Steegen, 2001). This stresses again the need to take land use into account when extrapolating erosion rates observed on agricultural land to entire catchments. 6. Conclusion A Holocene sediment budget was established for the 758 km 2 large Dijle catchment, situated in the Belgian loess belt. The resulting historical erosion volumes for agricultural land are consistent with the findings of other studies within the Belgian loess belt. The erosion and deposition calculations for the Nethen subcatchment are more realistic than those reported by previous studies (Rommens et al., 2006; Verstraeten et al., in press) as more augering data are used for the estimation of hillslope processes. Also morphometric mapping of colluvial dry valleys was improved, and the presence of historical forests was taken into account. The incorporation of historical land use into the budget results in large differences, showing that land use has major implications on the sediment dynamics. Augering data derived from agricultural land can thus not simply be extrapolated across the entire catchment. It is clear that erosion cannot only be explained by the local slope, and future research should concentrate on the refinement of the extrapolation of the augering datasets by testing the inclusion of other explanatory variables. Based on sedimentary properties, three sedimentological units can be recognised in the alluvial deposits throughout the catchment. Preliminary dating results show that the end of the peat development of unit 1 is rather variable (see Table 3), while the peat layer at the top of the second unit yields medieval dates. It is clear that the major part of alluvial sediment deposition took place after the early medieval period and can be attributed to the important agricultural activities that developed from this period on. These results are in accordance with the dating results from both alluvial and colluvial deposits in the Nethen subcatchment (Rommens et al., 2005, 2006, 2007). More detailed dating of alluvial and colluvial deposits is needed to see whether there is a time lag between both. The resulting sediment budget gives a detailed insight into the (historical) sediment sources and sinks within the catchment and their geographical distribution. Total erosion amounts to 817 ± 66 Mt, hillslope deposition 327±33 Mt (40%) and alluvial deposition 352±11 Mt (43%). This results in a HSDR of 60% and a SDR of 17%. Both the HSDR and SDR are variable within the different subcatchments, reflecting the differences in sediment dynamics. The presence of high amounts of colluvium and the variable (H)SDRs show that alluvial storage and export cannot be used as a proxy for erosion. The calculated sediment export rate falls well within the range of contemporary suspended sediment yield measurements, indicating that the applied methodology yields a realistic sediment budget. Acknowledgements This research is part of a project funded by the Fund for Scientific Research Flanders (research project G ). Their support is gratefully acknowledged. The authors would also like to thank Jeroen Monsieur, Bjorn Dieu and the several Msc. students in physical geography for their assistance during field work. We also thank Dr. Tom Vanwalleghem for providing soil augering data on the Meerdael forest, and Dr. Dirk Goossens for the discussion on Belgian loess soils. References Arnould-De Bontridder, O., Paulis, L., Etude du ravinement Holocène en forêt de Soignes. Acta Geographica Lovaniensia IV, Beach, T., The fate of eroded soil sediment sinks and sediment budgets of agrarian landscapes in southern Minnesota, Annals of the Association of American Geographers 84 (1), Bork, H.R., Bodenerosion, Holozäne und Pleistozäne Bodenentwicklung. Catena. Supplement 3, 183 pp. Bork, H.R., Soil erosion during the past millennium in Central Europe and its significance within the geomorphodynamics of the Holocene. Catena. Supplement 15,

19 Author's personal copy 162 B. Notebaert et al. / Catena 77 (2009) Bronk Ramsey, C., Deposition models for chronological records. Quaternary Science Reviews 27 (1 2), Clemens, G., Stahr, K., Past and present erosion rates in catchments of the Kraichgau area (SW Germany). Catena 22, de Moor, J., Verstraeten, G., Alluvial and colluvial sediment storage in the Geul River catchment (The Netherlands) Combining field and modelling data to construct a Late Holocene sediment budget. Geomorphology 95 (3 4), De Smedt, P., Paleogeografie en kwartair-geologie van het confluentiegebied Dijle-Demer. Acta Geographica Lovaniensia pp. Diriken, P., Geogids Huldenberg en Oud-Heverlee: Huldenberg, Loonbeek, Neerijse, Ottenburg, Sint-Agatha-Rode, Oud-Heverlee, Haasrode, Vaalbeek, Blanden, Sint-Joris-Weert. De Blauwe Vogel, Sint-Truiden. FAO, World reference base for soil resources. World Soil Resources Reports 84. Foulds, S.A., Macklin, M.G., Holocene land-use change and its impact on river basin dynamics in Great Britain and Ireland. Progress in Physical Geography 30 (5), Fryirs, K., Brierley, G.J., Variability in sediment delivery and storage along river courses in Bega catchment, NSW, Australia: implications for geomorphic river recovery. Geomorphology 38 (3-4), Geurts, M.-A., Genèse et stratigraphie des travertins de fond de vallée en Belgique. Acta Geographica Lovaniensia, 16. Gillijns, K., Poesen, J., Deckers, J., On the characteristics and origin of closed depressions in loess-derived soils in Europe a case study from central Belgium. Catena 60 (1), Goossens, D., Sedimentatiemechanismen bij natuurlijke stofdeeltjes in lucht. Unpublished Ph.D. Thesis, K.U. Leuven, Department Geography- Geology, Leuven, Belgium. Govers, G., Quine, T.A., Desmet, P.J.J., Walling, D.E., The relative contribution of soil tillage and overland flow erosion to soil redistribution on agricultural land. Earth Surface Processes and Landforms 21 (10), Hoffmann, T., Erkens, G., Cohen, K.M., Houben, P., Seidel, J., Dikau, R., Holocene floodplain sediment storage and hillslope erosion within the Rhine catchment. Holocene 17 (1), Houben, P., A Holocene sediment budget for the Rockenberg catchment. Pre-workshop field guide. Open Lucifs Workshop Houben, P., Geomorphological facies reconstruction of Late Quaternary alluvia by the application of fluvial architecture concepts. Geomorphology 86 (1-2), Knox, J.C., Floodplain sedimentation in the Upper Mississippi Valley: natural versus human accelerated. Geomorphology 79 (3-4), Lang, A., Phases of soil erosion-derived colluviation in the loess hills of South Germany. Catena 51 (3-4), Lang, A., Bork, H.-R., Soil erosion in an historical context. In: Boardman, J., Poesen, J. (Eds.), Soil Erosion in Europe. John Wiley & Sons. Langohr, R., L'antgropisation du paysage pédologique agricole de la belgique depuis le Néolithique ancien- Apports de l'archéopédologie. Étude et Gestion des Sols 8 (2), Langohr, R., Sanders, J., The Belgian loess belt in the last years: evolution of soils and relief in the Zonien forest. In: Boardman, J. (Ed.), Soils and Quaternary Landscape Evolution. John Wiley & Sons Ltd., pp Larue, J.P., Small valley bottom deposits in the sandy districts of the Sarthe basin (France): climate and/or human origin? Geomorphology 45, Larue, J.P., The status of ravine-like incisions in the dry valleys of the Pays de Thelle (Paris basin, France). Geomorphology 68, Lewis, D.T., Lepele, M.J., Quantification of soil loss and sediment produces from eroded land. Soil Science Society of America Journal 46 (2), Macaire, J.J., Bellemlih, S., Di-Giovanni, C., De Luca, P., Visset, L., Bernard, J., Sediment yield and storage variations in the Negron River catchment (South Western Parisian Basin, France) during the Holocene period. Earth Surface Processes and Landforms 27 (9), Miall, A.D., Architectural-element analysis a new method of facies analysis applied to fluvial deposits. Earth-Science Reviews 22 (4), Mullenders, W., Gullentops, F., Palynologisch en geologisch onderzoek in de alluviale vlakte van de Dijle te Heverlee-Leuven. Agricultura Band V, 2e reeks(1). Mullenders, W., Gullentops, F., Lorent, J., Coremans, M., Gilot, E., Le Remblement de la vallée de la Nethen. Acta Geographica Lovaniensia IV, Notebaert, B., Verstraeten, G., Govers, G., Poesen, J., submitted for publication. Qualitative and quantitative applications of regional scale LIDAR images in fluvial geomorphology. Earth Surface Processes and Landforms. Page, M.J., Trustrum, N.A., Dymond, J.R., Sediment budget to assess the geomorphic effect of a cyclonic storm, New Zealand. Geomorphology 9, Parsons, A.J., Wainwright, J., Brazier, R.E., Powell, D.M., Is sediment delivery a fallacy? Earth Surface Processes and Landforms 31 (10), Peeters, I., Spatial modeling of sediment redistribution patterns on a millennial time scale. Ph.D. Thesis, K.U. Leuven, Department of Geography- Geology, Leuven, Belgium. Poesen, J.W.A., Verstraeten, G., Soenens, R., Seynaeve, L., Soil losses due to harvesting of chicory roots and sugar beet: an underrated geomorphic process? Catena 43 (1), Reid, L.M., Dunne, T., Sediment budgets as an organizing framework in fluvial geomorphology. In: Kondolf, M.G., Piégay, H. (Eds.), Tools in Fluvial Geomorphology. Wiley, pp Reimer, P.J., Baillie, M.G.L., Bard, E., Bayliss, A., Beck, J.W., Bertrand, C.J.H., Blackwell, P.G., Buck, C.E., Burr, G.S., Cutler, K.B., Damon, P.E., Edwards, R.L., Fairbanks, R.G., Friedrich, M., Guilderson, T.P., Hogg, A.G., Hughen, K.A., Kromer, B., McCormac, G., Manning, S., Ramsey, C.B., Reimer, R.W., Remmele, S., Southon, J.R., Stuiver, M., Talamo, S., Taylor, F.W., van der Plicht, J., Weyhenmeyer, C.E., IntCal04 terrestrial radiocarbon age calibration, 0 26 ka BP. Radiocarbon 46, Rommens, T., Holocene Sediment Dynamics in a Small River Catchment in Central Belgium, PhD-thesis. K.U. Leuven, Department Geography- Geology, Leuven, Belgium. Rommens, T., Verstraeten, G., Poesen, J., Govers, G., Van Rompaey, A., Peeters, I., Lang, A., Soil erosion and sediment deposition in the Belgian loess belt during the Holocene: establishing a sediment budget for a small agricultural catchment. Holocene 15 (7), Rommens, T., Verstraeten, G., Bogman, P., Peeters, I., Poesen, J., Govers, G., Van Rompaey, A., Lang, A., Holocene alluvial sediment storage in a small river catchment in the loess area of central Belgium. Geomorphology 77 (1-2), Rommens, T., Verstraeten, G., Peeters, I., Poesen, J., Govers, G., Van Rompaey, A., Mauz, B., Packman, S., Lang, A., Reconstruction of late-holocene slope and dry valley sediment dynamics in a Belgian loess environment. The Holocene 17 (6), Ruysschaert, G., Poesen, J., Notebaert, B., Verstraeten, G., Govers, G., in press. Spatial and long-term variability of soil loss due to crop harvesting and its importance relative to water. Agriculture, Ecosystems and Environment. Slaymaker, O., The sediment budget as conceptual framework and management tool. Hydrobiologia 494 (1 3), Stankoviansky, M., Historical evolution of permanent gullies in the Myjava Hill Land, Slovakia. Catena 51, Steegen, A., Sediment deposition in and export from small agricultural catchments, Unpublished Ph.D. Thesis, K.U. Leuven, Department Geography- Geology, Leuven, Belgium. Trimble, S.W., A sediment budget for Coon Creek basin in the Driftless Area, Wisconsin, American Journal of Science 283 (5), Trimble, S.W., Contribution of stream channel erosion to sediment yield from an urbanizing watershed. Science 278 (5342), Trimble, S.W., Decreased rates of alluvial sediment storage in the Coon Creek Basin, Wisconsin, Science 285 (5431), Van Hove, D., Vanmontfort, B., Verstraeten, G., Van Rompaey, A., De Man, J., Van Peer, Ph., Mens & Landschap in het Dijlebekken Eindrapport Fase 2. Studie uitgevoerd in opdracht van het Vlaams Gewest Afdeling Monumenten en Landschappen. Van Oost, K., Van Muysen, W., Govers, G., Deckers, J., Quine, T.A., From water to tillage erosion dominated landform evolution. Geomorphology 72 (1 4),

20 Author's personal copy B. Notebaert et al. / Catena 77 (2009) Van Rompaey, A.J.J., Verstraeten, G., Van Oost, K., Govers, G., Poesen, J., Modelling mean annual sediment yield using a distributed approach. Earth Surface Processes and Landforms 26 (11), Vanmontfort, B., Van Hove, D., Verstraeten, G., Van Rompaey, A., De Man, J., Van Peer, Ph., Mens & Landschap in het Dijlebekken Eindrapport Fase 1. Studie uitgevoerd in opdracht van het Vlaams Gewest Afdeling Monumenten en Landschappen. Vanwalleghem, T., Van Den Eeckhaut, M., Poesen, J., Deckers, J., Nachtergaele, J., Van Oost, K., Slenters, C., Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium). Geomorphology 56 (1 2), Vanwalleghem, T., Poesen, J., Van Den Eeckhaut, M., Nachtergaele, J., Deckers, J., Reconstructing rainfall and land-use conditions leading to the development of old gullies. Holocene 15 (3), Vanwalleghem, T., Bork, H.R., Poesen, J., Dotterweich, M., Schmidtchen, G., Deckers, J., Scheers, S., Martens, M., Prehistoric and Roman gullying in the European loess belt: a case study from central Belgium. Holocene 16 (3), Vanwalleghem, T., Poesen, J., Vitse, I., Bork, H.R., Dotterweich, M., Schmidtchen, G., Deckers, J., Lang, A., Mauz, B., Origin and evolution of closed depressions in central Belgium, European loess belt. Earth Surface Processes and Landforms 32 (4), Vanwalleghem, T., Van Den Eeckhout, M., Poesen, J., Govers, G., Deckers, J., in press. Spatial analysis of factors controlling the presence of closed depressions and gullies under forest: application of rare event logistic regression. Geomorphology. Verstraeten, G., Poesen, J., Modelling the long-term sediment trap efficiency of small ponds. Hydrological Processes 15 (14), Verstraeten, G., Poesen, J., Goossens, D., Gillijns, K., Bielders, C., Gabrield, D., Ruysschaert, G., Van Den Eeckhaut, M., Vanwalleghem, T., Govers, G., Belgium. In: Boardman, J., Poesen, J. (Eds.), Soil Erosion in Europe. John Wiley & Sons, Ltd, pp Verstraeten, G., Rommens, T., Peeters, I., Poesen, J., Govers, G., Lang, A., in press. A temporarily changing sediment budget for a loess-covered catchment (central Belgium) during the Holocene. Geomorphology. Walling, D.E., Collins, A.L., Jones, P.A., Leeks, G.J.L., Old, G., Establishing fine-grained sediment budgets for the Pang and Lambourn LOCAR catchments, UK. Journal of Hydrology 330 (1 2), Walling, D.E., Quine, T.A., Using Chernobyl-derived fallout radionuclides to investigate the role of downstream conveyance losses in the suspended sediment budget of the River Severn, United Kingdom. Physical Geography 14, Walling, D.E., Russell, M.A., Hodgkinson, R.A., Zhang, Y., Establishing sediment budgets for two small lowland agricultural catchments in the UK. Catena 47 (4),

21 Netherlands Journal of Geosciences Geologie en Mijnbouw Fluvial architecture of Belgian river systems in contrasting environments: implications for reconstructing the sedimentation history B. Notebaert 1,3,*, G. Houbrechts 2, G. Verstraeten 1, N. Broothaerts 1, J. Haeckx, M. Reynders, G. Govers 1, F. Petit 2 & J. Poesen 1 1 Department Earth & Environmental Sciences, KU Leuven, Belgium 2 Hydrology and Fluvial Geomorphology Research Centre, Department of Geography, University of Liège, Belgium 3 Research Foundation Flanders FWO * Corresponding author. bastiaan.notebaert@ees.kuleuven.be. Manuscript received: August 2010, accepted: June 2011 Abstract Accurate dating is necessary to get insight in the temporal variations in sediment deposition in floodplains. The interpretation of such dates is however dependent on the fluvial architecture of the floodplain. In this study we discuss the fluvial architecture of three contrasting Belgian catchments (Dijle, Geul and Amblève catchment) and how this influences the dating possibilities of net floodplain sediment storage. Although vertical aggradation occurred in all three floodplains during the last part of the Holocene, they differ in the importance of lateral accretion and vertical aggradation during the entire Holocene. Holocene floodplain aggradation is the dominant process in the Dijle catchment. Lateral reworking of the floodplain sediments by river meandering was limited to a part of the floodplain, resulting in stacked point bar deposits. The fluvial architecture allows identifying vertical aggradation without erosional hiatuses. Results show that trends in vertical floodplain aggradation in the Dijle catchment are mainly related to land use changes. In the other two catchments, lateral reworking was the dominant process, and channel lag and point bar deposits occur over the entire floodplain width. Here, tracers were used to date the sediment dynamics: lead from metal mining in the Geul and iron slag from ironworks in the Amblève catchment. These methods allow the identification of two or three discrete periods, but their spatial extent and variations is identified in a continuous way. The fluvial architecture and the limitation in dating with tracers hampered the identification of dominant environmental changes for sediment dynamics in both catchments. Dating methods which provide only discrete point information, like radiocarbon or OSL dating, are best suited for fluvial systems which contain continuous aggradation profiles. Spatially more continuous dating methods, e.g. through the use of tracers, allow to reconstruct past surfaces and allow to reconstruct reworked parts of the floodplain. As such they allow a better reconstruction of past sedimentation rates in systems with important lateral reworking. Keywords: Belgium; climate change; dating; fluvial architecture; Holocene; land use change Introduction Soil erosion and sediment redistribution are important geo - mor pho logic processes during the Holocene in many West- and Central European catchments. An important component of sediment redistribution is (net) floodplain deposition, which provides a buffer between hillslope soil erosion and downstream sediment delivery at different spatial and temporal scales (e.g. Trimble, 2010). In many catchments floodplain deposition has varied during the Holocene, which is often attributed to changes in (anthropogenic) land use or climate (e.g. Dotterweich, 2008; Trimble, 2009; Verstraeten et al., 2009). When studying such relationships, it is essential that net floodplain sediment accu - mulation is dated, which requires a good insight in floodplain processes and the resulting sedimentary facies. The large variation in river types and dominant processes is reflected in different sedimentary floodplain structures (e.g. Nanson and Croke, 1992). While some floodplains are dominated by lateral accretion deposits, others are dominated by vertical aggradation deposits. The identification of the past deposition environments Netherlands Journal of Geosciences Geologie en Mijnbouw

22 of the different facies, based on sedimentologic properties, is often referred to as fluvial architecture (e.g. Miall, 1985). The fluvial architecture will help to identify the nature of the dated deposits and therefore the possibilities and interpretations for dating floodplain processes. Dates of overbank fines may produce totally different results from dates of other fluvial settings like channel beds or abandoned channel infillings. Radiocarbon dates from deposits which result from lateral accretion (e.g. channel bed) show other age distributions than dates from floodplain deposits or dates from floodbasins in the same catchment (e.g. Hoffmann et al., 2008; Macklin et al., 2010). Interpretation of dating results therefore requires a thorough understanding of the depositional environment of the dated material. For many large river (catchment >1000 km 2 ), the flood - plain type and associated processes (e.g. Nanson & Croke, 1992) can be derived from simple topographic information. For smaller rivers this is often difficult (e.g. Notebaert et al., 2009a), especially when floodplain aggradation occurred over the last few hundred years, like in many catchments in Western Europe. In addition, the extent and importance of processes may have changed through time, and determining the past environments may be necessary to evaluate dating possibilities.the main objective of this paper is to identify the fluvial architecture for three Belgian catchments and the implications the different styles of fluvial architecture have on dating possibilities of floodplain sedimentation. The catchments of the selected rivers, the Dijle, Geul and Amblève, differ in environmental settings. We focus on the different elements of the fluvial architecture, and stress the possibilities and limitations for dating floodplain sediment storage based on these elements. Study areas This paper discusses the floodplain of three Belgian catch ments: the Dijle, Geul and Amblève catchment (Fig. 1). The Dijle catch - ment (Fig. 2A) is situated in the central Belgian loess belt, and Fig. 1. Location of the study sites in Belgium. A: Amblève; D: Dijle; G: Gulp. in this study we consider the part of the catchment upstream the city of Leuven (760 km 2 ). The topography exists of an undu - lating plateau in which the rivers are incised. The floodplain width of the main valley varies between 200 and 1800 m, while the tributaries have smaller floodplains. The soils of the catch - ment are mainly Luvisols which developed in Pleistocene loess deposits. The first palynological traces of agriculture date from the Atlantic Period ( cal BP; Mullenders & Gullentops, 1957; Mullenders et al., 1966; De Smedt, 1973), and agricultural land use peaked during the Roman Period and from the Middle Ages on. Despite medieval and contemporary intensive land use, some large areas remained forested since at least the 14 th century (e.g. Vanwalleghem et al., 2006). Current land use is dominated by cropland and these historical forests on the plateaus and slopes, and grassland and forests on the floodplains. Large quantities of soil have been eroded and deposited in colluvial and alluvial valleys caused by the intensive land use history (e.g. Notebaert et al., 2009b). The Geul catchment (350 km 2 ; Fig. 3) is located in the northeast of Belgium and the southeast of the Netherlands. This study considers the Belgian part of the Geul floodplain (c. 120 km 2 upstream area) and its tributary, the Gulp (c. 47 km 2 upstream area). The topography of the catchment consists of an undulating plateau with deeply incised river valleys. Floodplains are up to 250 m wide. Soils are mainly Luvisols developed in loess, although some sand, gravel and bedrock outcrops occur. The land use history is comparable with that of the Dijle catchment, with the exception of the last few hundred years: a conversion of cropland into grassland started during the 17 th century in the south-western part of the catchment, and progressively spread towards the north (Mols, 2004). Hence, current land use is dominated by grassland and forests. The Amblève catchment (c km 2 ; Fig. 4a) is located in the Belgian Ardennes Hercynian massif. The topography consists of undulating plateaus, deeply incised (up to more than 250 m) by some large river valleys, often with very steep valley slopes (>15%). The floodplains of the upper reaches widen downstream to c. 350 m. The floodplains of the lower, deeper incised reaches are smaller and the width ranges between an almost absent floodplain (~0 m) and 330 m, depending on local geology. The Warche tributary flows through the Malmedy graben, where the floodplain is up to 800 m wide. The upper and lower parts are for most tributaries separated by a reach with a steep gradient (slope >1 m/m), where floodplains are very narrow or absent. The land use history of this catchment is much less intense compared to the other two catchments. The first traces of agri - culture in palynological records from the Hautes Fagnes date from the Neolithic period, but the anthropogenic influence remains very low until c AD (Damblon, 1969, 1978). Gullentops et al. (1966) report palynological evidence for agri - culture in the Lienne catchment from the Subboreal period onwards, while Houbrechts (2005) reports on the local start of 32 Netherlands Journal of Geosciences Geologie en Mijnbouw

23 a. b. Fig. 2. a. Overview of the Dijle cathment. The rectangle displays the extend of Fig. 2b. Main cities: CSE: Court-St- Etienne; Gen: Genappes; Leu: Leuven; Wa: Wavre. Cross sections: KOR: Korbeek-Dijle (Fig. 7); SJW: Sint-Joris-Weert (Fig. 8); SCL: Sclage (Fig. 9); b. Location of surface samples of the floodplain deposits for study of the contemporary texture as function of the depositional environment. Coordinates in this and following figures are in the Belgium Lambert 72 system. colluvial deposition related to agriculture in a subcatchment at 3195±30 BP ( BC; ages are calibrated using Oxcal 4.1 (Bronk Ramsey, 2001, 2009) and the Intcal 04 calibration curve (Reimer et al., 2004), with a 2σ uncertainty; non-calibrated radiocarbon ages are referred years BP, calibrated as years BC/AD). Large deforestations for iron industries occurred from the 14 th century on (e.g. Houbrechts & Petit, 2004). Historical maps indicate a far lesser extent of cropland during the 18 th century than for the other catchments (e.g. de Ferraris map, 1775), and a conversion from cropland to grassland occurred during the 20th century (Mols, 2004). The contemporary land use is dominated by forests and grassland. Methods Floodplain characterization Information on the nature of fluvial deposits is retrieved through an extended coring datasets, complemented with profile pit data. Corings are grouped in floodplain cross sections. For each coring a detailed in field description is made with a vertical resolution of 5 cm, providing information on approximate texture class, colour, quantity, nature and size of gravel, presence of plant material, peat or other inclusions, and soil horizons. Table 1 provides an overview of the number of corings for each catchment. In order to study the complete fluvial architecture of the floodplain, corings should be spaced at a distance that is smaller than the width of the smallest architectural element, which is often the river channel (e.g. Houben, 2007). Achieving Table 1. Overview of the collected field date for the different catchments. Fig. 3. Overview of the Geul catchment. Main cities: Gu: Gulpen; Ke: Kelmis; Va: Valkenburg. Cross sections: COT: Cotessen; KAR: Karsveld; Pl: Plombières; TEU: Teuven. Catchment Number Number Catchment Total of corings of cross area (km 2 ) floodplain sections area (km 2 ) Dijle Amblève Gulp Geul excluding Gulp Netherlands Journal of Geosciences Geologie en Mijnbouw

24 such a coring density is very time and labour demanding, especially for catchments larger than a few km 2 where multiple cross sections are required to tackle the spatial variability between sites. The coring densities in this study vary between less than 1 time the river channel width in the downstream valley sections of the Amblève catchment to more than 10 times the channel width in the underfit periglacial valleys of the upper reaches of the Amblève. For most cross-sections the distance between two cores is c. 2 times the channel width, which allows identifying most sedimentary units. In order to evaluate the relationship between the sediment texture and the depositional environments in the con - temporary Dijle catchment, detailed textural analysis was performed. Two study sites were selected in the main trunk valley: the floodplain immediately downstream Korbeek-Dijle, and the floodplain near Sint-Joris-Weert (Fig. 2b). Samples were taken from the upper 0.3 m of the deposits, and as such the samples represent the current depositional environment (Table 2). Samples were homogenized, sieved at 500 µm, and the grain size distribution was determined through laser diffraction particle-size analysis, using a Beckham Coulter LS For each sample the median, 90%, 95% and 99% percentile of the grain size distribution was determined. A CM pattern (Passega, 1957) is obtained by plotting the 99% percentile (C, from coarsest fraction) on the Y-axis and the median grain size (M, from median fraction) on the X-axis, using a logarithmic scale for both axes. The different depositional processes are reported to form separate groups on these diagrams (e.g. Passega, 1957). Table 2. Number of samples per depositional environment used for the grain size analysis. Depositional environment Korbeek-Dijle Sint-Joris-Weert Backswamps Floodplain 12 9 Levee (40m from channel) 10 9 Levee (top; 2-4m from channel) 9 9 Point Bar 9 8 Channel 9 9 Total Dating of floodplain deposits The chosen method for dating floodplain deposition depends on the depositional environment and the resulting nature and availability of datable material. In this study, we focus on the net floodplain aggradation, which requires dates from overbank deposits. Ages from in-channel deposits provide information on the former position of the river channels and the lateral migration. Ages of overbank deposits provide information on net sediment accumulation in the floodplain. Samples for radiocarbon dating were sieved, dried, and manually searched for datable material. Identified terrestrial plant remains were preferred above charcoal. Only a few samples were dated in the Amblève catchment, due to the lack of datable material in overbank deposits. Optical Stimulated Luminescence dating (OSL) was performed in the Dijle catchment and the used methodology and associated problems are discussed by Notebaert et al. (2011a). Tracers may provide another dating method for floodplain deposits (e.g. Brown et al., 2003): sediments deposited after the introduction of a tracer in a fluvial system will be contaminated with this tracer. Sediments deposited after the removal of the tracer sources may still be contaminated, as reworking of the (contaminated) floodplain sediments by the river may continue to provide a source of the tracer. In this case, the presence of the tracer provides one discrete age control point. Additional age control is derived if the temporal variations in tracer intro - ductions are known, as these can be linked to tracer concen - trations in the floodplain deposits (e.g. Stam, 1999, 2002). At least since the Middle Ages until the end of the 19 th century Pb and Zn mining occurred in the Geul catchment (Fig. 5), causing severe contamination of fluvial deposits. Previous studies (Stam, 1999, 2002) have shown that peaks in heavy metal concentrations in fluvial deposits can be linked with peaks in 19 th century mining activities. Where previous studies focused on profiles at cutbanks, we sampled within the alluvial plain to get a better spatial distribution of sampling sites over the floodplain. Samples were taken at two sites: at Plombières (Belgium) and 1700 m downstream at Cottessen (the Netherlands) (Fig. 3). Samples were taking using a percussion drill with a diameter of 5 cm, and samples have depth intervals of 5 to 15 cm. In total 12 percussion drillings were analysed. Samples were dried, sieved at 2 mm and grounded. Pb concentrations were determined after dissolution with concentrated HCl, HNO 3 and HF, using flam atomic absorption spectroscopy (F-AAS). Pb concentrations where plotted as function of sample depth, and these plots were correlated with the historical Pb mining activity intensity. For each coring the start of increased Pb concentrations and the peak was identified, corresponding with c and ca 1869 AD respectively. In the Amblève catchment, contamination of floodplain deposits by slag particles originating from metal industries (blast furnaces and bloomeries) has previously been used to study floodplain histories in the Belgian Ardennes (e.g. Henrottay, 1973; Brown et al., 2003; Houbrechts & Petit, 2003, 2004). Metal industries were set up in the Lienne valley (Fig. 4b), a tributary of the Amblève, starting at the end of the 14 th century AD (Houbrechts & Weber, 2007), and large amounts of ironwork waste products were dumped in the river channel and on the alluvial plain. Slag particles dumped in the floodplain are reworked by the river ever since, providing a continuous source of scoria. 34 Netherlands Journal of Geosciences Geologie en Mijnbouw

25 a. b. Fig. 4. a. Overview of the Amblève catchment. Main cities: Mal: Malmedy; Rem: Remouchamps; St: Stavelot. Main rivers: A: Amblève; B: Warche; C: Salm; D: Lienne. The rectangle displays the extend of Fig. 4b. Floodplain cross sections: BUL4: Bulingen 4; FDQ: Fond de Quarreux; WAR3: Warche 3; b. Overview of the Lienne subcatchment. Years indicate the year of initiation of metal industry for the given location, if known (Houbrechts and Weber, 2007). Neu: Neucy cross section; Mont: Monty cross section; Rah: Rahier cross section. Samples with a thickness of 0.05 to 0.15 m were taken at 6 sites (Fig. 4b) using a cleaned Edelman auger. Dry samples were sieved and the coarsest sand fraction was analysed for the presence of metal slag elements. Metal slag concentrations were plotted on the cross sections, which allowed the identification of sediments deposited after the initiation of metal industries. Small contaminations caused by bioturbation and sampling were taken into account: only when the mass percentage of scoria is larger than 2%, the sample is considered as containing such scoria. Results Dijle catchment Sediment grain size The analysis of the grain size of contemporary deposits shows that the different deposition environments can be differentiated based on the coarsest (C) and median (M) grain size. There is a clear decrease in median grain size when comparing deposi - tional environments in or close to the river channel with those further from the channel (Table 3). The channel lag deposits have a mode of about 200 µm (Fig. 6) and sand is the dominant fraction (Table 3 and 4), although some samples contain larger amounts of silt and clay (up to 35 and 50%). This is possibly caused by sampling a combination of channel lag and under - lying fine grained Holocene overbank deposits due to the limited thickness of channel lag deposits. This superposition of channel lag deposits above overbank deposits is the result of floodplain aggradation (e.g. Notebaert et al., 2011a), combined with a recent migration of the meander. Point bar deposits show typically a bimodal grain size distri - bution, with peaks around µm (sand) and µm (silt). The levee deposits still show a bimodal distribution, but because the coarsest peak becomes less important, they are dominated by silts. The histograms of overbank and back swamps deposits are dominated by a peak around µm (Korbeek- Dijle) or µm (St-Joris-Weert). They are dominated by silt and clay. T-tests show that the depositional facies can best be differentiated using median values combined with 90% or 95% percentiles (Broothaerts, 2008), while the 99% percentile is less fitting. We hypothesise that this is mainly due to the sensitive - ness of the 99% percentile, as this percentile relies on a very low amount of grains. Based on the t-tests on the 50% and 95% percentiles, all groups are significantly different. The CM pattern (Fig. 7) of both sites are in agreement with the patterns found by other studies (e.g. Passega, 1957, 1964; Bravard et al., 1989; Bravard & Peiry, 1999). These results confirm that grain size distribution can be used distinguish depositional environments. Fluvial architecture Different Holocene fluvial facies units could be distinguished based on texture and other sedimentologic properties like layering patterns and presence and nature of organic material. Figure 8 displays typical cross sections of the floodplains in the Dijle catchment, and the different facies units are listed in Table 5. Netherlands Journal of Geosciences Geologie en Mijnbouw

26 Table 3. Grain size properties for the different deposition environments: median texture (µm), average en standard deviation of the percentage clay, silt and sand. Korbeek-Dijle Sint-Joris-Weert Median (µm) % clay % silt % sand Median (µm) % clay % silt % sand Backswamps 17±8 35.6± ± ±2.9 9±5 45.7± ± ±2.9 Floodplain 21±6 25.3± ± ±5.6 11±3 44.8± ± ±3.5 Levee (~40 m from channel) 27±6 23.9± ± ±5.4 20±2 29.0± ± ±2.5 Levee (top; 2-4 m from channel) 31± ± ± ± ±3 24.9± ± ±4.3 Point Bar 55± ± ± ± ± ± ± ±14.3 Channel 169±86 6.9± ± ± ± ± ± ±32.6 Table 4. Grain size properties of the different deposition environments: range in percentage clay, sand and silt. Korbeek-Dijle Sint-Joris-Weert % clay % silt % sand % clay % silt % sand Backswamps Floodplain Levee (~40 m from channel) Levee (top; 2-4m from channel) Point Bar Channel Facies unit 1 exists of sandy sediments, deposited at the bottom of the cross sections. Its texture varies between loamy sand and sand with fine gravel (<2 cm), without organic material. This texture indicates a relatively high energy fluvial system. The top of this layer is not horizontal and has indications of shallow channels, which are possibly caused by a post-deposi - tional erosional phase (see also De Smedt, 1973). A single OSL age is available from this layer: 26000±4000 BP (2σ uncertainty; Notebaert et al., 2011a). This unit is interpreted as a Weichselian and Lateglacial braided river deposit. Facies unit 2 is found in comparable positions as unit 1, and exists of compact silty to loamy sediments. This layer grades horizontally and vertically in unit 1, often with an intermediate texture in between. Because of its position, this unit is con - sidered as a Late Weichselian to Lateglacial deposit. Possibly it was deposited in the braided river plain at a distance from the channels, although it may also be an in-situ loess deposit or a mixture of both. Fig. 5. Yearly production of galena at Plombières (after Dejonghe et al., 1993). Unit 3a exists of a complex of organic layers. This layer sometimes consists of reed peat or in some cases woody peat. But in most corings it consists of a very organic layer of decomposed peat or a very organic silty to clayey layer with plenty of organic material. Sometimes this layer exists of gyttja, especially the lower parts. This layer often contains small fresh - water shells. These different types grade into each other both vertically and horizontally. The organic matter content varies between c. 20 and 80% (Rommens, 2006; Notebaert et al., 2011a). Facies unit 3b exists of a complex of calciumcarbonate rich, often organic, layers. It varies from layers with a high content of calciumcarbonates (nodules, sometimes shells) to deposits of almost pure calciumcarbonate nodules. Like unit 3b, it can consist of gyttja. This unit is located at the same position and grades vertically and horizontally in facies unit 3a. In the Train tributary, this layer exists locally of almost pure calcium - carbonate nodules, with only a few organic layers. Here, this deposit was previously identified as travertine (Geurts, 1976). Unit 3a and 3b (= unit 3) form a large complex of organic and calciumcarbonate rich deposits and are deposited above units 1 and 2. Dates from the base of unit 3 range from 9500 BC to c BCE, ages of the top vary between c BC and c AD (Notebaert et al., 2011a). The thickness of unit 3 varies, but most often it is 1 to 3 m thick. It is interpreted as an early to middle Holocene organic and calcic vertical floodplain accumulation. The high calciumcarbonate content of unit 3b and the presence of gyttja indicate an environment with stagnant water. Downstream the studied catchment, comparable deposits are formed (De Smedt, 1973). Exposed profiles along 36 Netherlands Journal of Geosciences Geologie en Mijnbouw

27 the entire floodplain width indicate the absence of a channel facies formed synchronous with unit 3, which is explained by a diffuse water flow (De Smedt, 1973). The formation of this facies unit is related to a period with limited water and sediment discharge, during which floodplains were stable and mineral sediment deposition rates were low. Fig. 7. CM patterns (Passega, 1957) for Korbeek-Dijle and Sint-Joris-Weert. Fig. 6. Typical grain size distributions for the different depositional environments in the contemporary Dijle floodplain. Unit 4 exists of a silty clay loam and silt loam. There are often small organic layers, sometimes small peat layers, but in general the amount of organic matter is decreasing. The top of this layer is often a well developed organic or peat layer. Unit 4 covers unit 3, and the transition between both units is gradual. Ages from the top peat layer range between c. 700 and 1500 AD (Notebaert et al., 2011a). This unit is interpreted as an overbank sediment, deposited under conditions of increasing sediment load and floodplain deposition, and decreasing importance of aggradation of organics in the floodplain. The texture of the unit varies both laterally and vertically, as a function of the local stream power of the water, varying between the more distal parts of the floodplains and the parts close to the channel. Well developed levees cannot be identified for this unit. Netherlands Journal of Geosciences Geologie en Mijnbouw

28 Fig. 8. Floodplain cross section in the Dijle catchment at Korbeek-Dijle (Dijle River), St-Joris-Weert (Dijle River) and Sclage (Cala River). An explanation of the different units can be found in the text and Table 5. Vertical exag - geration 50 times for Korbeek- Dijle and St-Joris-Weert, and 10 times for Sclage. The location of the different cross sections is indicated on Figure 2. Unit 5 exists of complex of silty clay loam, silt loam and some loam, which are laterally and vertically grading into each other. This unit is located at the top of the floodplain, and contains in general no peat or organic-rich layers, except for the current A-horizon. Unit 5 is often between two and five metres thick, and is interpreted as the overbank deposits of the last 1000 years, which are related to large scale deforestations that triggered severe soil erosion on the loess plateau (Rommens et al., 2006; Notebaert et al., 2009b; Verstraeten et al., 2009). At some locations thin layers (<15 cm) of sandy loam to loam deposits interrupt the silty (clay) loam layers. These deposits are interpreted as being deposited by large floods, probably on crevasse splays. Other heterogeneous sandy and silty deposits which are situated close to a palace channel are interpreted as levees. More clayey deposits can be found at the distal parts of the floodplains (backswamps). This unit was deposited during the last 1000 to 1300 years, and makes up the largest part of the Holocene floodplain deposits of the catchment (Notebaert et al., 2011a). Unit 6 exists of sandy loam, sands with at some location <20% fine (<5 cm) gravel, and is found at the same absolute height as units 4 and/or 5, sometimes forming a depression in 38 Netherlands Journal of Geosciences Geologie en Mijnbouw

29 Table 5. Different facies units of the Dijle catchment. Unit Texture Position Other properties Interpreted deposition Age environment 1 Sands with fine gravels Bottom of the floodplain deposits - Braided river deposits Pleistocene to loamy sand 2 Compact silty and Bottom of floodplain deposits; grades - Braided river deposit: distal Pleistocene? loamy sediments laterally and vertically into unit 1 parts of the floodplain? 3a Organic: peat to very Above units 1 and 2, covered by - Organic floodplain infilling Start: early organic silt and clay unit 4 or unit 5; over the entire Holocene deposits floodplain width End: 4600 BC 3b Calciumcarbonate rich - Organic and calciumcarbonate to 1500 AD, deposits, often organic rich floodplain infilling with depending on stagnant water location 4 Silty clay loam and silt Covering unit 3 and covered by unit 5 Top is often a Overbank deposit Start: depending loam, contains some peat layer on location from organic or peat layers c BC onwards End: c AD 5 Silty clay loam to loam Top of floodplain - Overbank deposits Deposition from c AD onwards 6 Sandy loam, sands and At the same level of units 4 and 5, Contains River channel and point bar After c BC sands with some (<20%) with an erosive lower boundary; sometimes brick deposits fine(<5 cm) gravel covers <20% of floodplain width; fragments, twigs, often associated with current or or other organic known past channel belts remains 7 Fine sand to silty clay At location of colluvial fans or Sometimes contains loam, often arranged footslope deposits; grades laterally fragments of bricks in small layers into units 4 and 5 and charcoal Colluvial deposit After c BC unit 3. In some cases a textural fining up can be observed, but generally textural variations are hard to determine. The transition to the underlying unit is sharp, indicating an erosive boundary, while the upper transition to unit 4 or 5 is mostly gradual. Unit 6 often contains small wood fragments. This unit is interpreted as a combination of channel lag and point bar deposits. The differentiation between both facies is not possible in the field. The position of this unit is always confined to a part of the floodplain, and often located in its centre, forming stacked point bar deposits. In most cases less than 20% of the floodplain area in the cross section contains unit 6, but in some exceptional cases (like the Korbeek-Dijle cross section, Fig. 8) the Holocene channel belt crosses the cross-section several times. The texture of unit 7 contains fine sand to silty clay loam, often arranged in small horizontal layers. This layer sometimes contains fragments of bricks and charcoal and is always positioned along the floodplain edges, and connects to colluvial fans and footslope deposits. It is interpreted as a colluvial deposit or a mixture of colluvial and alluvial deposits (unit 7b). It only occurs in the upper parts of the floodplain deposits, at a higher altitude than unit 3, indicating that it was deposited later than unit 3. Due to the slight differences between units 4, 5 and 7, it is unclear whether unit 7 was deposited contemporary with unit 5 or with unit 5 and unit 4. In general, the Holocene floodplain deposition in the Dijle catchment can be divided in three phases. During phase 1 the organic and calciumcarbonate deposits of unit 3 were deposited. This phase started in the early Holocene, and the end varies from site to site. During phase 2 units 4 and the lower parts of unit 6 were deposited. This phase is the result of the increase of sediment load in the floodplain, which is related with anthropogenic land use (Notebaert et al., 2011a). During this phase peat growth in the floodplain was replaced by clastic aggradation. Phase 3 consists of unit 5 and the upper part of unit 6, and started approximately at 1000 CE. This is the major floodplain aggradation phase, and is related to an intense anthropogenic land use, causing severe soil erosion and sediment redistribution (Rommens et al., 2006; Verstraeten et al., 2009; Notebaert et al., 2011 a). This general pattern is not homogenously present in the catchment, and variations occur in the presence and thickness of units 3 to 5. Locally only two units (units 3 and 5) can be differentiated. Locally a deep channel belt incised before the Holocene in the Weichselian deposits (units 1 and 2), and this Netherlands Journal of Geosciences Geologie en Mijnbouw

30 depression was filled during the earlier parts of the Holocene (units 3 and 4). Only from the moment that this depression was filled up, the entire contemporary floodplain was covered with sediments (phase 3) (e.g. Fig. 8, cross section St-Joris-Weert). In its upper reaches, the Train River is incised in Early and Middle Holocene floodplain deposits, forming a gorge with steep, up to six meter high, banks, while a contemporary floodplain is absent. The (terraced) floodplain is build up of unit 3b, and the subsequent deposition of units 4 to 6 did not occur here. At a depth of 0.2 m, this terraced floodplain was dated at c BCE, providing the only dating control of the start of the incision phase (Notebaert et al., 2011a; location Bonlez U ). There are no indications why late Holocene aggradation is absent for this location. Thick Early and Mid Holocene travertine deposits are, however, present here (see also Geurts, 1976), and we hypothe - sise that breaching of a travertine dam and a subsequent lowering of the base level is responsible for the incision phase. Geul catchment Fluvial architecture In the Geul catchment, the floodplains of the Gulp tributary and of the Belgian part of the Geul itself were studied. De Moor et al. (2008) discuss the different sedimentary units encountered in the Geul floodplain, and these units largely agree with the units encountered in the Gulp catchment. But not all units that are described by De Moor et al. (2008) were encountered in this study, mainly because some units occur only downstream of the studied stretches. A summary of all units is provided in Table 6, while typical floodplain cross sections are provided in Fig. 9. There are also some other differences with the descriptions of De Moor et al. (2008): they make a distinction between units with a silty loam (their units 3 and 4) and a silty clay loam (their unit 5) texture, while this was not possible for most corings in this study, and hence these units are merged (unit 3). But in addition, we were able to make a further distinction within the overbank deposits (unit 3): the upper parts (unit 3b) have characteristic dark brown and dark grey colour, which coincides with high lead concentrations (see below). This unit can only be found in the floodplain of the Geul, downstream of the mining sites near Plombières, and it is interpreted as being deposited since the initiation of large scaled lead mining in the catchment (1842 CE). In general, the floodplains of the Gulp and Geul show a pattern of a basal gravel layer (unit 1), covered with finer sediments (units 2 to 7) deposited on point bars or as overbank floodplain deposits. The accumulated thickness of these fine deposits increases downstream, for the Geul from c. 0.5 m to more than 3 m near the Belgian-Dutch border, and for the Gulp from c. 0.5 m to c. 2.6 m. The occurrence of point bar deposits over the entire floodplain width (Fig. 9) indicates the importance of lateral migration of the river channel during the Holocene. The thickness of these point bars varies within cross sections, which may be partially due to the fuzzy delineation of this unit, both with respect to units 1 and 3. Locally gravels are deposited on the lower parts of point bars, and as a result the upper parts of unit 1 may have been deposited on such point bars. In addition, the thickness of the point bars may have increased over time, simultaneous with floodplain aggradation (see also De Moor et al., 2008). Dates from the organic material directly on top of unit 1 in the Dutch Geul floodplain indicate that the majority of the floodplain was reworked during the Holocene period (De Moor et al., 2008). Dating floodplain deposition Radiocarbon datable material found in the Geul floodplain is always situated in units 1 and 2, and most often at the transition between both units. The resulting ages would provide information on the moment of floodplain reworking (and lateral channel movement) but not on floodplain aggradation. But as point bar deposits mainly consist of reworked material, the value of dating such material is further reduced. As a result, dating based on Pb as a tracer is used to identify net floodplain sedimentation within the studied section of the Geul catchment. Lead concentrations for sediments deposited before 1842 show large variations between corings for the Plombières site (150 to 700 mg Pb per kg soil) while they are more constant at Cottessen (around 200 mg Pb per kg soil) (Table 7). This can be explained by small scaled and localised mining activities before this period. Peak values reach 6000 mg Pb per kg soil at Plombières and 1600 mg Pb per kg soil at Cottessen, which is a function of downstream dilution of the pollution. Peaks in lead concentration corresponding with ~1842 AD and ~1869 AD are determined for the different corings (Fig. 10). The results of the sediment deposition per time span (Table 8) show that for both studied sites on average 17% of the total Holocene floodplain deposition occurred after the initiation of the metal mining (~1842 AD). There are some differences between both sites concerning the deposited fractions in the time frames 1842 AD AD and 1869 AD - present (Table 8). Results are influenced by the vertical sampling resolution and difficulties in interpreting the peaks in lead concentrations, which hampers a detailed reconstruction of the sedimentation history. Nevertheless, for both sites the sedimentation rate is higher for time frame 1842 AD AD. Amblève catchment Fluvial architecture The floodplains of the Amblève catchment can be distinguished into three reaches: the upper reaches (e.g. cross section Bullingen 4, Fig. 11), the lower reaches (e.g. cross section Warche 4, 40 Netherlands Journal of Geosciences Geologie en Mijnbouw

31 Fig. 9. Floodplain cross section in the Geul catchment at Teuven (Gulp River), Karsveld (Gulp River) and Plombières (Geul River). An explanation of the different units can be found in the text and Table 6. Vertical exaggeration: 10 times. LR: lateral reworked since increased lead concentrations (c AD). NFA: net floodplain sediment accumulation since increased lead concentrations. The boundary between unit 3 and unit 3a is based on laboratory measurements and field observations, while the increased lead concentrations in units 1 and 2 are based on laboratory analysis. The location of the different cross sections is indicated on Figure 3. Fig. 11), both having relatively gentle slopes, separated by reaches with steeper gradients which are associated with a river long profile knickpoint. Also the lower reaches contain one steeper reach (Fonds de Quarreux; cross section depicted in Fig. 11), which has a comparable valley morphology and fluvial architecture as the other steep reaches. The upper reaches often have broad floodplains along a meandering river. Here, the river has a low width/depth ratio (<10). The width of the floodplain of the lower reaches is highly variable, depending on local geology, and the rivers have a straight pattern within their inherited meandering valleys. The width/depth ratio of these lower reaches is high (>10, most often >20). Steep reaches occur just downstream the knickpoints which separate these upper and lower reaches, and at a lithological knickpoint at Fonds de Quarreux. Here floodplains are almost absent. The fluvial architecture between upper and lower reaches differs only slightly, except for the steep reaches. Unit 1 consists of a poor sorted basal gravel layer, sometimes mixed with some sands or organic material. The depth of the top of this gravel layer varies between and within cross sections. In the lower reaches, abandoned river channels are still visible in the floodplains as depressions, and form often also depressions in the top of this gravel layer (Notebaert et al., 2009a). This gravel layer was most probably deposited during the Weichselian, and is currently being reworked in the river channel and deposited as channel lag deposit or on bars. A Netherlands Journal of Geosciences Geologie en Mijnbouw

32 Table 6 Different facies units of the Geul and Gulp catchments. Unit Unit Texture Position Other properties Interpreted Age De Moor deposition et al. (2008) environment 1 1 Poor sorted gravel, Base of Holocene fluvial Contains sometimes Channel lag and lower Weichselian mixed with some sand deposits,underlying the organic material bar deposits; reworked and reworked entire valley width (twigs, nuts) Pleistocene material Holocene 2 2 Fining up sequence of Above unit 1 and covered Lower part may contain Point bar Holocene sand and sandy loam to by unit 3, over the entire organic material like silt loam or silty clay loam valley width twigs and nuts Overbank fines; lower Holocene 3 3, 4 and 5 Silt loam and silty clay On top of unit 2, at the surface - part loam or covered by unit 4 or 7 3b - Silt loam and silty clay Downstream the mining sites Distinctive grey color Overbank deposit Since loam in the mean Geul valley, at c AD the surface 6 6 Silty or loamy organic In the floodplain, with a - Organic infilling of Holocene deposits, often containing limited width cut off channel decomposed peat 7 7 Heterogeneous mixture (contemporary) levees Often a distinctive grey Levee deposit Holocene of silty clay loam to sand; color similar to unit 4 at cutbanks it shows a structure of small layers with differing texture 9 9 Heterogenous silty clay Footslopes; grades laterally Contains sometimes Colluvial deposit Holocene loam to loam into unit3 bricks and charcoals profile pit in the Lienne catchment shows that this unit is locally at least three meter thick, while locally surfacing bedrock in the river bed indicates that it can also be just a few cm thick. Unit 2 is situated directly on top of this gravel layer, and consists of a textural fining up from sandy deposits to silty clay loam. The thickness varies, and it grades into the overlying unit 3 from which it can be hardly separated. It contains often organic material like twigs, wood or nuts and also some rounded gravels, with a diameter varying between 2 and 20 cm. This unit is interpreted as deposited on bars. The limited thickness of these point bars is explained by the limited depth of the river channel and the sheet like nature of point bars. The plan view position of contemporary deposits of this unit, in Table 7. Average depth of different marker horizons (m), standard deviation (m) and number of used corings (n) for the two study sites in the Geul catchment. Plombières Cottessen Bottom of Holocene 2.36±0.43 (n = 8) 2.70±0.25 (n = 4) deposition (m) Bottom of increased Pb 0.41±0.09 (n = 6) 0.45±0.09 (n = 3) pollution (~1842 AD) (m) Peak in Pb pollution 0.24±0.13 (n = 5) 0.18±0.05 (n = 4) (~1869 AD) (m) the meander inner bends, provides additional indications for its genesis. Unit 3, situated on top of unit 2, consist of rather homogenous silty loam to silty clay loam. It contains sometimes charcoals and a few gravels with a diameter of 2 to 20 cm. At most locations, the combination of units 2 and 3 cover unit 1 over the entire floodplain. This unit is interpreted as overbank fines, but the lower part is probably deposited on bars like unit 2. Unit 4 consists of small (often <15 cm thick) layers of sands, sandy loams and gravels, and can sometimes be found within units 2 or 3. These deposits are interpreted as flood deposits. Unit 5 consist of poor sorted sands, sometimes mixed with gravels. This unit occurs in the steep reaches and makes up the entire non-gravel fraction of the floodplain sediments. This unit is interpreted as a bank deposit. The combined thickness of units 2 to 5 (the fine deposits) ranges from c. 0.2 m to more than 1.5 m, and is between 0.6 and 1.2 m for most locations. Unit 6a consists of peat and silty deposits with a high organic matter content and decomposed plant remains. This unit occurs in distal parts of the floodplain at a few locations in the upper reaches, covering unit 1 or a layer of unit 2, and is covered by some centimetres of unit 3. The thickness of this unit ranges from some centimetres to c. 0.4 m. Given its position and nature, it is being interpreted as a back swamp facies of the distal part of the floodplain. Unit 6b consists, like unit 6a, of 42 Netherlands Journal of Geosciences Geologie en Mijnbouw

33 Table 8. Average fraction of sediments deposited during different time periods compared to the total Holocene deposition and sedimentation rate for the two study sites in the Geul catchment. For the sedimentation rate prior to 1842 AD, different sedimentation periods (with a start varying between 9000 BC and 6000 BC) are used, in order to incorporate the uncertainty on the start of the deposition. Fraction (%) of Sedimentation Holocene deposits rate (mm/a) Time period Plombières Cottessen Plombières Cottessen 9000 BC AD BC AD BC AD BC 1842 AD AD AD AD - present Fig. 10. Pb concentrations in function of depth for coring Pb04 at Plombières and cot41 at Cottessen (Geul catchment). The interpreted depths of the deposits of ~1842AD (A) and ~1869AD (B) are indicated, as well as the in field observed lower border of the dark-grey upper layer (C). Both corings are located in the floodplain, not on levees. peaty deposits and silty deposits with a high organic matter content, but has a different position. This unit is only encoun - tered at a few locations, and is found in the middle of the floodplain, surrounded by places where units 2 and 3 make up the entire thickness of the Holocene deposits. In the lower reaches this unit is always associated with depressions related with former channels (Notebaert et al., 2009b). This unit is positioned above unit 1 or 2 and is covered by unit 3, and has a limited lateral extend (maximal about the width of the current channel). Given its texture and position in the floodplain, unit 6b is interpreted as an organic infilling of abandoned (cut off) channels. Fig. 11. Floodplain cross section in the Amblève catchment at Bullingen (cross section Bullingen 4; Warche River), Fonds de Quarreux (Amblève River) and the village Warche (cross section Warche 3; Amblève River). Vertical exaggeration 10 times for the cross sections at Bullingen and Fonds de Quarreux, and 30 times for the cross section at Warche. An explanation of the different units can be found in the text and Table 9. The location of the different cross sections is indicated on Figure 4. Netherlands Journal of Geosciences Geologie en Mijnbouw

34 Table 9. Different facies units of the Amblève catchment. Unit Texture Position Interpreted deposition environment Age 1 Poor sorted gravel Base of floodplain deposits, in the channel and Channel and bar deposits; reworked Pleistocene on the bars; over the entire width of the valley Pleistocene material and Holocene 2 Textural fining up from sands to On top of unit 1; often over the entire width (Point) bar Holocene silty clay loam. Bottom sometimes of the valley contains twigs, wood and nuts. 3 Loam to silty clay loam Floodplain surface, on top of unit 2; often over Overbank fines; lower part possibly Holocene the entire width of the valley on point bars 4 Small layers of sand or fine gravel Floodplain, within units 2 and/or 3 Flood deposits Holocene 5 Poor sorted sands, sometimes Floodplain of the steep reaches River banks and overbank Holocene with gravels 6a Silt with high organic matter Distal parts of the floodplains of the upper Distal parts of the floodplain with Holocene content and peat reaches; covers unit 1 or 2 and is covered peatland by unit 3 6b In the middle of the floodplain of the lower Organic cut-off channel infilling Holocene and upper reaches, often associated with former channels; above unit 2, covered by unit 3 7 Silty clay loam to loamy sand Lower terrace level in the lower reaches Terrace deposit?? with >5% gravel 8 Silt to loam, >10% gravel Colluvial footslope; grades laterally into Colluvial deposits Holocene units 2 and 3 Unit 7 consist of heterogeneous silty clay loam to loamy sand, with a domination of silty clay loam, and contains >10%, but often >30%, large, well rounded gravels (longest side >3 cm). This unit is positioned above unit 1 at a lower terrace level of unknown age in the lower reaches (see Fig. 11, Warche 3 cross section). The upper part of this unit is more fine grained, and sometimes a thin (<0.4 m) cover of unit 3 can be distinguished. The thickness of unit 7 varies from 0.1 m to 0.6 m. The terrace level is situated 0.2 to 0.8 m above the rest of the floodplain, while the top of unit 1 is here also elevated distinctively higher (0.2 to 1 m) than in the surrounding floodplain. Due to the low elevation of this terrace, it is still flooded occasionally and recent floodplain sediments (unit 3) are still being deposited. This unit was deposited before the formation of the current floodplain, but given the contemporary flooding, the upper parts may be deposited more recently. Unit 8 has a silty to loamy texture, and contains often >10% often angular stone fragments and also often some charcoals. It is always encountered at the footslope, and is interpreted as a colluvial deposit. It grades often in the fine floodplain deposits of units 2 and 3. Dating results Radiocarbon datable material within this catchment was mainly found at the contact of the basal gravel layer and the finer deposits, or within the sandy deposits just above the gravels, indicating that these wood and plant remains were deposited on a point bar. Dating these deposits provides information on past positions and lateral movement of the channel, if it is assumed that the remains were not reworked. The presence of iron slag proved to be the most useful dating method to yield net sediment accumulation rates. This technique was applied in the Lienne subcatchment (Table 10). In total 6 sites were examined and sampled, while data from a site on the Chavanne tributary are available from Houbrechts & Petit (2004; pers. comm.). Metal slag concentrations in the coarsest sand textural class were plotted for the different cross sections (Fig. 12). A slag concentration of 2% was used to differentiate the sediments that were deposited before or after the initiation of metal industries, in this way taking into account bioturbations and minor contaminations during coring. Results show that a disproportional part of net floodplain accumulation occurred during the last years. In the Chavanne floodplain (Fig. 4), which is located in the upper reaches, about 50% of net sediment accumulation occurred after 1537 (Houbrechts & Petit, 2004, pers. comm.). One site was sampled on a small tributary near Monty, and here only the upper samples contain slag. Due to the thickness (0.2 m) of these upper samples, only a maximum estimate can be made: 40% or less of the net sediment accumulation occurred since In the lower Lienne floodplain, 5 sites were studied (Fig. 4). The coring site near Rahier (Fig. 4) is situated at a former scoria dumping site associated with a local blast furnace, and is considered to be not representative for natural floodplain accumulation. The other studied sites show a net floodplain 44 Netherlands Journal of Geosciences Geologie en Mijnbouw

35 Table 10. Results for the different cross sections in the Lienne catchment. The Rahier site is located at a historical scoria dump site and is not included in further analysis. Data from the Chavanne come from Houbrechts and Petit (2004). Coring site Initiation of metal industry Mean thickness of Net floodplain accumulation since Part of the floodplain reworked upstream the site (year AD) floodplain fines (m) initiation of metal industries since initiation of metal industries (m) (%) (%) Targnon Chession Rahier Neucy Trou de Bra Monty ~ Chavanne accumulation of 17 to 51% during the last years, a period comprising ~5% of the total Holocene period. Results also show the importance of lateral reworking of the floodplain by the river: between 14% and 50% of the floodplain has been reworked in the last years, as indicated by the presence of iron slag in point bar deposits. Discussion Dijle catchment The fluvial architecture of the Dijle catchment demonstrates that vertical floodplain aggradation is the main Holocene floodplain process in this catchment, while lateral reworking of the floodplain by the river affects only a limited part of the floodplain. Dating of the parts of the floodplain not affected by such lateral reworking provides a tool for getting insight in the Holocene sediment dynamics and especially in the net sediment accumulation in the floodplains. The high synchronicity between cultural phases and floodplain aggradation indicate that these sediment dynamics were, during the Holocene, mainly driven by land use changes (Notebaert et al., 2011a). The fluvial architecture of the Dijle catchment indicates several changes in the fluvial system, both before and during the Holocene. First, a braided system occurred during the Weichselian, although the end of this phase is not well dated. Downstream of Leuven, incision of large meanders is reported from the Younger Dryas (De Smedt, 1973), and although such incision also occurred upstream Leuven, they are not yet well dated. During the early Holocene, vertical organic and calciumrich accumulation dominated the floodplain. Water discharge was probably diffuse (De Smedt, 1973). This system changed radically with the introduction of agriculture, and silty to clayey floodplain deposition started, combined with the formation of river channels which formed sandy facies. Deposition of these clastic sediments occurred during two phases: between Fig. 12. Cross section of the Lienne floodplain at Neucy with indication of the metal slag concentrations. The metal industry upstream this cross section were probably started in 1421 AD or slightly earlier. A part of the floodplain that was laterally reworked since 1421 AD. Several depressions which are remnants of former river channels are still present; B part of the floodplain with net sediment accumulation (20-30% of total net Holocene (non-gravel) deposition) since 1421 AD; C part of the floodplain with colluvial deposition above overbank deposits which lack scoria. D former river channel, partially infilled before and partially infilled after 1421 AD. The radiocarbon age of the bottom of the channel does not necessarily correspond with the latest occupation of this channel. Netherlands Journal of Geosciences Geologie en Mijnbouw

36 c BC and 1000 AD deposition rates were relatively low, but under increasing land use pressure these rates increased after 1000 AD, resulting in an important deposition phase. The first formation of textural clearly identifiable levees occurs during this last phase, and also downstream levees are reported to start forming during this period (De Smedt, 1973). This implies the change from a flat floodplain type to a convex floodplain, resulting from the availability of sand outside the river channel. This is related to an increase in flood discharge and/or an increase in sediment supply, two processes which can be related to an increased anthropogenic land use. Geul catchment The fluvial architecture of the Geul catchment indicates the importance of lateral reworking of the floodplain by the river. The presence of bed and bar deposits over the full width of the floodplain indicates that the entire floodplain was reworked at least once over the course of the Holocene. This is confirmed by dates from these bar deposits (De Moor et al., 2008). Modelling results from a meander model suggest a floodplain reworking time of a couple hundred years (De Moor, 2006). As a consequence, only the upper parts of the Holocene deposits, the overbank fines, are suited for dating the sedimentation history of the catchment. In addition, information from the early Holocene will be missing, as they are eroded during the lateral reworking. The fluvial architecture of the Geul catchment demonstrates only one distinct change in the fluvial style of this river system, from a Pleistocene braided river to a straight or meandering river in the Holocene with a river channel and a floodplain. It can however not be excluded that channel contraction occurred during the Holocene and that a multiple channel system was transformed to a single channel system (see also De Moor et al., 2008). Due to lateral reworking of the floodplain by the meandering river, information on the past fluvial styles is not well preserved. In this study, the interpretation of the floodplain sedimen - tation history of the Geul catchment suffers from an averaging effect over time: the use of lead as a tracer allows the recon - struction of sedimentation rates of three different timeframes with contrasting lengths: early Holocene to c AD, 1842 AD to 1869 AD and 1869 AD to 2008 AD. It is clear that a dispropor - tional part of sedimentation (~17%) took place after 1842 AD, which equals c. 1.5% of the total Holocene timeframe. The presence of a dark grey upper soil layer was observed during sampling, and lab analysis proved that this layer coincides with the peak lead contamination situated after c AD. Because of the coarse temporal resolution, it cannot be excluded that phases with a comparable high sedimentation rate also occurred earlier during the Holocene, as only an average value of a very long time interval is available. The results of this study are comparable with results of exposed cut banks of the Geul River dated with lead contami - nation and cosmogenic tracers (Stam 1999, 2002), which also indicate an increased sedimentation during the mining period, followed by a sharp decline in sedimentation and again higher sedimentation in the 20 th century. The higher sedimentation rates in the 19 th century can partially be explained by the mining activities, because large amounts of soil and rocks originating from mining were deposited in the alluvial plain. Information is also provided from the height of point bars, which is reported to increase over the Holocene, associated with an aggrading floodplain which related to land use changes (De Moor et al., 2008). In addition, the sedimentation history of alluvial fans indicates the causal relationship between land use changes and alluvial fan sedimentation (De Moor et al., 2008). These different observations suggest the important influence of land use changes on the sedimentation history, while the influence of the climate remains unclear, because of the limited temporal resolution of the records (Notebaert et al., 2011a, 2011b). Amblève catchment Comparable to the Geul catchment, the fluvial architecture of the Amblève catchment indicates that at most sites the entire width of the floodplain was reworked by the river during the Holocene, and a sequence that spans the full Holocene is missing. The methodology used in this study provides a solution, as it allows the identification of zones which are affected by lateral reworking and the identification of the net sediment accumulation over the last c. 600 years. The fluvial architecture of the Amblève catchment demon - strates one important change in fluvial style: from a braided river system in the Pleistocene to a single thread meandering or straight channel in the Holocene, in the lower reaches sometimes with islands. For the lower reaches depressions can be observed at many locations, which are linked to former cut off channels. For most locations of these lower reaches, at least one former cut off channel is present, often near one valley edge while the river is located at the other one. Locally, several former cut off channels resemble a braided pattern (e.g. fig. 10 in Notebaert et al., 2009a). There are however no dates available to prove the co-existence of several channels. The averaging of sedimentation rates over timeframes with totally different lengths form a major problem in calculating the sedimentation rates for the Amblève catchment. These time frames are, however, less contrasting than for the Geul catchment, and provide therefore a better interpretation framework. Reported incision rates for rivers in the Ardennes during the Late Pleistocene and Holocene are in the order of m/ma (e.g. Van Balen et al., 2000), which corresponds with c m for the entire Holocene. As a consequence, the c. 0.2 to 1 m high lower terrace, formed by units 1 and 7, 46 Netherlands Journal of Geosciences Geologie en Mijnbouw

37 dates most likely from before the Holocene. There are no indications for the studied catchment, nor reports for other catchments in the Ardennes, of Holocene floodplain incision and terrace formation. We hypothesize that the sediments found in the contemporary floodplain combined with a thin layer of fine sediments deposited above these lower terraces (unit 7), represent the total Holocene floodplain sediment deposit. The dating results show that a disproportional large part of the total net floodplain sediment accumulation (17-69%, c. 40% on average) was deposited during the last 400 to 600 year, a period which equals c. 5% of the entire Holocene. The limitations of the applied dating methodology put some constraints on the identification of the environmental parameters which influence the sediment dynamics. The increased sedimentation rate of the last 400 to 600 a can be explained by the first major deforestation and anthropogenic land use which occurred for the first time during this period. But climatic events, like the little ice age, can also have influenced sedimentation rates through changes in precipitation patterns. The dating resolution does not allow identifying the influence of such climatic events or the interplay between climate and anthropogenic factors. The large difference in the relative importance of recent sedimentation between sites (17-69%) can only be explained by the importance of local factors controlling local sediment deposition. Possibly the position of anthropogenic (hydro)engineering structures (e.g. mills, milldams, bridges,...) in the floodplain has an influence, while also topographic variations in the floodplain may have an influence. Fluvial architecture and dating methods The fluvial architecture of the three studied catchments shows great differences, with a dominance of vertical aggradation in the Dijle catchment and a dominance of lateral reworking in the Amblève and Geul catchments. These differences have important implication for the use of dating techniques to identify the dynamics of sediment accumulation on the floodplains. The used dating methods fall apart in two main categories, based on the spatial information they provide: discrete dating methods like radiocarbon and OSL dating, and continuous dating methods based on the presence of a tracer. Where discrete dating methods provide an age control for discrete points in a core, continuous dating methods allow the reconstruction of palaeosurfaces and provide information on preservation of past deposits. When continuous vertical aggradation profiles are present, like in the Dijle catchment, reliable sedimentation rates can be expected on any core which does not contain point bar or river bed deposits, or other indications for a hiatus or incision phase. Discrete dating methods like AMS radiocarbon dating and OSL dating provide age information on individual cores and allow reconstructing a site specific sedimentation history (see Notebaert et al., 2011a). When it is assumed that the dated core(s) are representative for the entire floodplain, such discrete dates provide information on floodplain aggradation. In this case, the fluvial architecture of the floodplain makes an important contribution as it allows identifying cores where the fluvial archive has not been influenced by erosion phases. When the fluvial architecture is dominated by lateral accretion and the river valley was (almost) entirely reworked by the river channel during the Holocene, other dating methods are required, as for example in the Geul and Amblève catchment. As the transition between the lower point bar deposits and the upper floodplain deposits is often hard to identify, the inter - pretation of discrete dates is difficult, as it remains unclear which process is dated. As a result it is often uncertain whether the dating results provides information on lateral accretion or vertical aggradation, and such discrete data rather provide core specific information than data on overall floodplain aggradation. Core specific data may be influenced by local point bar formation. Using a spatially continuous tracer, like metal slag or the presence of lead pollution, allows the reconstruction of past surfaces. As such, two sedimentary bodies are identified, those deposited before and after introduction of the tracer. The combination of the dating information with the fluvial architecture allows identifying which parts of the floodplain are laterally reworked since the introduction of the tracer. For the parts which are not laterally reworked, the vertical aggradation can be assessed. Using tracers as a dating method provides more or less continues information over space, but only a limited number of periods, two and three respectively in the studied catchments, can be differentiated. Using radiocarbon or optical dating can result in different ages for the same coring and, depending on the availability of data, in more or less continuous data over time. In order to reconstruct past surfaces like with continuous dating methods, discrete data from more or less the same age should be available for each coring. This would require a much extended dataset or is even impossible due to the limited availability of datable material. Therefore it is important to get insight in the fluvial architecture to optimise the use of dating techniques. The value of sedimentary archives to study the influence of environmental changes on the (fluvial) sediment dynamics depends largely on the fluvial architecture of the catchment. With a continuous aggrading system where lateral reworking is limited to parts of the floodplain, a sedimentation history can be constructed and linked to past land use and climatic changes. When the fluvial architecture is dominated by lateral reworking, like in the Geul and Amblève catchments, parts of the fluvial archive are missing due to erosion, and there is a low preservation potential (Lewin & Macklin, 2003). The floodplain sedimentation history can only be studied for the overbank deposits that are present and make up the upper part of the floodplain, creating a potential bias towards the most recent sedimentation period. The results of the Dijle catchment allow a more detailed correlation with the past environmental Netherlands Journal of Geosciences Geologie en Mijnbouw

38 changes than the results of the Amblève and Geul catchment which are less detailed and have a larger uncertainty in the correlation with environmental changes. When dating flood - plain deposition in a lateral reworking river system without understanding the fluvial architecture, incorrect temporal variations of the sedimentation rate will be concluded. For any single core or spatial point in the floodplain, the point bar deposits are deposited during a short time period, leading to an overestimation of the sedimentation rates for this period. Figure 13 provides a conceptual model for a floodplain where lateral movement of the channel is dominant, based on the Neucy site of the Lienne River (Fig. 12) but applicable on any studied site in the Geul and Amblève catchments. When dating radiocarbon datable material originating from the point bar deposits, and not taking into account the fluvial archi - tecture (Fig. 13B), the sedimentation rate for the period after the deposition of the dated material is overestimated, while the rate for the preceding period is underestimated. Additional problems arise due to the increased possibility for datable material in point bars to have an age which does not correspond with the sedimentation moment, due to the accumulation of fluvial transported material on pointbars. When the fluvial architecture is taken into account when dating (Fig. 13C), the average sedimentation rate of the period after introduction of the tracer is well established. But the sedimentation rate before the introduction is suffering from an unknown start of the sedimentation. This paper it is supposed that the sedimen - tation started at the beginning of the Holocene. In addition, as the tracers are introduced rather late in the Holocene, the calculated sedimentation rate represents an average value for a very long period. It is possible that over such a long timescale, comparable sedimentation rates as the post tracer introduction rate occurred. same model results show that the influence of climatic variations during the Holocene is very low compared to land use changes. When comparing the total masses of Holocene sediment deposition between the three studied catchments, largest area specific quantities are present in the Dijle catchment (0.40 Tg km 2 = g km 2 ), followed by the Gulp catch - ment (0.10 Tg km 2 ) and the Amblève catchment (0.03 Tg km 2 ) (Notebaert et al., 2010). The values for the Gulp catchment are in accordance with previously published data for the entire Geul catchment (c Tg km 2 ) (De Moor & Verstraeten, 2008). The differences between the three catchments can be explained The influence of environmental changes The dating results of three studied catchments show that periods with increased floodplain sedimentation coincide with periods with increased land use changes, which suggests a relationship between both. The dating results do not allow identifying an influence of climatic events. Due to synchronous variations in land use and climate the individual effect of both parameters is often difficult to assess, also because of the possible occurrence of a lag time between variations in environmental setting and the fluvial response (e.g. Vandenberghe, 1995). The construction of a sediment budget incorporating the different sinks and sources of sediment in a catchment may allow to further establish the link between environmental changes and sedimentary response (e.g. Trimble, 1999; Notebaert et al., 2009b, 2011a). Modelling results from the Dijle catchment show indeed the important influence of land use changes on soil erosion and colluviale and alluvial sediment deposition (Notebaert et al., 2011b). The Fig. 13. Conceptual model of the importance of fluvial architecture for dating floodplain aggradation. This conceptual model is based on the Neucy site (Fig. 12) but is applicable for the studied sites in the Amblève and Geul floodplains. A temporal development of the floodplain through the Holocene, with three time frames: A1 hypothetical surface before the introduction of the tracer; A2 surface at the time of the introduction of the tracer; A3 contemporary surface with indication of the sediment containing the tracer (red). B and C hypothetical corings in the floodplain, with an indication of the different architectural units. For each coring a depth/age and a time/sedimentation rate curve are plotted. Coring B is dated using a radiocarbon age from the lower parts of the floodplain deposits, and the resulting sedimentation rates are not in accordance with net floodplain aggradation. Coring C is dated using the tracer horizon and taking into account the fluvial architecture. The resulting sedimentation rate after the introduction of the tracer provides an average floodplain aggradation rate, while the rate for the period before the introduction is influenced by the unknown start of the sedimentation. 48 Netherlands Journal of Geosciences Geologie en Mijnbouw

39 by differences in land use history and connectivity between slopes and floodplains (Notebaert et al., 2010), but the fluvial architecture may possibly also have influenced floodplain storage. The floodplain processes which become clear from the fluvial architecture may influence the possibility for a flood - plain to store sediment, especially when aggradation rates differ between the different depositional environments. In the Dijle river, the river bed and the flood plain have aggraded, although no data are available to estimate the evolution of the channel depth over a long timescale. For the Geul and Amblève river, there are no traces of a change of the absolute height of the river bed over the Holocene. With an aggrading floodplain and a constant absolute river bed height, the river channel increases in absolute depth, greater flows will remain restricted to the channel and stream power increases (see e.g. Trimble, 2009). As a result floodplain sedimentation slows down as only larger events will cause overbank flooding. Similar processes in the UK are described by Brown & Keough (1992) as the stablebed aggrading-banks model (SBAB). This may particularly be true for the Geul catchment, as river banks in the Amblève catchment still have a limited height. Conclusions In this study the fluvial architecture was studied for three catchments in Belgium, the Dijle, Geul and Amblève catchment, and this was combined with different dating methods in order to derive the Holocene fluvial sedimentation history. The fluvial architecture of the Dijle catchment is dominated by vertical aggradation, which allows dating aggradation profiles of the entire Holocene using radiocarbon or optical dating. In the Amblève and Geul catchment, lateral reworking dominates, and vertical aggradation deposits from the early Holocene are eroded. The upper parts of the floodplain contain vertical aggradation overbank deposits, which were dated using tracers. In the Geul catchment Pb contamination resulting from 19 th century mining activities was used, while in the Amblève catchment contami - na tion with metal slag from medieval metal industries was used. These dating methods allow the identification of postcontamination vertical aggradation and of post-contamination lateral reworking deposits. As such, only two (or three) discrete periods can be identified, but the spatial variation is more easily identified. Linking environmental changes with variations in floodplain deposition is most straightforward for the Dijle catchment, due to a denser temporal resolution. Establishing such links is hampered by the limited temporal resolution for the other two catchments. Nevertheless, the sedimentation history of all three catchments indicates a major influence of anthropogenic land use changes, which caused an increase in floodplain deposition. Acknowledgements This research is part of a project funded by the Fund for Scientific Research Flanders (research project G ). Their support is gratefully acknowledged. References Bravard, J.P. & Peiry, J.L., The CM pattern as a tool for the classification of alluvial suites and floodplains along the river continuum. Floodplains: interdisciplinary approaches: Bravard, J.-P., Burnouf, J. & Verot, A., 1989, Géomorphologie et archeology dans la region lyonnaise: Questions et réponses d un dialogue interdisciplinaire. Bulletin de la Société Préhistorique Française 10-12: Bronk Ramsey, C., Development of the radiocarbon calibration program OxCal. Radiocarbon 43: Bronk Ramsey, C., Bayesian analysis of radiocarbon dates. Radiocarbon 51: Broothaerts, N., 2008, Geomorfologische opbouw van de Dijle vallei. Bachelor thesis, KU Leuven, 62 pp. Brown, A.G. & Keough, M., Holocene floodplain metamorphosis in the Midlands, United Kingdom. Geomorphology 4 (6): Brown, A., Petit, F. & James, A., Archaeology and Human Artefacts. In: Kondolf, M., Piégay, H. (eds): Tools in Fluvial Geomorphology, Wiley, New York: Damblon, F., Etude palynologique comparée de deux tourbières du plateau des Hautes Fagnes de Belgique: la Fagne Wallonne et la Fagne Clefay. Bulletin du Jardin botanique national de Belgique / Bulletin van de National Plantentuin van België 39: Damblon, F., Etudes paléo-écologiques de tourbières en haute ardenne. Ministère de l agriculture, Administration des eaux et forêts. Service de la conservation de la Nature. Traveaux, No. 10. De Moor, J., Human impact on Holocene catchment development and fluvial processes the Geul River catchment, SE Netherlands. PhD thesis, VU Amsterdam: 142 pp. De Moor, J. & Verstraeten, G., Alluvial and colluvial sediment storage in the Geul River catchment (the Netherlands) Combining field and modelling data to construct a Late Holocene sediment budget. Geomorphology 95: doi: /j.geomorph De Moor, J.J.W., Kasse, C., Van Balen, R., Vandenberghe, J. & Wallinga, J., Human and climate impact on catchment development during the Holocene Geul River, the Netherlands. Geomorphology 98 (3-4): De Smedt, P., Paleogeografie en kwartair-geologie van het confluentie - gebied Dijle-Demer. Acta Geographica Lovaniensia 11, 141 pp. Dejonghe, L., Ladeuze, F. & Jans, D. et al., Atlas des gisements plombozincifères du Synclinorium de Verviers (Est de la Belgique). Mém. Explic. Cartes Géol. Min. Belgique 33: Dotterweich, M., The history of soil erosion and fluvial deposits in small catchments of central Europe: Deciphering the long-term interaction between humans and the environment A review. Geomorphology 101: doi: /j.geomorph Geurts, M.-A., Genèse et stratigraphie des travertins de fond de vallée en Belgique. Acta Geographica Lovaniensia 16. Netherlands Journal of Geosciences Geologie en Mijnbouw

40 Gullentops, F., Mullenders, W., Schaillee, L., Gilot, E. & Bastin-Servais, Y., Observations géologiques et palynologiques dans la vallée de la Lienne. Acta Geographica Lovaniensia 4: Henrottay, La sédimentation de quelques rivières belges au cours des sept derniers siècles. Bulletin de la Sociète Géographique de Liège 9: Hoffmann, T., Lang, A. & Dikau, R., Holocene river activity: analysing 14Cdated fluvial and colluvial sediments from Germany. Quaternary Science Reviews, 27: doi: /j.quascirev Houben, P., Geomorphological facies reconstruction of Late Quaternary alluvia by the application of fluvial architecture concepts. Geomorphology 86: doi: /j.geomorph Houbrechts G., Utilisation des macroscories et des microscories en dynamique fluviale: application aux rivières du massif ardennais. PhD thesis, University of Liège, 328 pp. Houbrechts, G. & Petit, F., Utilisation des scories métallurgiques en dynamique fluviale: détermination de la compétence effective des rivières et estimation des vitesses de progression de leur charge de fond. Géomorphologie: relief, processus, environnement 2003 No. 1: Houbrechts, G. & Petit, F., Etude de la dynamique fluviale des rivières ardennaises grâce aux scories métallurgiques. De la Meuse à l Ardenne 36: Houbrechts, G. & Weber, J.-P., La sidérurgie proto-industrielle dans le bassin de la Lienne. De la Meuse à l Ardenne 39: Lewin, J. & Macklin, M., Preservation potential for Late Quaternary river alluvium. Journal of Quaternary Science 18: Doi: /jqs.738 Macklin, M., Jones, A. & Lewin, J., River response to rapid Holocene environ - mental change: evidence and explanation in British catchments. Quaternary Science Reviews 29 (13-14): doi: /j.quascirev Miall, A.D., Architectural-element analysis a new method of facies analysis applied to fluvial deposits. Earth-Science Reviews 22: doi: / (85) Mols, J., Dynamique fluviale en réponse aux changements d affectation du sol des bassins versant de l Euregio Meuse-Rhin. ULg-LHGF, Mémoire de DEA, 54 pp. Mullenders, W. & Gullentops, F., Palynologisch en geologisch onderzoek in de alluviale vlakte van de Dijle te Heverlee-Leuven. Agricultura Band V 2 e reeks(1): Mullenders, W., Gullentops, F., Lorent, J., Coremans, M. & Gilot, E., Le Remblement de la vallée de la Nethen. Acta Geographica Lovaniensia IV: Nanson, G. & Croke, J., A genetic classification of floodplains. Geomorphology 4: doi: / x(92)90039-q Notebaert, B., Verstraeten, G., Govers, G. & Poesen, J., 2009a. Qualitative and quantitative applications of LiDAR imagery in fluvial geomorphology. Earth Surface Processes and Landforms 34 (2): Doi: /esp.1705 Notebaert, B., Verstraeten, G., Rommens, T., Vanmontfort, B., Govers, G. & Poesen, J., 2009b. Establishing a Holocene sediment budget for the river Dijle. Catena 77 (2): doi: /j.catena Notebaert; B., Verstraeten, G., Govers, G. & Poesen, J., Quantification of alluvial sediment storage in contrasting environments: methodology and error estimation. Catena 82: Notebaert, B., Verstraeten, G., Vandenberghe, D., Marinova, E., Poesen, J. & Govers, G., 2011a. Changing hillslope and fluvial Holocene sediment dynamics in a Belgian loess catchment. Journal of Quaternary Science 26 (1): Notebaert, B., Verstraeten, G., Ward, P., Renssen, H. & Van Rompaey, A., 2011b. Modeling the sensitivity of sediment and water runoff dynamics to Holocene climate and land use changes at the catchment scale. Geomorphology 126: Passega, R., Texture as Characteristic of Clastic Deposition. AAPG Bulletin 41. Passega, R., Grain size representation by CM patterns as a geologic tool. Journal of Sedimentary Research 34 (4): Reimer, P.J., Baillie, M.G.L., Bard, E., Bayliss, A., Beck, J.W., Bertrand, C.J.H., Blackwell, P.G., Buck, C.E., Burr, G.S., Cutler, K.B., Damon, P.E., Edwards, R.L., Fairbanks, R.G., Friedrich, M., Guilderson, T.P., Hogg, A.G., Hughen, K.A., Kromer, B., McCormac, G., Manning, S., Ramsey, C.B., Reimer, R.W., Remmele, S., Southon, J.R., Stuiver, M., Talamo, S., Taylor, F.W., van der Plicht, J. & Weyhenmeyer, C.E., IntCal04 terrestrial radiocarbon age calibration, 0-26 ka BP. Radiocarbon 46: Rommens, T., Holocene Sediment Dynamics in a Small River Catchment in Central Belgium, Phd-thesis. K.U. Leuven, Department Geography-Geology, Leuven, Belgium. Rommens, T., Verstraeten, G., Bogman, P., Peeters, I., Poesen, J., Govers, G., Van Rompaey, A. & Lang, A., Holocene alluvial sediment storage in a small river catchment in the loess area of central Belgium. Geomorphology 77 (1-2): doi: /j.geomorph Stam, M.H., The dating of fluvial deposits with heavy metals, 210Pb and 137Cs in the Geul catchment (the Netherlands). Physics and Chemistry of the Earth. Part B: Hydrology, Oceans and Atmosphere 24: doi: / S (98) Stam, M.H., Effects of land-use and precipitation changes on floodplain sedimentation in the nineteenth and twentieth centuries (Geul River, the Netherlands). Special Publications of the International Association of Sedimentologists 32: Trimble, S.W., Decreased rates of alluvial sediment storage in the Coon Creek Basin, Wisconsin, Science 285 (5431): Doi: / science Trimble, S.W., Fluvial processes, morphology and sediment budgets in the Coon Creek Basin, WI, USA, Geomorphology 108 (1-2): doi: /j.geomorph Trimble, S.W., Streams, valleys and floodplains in the sediment cascade. In: Burt, T. & Allison, R. (eds): Sediment cascades. An integrated approach. Wiley-Blackwell, Chichester. Van Balen, R., Houtgast, R., Van der Wateren, F., Vandenberghe, J. & Bogaart, P., Sediment budget and tectonic evolution of the Meuse catchment in the Ardennes and the Roer Valley Rift System. Glob. Planet. Change 27: doi: /s (01) Vandenberghe, J., Timescales, climate and river development. Quaternary Science Reviews 14 (6): pp Vanwalleghem, T., Bork, H. R., Poesen, J., Dotterweich, M., Schmidtchen, G., Deckers, J., Scheers, S. & Martens, M., Prehistoric and Roman gullying in the European loess belt: a case study from central Belgium. Holocene 16(3): doi: / hl935rp Verstraeten, G., Rommens, T., Peeters, I., Poesen, J., Govers, G. & Lang, A., A temporarily changing Holocene sediment budget for a loess-covered catchment (central Belgium). Geomorphology 108: doi: / j.geomorph Netherlands Journal of Geosciences Geologie en Mijnbouw

41 Geomorphology 126 (2011) Contents lists available at ScienceDirect Geomorphology journal homepage: www. elsevie r. com/ locate/ geomor ph Modeling the sensitivity of sediment and water runoff dynamics to Holocene climate and land use changes at the catchment scale Bastiaan Notebaert a,, Gert Verstraeten a, Philip Ward b, Hans Renssen b, Anton Van Rompaey a a Department Earth & Environmental Sciences, K.U. Leuven, Belgium b Department of Earth Sciences, Vrije Universiteit, Amsterdam, The Netherlands a r t i c l e i n f o abstract Article history: Received 5 March 2010 Received in revised form 30 July 2010 Accepted 4 August 2010 Available online 17 September 2010 Keywords: Sediment dynamics Climate change Land use change Water runoff Modeling Belgium An increasing number of studies have indicated that soil erosion, sediment redistribution and water discharge during the Holocene have varied greatly under influence of environmental changes. In this paper we have used a modeling approach to study the driving forces for soil erosion and sediment redistribution on the one hand, and water discharge on the other hand, during the Holocene for the Belgian Dijle catchment. Soil erosion and sediment redistribution was modeled using the spatially distributed Watem/Sedem model. Several scenarios of past land use and climate were used to model soil erosion, colluvial sedimentation and sediment export to the fluvial system, and those scenarios were combined with a sensitivity analysis. Modeling results are validated based on an available time differentiated field-based sediment budget. Water discharge was modeled using the spatially distributed STREAM model with a daily time step. The model was calibrated based on contemporary discharge data. The results indicate that soil erosion has increased between the early Holocene and the current situation by ca. 9% due to climatic variations, and by ca. 6000% due to changes in land use. The average discharge has increased by ca. 7%, mainly due to land use changes, and the discharge also shows more peaks. The decrease in cropland area and increase in built up area since 1775 CE has resulted in a decrease in soil erosion and in a further increase in discharge, showing the difference in sensitivity of both processes to land use changes Published by Elsevier B.V. 1. Introduction Soil erosion and sediment redistribution are important processes on a Holocene time scale in many temperate regions, moulding the landscape and providing an important link between the hillslopes and the fluvial system. Several studies have demonstrated the important influence of environmental variations on sediment dynamics, especially of land use changes and climatic changes (e.g. Trimble, 1999; Dotterweich, 2008; Verstraeten et al., 2009a). Understanding Holocene sediment dynamics often relies on extended sets of field data. Studies based on these datasets suffer from several limitations, resulting from the spatial and temporal scale of the field data and the involved errors. Identification of driving factors often relies on synchronicity between variations in these driving factors and variations in the sediment dynamics. The use of sediment budgets can give a deeper insight into the sediment pathways, and in the entire soil erosion and sediment redistribution system (e.g. Slaymaker, 2003; Reid and Dunne, 2003). Nevertheless, there remain large uncertainties, due to for example synchronous evolution in driving forces and the lack of high-resolution chronologies of many Corresponding author. Tel.: address: bastiaan.notebaert@ees.kuleuven.be (B. Notebaert). sedimentary archives (Verstraeten et al., 2009b). As a result, it is often impossible to disentangle the respective role of natural and humaninduced environmental change or to understand the impact of complex interactions between climate and land use change on sediment dynamics. This also implies that we should be very cautious when the potential impact of future environmental change on fluvial system behavior is assessed using low-resolution fluvial palaeoenvironmental data. Modeling can address some of these problems and provide a better understanding of the involved processes. When using models, the sensitivity of the system to one single parameter can be studied by varying one of the parameters and keeping the others constant. Several studies have modeled the potential role of land cover and climatic variations on hillslope soil erosion (e.g. Boardman and Favis- Mortlock, 1993; Yang et al., 2003), mainly in the framework of potential future climate changes, but only few studies (e.g. Ward et al., 2009) also incorporate sediment transport over a Holocene period. Most studies on sediment redistribution and sediment yield are based on contemporary measurements (e.g. Milliman and Syvitski, 1992; Syvitski and Milliman, 2007; de Vente et al., 2007) and only a limited number of studies have tried to model soil erosion and sediment redistribution over longer time scales (e.g. de Moor and Verstraeten, 2008; Ward et al., 2009). Several approaches have been used to model past soil erosion and sediment redistribution processes. Landscape X/$ see front matter 2010 Published by Elsevier B.V. doi: /j.geomorph

42 B. Notebaert et al. / Geomorphology 126 (2011) evolution models have been used to model landscape evolution and landscape reconstruction (e.g. Coulthard, 2001; Peeters, 2007), incorporating erosion and deposition processes. Some models focus on soil erosion processes and related sediment redistribution (e.g. WatemLt model; Peeters, 2007) in order to model the influences of varying climate and land cover. Other models (e.g. Caesar model; Coulthard et al., 1997) are based on hydrological modeling and concentrate on processes like mass movement, creep and alluvial erosion, allowing interactions between neighboring cells. Models and model applications vary in their spatial and temporal scale, and as such also in the details to which the processes are studied. Hancock et al. (2010) show that the Caesar model is better in simulating event based fluctuations. When running them on large (10,000 a) timescales, such event based models result in comparable landscapes as those from models based on average rates, although erosion and deposition rates may differ by an order of magnitude (Hancock et al., 2010). Peeters et al. (2008) show that an approach based on average erosion rates produces realistic results over a Holocene timescale. Alternatively to such dynamical models, soil erosion and sediment redistribution have been modeled successfully for several scenarios of past land use and climate, without changing landscape topography (e.g. de Moor and Verstraeten, 2008). Although the topography of many catchments has changed over the Holocene, such an approach has several advantages. First of all it avoids numerical problems with landscape evolution and errors are aggregated over larger spatial units instead of accumulating them over time and over small spatial units like pixels. In addition, the input data which suffer large uncertainties do not have to be provided on a continuous temporal scale. A main disadvantage is the non-dynamical character of the environmental variables, resulting in a steady state approach. Also, threshold changes triggered by changes in land use, climate or topography are not modeled. Nevertheless, such a scenario based approach allows us to understand system dynamics in relation to reconstructed land cover and precipitation data. Apart from soil erosion and sediment redistribution, river discharge dynamics plays an important role in catchments on a Holocene timescale. The combined effect of water and sediment discharge controls the fluvial style of river systems and determines the dimensions of the river channel. Field evidence for past discharge is often lacking, resulting in a lack of discharge information beyond the often relatively short period of measurements. Palaeodischarge modeling provides a tool to address the lack of long-term observed data and sedimentary evidence. In this way, not only can the past discharge be modeled, but also the sensitivity of the discharge to variations in the landscape (both land cover and climate) can be assessed. This is important in the framework of possible future climate variations and associated flooding control (e.g. Ward et al., 2008). The objective of this paper is to analyze the sensitivity of a catchment's sediment delivery and water discharge to climatic variability and land use change using a modeling approach. Furthermore, the impact of intra-holocene changes in climate and land use on fluvial system dynamics for the intermediate-sized Dijle catchment (758 km²), situated in the Belgian loess region, will be quantified. For this catchment, field data revealed important variations in sediment dynamics during the Holocene (e.g. Notebaert et al., 2009), while there are also indications for variations in water discharge (De Smedt, 1973; Notebaert, 2009). The chronology of the sedimentary record suggests that sediment dynamics vary mainly under the influence of land use changes, whilst no straightforward impact of climatic variations could be detected, possibly due to the insufficient temporal resolution (Notebaert et al., in press). 2. Study area In this study, the Dijle catchment (758 km², Fig. 1) upstream from Leuven is considered. The catchment consists of a loess-covered undulating plateau ranging between 80 and 165 m a.s.l., in which river valleys have incised. Based on a 20-m resolution DTM, the catchment's average gradient is 3.61%, while about 5.4% of the landscape has a gradient of at least 10%. Current land use is dominated by cropland on the plateaus and slopes, and grassland and forest in the floodplains. Some large forests have existed on the plateaus since at least the 14th century CE (for example Vanwalleghem et al., 2006), and extended built up areas are situated in the center of the catchment. Archeological, historical and palynological data show a rather intense land use history during the second half of the Holocene. There is plenty of field evidence for current and past soil erosion in the catchment, for example the presence of large gullies and large quantities of Holocene floodplain and colluvial deposits (e.g. Vanwalleghem et al., 2006; Notebaert et al., 2009). First traces of agriculture date from the Atlantic Period (ca BCE; De Smedt, 1973). A first peak in land use intensity occurred during the Roman Period. Land use intensity peaked again from the Middle Ages onwards. A more detailed description of the catchment can be found in Notebaert et al. (2009). The fluvial architecture indicates that vertical floodplain aggradation is the most important Holocene fluvial process (Notebaert et al., 2009). Field data show that sediment deposition was limited during the early Holocene, with stability in the colluvial valleys and peat formation in the floodplains. A first increase in sedimentation rates occurred sometime after ca BCE, depending on location (Notebaert et al., in press). This first deposition phase was more important for the colluvial valleys than for the alluvial valleys. The majority of the sediment was, however, deposited during the last 1000 years (Notebaert et al., in press, 2009). The relative importance of floodplain deposition increased during this phase. A timedifferentiated sediment budget for this catchment (Notebaert et al., submitted) divides the Holocene in three periods: BCE, 2000 BCE 1000 CE and CE. This time-differentiated sediment budget offers an ideal tool to validate sediment dynamics modeling results. 3. Methods Fig. 1. Location of the Dijle catchment in Belgium. When modeling geomorphic processes on a spatial and temporal scale such as that used in this study (N10 2 km² and N10 3 a respectively), the selected model should fulfill some prerequisites. Preference goes to a model with relatively few calibration parameters. Introducing more calibration parameters can increase the uncertainty (Van Rompaey and Govers, 2002). Additionally, the model requires relatively few input data, which is important when modeling past processes: often input parameters are difficult to estimate and the uncertainties are high. As land use is an essential part of the objectives

43 20 B. Notebaert et al. / Geomorphology 126 (2011) of this paper, and the spatial pattern of land use may play an important role, it is necessary to use a spatially distributed approach. Based on these prerequisites, two models were selected: the Watem/ Sedem model for soil erosion and sediment redistribution, and the STREAM (Spatial Tools for River Basins and Environment and Analysis of Management Options) model for discharge. Both models were applied for a number of scenarios for which the past modeled climate data are combined with reconstructed land use, representing past environments. We used the Watem/Sedem model, which is based on the understanding of contemporary processes, for modeling soil erosion and sediment redistribution. This model has been successfully applied in the central Belgian Loess belt to model contemporary erosion and sediment delivery (e.g. Van Rompaey et al., 2001; Verstraeten, 2006). It requires relatively few input and calibration parameters, and employs widely used relationships between soil erosion and catchment properties (relief, land cover, soil properties, and climate). Past soil erosion and sediment redistribution was previously modeled using the Watem/Sedem for the Dutch Geul catchment (240 km²) (de Moor and Verstraeten, 2008) and the Meuse catchment (33,000 km²) (Ward et al., 2009), but a quantitative model validation is still missing as only limited basin wide sediment observations exist for these catchments (Ward, 2008). While the Watem/Sedem can model hillslope processes, sediment transport within the fluvial system is largely ignored. This is mainly controlled by the stream power of the river flow, and thus related to discharge. In order to have a better insight into the Holocene variability in water discharge, we used the STREAM model (Aerts et al., 1999). This model is a spatially distributed discharge model that works with a number of storage compartments. It requires relatively few calibration parameters and input data and has previously been used to model palaeodischarge in several catchments around the world (Aerts et al., 2006; Renssen et al., 2007; Ward et al., 2007, 2008), including the Meuse and Rhine Rivers. Both models were applied to describe the relationships between the system dynamics and past variations in land use and climate. A reconstruction of past land cover and climate is necessary in order to apply the models. The reconstruction of past climate was based on a modeling approach, while land cover was reconstructed using a parameter set based on archeological and other data Reconstruction of land cover and climate The contemporary land cover for the model runs was based on the Corine Land Cover map 2000 (CLC2000, 100 m resolution) of the European Environment Agency ( EEA, Copenhagen 2000, The original 46 land cover categories were regrouped into five categories: urban, water, forest, pasture and arable land. Detailed historical land cover data for the Dijle catchment are available for the period since the publication of the first topographic maps, around The land cover map of 1775 CE was based on the de Ferraris map (1775, reprinted by Gemeentekrediet). From this map, the following land cover classes were digitized: cropland, grassland, forest, orchards, gardens/houses, built-up area, heathland, marshes, and water; and these classes were then again simplified into the same five categories as the CLC2000 map. This paper, however, focuses on the entire Holocene time period, during which land cover variations are far more important than for the last few centuries, as is evidenced for many West- and Central European catchments (e.g. Zolitschka et al., 2003). Thus, a modeling approach was followed in order to simulate land use patterns before 1775 CE, based on the methodology developed by Peeters (2007) for the Nethen, a subcatchment of the Dijle catchment. This methodology is a three-step procedure: first, population density and agricultural productivity per time period were estimated using historical data; next, the location of settlements was simulated; and finally, land cover was allocated taking into account spatial decision criteria established by Van Rompaey et al. (2002). They analyzed land use changes for part of the Dijle catchment using a series of historical land cover maps, dating from 1775, 1840, 1930 and From that study, spatial decision rules could be derived, indicating which parts of the catchment would be transformed from, for instance, cropland to forest or vice versa. A more detailed description of the land use reconstruction can be found in Appendix A. Land use maps for the resulting scenarios are provided in Fig. 2 and Table 1. Scenarios were developed for a pristine landscape (i.e. a complete forest cover), the Neolithic period, the Roman Period, 1300 CE, 1650 CE, 1775 CE and 2000 CE. The selection of these scenarios depended on the available information on past land use, and the contrast in land use between the periods. The existing land cover reconstructions of Pongratz et al. (2008) were not used because the spatial resolution (0.5 ) is too limited, and does not cover the older periods of our simulations. Moreover, it was preferred to use consistently the same method for land use reconstruction in this study. A continuous analysis (i.e. with one simulation for each year since the early Holocene) was not possible due to the limited data on past land use. Daily climate data were obtained from the ECBilt-CLIO-VECODE model (Renssen et al., 2005). This is a three-dimensional coupled climate model consisting of three components describing the atmosphere, ocean and vegetation (Opsteegh et al., 1998; Goosse and Fichefet, 1999; Brovkin et al., 2002). For this study a transient run of the last 9000 years was used, forced by annually varying orbital parameters and atmospheric greenhouse gas concentrations (Renssen et al., 2005), as well as atmospheric volcanic aerosol content and fluctuations in solar activity (Goosse et al., 2005). The outputs used in this study are daily temperature and precipitation data with a resolution of ca These simulated palaeoclimate data were used in this study to evaluate the rainfall erosivity for soil erosion modeling (Watem/Sedem), and for the calculations of water discharge (STREAM). Downscaling of the climatic data was necessary due to the coarse spatial resolution at which the ECBilt-CLIO-VECODE model operates. In a first step the climatic data were spatially downscaled by resampling to a m resolution, as required for the different model runs (see Sections 3.2 and 3.3, Appendix A). The resulting maps have a downscaled resolution but the data show only variability on the original scale. In order to introduce spatial variability in the climate data at this finer spatial scale, a second downscaling step was necessary. This was performed using a statistical approach, following Bouwer et al. (2004) and Ward (2007), for which the CRU TS 1.2 dataset with a spatial resolution of was used (Mitchell and Jones, 2005). Observed average monthly data for the period for the meteorological station of Uccle (Brussels; see Verstraeten et al., 2006), which is situated ca. 10 km to the west of the study area, were used to calculate the rainfall erosivity for the Watem/Sedem model (see Appendix A) Soil erosion and sediment redistribution modeling The Watem/Sedem model was used to model past soil erosion and hillslope sediment delivery using a spatially distributed approach. A detailed description of the model is provided by Van Oost et al. (2000), Van Rompaey et al. (2001), and Verstraeten et al. (2002), and the applied methodology for this study is summarized in Appendix A. The model uses the RUSLE formula to calculate soil erosion, and a transport capacity factor to route eroded sediment towards the fluvial system. It also simulates hillslope sediment deposition. Processes within the fluvial system are not modeled, and as such a differentiation between floodplain deposition and sediment export from the catchment is not possible. The climatic factor is taken into account through the RUSLE rainfall erosivity (R) factor, while the land use is taken into account through the RUSLE crop (RUSLE C) factor and a differentiation in the transport capacity for different land use classes.

44 B. Notebaert et al. / Geomorphology 126 (2011) Fig. 2. Different simulated land cover patterns for the Dijle catchment. Numbers of the scenarios refer to Table 1. Soil properties are modeled through the soil erodibility or K factor. For this paper, the model calibration of Verstraeten (2006) was used, based on the CLC2000 land cover data and the SRTM digital elevation model. A detailed description of the used data can be found in Appendix A. As a consequence of the resolution of the input data, the Watem/Sedem model was applied on a pixel resolution of 90 m. Climate data were downscaled to this resolution, without taking into account spatial patterns. The annual rainfall erosivity factor (R factor) was derived from daily precipitation data generated by the ECBilt-CLIO-VECODE model.

45 22 B. Notebaert et al. / Geomorphology 126 (2011) Table 1 Land cover scenarios for various time periods. Scenario 9 (1775 CE) is based on the de Ferraris map; scenario 10 is based on the CLC2000 map. The remaining land cover for scenario 10 is urbanized land. Scenario Period Number of settlements Here, we used a relationship between the daily rainfall data and daily erosivity values established for Uccle based on 10 minute rainfall data for the period , with monthly correction parameters (Verstraeten et al., 2006): R = d = 360 d =1 a pre 1:8067 d Where a is an empirical derived monthly calibration factor and pre d the daily precipitation. This regression analysis was based on the period The modeled climate data count only 360 days in one year. An important tool to evaluate the sediment dynamics is the hillslope sediment delivery ratio (HSDR), which is calculated as the fraction of sediment that is delivered by the hillslopes compared to the total soil erosion. The HSDR parameter provides additional information on the fluxes of sediment between the hillslopes and the fluvial system, and as such also on the connectivity between them Water discharge modeling % forest % arable land 1 Pristine Neolithic (~3000 BCE) Neolithic (~3000 BCE) Neolithic (~3000 BCE) Roman (200 CE) Roman (200 CE) CE CE CE Current % grassland The STREAM hydrological model was selected to model the palaeodischarge of the Dijle catchment. This model has been successfully applied to the palaeodischarge modeling of several catchments around the globe (Aerts et al., 2006; Renssen et al., 2007; Ward et al., 2007, 2008), including the central catchments of the Meuse and Rhine Rivers, but the model was always applied to catchments of at least some 10 4 km 2. The STREAM model is a gridbased spatially distributed water balance model that simulates the hydrological cycle of a catchment through a series of storage compartments and flows (Aerts et al., 1999). The water balance is calculated using the Thornthwaite (1948) equations for potential evapotranspiration and the Thornthwaite and Mather (1957) equations for actual evapotranspiration; these equations use temperature and precipitation as the major input parameters. For each time-step the model generates runoff, groundwater storage (shallow and deep), snow cover, and snow melt. The direction of water flow between cells is based on the steepest descent for the eight surrounding grid cells on a digital elevation model (DEM). In this study it was run on a daily basis, which is important when calculating extreme events in the past. Although the original model has been successfully applied to many European catchments, it has some shortcomings for application in the Dijle catchment, namely in the calculations of base flow and direct runoff. Therefore, in this study a baseflow parameter was added, and the direct runoff calculations were altered. The baseflow parameter is based on extrapolation based on available baseflow data, while the direct runoff is calculated based on the curve number (CN) method (soil conservation service, SCS; e.g. Haan et al., 1994). As a tradeoff between computing time and spatial resolution, a pixel resolution of ð1þ 500 m was used for the STREAM model. A detailed description of the data preparation can be found in Appendix A. The adapted STREAM model holds in total seven parameters that require calibration: temperature threshold for which precipitation falls as snow and for which snow melts (SNOW-temp), a factor determining the rate of snowmelt (MELT), a cropfactor calibration parameter, a heat calibration parameter, a water holding capacity factor (whold), a factor determining the proportion of groundwater that contributes to baseflow (c-factor) and a baseflow parameter. Calibration of these parameters was first based on independent data (Gellens-Meulenberghs and Gellens, 1992) for the evapotranspiration, by using the average monthly potential evapotranspiration for different land use classes. The remaining parameters (c-factor and whold) were calibrated using observed discharge records. A detailed description of the calibration procedure can be found in Appendix A. We also fitted the Gumbel distribution (e.g. Bridge, 2003) to the simulated and observed time-series of annual maximum daily discharges to estimate the recurrence times of extreme discharge events. For each scenario we estimated the recurrence intervals of daily discharge events corresponding to the 95, 99, 99.9 and % percentiles of the current scenario (i.e. model results with CLC2000 land cover and climate), i.e. discharge events of the following magnitudes: 9.7, 12.6, 25.7 and 46.6 m³s 1 respectively Scenario and sensitivity analysis Besides the 10 palaeo-environmental scenarios for different periods, a sensitivity analysis was performed in order to analyze the sensitivity of the model results to changes in land cover and climate. Such a sensitivity analysis allows a deeper insight into the system dynamics, as some of the parameters can be held constant while others vary between modeling runs. For the Watem/Sedem model, we developed several scenarios by varying land cover, climate or soil properties. Different sensitivity scenarios were developed to test the influence of land cover: by varying the amount and allocation procedure of settlements, and by varying the fraction of agricultural land cover. Two procedures for the allocation of settlements were used based on the formulas of Peeters (2007): first the probability of settlement allocation is set random, i.e. independent from geomorphic parameters, whilst a second set of scenarios assumed that settlement location depends on soil texture and distance to rivers, as can be observed for (post-) Medieval settlements (e.g. Peeters, 2007). Each of these scenarios was run with different settlement densities, i.e. a low density of 15 settlements in the entire catchment, and a high density of 130 settlements. These numbers agree with the number of settlements for the scenarios for the Roman Period on the one hand, and the scenarios for 1300 and 1650 CE on the other hand. Since land use is allocated around settlements, the scenario with low densities corresponds with a few but large (and presumably more or less constant through time) patches where the forest is cleared, whereas the high settlement density corresponds to many small exploitations. Finally, for these four possibilities of quantity and location of settlements, the fractions of the different land cover types were varied. The fraction of forests varies between 0 and 100%, while the remainder area was divided between cropland and pasture, with a 9/1 ratio. For example, a land cover with 80% forests has 18% arable land and 2% grassland. This ratio corresponds with the ratio on the 1775 CE de Ferraris map, but it is unclear whether this ratio has changed over time. All of the scenarios used are summarized in Table 2, with their abbreviations which are used hereafter. For instance, MP 20 is a scenario with 130 settlements ( M, from many) that are allocated based on geomorphic parameters ( P, from parameter), and where 20% of the catchment is covered by forest ( 20 ). Next, for some of these scenarios the R and K factors were also varied to evaluate the influence of variations in rainfall erosivity and soil erodibility. Finally, scenarios were tested for which the simulated Neolithic land use

46 B. Notebaert et al. / Geomorphology 126 (2011) Table 2 Different scenarios for the allocation of settlements used to produce land cover maps for the sensitivity analysis of the Watem/Sedem model. Name Number of settlements Location of settlements MP 130 Parameterized (1300 CE, Table 1) MR 130 Random (200 CE, Table 1) FP 15 Parameterized (1300 CE, Table 1) FR 15 Random (200 CE, Table 1) pattern was used, but the RUSLE C factor (representing the erosional sensitivity of land cover) of the forests was increased from to 0.01, while the RUSLE C factor for cropland was decreased from 0.37 to 0.1. Values of and 0.37 for forest and cropland, respectively, are typical values for present-day vegetation covers under these land use conditions in central Belgium (Verstraeten et al., 2003). However, it is likely that Neolithic fields were less susceptible to erosion due to higher roughness values (no plows or harrows) and higher organic matter contents. Therefore, a lower value for Neolithic cropland (i.e. 0.1) was also evaluated. On the other hand, Neolithic forests could have been used more intensively, for instance for cattle grazing (e.g. Tack et al., 1993; Williams, 2003), thus reducing the ground vegetation cover, resulting in higher RUSLE C values. For each sensitivity analysis scenario run, total hillslope soil erosion, hillslope sediment deposition, and sediment delivery to the fluvial system are evaluated. Given the long computational times for the STREAM model compared to the Watem/Sedem model, an extended sensitivity analysis was not possible for the former model. Therefore the sensitivity analysis was limited to the combination of the climate data of BCE, the period corresponding with the runs for a pristine situation, and CE with the different land cover scenarios, in order to identify the individual contribution of land cover changes and climate variations on water discharge. 4. Results 4.1. Climate data Downscaled model precipitation and temperature data were taken for the periods and BCE, as well as , , , and CE. The precipitation data show an increase in summer precipitation after 0 CE, when compared to the data for and BCE. Winter precipitation is higher for the runs of and BCE and CE than those of the other periods. Average annual precipitation varies between 754 and 787 mm. The average precipitation for the period CE is highest, although there is no statistically significant difference with the average precipitation in , and CE. Climatic data for the different considered periods are represented in Fig. 3. These show a gradual decrease in average temperature from the Mid-Holocene towards the 18th century CE, and again a higher temperature for the 20th century CE. Differences are, however, limited, with less than 1 C difference between the average temperature for the warmest century ( BCE). Variability in annual temperature between the different periods is Fig. 3. Climatic data produced by the ECBilt model for the different considered periods.

47 24 B. Notebaert et al. / Geomorphology 126 (2011) in most cases statistically significant (t-test, pb0.05), while that within a single period is in most cases not significant. These modeled temperature and precipitation curves show less variation over the Holocene than climate reconstructions from proxy data from northwestern France (e.g. Naughton et al., 2007). There is, however, an important spatial variability in Holocene climate changes within Europe. Pollen data indicate only minor temperature variations for the surroundings of Belgium during the last 8000 years (e.g. Davis et al., 2003). Magny et al. (2003) pointed out that the hydrological response to climate change varies over Europe, whereby early Holocene precipitation changes were opposite between south of and north of 50 N, at which Belgium is located Scenarios for past soil erosion and sediment redistribution Results of the Watem/Sedem model runs are presented in Table 3. The highest soil erosion amounts are simulated for scenarios with the largest cropland areas, i.e. for the scenarios of 1775 CE and 1300 CE, with a more than 100 fold increase compared to a pristine land cover. For the current scenario, soil erosion amounts are ca 50% lower compared to the 1775 CE scenario due to the decrease in cropland area. HSDR for a pristine land cover is very high (92%), while it is much lower for other scenarios (45 60%). This implies that, when parts of the catchment are deforested, colluvial deposition increases even more than soil erosion, relative to a catchment with a pristine land cover. Soil erosion amounts and HSDR values vary only slightly between the four basic scenarios of the sensitivity analysis (MP, MR, FP, and FR; Fig. 4). For most tested land cover quantities the lowest erosion amounts were yielded for scenarios based on FP. Differences are small (often b10%, maximal 23%) except for the scenarios with 95% forest cover: here the location of the cropland area within the catchment plays a more important role. For these scenarios, the cropland area is limited, and as such there is a higher probability that the major part of the cropland area is situated on one slope class (e.g. steep slopes), creating a larger variability between the various scenarios. In general, varying the fraction of forest (and thus also the fraction of agricultural land) has a far more important influence on the soil erosion amounts and HSDR values than varying the location of settlements and land use patterns (Figs. 4 and 5). In a pristine situation soil erosion is very low, and almost all the produced sediment can be transported to the fluvial system. Soil erosion increases exponentially with increasing cropland area (Fig. 4), and HSDR values decrease from about 90% for the total forest cover to about 45% when a small agricultural area (2% arable land) is introduced in the catchment, and then increases again to about 60% with the cropland area increasing to about 50% (Fig. 5). In addition, Yearly erosion (10 3 t a -1 ) Fraction cropland in catchment (%) Fig. 4. Yearly soil erosion (10³Mg a 1 ) as a function of the fraction of cropland in the catchment (%) for different tested scenarios. Differences between the scenarios are explained in the text and summarized in Table 2. adapting the RUSLE C factor for the Neolithic Period results in higher average erosion rates (ca. +15%) while the HSDR value also increases (to ca. 65%) because of an increased transport capacity of the forests. Increasing the RUSLE C factor of the Watem/Sedem model for forest and decreasing it for cropland for the Neolithic scenarios results in a total increase of up to 20% in soil erosion, while HSDR values increase to ca. 65%. This is due to the coupling of the transport capacity to the RUSLE C, which results in an increased sediment transport through forests. The sensitivity analysis indicates a one-to-one relation between erosion and the R factor (Fig. 6), and between erosion and the K factor, which are logical consequences of the used RUSLE equation in the Watem/Sedem. However, the R and K factors also show no influence on HSDR in the tested range, and the sensitivity for catchment-scale soil erosion to the factors is independent of the land cover in relative terms: an increase of one of the factors by 50% will increase the erosion by 50% for each land cover scenario Water discharge modeling STREAM model calibration Calibrations of the crop factor and the heat calibration parameter are based on potential evapotranspiration. The selected values have a model efficiency of 0.81 but the potential evapotranspiration values modeled by the (calibrated) STREAM model show a time lag with the observed values (Appendix A), caused by the calculation methods. The calibration of discharge is based on discharge data measured at the catchment outlet for ( MR FP FR Table 3 Modeling results of the Watem/Sedem for different scenarios representing past land cover and climate. Land cover scenario Period % forest % arable land % grass-land Average yearly rainfall (mm a 1 ) R factor (MJ mm ha 1 h 1 a 1 ) Erosion (Mg a 1 ) Deposition (Mg a 1 ) Export (Mg a 1 ) 1 Pristine Neolithic , , (~3000 BCE) 3 Neolithic , (~3000 BCE) 4 Neolithic , (~3000 BCE) 5 Roman (200 CE) ,038 10, Roman (200 CE) ,024 72,710 72, CE , , , CE ,609 87,330 73, CE , , , Current ,342 94, , HSDR (%) Erosion rate (Mg km 2 a 1 )

48 B. Notebaert et al. / Geomorphology 126 (2011) HSDR (%) MP MR FP FR Fraction cropland in catchment (%) Daily discharge (m 3 /s) BCE BCE CE CE CE CE CE observed CE Fig. 5. Hillslope sediment delivery ratio (HSDR,%) as a function of the fraction of cropland in the catchment (%) for different tested scenarios. Differences between the scenarios are explained in the text. station Korbeek Dijle; see Appendix A). Average monthly discharge data were used for calibration, and a lag time between observed and modeled discharge was taken into account to correct for the effects of the differences in potential evapotranspiration (see Appendix A). The model efficiency for the calibrated model is Nevertheless, there is a statistically significant difference between the distributions of the modeled and observed daily discharges (Kolmogorov Smirnov test, pb0.05). The modeled mean daily discharges are not significantly different from the observed values (Student t-test, pn0.05), but the variance of the modeled discharge data is significantly higher (modeled: σ=2.8 m³s 1 ; observed: σ=2.1 m³s 1 ; F-test pb0.05) Modeling results Modeling results for mean daily discharge at Korbeek Dijle show an initial slight, but statistically significant, decrease between the periods and BCE, and then an increase towards the end of the Holocene (Figs. 7 and 8, Table 4). The differences in mean discharge for each period in chronological order are significant in each case (Student t-test, pb0.05, Appendix A), except for between the periods BCE and CE. The variability in the daily discharge shows a comparable pattern, with higher variability towards the end of the Holocene, and this variability being statistically different between most of the scenarios in chronological order (F-test, pb0.05). The frequency distributions of the different scenarios are also significantly different from each other (KS-test, pb0.05). Whilst changes in average daily discharge, and thus total discharge, are limited, changes in peak discharges are greater. The estimated recurrence intervals of the extreme discharge events (Table 4) show a comparable pattern, with very long recurrence Yearly erosion (10 3 t a -1 ) MP 20 MP 80 MP Rainfall erosivityfactor (R-factor, MJ mm ha -1 h -1 a -1 ) Fig. 6. Yearly soil erosion (10³Mg a 1 ) as a function of the rainfall erosivity factor (MJ mm ha 1 h 1 a 1 ) for different tested scenarios. Scenarios are based on the basic scenario MP (see text) with 20% (MP 20), 80% (MP 80) and 100% (MP 100) forest cover Recurrence interval (years) Fig. 7. Recurrence interval of mean daily discharges at Korbeek Dijle for the observed data and the different modeled scenarios for past climate and land cover. Recurrence intervals are determined based on a Gumbel max distribution which was fitted through the modeled and observed yearly maximal daily discharges. intervals for high discharge values when large parts of the catchment are under forest Sensitivity analysis The results of the sensitivity analysis (Fig. 9) show an increase in average discharge with increasing land use intensity under constant climate ( CE): for a pristine land cover the average discharge is 7.2 m³s 1, while it increases up to 7.6 m³s 1 for the CLC2000 land cover map (Table 5). Only the average discharges using the land cover maps of 1300 and 1775 CE are not significantly different to the average discharge using the CLC2000 land cover (Student t-test, pn0.05; Appendix A). The variance in daily discharge is highest for the run with the CLC2000 land cover configuration (σ=2.80 m³s 1 ) and lowest for the run with a pristine land cover (σ=1.86 m³s 1 ), and the variance is significantly different between most scenarios. Small changes in land cover have no significant influence on average discharge and its variation, for example when the pristine and Neolithic Period scenarios are compared. The return times of extreme discharge events are also strongly influenced by land cover, with the shortest return times for the CLC2000 land cover and the longest for the pristine land cover (Table 5). The frequency distributions of daily discharge for the different scenarios are in most Main daily discharge (m 3 /s) BCE BCE CE CE CE CE CE Month Fig. 8. Modeled mean daily discharge for each month at Korbeek Dijle for the different scenarios.

49 26 B. Notebaert et al. / Geomorphology 126 (2011) Table 4 Results of the modeled scenarios for the STREAM model. Average and maximal discharges are for each scenario combined with recurrence time for some selected discharges. These discharges correspond with the recurrence times of the 95, 99, 99.9 and % percentiles of the observed discharges. a=year(s), d=day(s). Land cover scenario Climatic scenario Average discharge (m³s 1 ) Maximal discharge (m³s 1 ) Recurrence times Q=9.7 m³s 1 Q=12.6 m³s 1 Q=25.7 m³s 1 Q=46.6 m³s 1 1. Pristine BCE 6.8± d 18.4 a 25 10³ a a 3. Neolithic BCE 6.5± d 6.5 a 267 a ³ a 6. Roman CE 6.5± d 3.1 a 31.8 a 1688 a CE CE 6.6± d 261 d 6.2 a 86 a CE CE 7.0± d 2.1 a 25.7 a 92 a CE CE 7.2± d 242 d 4.7 a 47 a 0. CLC CE 7.2± d 100 d 3.4 a 24.8 a cases significantly different (KS-test, pb0.05), although they indicate that minor differences in land cover do not result in significant changes in the frequency distribution. Additional model runs were carried out combining the climatic data for BCE with the pristine land cover map, and the CLC2000 land cover map (Table 5). The results show that the average daily discharge for these runs differs from the average daily discharge modeled with the same land cover maps but the contemporary climate, but these differences are not statistically significant (Student t-test; pn0.05). The variance is, however, significantly lower when the BCE climatic data are used (F-test, pb0.05), and also the frequency distributions differ between the runs with different climate (KS-test, pb0.05). The A Mean daily discharge (m 3 /s) B Mean daily discharge (m 3 /s) pristine Neolithic Roman Period 1300 CE 1650 CE de Ferraris map CLC Month pristine, climate CE CLC2000, climate CE CLC2000, climate BCE pristine, climate BCE Month Fig. 9. Modeled average daily discharge per month at Korbeek Dijle for the scenario analysis. A) with constant climatic data ( CE) and varying land cover; B) with varying land cover and climatic data. recurrence intervals for high discharge events are higher for the BCE climate scenario. 5. Discussion 5.1. Soil erosion and sediment redistribution modeling The time-differentiated sediment budget of the Dijle catchment (Notebaert et al., in press) provides a unique tool for validating the sediment modeling results (Table 6 and Fig. 10), although the time resolution of this budget and the modeling results do not allow a direct comparison. Erosion rates of the sediment budget for 9000 to 2000 BCE (period 1) and 2000 BCE to 1000 CE (period 2) are in correspondence with modeled values. For the last 1000 years (period 3), however, the observed erosion rate (704±96 Mg km 2 a 1 ) is much higher than the modeled rates ( Mg km²a 1 ). Nevertheless, there is an overall good correspondence between modeled erosion rates and erosion rates derived from the field-based approach. This is also evidenced by the modeled HSDR values which are in close agreement with those derived from the sediment budget. For the pristine scenario and period 1, HSDR is very high (92% and 96%, respectively). For period 2, the observed HSDR value decreases to 44%, which is consistent with the HSDR values of the sensitivity runs for forest covers between 98% and 60%. The observed HSDR value increases again to 61% for period 3, which is consistent with the higher HSDR values for sensitivity runs with 40% or less forest cover, and with scenarios for 1300, 1775 and 2000 CE. The results indicate that soil erosion and hillslope sediment delivery are mainly sensitive to changes in land cover, and that HSDR is not influenced by the R and K factors. Changes in the RUSLE C factor of the different land cover types have only a minor influence on the soil erosion values, but the influence on the HSDR values is somewhat larger. These observations indicate that the modeled changes in HSDR are not sensitive to uncertainties in reconstructed climate data. When there is complete forest cover, soil erosion is very limited and almost all eroded sediment can be transported towards the fluvial system as the transport capacity for overland flow is not exceeded (i.e. the hillslope system is supply-limited). When the forest cover decreases and cropland area increases, soil erosion also increases, and at these locations the hillslope systems changes to a capacity-limited system where transport capacity is exceeded and sediment is deposited as colluvium, leading to a lower HSDR value. As cropland area increases, larger parts of cropland become situated on steeper slopes, while also the slope length under cropland increases, which results in an exponential increase in soil erosion. The connectivity between the cropland area and the fluvial system also increases, as more and more cropland area is directly connected to the fluvial system, resulting in more efficient sediment transport between the hillslopes and the fluvial system. Soil erosion values of the sensitivity analysis are largely in agreement with those of the past soil erosion scenarios (Fig. 11), and differ only slightly between scenarios. This shows that the location of settlements is less important than the total amount of

50 B. Notebaert et al. / Geomorphology 126 (2011) Table 5 Sensitivity analysis of the STREAM model. A) Modeled average discharge and maximal average daily discharge using the climatic data for the period CE and different land cover scenarios; B) Modeled average discharge and maximal average daily discharge using the climatic data for the period BCE and different land cover scenarios. Land cover scenario Average discharge m 3 s 1 Maximal discharge m 3 s 1 Recurrence time Q=9.7 m³s 1 Q=12.6 m³s 1 Q= m³s 1 Q=46.6 m³s 1 Climate Pristine d 44 d 63 a ³ a 3. Neolithic d 44 d 40 a ³ a 6. Roman d 40 d 5.4 a 34.2 a CE d 31 d 2.7 a 7.0 a CE d 35 d 3.8 a 15.1 a CE d 34 d 2.4 a 5.6 a 0. CLC d 30 d 2.1 a 4.3 a Climate BCE 1. Pristine d 13.5 a 27 10³ a a 0. CLC d 110 d 2.2 a 6.0 a agricultural land for the quantities of eroded sediment. Small differences can be explained by lower rainfall erosivity values and larger forest and grassland areas on steep slopes. Several scenarios were used for the land use reconstruction of the Neolithic and Roman Periods, due to the large uncertainties of past land use for these periods. The available sediment budget can be used in order to evaluate the different scenarios and the modeled soil erosion quantities for the same period. However, due to the temporal resolution of the sediment budget, such an evaluation is not straightforward. Erosion rates for the Neolithic Period fall within the first period of the sediment budget, and are averaged out by a rather long period of no, or very limited, human impact. The results show that on average, in this first period, human impact on the landscape was low, with a forest cover of ca. 95%, or even more if part of the forest was degraded due to grazing and thus producing more sediment than modeled. A part of the second period of the sediment budget corresponds with the Roman Period. Our results indicate that the scenario for the Roman Period based on Peeters (2007) underestimates the anthropogenic impact. A scenario with a 10 25% cropland area better corresponds to the sediment budget results; because of the temporal resolution a more precise fraction of cropland cannot be assessed. Based on the sensitivity analysis, the individual contribution of climate and land use can be modeled (Table 6). When the current R factor is applied in combination with the pristine land cover, soil erosion and hillslope sediment delivery increase by 9%. Combining the R factor of BCE with the current land cover leads to a 60 fold increase in soil erosion and a 35 fold increase in hillslope sediment delivery compared to the pristine land cover. Contemporary soil erosion is 66 times higher than pristine erosion, while hillslope sediment delivery increased with a factor of 38. Soil erosion rates decreased with about 49% between the 1775 CE scenario and the current scenario, mainly due to a decrease in arable land. All of these observations indicate that land use has a far more important influence on soil erosion and sediment redistribution than climate, within the variations that occurred during the Holocene. The use of monthly precipitation data for the calculation of the R factor implies that climatic variations with a temporal resolution of less than a month, for example variations in extreme events, are not incorporated in the results. In addition, the RUSLE based approach combined with scenarios means that only variations in average annual values for longer periods are represented, rather than short timed variations, and hence the results may seem to represent a steady state. It was previously shown that simulating long term soil erosion using models based on average rates fails to predict the contribution of individual events, which may vary largely (e.g. Hancock et al., 2010). Soil erosion increased 64-fold between ~4800 BCE and ~1990 CE, while colluvial sediment deposition increased 362 times, and export to the fluvial system increased with a factor of only 38. Thus, a strong scaledependency can be observed in the geomorphic response to environmental change, which is related to the non-linear behavior of the system and complex long-term interactions between sediment sources and sinks, as demonstrated by other studies (e.g. Brierley and Fryirs, 1999; Lang and Hönscheidt, 1999; Trimble, 1999; Lang et al., 2003; Verstraeten et al., 2009a). This scale-dependency is also found when looking at the relative role of climate and land use in controlling sediment fluxes. Only 9% of the increase in erosion and sediment delivery between a pristine mid-holocene environment and the present can be attributed to climate change, thus pointing to the overwhelming importance of land use. However, colluvial deposition rates are to a much higher extent controlled by land use change (+ ca. 34,000%) compared to sediment delivery to the river system (+ ca. 3550%). This corroborates findings from other studies that state that colluvial sediment archives are more suited to identifying signals of human impact, whereas climate signals are more easily reported in fluvial deposits (e.g. Zolitschka et al., 2003, Dotterweich, 2008). Table 6 Modeling results for different model runs of the Watem/Sedem model, indicating the influence of rainfall erosivity and land use changes. Land cover scenario R-factor (MJ mm ha 1 h 1 a 1 ) Erosion (Mg a 1 ) Deposition (Mg a 1 ) Hillslope export (Mg a 1 ) Compared to pristine situation a Erosion Deposition Export Pristine de Ferraris map , , ,793 11,858 63, , , ,632 12,741 68, , , ,488 12,956 69, Current land cover ,777 86, , , ,342 94, , , Sed. Budget 1000 CE 2000 CE 523, , ,000 16,535 79,528 11,004 a Value normalized over the modeled value for a pristine land cover and rainfall erosivity.

51 28 B. Notebaert et al. / Geomorphology 126 (2011) Fig. 10. Soil erosion rates, sediment deposition rates, hillslope sediment delivery rates and HSDR for the Dijle model. Values from the sediment budget (Notebaert et al., submitted) are displayed as boxes. Values from the model scenarios are displayed as points with bars indicating the approximate period for which the values are indicative. For lake sediments, Dearing and Jones (2003) already pointed out that the ratio between maximum and minimum sedimentation rates, which can be related to catchment disturbance, decreases with increasing catchment area. The observed and modeled scaledependency of a catchment's response to environmental change has important implications when sedimentary archives are used as a proxy for environmental change. No straightforward reconstruction of past soil erosion rates, or a chronology of human impact, can be established when relying on one single sedimentary archive. With an approach comparable to that used in the present study, previous modeling results have demonstrated the importance of land cover for soil erosion rates and sediment delivery to the fluvial system of the Geul and Meuse catchments (de Moor and Verstraeten, 2008; Ward et al., 2009). It was demonstrated for the Meuse catchment that the effect of land cover changes on sediment delivery to the fluvial system is far more important than variations in rainfall erosivity (Ward et al., 2009). In this catchment, sediment delivery to the fluvial system increased by ca. 333% from the period before human impact to the 20th century. Sediment delivery decreased with ca. 28% between the 19th and 20th centuries, due to urbanization and reforestation. The lower impact compared to the Dijle catchment can be explained by the lower intensity of human land use in the former basin, and by a lower sensitivity of the landscape to erosion (e.g. due to less erodible soils) Water discharge modeling The calibrated STREAM model results show a higher variability of discharge than the observed data for at the Korbeek Dijle hydrological station. These differences may in part be attributed to errors in the observed discharge series. The measuring period (27 years) is too short to accurately represent extreme events. Whilst the highest observed daily discharge at Korbeek Dijle in the period is 38 m 3 s 1, past discharges near Leuven are reported to equal up to 126 m³s 1 (23 January 1891; Raymaeckers et al., 2002), although it is unclear how long the recurrence interval of such an event is. The recurrence interval for discharges with a magnitude of 100 m³s 1 at Korbeek Dijle, based on the measured discharge data for (Fig. 7), equals more than 10 9 years, although it should be noted that the use of the Gumbel distribution in this estimation in itself introduces large uncertainty (e.g. De Wit and Buishand, 2007). On the other hand, the model assumes that all excess precipitation contributes to discharge at the basin outlet on the same day, while in reality a spatially distributed time lag between precipitation and discharge may occur. The water discharge at the catchment outlet of a given precipitation event will therefore be spread in time (i.e. over two to three consecutive days), and as a result, modeled discharge can be expected to show more peaks than observed discharge. However, the average duration of a flood event in the Dijle catchment is seldom Yearly erosion (10 3 t a -1 ) LC SENS HSDR (%) LC SENS Fraction cropland in catchment (%) Fraction cropland in catchment (%) Fig. 11. Yearly soil erosion (10³ Mg a 1 ) as a function of the fraction of cropland in the catchment (%) and hillslope sediment delivery ratio (HSDR,%) as a function of the fraction cropland in the catchment (%). LC are the historical land cover reconstruction, while SENS are the MP sensitivity scenarios.

52 B. Notebaert et al. / Geomorphology 126 (2011) larger than one day. Thus, we believe that the calibrated STREAM is capable of predicting the hydrological response to changes in land use and climate in the past. Validation of the results obtained with the STREAM for the historical scenarios is, however, not possible as no detailed long time discharge records are available from the catchment, and sedimentary evidence for past discharges is missing. Results of the sensitivity analysis and for the different past scenarios provide insights into the mechanisms and driving forces of water discharge in the Dijle catchment. The sensitivity analysis indicates that a major decrease of the forest cover is responsible for a slight increase in average yearly discharge. However, this change in land cover has a far more important effect on the peak discharges, the recurrence intervals of which decrease dramatically over the study period (Fig. 7). When considering climate alone, changes between BCE and CE did not result in a statistically significant difference in runoff under the same land cover scenarios. However, these climatic changes do result in a statistically significant increase in the variance of daily runoff. The results of the STREAM modeling indicate that, over the studied period, land cover has a large influence on average discharge and variability in daily discharge, while climate only has a significant influence on discharge variability. This can be explained by the limited variations in average precipitation, while the variability in precipitation has undergone more important changes. When the periods BCE and CE are compared, increases in mean and peak discharges are indicated (Table 7). The combined effect of land use and climatic variations is not just simply the sum of both effects (Table 7). Land cover changes contribute the most to changes in discharges and peak discharges over this time period, a consequence of the relatively large changes in land use and relatively small changes in climate. Whilst soil erosion and sediment export decrease again in the 20th century CE, due to the decrease in cropland, water discharge and peak discharges still increase. The results obtained here are in accordance with several other studies. For instance, Leopold (1968) and Arnold et al. (1982) similarly found a strong increase in direct runoff in urbanized areas, which contributes largely to the direct runoff. Comparably, land use changes between the 19th and 20th centuries CE in the Meuse basin resulted only in minor changes in discharge (Ward et al., 2008). Important changes in annual and peak discharges occur in the Meuse catchment between the Mid-Holocene and contemporary scenarios (Ward et al., 2008), comparable to this study. Other studies also report a strong link between deforestation and increased mean discharge in other basins (e.g. Bosch and Hewlett, 1982; Andréassian, 2004; Keesstra, 2006). Variations in discharge related to land cover are often reported to be resulting from alterations in evapotranspiration (e.g. Calder, 1993) or changes in soil water holding capacity (e.g. Mahe et al., 2005). These changes in soil water holding capacity are not incorporated in this application of the STREAM model, and as a result only the changes in evapotranspiration are considered. There are also other reports on increases of peak floods due to deforestation (e.g. Gentry and Parody, 1980; Brown et al., 2005), although some authors report that this effect is much more varied in time and space (e.g. Andréassian, 2004). This study has indicated that during the Holocene, land use was the most important factor influencing soil erosion, sediment deposition, and water discharge. Climatic variability, however, was relatively limited during this period. In the prospect of future climatic changes, Table 7 Contribution of climate and land use to changes in average annual discharge (Q ann ) and various highflow discharge percentiles (Q k with k the percentile) between BCE and CE. The differences are expressed as a percentage increase or decrease compared to the values for BCE with a pristine land cover. Q ann Q 75 Q 90 Q 95 Q 99 Q 99.9 Climate and land use Climate Land cover the influence of these changes on the studied processes may become more important. It may be the case that land use is much more dominant over the late Holocene, but this may not hold for the future, especially on already degraded land, causing important changes (e.g. Ward et al., in press). The sensitivity analysis of the soil erosion modeling (Fig. 6), however, indicates that even with major climatic changes, land use remains the dominant factor Uncertainties Although the model results correspond rather well with the available proxy records of past sediment dynamics and river discharge, there are several uncertainties in both model applications. These uncertainties can account for at least part of the difference in erosion rates that exist between the model and sediment budget approach, especially for period 3. First of all, cropland area may be underestimated, and grassland area is probably overestimated. Almost the entire grassland area of the de Ferraris map is located in floodplains, while on the reconstructed maps large areas of grassland are also situated on (often steep) slopes. The sensitivity analysis shows that the observed erosion rates correspond with a cropland area of ca. 75% if the current rainfall erosivity is valid for the entire period since 1000 CE. Other factors may have played a role as well. The RUSLE C factor for some land use types may also be underestimated. For instance, in the Middle Ages forests were often used as grazing grounds (e.g. Tack et al., 1993), and would therefore probably have had a higher susceptibility to erosion compared with contemporary forests. Also, the K factor may be higher due to the dynamic nature of the soil profile over longer time periods. Indeed, with intense soil erosion, the soil profile becomes truncated, bringing more erodible soil horizons to the surface (e.g. Rommens et al., 2007). Rainfall erosivity may be underestimated as well. Several authors report on large rainfall events for the 13th and 14th centuries CE in central Europe (e.g. Pfister et al., 1996; Glaser, 2001; Dawson et al., 2007), and major soil erosion and sediment deposition events are linked to this period (e.g. Bork et al., 1998; Dotterweich, 2008). Such important climatic events are, however, not found in the modeled climate data. Even if they were present in the modeled climate data, they would only influence modeled soil erosion amounts when they influence the climate on a monthly temporal resolution. Although no geomorphic imprints of such events can be found in central Belgium, it cannot be excluded that some of these events also occurred in our study area, but that these are not simulated by the climate model. Finally, the errors in the sediment budget approach also need to be considered. Several assumptions had to be made in order to estimate the erosion of the sediment budget and to differentiate the different periods (Notebaert et al., in press), leading to uncertainties in the budget. Peak flows in the used version of the STREAM model are influenced by the actual evapotranspiration (e.g. Ward, 2007) and the direct runoff, which vary with the land cover and antecedent moisture conditions. Comparable with the soil erosion modeling, uncertainties may evolve from the land use reconstruction and the used climatic data. The highly simplified evapotranspiration calculations in the STREAM model may also influence the results (see Ward et al., 2008). Finally, influences of anthropogenic water flow regulations are not incorporated in the model. Due to these limitations, the modeling results should be treated with care. Despite these limitations, the model is based on robust and generally accepted equations, and modeling results are in agreement with reported variations in discharge and peak discharge for comparable land cover changes. 6. Concluding remarks Comparison of model results with the time-differentiated sediment budget (Notebaert et al., in press) indicates that a spatially distributed erosion and sediment delivery model like the Watem/ Sedem, which was initially calibrated for contemporary erosion

53 30 B. Notebaert et al. / Geomorphology 126 (2011) processes, provides realistic estimates of past erosion and sediment delivery rates. Absolute masses of soil erosion are slightly underestimated, especially for the period between 1000 and 2000 CE, which is probably related to uncertainties in reconstructed land use and climate scenarios, although inaccuracies in the field-based sediment budget approach cannot be excluded. The soil erosion and sediment redistribution modeling indicates that: Soil erosion increased 64-fold between ~4800 BCE and ~1990 CE. Only 9% of this increase can be attributed to climatic changes, while the remainder can be dedicated to land use changes. Modeling stresses the differentiated response of colluvial and alluvial deposition to environmental changes, with a more pronounced increase in colluvial deposition rates. Due to this scale dependency, no straightforward reconstruction of past soil erosion rates can be established when relying on one single sedimentary archive. In order to fully understand the response of hillslopes and fluvial systems to environmental change, a systembased approach is necessary. Discharge of the Dijle River was modeled using the STREAM model, showing the land use is dominant over climate as a controlling factor using the time-periods studied. Climate change is responsible for a decrease in average discharge of ca. 0.2% (early Holocene compared with 20th century CE), while land cover changes are responsible for an increase of about 6%. Both land use and climatic variations have a more pronounced influence on peak discharges than on average discharge, with the shortest recurrence intervals for extreme events for the simulation of the period When both models are compared, it becomes clear that: Overall, soil erosion and sediment redistribution are more sensitive to land use changes than discharge is, as demonstrated by the more important variations in sediment export during the Holocene compared to fluctuations in (peak) discharge. During the last few hundred years, sediment and runoff dynamics became more and more decoupled. The current land use configuration is not responsible for the highest erosion rates; soil erosion and sediment deposition rates were highest in the 1775 CE scenario, when cropland areas were more extensive than today. However, in contrast to sediment dynamics, changes in land use since 1775 resulted in a further increase of average and peak discharges, showing the different response of discharge and soil erosion and redistribution to recent land cover changes. Whereas conversion of cropland to urban and residential areas decreases average soil erosion rates, direct runoff increases. The different response of discharge and sediment dynamics to these recent changes has important implications for future dynamics within the fluvial system. Rivers deprived of sediment, but with more frequent and more intense flood events, will adjust to a new equilibrium (Schumm, 1977). Indeed, detailed topographciic measurements of river channel morphometry already point out that the Dijle River has been widening and deepening since at least 1968 (Notebaert, 2009). Appendix A. Supplementary data Supplementary data to this article can be found online at doi: /j.geomorph References Aerts, J.C.J.H., Kriek, M., Schepel, M., STREAM (Spatial Tools for River Basins and Environment and Analysis of Management Options): set up and requirements. Physics and Chemistry of the Earth part B Hydrology Oceans and Atmosphere 24, Aerts, J.C.J.H., Renssen, H., Ward, P.J., De Moel, H., Odada, E., Bouwer, L.M., Goosse, H., Sensitivity of global river discharges under Holocene and future climate conditions. Geophysical Research Letters 33. doi: /2006gl Andréassian, V., Waters and forests: from historical controversy to scientific debate. Journal of Hydrology 291, Arnold, C., Boison, P., Patton, P., Sawmill Brook: an example of rapid geomorphic change related to urbanization. The Journal of Geology 90, Boardman, J., Favis-Mortlock, D., Climate change and soil erosion in Britain. The Geographical Journal 2, Bork, H.-R., Bork, H., Dalchow, C., Faust, B., Piorr, H.-P., Schatz, T. (Eds.), Landschaftsentwicklung in Mitteleuropa Wirkungen des Menschen auf Landschaften. Klett-Perthes, Gotha. Bosch, J.M., Hewlett, J.D., A review of catchment experiments to determine the effect of vegetation changes on water yield evaporation. Journal of Hydrology 55, Bouwer, L.M., Aerts, J.C.J.H., Van de Coterlet, G.M., Van de Giesen, N., Gieske, A., Mannaerts, C., Evaluating downscaling methods for preparing Global Circulation Model (GCM) data for hydrological impact modelling. In: Aerts, J.C.J.H., Droogers, P. (Eds.), Climate Change in Contrasting River Basins. Adaptation Strategies for Water, Food and Environment. CABI Publishing, Wallingford, pp Bridge, J.S., Rivers and Floodplains. Forms, Processes and Sedimentary Record. Blackwell, Oxford. Brierley, J., Fryirs, K., Tributary-trunk stream relations in a cut-and-fill landscape: a case study from Wolumla catchment, New South Wales, Australia. Geomorphology 28, Brovkin, V., Bendtsen, J., Claussen, M., Ganopolski, A., Kubatzki, C., Petoukhov, V., Andreev, A., Carbon cycle, vegetation, and climate dynamics in the Holocene: experiments with the CLIMBER-2 model. Global Biogeochemical Cycles 16, doi: /2001 GB Brown, A.E., Zhang, L., McMahon, T.A., Western, A.W., Vertessy, R., A review of paired catchment studies for determining changes in water yield resulting from alterations in vegetation. Journal of Hydrology 310, Calder, I.R., Hydrologic effects of land use change. In: Maidment, D. (Ed.), Handbook of Hydrology. McGraw-Hill Inc, New York, pp Coulthard, T., Landscape evolution models: a software review. Hydrological Processes 15, Coulthard, T., Kirkby, M., Macklin, M., Modelling hydraulic sediment transport and slope processes, at a catchment scale, using a cellular automaton approach. Proc. GeoComputation '97 & SIRC '97, pp Otago, New Zealand. Davis, B., Brewer, S., Stevenson, A.C., Gujot, J., The temperature of Europe during the Holocene reconstructed from pollen data. Quaternary Science Reviews 22, Dawson, A.G., Hickey, K., Mayewski, P.A., Nesje, A., Greenland (GISP2) ice core and historical indicators of complex North Atlantic climate changes during the fourteenth century. The Holocene 17, de Moor, J., Verstraeten, G., Alluvial and colluvial sediment storage in the Geul River catchment (The Netherlands) combining field and modelling data to construct a Late Holocene sediment budget. Geomorphology 95, De Smedt, P., Paleogeografie en kwartair-geologie van het confluentiegebied Dijle-Demer. Acta Geographica Lovaniensia 11, 141. de Vente, J., Poesen, J., Arabkhedri, M., Verstraeten, G., The sediment delivery problem revisited. Progress in Physical Geography 31, De Wit, M.J.M., Buishand, T.A., Generator of rainfall and discharge extremes (GRADE) for the Rhine and Meuse basins. RIZA Report /. KNMI publication 218, RIZA/KNMI, Lelystad. Dearing, J.A., Jones, R.T., Coupling temporal and spatial dimensions of global sediment flux through lake and marine sediment records. Global Planet Change 39, Dotterweich, M., The history of soil erosion and fluvial deposits in small catchments of central Europe: deciphering the long-term interaction between humans and the environment a review. Geomorphology 101, Gellens-Meulenberghs, F., Gellens, D., L'évapotranspiration potentielle en Belgique: variabilité spatiale et temporelle. Institut Royal Meteorologique de Belgique, publication série A no Gentry, A.H., Parody, J.L., Deforestation and increased flooding of the Upper Amazon. Science 210, Glaser, R., Klimageschichte Mitteleuropas: 1000 Jahre Wetter, Klima, Katastrophen. Primus-Verlag, Darmstadt. Goosse, H., Fichefet, T., Importance of ice ocean interactions for the global ocean circulation: a model study. Journal of Geophysical Research Oceans 104, Goosse, H., Renssen, H., Timmermann, A., Bradley, R.S., Internal and forced climate variability during the last millennium: a model-data comparison using ensemble. Quaternary Science Reviews 24, Haan, C.T., Barfield, B.J., Hayes, J.C., Design Hydrology and Sedimentology for Small Catchments, XIV. Academic Press, San Diego, CA. 588 pp. Hancock, G., Lowry, J., Coulthard, T., Evans, K., Moliere, D., A catchment scale evaluation of the SIBERIA and CAESAR landscape evolution models. Earth Surface Processes and Landforms 35, Keesstra, S.D., The effects of natural reforestation on the hydrology, river morphology and sediment budget of the Dragonja River catchment, SW Slovenia. Ph.D thesis, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands. Lang, A., Hönscheidt, S., Age and source of colluvial sediments at Vaihingen-Enz, Germany. Catena 38, Lang, A., Bork, H.-R., Mäckel, R., Preston, N., Wunderlich, J., Dikau, R., Changes in sediment flux and storage within a fluvial system: some examples from the Rhine catchment. Hydrological Processes 17,

54 B. Notebaert et al. / Geomorphology 126 (2011) Leopold, L.B., Hydrology for Urban Land Planning, a Guidebook on the Hydrologic Effects of Urban Land Usesb:maintitle. Geological Survey Circular 554, U.S. Geological Survey, Washington. Magny, M., Bégeot, C., Guiot, J., Peyron, O., Contrasting patterns of hydrological changes in Europe in response to Holocene climate cooling phases. Quaternary Science Reviews 22, Mahe, G., Paturel, J.-E., Servat, E., Conway, D., Dezetter, A., The impact of land use change on soil water holding capacity and river flow modelling in the Nakambe River, Burkina-Faso. Journal of Hydrology 300, Milliman, J., Syvitski, J., Geomorphic tectonic control of sediment discharge to the ocean the importance of small mountainous rivers. Journal of Geology 100, Mitchell, T.D., Jones, P.D., An improved method of constructing a database of monthly climate observations and associated high-resolution grids. International Journal of Climatology 25, Naughton, F., Bourillet, J.-F., Sánchez Goñi, M., Turon, J., Jouanneau, J., Long-term and millennial-scale climate variability in northwestern France during the last 8850 years. The Holocene 17, Notebaert, B., Sensitivity of river systems to human actions and climatic events across different environments: a Holocene perspective. PhD-thesis. K.U. Leuven, Department Earth & Environmental Sciences, Leuven, Belgium, 334 pp. Notebaert, B., Verstraeten, G., Vandenberghe, D., Marinova, E., Poesen, J., Govers, G., in press. Changing hillslope and fluvial Holocene sediment dynamics in a Belgian loess catchment. Journal of Quaternary Science. Notebaert, B., Verstraeten, G., Rommens, T., Vanmontfort, B., Govers, G., Poesen, J., Establishing a Holocene sediment budget for the river Dijle. Catena 77, Opsteegh, J.D., Haarsma, R.J., Selten, F.M., Kattenberg, A., ECBILT: a dynamic alternative to mixed boundary conditions in ocean models. Tellus 50A, Peeters, I., Spatial modeling of sediment redistribution patterns on a millennial time scale. Ph.D. Thesis, K.U. Leuven, Department of Geography-Geology, Leuven, Belgium. Peeters, I., Van Oost, K., Govers, G., Verstraeten, G., Rommens, T., Poesen, J., The compatibility of erosion data at different temporal scales. Earth and Planetary Science Letters 265, Pfister, C., Schwarz-Zanetti, G., Wegmann, M., Winter severity in Europe: the fourteenth century. Climatic Change 34, Pongratz, J., Reick, C., Raddatz, T., Claussen, M., A reconstruction of global agricultural areas and land cover for the last millennium. Global Biochemical Cycles 22, Raymaeckers, F., La Rivière, J., Bogaert, J., Waterbeheersing in de Dijlevallei: techniek maakt plaats voor natuur. In: Bogaert, J., Water in natuurlijke hoeveelheden. De vrienden van Heverleebos en Meerdaalwoud, Jaarbulletin Leuven, Belgium. Reid, L., Dunne, T., Sediment budgets as an organizing framework in fluvial geomorphology. In: Kondolf, M., Piégay, H. (Eds.), Tools in Fluvial Geomorphology. Wiley, New York, pp Renssen, H., Goosse, H., Fichefet, T., Brovkin, V., Driesschaert, E., Wolk, F., Simulating the Holocene climate evolution at northern high latitudes using a coupled atmosphere sea ice ocean vegetation model. Climate Dynamics 24, Renssen, H., Lougheed, B.C., Aerts, J.C.J.H., De Moel, H., Ward, P.J., Kwadijk, J.C.J., Simulating long-term Caspian Sea level changes: the impact of Holocene and future climate conditions. Earth and Planetary Science Letters 261, Rommens, T., Verstraeten, G., Peeters, I., Poesen, J., Govers, G., Van Rompaey, A., Mauz, B., Packman, S., Lang, A., Reconstruction of late-holocene slope and dry valley sediment dynamics in a Belgian loess environment. The Holocene 17, Schumm, S.A., The Fluvial System. Wiley-Interscience, New York. 338 pp. Slaymaker, O., The sediment budget as conceptual framework and management tool. Hydrobiologia 494, Syvitski, J., Milliman, J., Geology, geography, and humans battle for dominance over the delivery of fluvial sediment to the coastal ocean. Journal of Geology 115, Tack, B., Van Den Bremt, P., Hermy, M., Bossen van Vlaanderen. Davidsfonds, Leuven. 320 pp. Thornthwaite, C.W., An approach toward a rational classification of climate. Geographical Review 38, Thornthwaite, C.W., Mather, J.R., Instructions and tables for computing potential evapotranspiration and the water balance. Publications in Climatology 10, Trimble, S.W., Decreased rates of alluvial sediment storage in the Coon Creek Basin, Wisconsin, Science 285, Van Oost, K., Govers, G., Desmet, P., Evaluating the effects of changes in landscape structure on soil erosion by water and tillage. Landscape Ecology 15, Van Rompaey, A., Govers, G., Data quality and model complexity for regional scale soil erosion prediction. International Journal for Geographical Information Science 16, Van Rompaey, A.J.J., Verstraeten, G., Van Oost, K., Govers, G., Poesen, J., Modelling mean annual sediment yield using a distributed approach. Earth Surface Processes and Landforms 26, Van Rompaey, A., Govers, G., Puttemans, C., Modelling land use changes and their impact on soil erosion and sediment supply to rivers. Earth surface processes and landforms 27, Vanwalleghem, T., Bork, H.R., Poesen, J., Dotterweich, M., Schmidtchen, G., Deckers, J., Scheers, S., Martens, M., Prehistoric and Roman gullying in the European loess belt: a case study from central Belgium. The Holocene 16, Verstraeten, G., Regional scale modelling of hillslope sediment delivery with SRTM elevation data. Geomorphology 81, Verstraeten, G., Van Oost, K., Van Rompaey, A., Poesen, J., Govers, G., Evaluating an integrated approach to catchment management to reduce soil loss and sediment pollution through modelling. Soil Use and Management 18, Verstraeten, G., Van Rompaey, A., Poesen, J., Van Oost, K., Govers, G., Evaluating the impact of watershed management scenarios on changes in sediment delivery to rivers? Hydrobiologia 494, Verstraeten, G., Poesen, J., Goossens, D., Gillijns, K., Bielders, C., Gabrield, D., Ruysschaert, G., Van Den Eeckhaut, M., Vanwalleghem, T., Govers, G., Belgium. In: Boardman, J., Poesen, J. (Eds.), Soil Erosion in Europe. John Wiley & Sons, Ltd, Chichester, pp Verstraeten, G., Rommens, T., Peeters, I., Poesen, J., Govers, G., Lang, A., 2009a. A temporarily changing Holocene sediment budget for a loess-covered catchment (central Belgium). Geomorphology 108, Verstraeten, G., Lang, A., Houben, P., 2009b. Human impact on sediment dynamics quantification and timing. Catena 77, Ward, P.J., A coupled climate-hydrological model for Meuse palaeodischarge modeling: set-up and calibration. Technical report July 2007, Department of Palaeoclimatology and Geomorphology, Faculty of Earth and Life Sciences, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands, adaptation/project/files/file/documents/report Meusepalaeo Methods.pdf. Ward, P.J., River Meuse suspended sediment yield: a new estimate and past estimates revisited. Netherlands Journal of Geosciences 87, Ward, P.J., Renssen, H., Aerts, J.C.J.H., Verburg, P.H., in press. Sensitivity of discharge andflood frequency to 21st Century and late Holocene changes in climate and land use (River Meuse, northwest Europe). Climatic Change, in press, doi: /s Ward, P.J., Aerts, J.C.J.H., De Moel, H., Renssen, H., Verification of a coupled climate-hydrological model against Holocene palaeohydrological records. Global Planetary Change 57, doi: /j.gloplacha Ward, P., Renssen, H., Aerts, J., van Balen, R., Vandenberghe, J., Strong increases in flood frequency and discharge of the River Meuse over the late Holocene: impacts of long-term anthropogenic land use change and climate variability. Hydrology and Earth System Sciences 12, Ward, P., van Ralen, R., Verstraeten, G., Renssen, H., Vandenberghe, J., The impact of land use and climate change on late Holocene and future suspended sediment yield of the Meuse catchment. Geomorphology 103, Williams, M., Deforesting the Earth: From Prehistory to Global Crisis. The University of Chicago Press, Chicago. Yang, D.W., Kanae, S., Oki, T., Koike, T., Musiake, K., Global potential soil erosion with reference to land use and climate changes. Hydrological Processes 17, Zolitschka, B., Behre, K.-E., Schneider, J., Human and climatic impact on the environment as derived from colluvial, fluvial and lacustrine archives examples from the Bronze Age to the Migration period, Germany. Quaternary Science Reviews 22,

55 JOURNAL OF QUATERNARY SCIENCE (2011) 26(1) ISSN DOI: /jqs.1425 Changing hillslope and fluvial Holocene sediment dynamics in a Belgian loess catchment BASTIAAN NOTEBAERT, 1 * GERT VERSTRAETEN, 1 DIMITRI VANDENBERGHE, 2 ELENA MARINOVA, 1,3 JEAN POESEN 1 and GERARD GOVERS 1 1 Department of Earth and Environmental Sciences, KU Leuven, Heverlee, Belgium 2 Vakgroep geologie en bodemkunde, Universiteit Gent, Gent, Belgium 3 Centre for Archaeological Sciences, KU Leuven, Heverlee, Belgium Received 22 December 2009; Revised 12 May 2010; Accepted 2 June 2010 ABSTRACT: Floodplain deposition is an essential part of the Holocene sediment dynamics of many catchments and a thorough dating control of these floodplain deposits is therefore essential to understand the driving forces of these sediment dynamics. In this paper we date floodplain and colluvial deposition in the Belgian Dijle catchment using accelerator mass spectrometric radiocarbon and optical stimulated luminescence dating. Relative mass accumulation curves for the Holocene were constructed for three colluvial sites and 12 alluvial sites. A database was constructed of all available radiocarbon ages of the catchment and this database was analysed using relative sediment mass accumulation rates and cumulative probability functions of ages and site-specific sedimentation curves. Cumulative probability functions of ages were split into different depositional environments representing stable phases and phases of accelerated clastic deposition. The results indicate that there is an important variation between the different dated sites. After an initial stable early and middle Holocene phase with mainly peat growth in the floodplains, clastic sedimentation rates increased from 4000 BC on. This first phase was more pronounced and started somewhat earlier for colluvial deposits then for alluvial deposits. The main part of the Holocene deposits, both in colluvial and alluvial valleys, was deposited during the last 1 ka. The sedimentation pattern of the individual dated sites and the catchment-wide pattern indicate that land use changes are responsible for the main variations in the Holocene sediment dynamics of this catchment, while the field data do not provide indications for a climatological influence on the sediment dynamics. Copyright # 2011 John Wiley & Sons, Ltd. KEYWORDS: dating; alluvium; colluvium; Holocene; land use change; climate change. Introduction Soil erosion, sediment transport, deposition and redistribution are important geomorphological processes during the Holocene in many temperate regions. A large number of studies have indicated the strong link between anthropogenic land use and soil erosion and colluvial deposition, while also climatic events may have a determining influence (e.g. review in Dotterweich, 2008). These driving forces can also have an important influence on sediment storage in floodplains (e.g. Trimble, 2009; Verstraeten et al., 2009a), while river systems and floodplains may also buffer the influence of such catchment perturbations for the downstream parts of the catchment (e.g. Knox, 2006; Walling et al., 2006). Over time spans of centuries to millennia, river systems are not only transporting sediment but also play an important role in the storage of sediments: floodplain sedimentation often makes up an important fraction of the total eroded sediment, and the combined amount of sediments stored as colluvium and in floodplains most often exceeds the amount of sediment exported from catchments of at least some square kilometres (e.g. Hoffmann et al., 2007; Rommens et al., 2006). In the perspective of the transition from nature-dominated to humandominated environmental changes during the Holocene (e.g. Messerli et al., 2000; Meybeck, 2003), the changing influence of anthropogenic land use and climatic variations on the sediment dynamics is of particular interest. Holocene climatic and land cover variations have caused large variations in sediment dynamics in many temperate regions, but the individual contribution of both factors to the variations in the sediment dynamics is often difficult to quantify. Establishing links between forcing factors and sediment deposition relies on an accurate and detailed dating control of *Correspondence: Bastiaan Notebaert, Division of Geography, Department of Earth and Environmental Sciences, KU Leuven Celestijnenlaan 200E, 3001, Heverlee, Belgium. bastiaan.notebaert@ees.kuleuven.be floodplain deposition. Dating of floodplain deposits is, however, often problematic, e.g. due to the unavailability of datable material or methodological problems with the used dating techniques (Verstraeten et al., 2009b). Dating of floodplain sediments is often based on radiocarbon dating, although there are recent developments in the dating of sediment chronologies using optical dating and the application of terrestrial nuclides (e.g. Lang, 2008). One of the major problems with radiocarbon dating is the unavailability of datable material in many floodplain deposits, leading to low temporal resolutions. Several approaches have been applied to enhance the use of radiocarbon dating: statistical analysis of radiocarbon dates (e.g. Macklin and Lewin, 2003; Hoffmann et al., 2008) and derivative parameters (e.g. Hoffmann et al., 2009), and the use of replicate samples for accuracy testing and Bayesian modelling of resulting probability distributions (e.g. Chiverell et al., 2009). Verstraeten et al. (2009b) stress the importance of establishing precise chronologies and the construction of temporally differentiated sediment budgets in order to derive the relationships between environmental variability and sediment dynamics. Such budgets are the accounting of sources, sinks and pathways of sediment (Reid and Dunne, 2003; Slaymaker, 2003). According to Foulds and Macklin (2006), sediment budgeting is a necessary tool to understand the role of land use change on catchment stability, as it identifies reach-scale zones of sediment transfer and storage. Several studies have constructed catchment-wide sediment budgets for relative short periods ranging from days to a few decades (e.g. Beach, 1994; Page et al., 1994; Fryirs and Brierley, 2001; Trimble, 1983, 1999; Walling and Quine, 1993; Walling et al., 2002, 2006), while other studies have concentrated on long-term sediment budgets (e.g. spanning the entire Holocene). Only few studies have constructed a time-differentiated sediment budget for periods covering at least the period since the introduction of agriculture (e.g. de Moor and Verstraeten, 2008; Trimble, 2009; Verstraeten et al., 2009a). Such time- Copyright ß 2011 John Wiley & Sons, Ltd.

56 FLUVIAL HOLOCENE SEDIMENT DYNAMICS IN A BELGIAN LOESS CATCHMENT 45 differentiated sediment budgets provide insight into the temporal and spatial variability in sediment fluxes, allowing more robust relationships between driving forces and the sediment dynamics. The objective of this paper is to date sediment deposition in alluvial and colluvial environments in the medium-scaled Dijle catchment (758 km 2 ), located in the central Belgian Loess Belt. Radiocarbon dating and optically stimulated luminescence (OSL) dating are used and evaluated. Based on the available dates, a catchment-wide pattern of sedimentation is derived and this is being qualitatively linked to environmental changes as driving forces. Finally, the available dates are combined with an existing Holocene sediment budget (Notebaert et al., 2009) in order to construct a time-differentiated sediment budget. This sediment budget can be used to establish quantitative relationships between the environmental changes and variations in sedimentation rates. Study area Within this study the Dijle catchment upstream of Leuven (758 km 2 ) is considered (Fig. 1). The catchment consists of a loesscovered undulating plateau ranging between 80 and 165 m above sea level (a.s.l.), in which river valleys have incised. There is a long history of land use, with first traces of agriculture from the Atlantic Period (ca BC), and intensive anthropogenic land use during the Roman Period and from the Middle Ages on. A more detailed description of the catchment can be found in Notebaert et al. (2009). The fluvial architecture of the Dijle catchment indicates that overbank deposition in floodplains is the most important Holocene process (Notebaert, 2009). Floodplain and backswamp sediments are loamy and silty, while point bar and river bed sediments are sandy. The sandy sediment which is related to the lateral migration of the riverbed is limited to parts of the floodplain, and the majority of the floodplain consists of a continuous aggradation profile. Methods Corings were located along cross-sections through the floodplain (Notebaert, 2009), and cross-sections and corings were selected for dating using several criteria: representativeness according to the local fluvial architecture, suitability to date floodplain aggradation and availability of datable material. Accelerator mass spectrometry (AMS) radiocarbon dating and OSL dating were combined and typically one coring per cross section was dated. Corings were selected based on their suitability to provide insight into net floodplain aggradation: they lack traces of erosional boundaries or river channel and point bar deposits, and thus represent continuous Holocene aggradation profiles. Based on the available OSL and radiocarbon ages, sedimentation curves are constructed for the individual dated sites. The available sedimentation rates were transformed into mass accumulation rates, giving the mass accumulation per floodplain or colluvial valley area, by applying a bulk density value and correcting for the presence of organic material (see below). Next, these mass accumulation curves where normalised by the total Holocene deposited mass at the study site in order to allow direct comparison between the different dated sites. AMS radiocarbon dating Samples for radiocarbon dating were taken using hand augering, percussion drilling and profile pits. In this study in total 12 sites were dated using radiocarbon and OSL dating, ranging from small colluvial valleys to the main alluvial plain. Colluvial deposits at two sites were dated in detail by Rommens et al. (2007) and Rommens (2006), yielding 12 dates. In this study detailed dating was performed at the Bilande site, with 18 radiocarbon dates. Dating of floodplain deposits is available from the Nethen catchment from Mullenders et al. (1966) and Rommens et al. (2006), with three and 10 radiocarbon dates respectively. De Smedt (1973) dated peat accumulation at one location in the main Dijle valley using five radiocarbon dates. Within this study, 44 samples were dated: one site (Korbeek- Dijle) was dated in detail, while for 10 other sites at least two samples were dated. In total 29 radiocarbon dates from colluvial deposits and 62 from alluvial deposits are available (see Fig. 1 for sampling locations). Except for the bulk samples, each sample was sieved and datable material was handpicked from the fraction larger than 125 mm. Terrestrial plant remains and wood were selected for dating, in order to avoid hard-water effects (e.g. Moor et al., 1998) and reservoir effects (e.g. Törnqvist et al., 1992). Resulting conventional radiocarbon Figure 1. Location of the Dijle catchment within Belgium and sampling sites for radiocarbon and OSL dating in the Dijle catchment. B, Brussels; L, Leuven. Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

57 46 JOURNAL OF QUATERNARY SCIENCE ages (BP) were calibrated using Oxcal 4.1 (Bronk Ramsey, 2001, 2009) and the Intcal 04 calibration curve (Reimer et al., 2004). Calibrated ages are here referred to as cal a BP (calibrated years before 1950) or as cal a BC/AD or BC/AD, and uncalibrated ages as 14 C a BP. OSL dating OSL dating was used for the floodplain of the Dijle River near Korbeek-Dijle. Sampling of OSL datable material was carried out with percussion drilling equipment, using PVC tubes of length 1 m and an inner diameter of 5 cm. The PVC tubes were opened in the laboratory under subdued orange light. Samples for luminescence analysis were taken about every 13 cm, with each collected sample representing a vertical interval of 5 8 cm. In this manner, 32 samples were collected, of which 15 were selected for dating in the frame of this work. The samples were relatively fine grained, with grains in the range of mm generally representing the coarsest fraction that was still practicable for luminescence analysis. Luminescence analysis was performed at Ghent University. Quartz grains from this fraction were extracted using conventional sample preparation techniques (HCl, H 2 O 2, sieving, heavy liquids, HF). A sufficient amount of quartz grains could be extracted from 11 of the 15 samples. The purity of the quartz extracts was confirmed by the absence of a significant infrared stimulated luminescence (IRSL) response at 608C to a large regenerative beta dose. The sensitivity to infrared stimulation was defined as significant if the resulting signal amounted to more than 10% of the corresponding blue light stimulated luminescence (BLSL) signal (Vandenberghe, 2004) or if the IR depletion ratio was not within 10% of unity (Duller, 2003). For measurement, quartz grains were spread out on the inner 7 mm of 9.7 mm diameter stainless steel discs. The luminescence measurements were performed using an automated Risø-TL/OSL- Da-15 reader, equipped with blue ( nm) LEDs and IR (875 nm) diodes; all luminescence emissions were detected through a 7.5 mm thick Hoya U-340 UV filter. Details on the measurement apparatus can be found in Bøtter-Jensen et al. (2003). The equivalent dose (D e ) was determined using the singlealiquot regenerative-dose (SAR) protocol (Murray and Wintle, 2000). Optical stimulation with the blue diodes was for 38 s at 1258C; the initial 0.3 s of stimulation was used in the calculations, minus a background evaluated from the following 0.5 s of stimulation. A preheat of 10 s at 1808C and a test-dose cut heat to 1608C were adopted; the preheat at 1808C was chosen to minimise thermal transfer (i.e. transfer of charge by heating from light-insensitive but thermally stable traps, into light-sensitive traps). After the measurement of each test dose signal, a high-temperature bleach was performed by stimulating with the blue diodes for 40 s at 2808C (Murray and Wintle, 2003). Measured aliquots were accepted if the recuperation and IRSL/OSL ratio did not exceed a threshold set at 10%, and if both the recycling ratio and the IR depletion ratio were within 10% of unity. The suitability of the experimental procedure was confirmed through a dose recovery test as outlined by Murray and Wintle (2003; see also Table 1). The overall (n ¼ 11) average recovered to given dose ratio ( 1 standard error, s) is , which gives confidence in the reliability of the laboratory measurement procedure. For each sample, at least 15 replicate measurements of D e were made. The dosimetry (i.e. beta and gamma dose rates) is based on low-level gamma ray spectrometric analysis of sediment collected above and below each OSL sample, and used dose rate conversion factors derived from the nuclear energy releases tabulated by Adamiec and Aitken (1998). Water contents were Table 1. Summary of radionuclide activity concentrations, estimates of the time-average water content (w.c.), calculated dose rates, De values, optical ages and dose recovery data for the OSL samples of the Korbeek- Dijle site in the Dijle catchment. The uncertainties mentioned with the dosimetry, De and dose recovery data are random; the uncertainties on the optical ages represent the total uncertainty, which includes the contribution from systematic sources of error (Aitken, 1985). All uncertainties are 1s. The dose rate includes the contribution from cosmic rays and internal radioactivity. The number of aliquots used in luminescence analyses (D e determination and dose recovery) is given in parentheses. Columns headed All aliquots refer to values obtained by averaging the D e over all aliquots; columns headed Aliquots rejected refer to values obtained by calculating De following the procedure based on Fuchs and Lang (2001; see text for details). The dose recovery test consisted of bleaching fresh aliquots using blue diodes (2 250 s with a 10 ks pause in between), giving them a dose close to the expected natural dose, and measuring them using the SAR protocol. Aliquots rejected All aliquots De (Gy) Age (ka) Dose recovery De (Gy) Age (ka) Dose rate (Gy ka 1 ) w.c. (%) 40 K (Bq kg 1 ) 232 Th (Bq kg 1 ) 210 Pb (Bq kg 1 ) 226 Ra (Bq kg 1 ) 238 U (Bq kg 1 ) Depth (cm) GLL code (15) (7) (6) (17) (11) (6) (18) (3) (5) (17) (4) (6) (16) (8) (6) (17) (9) (4) (16) (11) (5) (17) (6) (6) (19) (3) (6) (18) (16) (6) (19) (16) (6) Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

58 FLUVIAL HOLOCENE SEDIMENT DYNAMICS IN A BELGIAN LOESS CATCHMENT 47 derived from laboratory measurements of saturated samples. The uppermost sample, taken at 0.3 m depth (GLL , see Table 1), was assumed to have been saturated with water for 50% of its burial period; for the remainder of the samples, the time-averaged moisture content was estimated to be equal to 90% of the saturation content. The contribution from cosmic rays was calculated following Prescott and Hutton (1994), and an internal dose rate of Gy ka 1 was adopted (Vandenberghe et al., 2008). Sediment mass accumulation rates Sedimentation rates (SR i,obs ) are normally calculated for each sample of the radiocarbon database using the following equation: SR i;obs ¼ðD i D i-1þ=ðt i T i-1þ (1) with D i and T i the depth and age of the sample and D i 1 and T i 1 the age and depth of the stratigraphic following (younger) sample. However, this approach has some drawbacks. Firstly, the above calculated sedimentation rates represent total floodplain accumulation rates, including both clastic sedimentation as well as organic sedimentation (peat growth and organic-rich sediments). Especially for the Dijle floodplain, where large variations occur in the organic matter concentrations within and between corings, a correction needs to be made in order to obtain clastic sediment mass accumulation rates. Secondly, at some floodplain locations more sediment has been deposited during the entire Holocene than at other locations, thus resulting in higher sedimentation rates. This site dependence on sedimentation rates makes it troublesome to identify periods with higher or lower sedimentation rates, especially when available sediment datings are retrieved from a variety of locations. Therefore relative Holocene floodplain mass accumulation rates were calculated. These are defined as the fraction of sediment mass per unit floodplain area deposited after the moment of sample deposition compared to the total Holocene sediment accumulation for the considered coring. The use of relative amounts allows a more straightforward comparison of corings with different total Holocene sediment accumulation. The resulting relative sedimentation rates were plotted vs. age for the different dated corings, giving a first insight into the sedimentation history of the catchments. Separate relationships were constructed for colluvial and alluvial ages, the last group containing both dates from floodplain fines and peat and organic rich layers but excluding the river channel and point bar deposits. In order to calculate these relative sediment mass accumulation rates, floodplain sediments were first divided into three units. Comparable to Rommens et al. (2006), soil units were in-field classified into three OM classes: clastic deposits, organic deposits and peat layers. The thickness of these units was converted into mass of sediment using (Verstraeten and Poesen, 2001) 1 M layer;i ¼ d layer;i (2) 1 OM þ DBD MS ð1 OMÞDBD OM with M layer,i the sediment mass (mg) per m 2 floodplain area of layer i, d layer,i the thickness of layer i (m), OM the percentage organic matter of the unit to which the layer belongs, DBD OM the dry bulk density of the organic matter (mg m 3 ) and DBD MS the dry bulk density of the clastic component (mg m 3 ). Values for OM were determined for 14 samples by using the combustion method (e.g. Bisutti et al., 2004) combined with data from Rommens et al. (2006), whereas the values for DBD were taken from Rommens et al. (2006) (DBD OM ¼ 0.35 mg m 3 ; DBD MS ¼ 1.42 mg m 3 ). Based on the laboratory results, average OM% values of 4%, 8% and 70% were defined for clastic deposits, organic deposits and peat layers. Although we recognise that the use of such averaged values introduces an additional error, we believe that these corrections are necessary to come to realistic accumulation rates. Additionally, the total mass per unit floodplain area of sediment that was deposited after the deposition of the dated material (M sample, mg m 2 floodplain area) was calculated for each sample using M sample ¼ Xn i¼1 M layer;i (3) with n the number of alluvial sedimentary layers situated above sample s, and deposited since deposition of the sample up to the present, and M layer i the mass of layer i (mg m 2 floodplain area) calculated using equation (2). The entire mass per unit floodplain area (M coring in mg m 2 floodplain area) can also be calculated for each coring. The relative mass (M relative, %) of sediment accumulation after the deposition of the sample up to the present can be compared to the mass of the total Holocene sediment accumulation using M relative ¼ M sample =M coring (4) The relative mass accumulation rate (MR, % a 1 ) for sample i can then be calculated using MR i ¼ðM relative;i M relative;i-1þ=ðt i T i-1þ (5) with T i the time since deposition, and index i 1 corresponding to the stratigraphic younger sample deposited above this deposit. In order to compare samples from different corings with different sampling and dating densities, the relative mass accumulation and sedimentation rate are calculated compared to the present, thus setting T i 1 ¼ 0, from which follows M relative,i 1 ¼ 0 and D i 1 ¼ 0. Hoffmann et al. (2009) calculated sedimentation rates and applied models to evaluate these sedimentation rates. Therefore we also compared our MR curves with modelled values, using five scenarios. Based on the analysis results, each scenario starts with an MR of 0.001% a 1 at BC, followed by a linear increase of the MR from a given date to the present. The linear increase is fitted in such a way that the total accumulated relative mass accumulation (M relative,total ) equals 100%: M relative;total ¼ X0 i¼ MR i ¼ 100% (6) with i the considered age with an interval of 1 AD. As a result, the maximal MR i will depend upon the onset of the increase in sedimentation. Scenarios were simulated with a linear increased MR starting at 2000 BC, 1000 BC, 1 AD, 500 AD and 1000 AD. The scenarios were evaluated using the Nash and Sutcliffe model efficiency (ME) statistic (Nash and Sutcliffe, 1970). The applied scenarios all yield an identical total mass accumulation (100%), while the early Holocene mass accumulation is also constant (0.001% a 1 ), which means that the magnitude of the accumulation for the most recent periods will depend on the length of the period with increased sedimentation. This is in contrast to the scenarios calculated by Hoffmann et al. (2009) which use a constant sedimentation rate of 0.03 cm a 1 for the early and mid Holocene period, followed by a linear increase to 0.4 cm a 1, with a start of this increase at 1000 BC, 500 AD or 1800 AD. Depending on the moment of the start of the linear increase, the total sedimentation (mm) will vary, with a larger total sedimentation when the increase starts earlier. Our method avoids the differences in total sediment accumulation between the different models. Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

59 48 JOURNAL OF QUATERNARY SCIENCE Cumulative probability distributions Recently, several studies use cumulative probability distributions (CPDs) of radiocarbon ages obtained from alluvial and/or colluvial deposits for the analysis and comparison of Holocene fluvial dynamics within Europe. Techniques were compiled and applied on data from Great Britain (e.g. Macklin and Lewin, 2003; Lewin et al., 2005) and later applied in Poland (Starkel et al., 2006), Spain (Thorndycraft and Benito, 2006a,b), Germany (Hoffmann et al., 2008) and in the Rhine catchment (Hoffmann et al., 2009). A comparison is also made between datasets obtained from Great Britain, Poland and Spain (Macklin et al., 2006). For the analysis of German datasets (Hoffmann et al., 2008) several methodological improvements were introduced. The radiocarbon ages used are grouped in a different way: where Lewin et al. (2005) focus mainly on changes in fluvial style or sedimentation rates, Hoffmann et al. (2008) calculate frequency distributions of ages which are yielded from deposits corresponding to active sedimentation phases. Hoffmann et al. (2008) argue that this results in a better proxy of the response to external impacts. The frequency distributions of German dates were also normalised, which limits the effects of the calibration curve, different preservation potentials and sampling bias. Recently Macklin et al. (2010) further improved the analysis of the frequency distributions of radiocarbon ages from the UK. In this study a database of radiocarbon ages from the Dijle catchment was compiled, holding all available radiocarbon dates from colluvial and alluvial deposits, based on this study and previous studies (Mullenders et al., 1966; De Smedt, 1973; Rommens, 2006; Rommens et al., 2006). Radiocarbon ages which are considered problematic in the original publication are not included in the database. The ages of this database were grouped and for the different groups cumulative probability distributions (CPDs) were constructed by the addition of the probability functions of the individual calibrated radiocarbon ages. A distinction was made between different sedimentary facies (e.g. Hoffmann et al., 2008): colluvial deposits, floodplain fines (overbank deposits), river channel and point bar deposits, and peat and organic-rich layers. Samples from channel and point bar deposits are not included in this analysis due to the low number of dated samples (Table 2). These facies types were then again grouped in terms of their fluvial activity (Table 3; based on Hoffmann et al., 2008): the first two are considered to represent phases of active colluvial or alluvial sediment deposition; the last one is considered to represent phases of relative stability in the clastic floodplain accumulation phases. Most sampling of peat and organic-rich layers occurred at the bottom or the top of these layers, in order to date the transition between clastic and organic floodplain accumulation, which can bias the distributions of peat ages. The shape of the calibration curve used also has an influence on the resulting CPDs. This effect is illustrated by the CPD of 100 equally spaced ages with an (uncalibrated) standard Table 2. Groups of sedimentary facies, available number of radiocarbon ages and number of radiocarbon ages used for the analysis of radiocarbon ages in the Dijle catchment. Sedimentary facies Number of 14 C ages Number of ages used Colluvial deposits Floodplain fines (overbank deposits) River channel and point bar deposits 1 1 Peat and organic rich layers Top of peat layer Bottom of peat layer Total Table 3. Different activity groups, available number of radiocarbon ages and number of radiocarbon ages used for the analysis of radiocarbon ages in the Dijle catchment. Activity group Number of 14 C ages deviation of 45 a. The used standard deviation of 45 a equals the average standard deviation of the radiocarbon ages in the database. Each of the calculated CPDs can be divided by the CPD of equally spaced ages in order to correct for the effects of calibration curve (Hoffmann et al., 2008). Alternatively, a correction can be made by dividing a CPD by the CPD of all available ages (Hoffmann et al., 2008). As preservation potential does not present a problem for the samples of the Dijle catchment due to the sampling techniques, where only samples were taken from corings with a continuous aggradation profile during the Holocene, this method was not applied here. Only for colluvial depots might the preservation potential play a role, as sedimentary evidence for intact aggradational colluvial archives is not available. Apart from the problems with the calibration curve and the preservation potential, other limitations may arise with the methodology used (Hoffmann et al., 2008), related to the variable precision of radiocarbon ages and the possible time difference between the age of deposition and the age of the dated material. Time-differentiated sediment budget Number of used ages Aggradation Stability A Holocene sediment budget was previously constructed for the Dijle catchment (Notebaert et al., 2009). A detailed description of the methodology can be found in Notebaert et al. (2009). The calculations were adapted in this study to the more accurate measurements of organic matter content of the peat layers which are mentioned above. In order to get a deeper insight into the sediment dynamics of a catchment, it is important to incorporate also the temporal dynamics of the system. Based on the available ages, and comparable with other studies (e.g. Trimble, 1999, 2009; de Moor and Verstraeten, 2008; Verstraeten et al., 2009a), the sediment budget was differentiated in three time periods: early Holocene to 2000 BC, 2000 BC to 1000 AD and 1000 AD to present. These periods were selected based on the sedimentation history of the catchment. A more detailed time differentiation of the sediment budget would require a more detailed dating of the alluvial and colluvial deposits. The time differentiation of sediment deposition is based on the relative mass accumulation rate (MR, equation (5)) for the dated sites. Based on the site-specific values, a weighted average mass accumulation for each period was calculated for the trunk valley, tributary valleys and the colluvial valleys. Only (dated) sites were used where a date was available within a time frame of ca a (for 2000 BC) or ca. 500 a (for 1000 AD) from the starting date of the considered time periods. Sites where no dates for these start points are available but where the uncertainty on the relative accumulated mass for these points is smaller than 10%, e.g. due to a small relative mass accumulation increase between two available dates, are also used. In this way the averaging effects between the dated points are minimised. Weighting was based on the floodplain width and Holocene sediment thickness. The selecting of sites and weighting result in a difference in the catchment averaged sedimentation curve compared to a pure mathematical curve. Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

60 FLUVIAL HOLOCENE SEDIMENT DYNAMICS IN A BELGIAN LOESS CATCHMENT 49 Dating information is only available for deposition environments, while temporal information on the sediment sources (erosion rates) and catchment export is absent. In order to come to a catchment-wide budget incorporating these sources and sinks, some assumptions are required: 1. Verstraeten et al. (2009a) assume that catchment export follows the same trends as floodplain storage. They state that sediment deposition is controlled by inundation frequency and sediment concentration factors which also control sediment yield. As a result, periods with higher floodplain storage will also be periods with higher export from the catchment. For the Dijle catchment it is demonstrated that floodplain storage was very low during the early and mid Holocene, while during this same period a fixed river channel is absent and water is transported in a diffuse way over a broad floodplain, without forming typical river bed deposits (De Smedt, 1973). Such a diffuse water transport is most probably accompanied by very low sediment export rates. 2. For each period mass is preserved. As a result the sediment production equals the sum of the export and the colluvial and alluvial storage. 3. As no dates are available for sediment deposition on the hillslopes (excluding colluvial valley deposition), the temporal framework for this sink should be based on some assumptions. This hillslope sediment deposition can be assumed to follow an equal trend as sediment deposition in the colluvial valleys. This is the first scenario ( scenario 1 ) that is calculated. For the Nethen catchment, however, Verstraeten et al. (2009a) argue that this would result in an unrealistic sediment delivery ratio of nearly 100% for the period before colluvial deposition starts (early Holocene). Modelling results in the Geul catchment (de Moor and Verstraeten, 2008) show indeed a high sediment delivery ratio for the early and mid Holocene, but still much lower than 100%. Therefore, scenario 2 was developed by assuming that sediment deposition on the slopes is assumed to follow the same trend as the combined storage in colluvial and alluvial deposits, an approach which yielded more realistic values in the Nethen catchment according to Verstraeten et al. (2009a). Results Dating floodplain and colluvial accumulation In this study the colluvial valley at Bilande (for location see Fig. 1) was dated in detail. In total 16 samples were dated using radiocarbon dating and sample depth is plotted against calibrated age in Fig. 2. Radiocarbon dating performed on charcoal samples or samples containing both charcoal and terrestrial plant remains or wood is problematic: six out of eight samples show an age depth inversion. Possibly the two lower charcoal-based dates suffer also an age depth inversion, but sampling and dating resolution is insufficient for conclusions. The overestimation of age for those charcoal dates can easily be explained by the cascade model of colluvial formation (Lang and Hönscheidt, 1999). This implies that older colluvial deposits have been eroded, and material from these older colluvial deposits is thus reworked and again deposited at their current location. The dating from the Bilande site shows two important sedimentation phases and one stability phase: a first sedimentation phase starts around 2000 cal a BC and is followed by the formation of an organic layer of which the top is dated around 1 AD. Parts of this first depositional phase show layers, indicating Figure 2. Overview of the coring data at Bilande (Dijle catchment) with radiocarbon dates. Sampling depths and calibrated radiocarbon ages are plotted on the right. Boxes indicate the calibrated radiocarbon ages with a 2s error: black boxes are ages based on charcoals; grey boxes are ages based on terrestrial plant remains and wood. The dashed line indicates the suggested sedimentation history. The width of the boxes in the left column varies with the texture. The presence of two boxes indicates an alternation of fine layers with different texture. Layering is indicated by a light-grey filling; organic deposits are indicated by dark-grey filling. that it does not originate from a single event. After the formation of the organic layer again clastic sediment was deposited. Between 1 AD and about 1000 AD, sedimentation rate appears to be low, although this can be caused by the limited dating resolution, which causes an averaging effect. A major sedimentation phase occurred between 1230 AD and 1295 cal a AD (1s calibrated interval, cal a AD on a 2s interval), with the deposition of at least 1.87 m sediment. Some parts of these deposits show layering, including small (1 3 mm) layers containing plant remains and traces of in situ growing plants, indicating that this depositional phase is not caused by a single event. Due to the absence of dates in the upper part of the coring, the extent of this major deposition phase and the subsequent deposition rates cannot be derived. Drainage tubes from ca AD are located at a depth of m, indicating that deposition in the last 150 a was less than 0.2 m. Dating results of the floodplain of the Dijle near Korbeek- Dijle (for location see Fig. 1) are represented in Fig. 3. Table 1 summarises the results from D e and dose rate determination of the OSL dating, and shows the calculated optical ages (see also Fig. 4, squares). It can be seen that not all age results are consistent with the stratigraphic position of the samples. We interpret this inconsistency as indicative for incomplete resetting of the OSL clock in the quartz grains. A large spread in D e values was observed for the majority of the samples, with relative standard deviations ranging from 20% (sample GLL ) to 50% (sample GLL ). This also indicates that the aliquots consist of a mixture of grains which have been reset to various degrees. Interestingly, the variability is already Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

61 50 JOURNAL OF QUATERNARY SCIENCE Figure 3. Overview of the coring data at Korbeek-Dijle (Dijle catchment) with radiocarbon and OSL dates. Peat layers are indicated by grey shading. Boxes indicate ages with a 2s error: black boxes indicate radiocarbon ages; light-grey boxes indicate corrected OSL ages. A radiocarbon date obtained from charcoal is indicated by 1. The dashed line represents the proposed sedimentation history. observed for large (i.e. 7 mm diameter) aliquots, which may imply that only a small fraction of the grains contributes to the measured luminescence. The D e values for the three lowermost samples (samples GLL , -49 and -67) have a higher accuracy, with values of the relative standard deviation in the range of 9 13%. For sample GLL , the relative standard deviation of the D e data is 16%. As all samples exhibit a similar overall luminescence sensitivity, this variability in precision indicates that resetting in the lowermost samples was more homogeneous (and possible more complete), and/or that they contain more grains that contribute to the measured signal, with each grain emitting less intense luminescence (so that grain-tograin variations are averaged out to a greater extent). As the uppermost 4 5 m of sediment is probably derived from the same source, samples GLL to -49 are expected to exhibit the same material characteristics. Therefore, the value of the relative standard deviation is taken as a measure for the degree of resetting. Fuchs and Lang (2001) suggested a procedure to improve D e determination in incompletely reset samples using a limited amount of replicate measurements. This procedure consists of arranging the D e values in ascending order, and calculating a running mean (starting with the two lowermost D e values) until Figure 4. Plot of age vs. depth for the OSL ages at Korbeek-Dijle (see also Fig. 3). The error bars represent the overall random uncertainty (1s) only. As such, the figure allows evaluation of the internal consistency of the optical ages. The squares refer to ages based on the mean D e over all measured aliquots; the circles refer to the ages obtained after rejecting D e values following a procedure similar to the one proposed by Fuchs and Lang (2001; see text for details). This figure is available in colour online at wileyonlinelibrary.com. the relative standard deviation exceeds the value of what is defined as the precision of the method. We applied this procedure to our dataset, with the precision of the measurement method being defined as that observed in the dose recovery tests. The resulting ages are shown in Fig. 4 (circles; see Table 1 for the analytical data). It can be seen that the OSL ages are now stratigraphically more consistent, with the dataset containing only one clear outlier (sample GLL ). Concerning the D e selection procedure outlined above, it should be pointed out that (i) it may yield ages that still overestimate the true burial age (as they are derived from OSL signals originating with more than one grain), (ii) it assumes that the precision obtained in measurements of artificially bleached and irradiated aliquots accurately represents the precision that would be observed for well-bleached natural samples, and (iii) it assumes that also a very limited number of replicate measurements yields a reliable estimate for the burial dose. These considerations should be taken into account when interpreting the age results and limit the conclusions that can be drawn. The ages obtained for the three lowermost samples (GLL and -49 situated in overbank or backswamp silts and GLL situated in the braided river sandy deposits) are based on the most reproducible measurements of the D e. The date of ka for sample GLL confirms the presumed Late Pleistocene age of the sandy braided river deposits below the peat layer. The set of OSL ages for the overlying unit suggests that it was deposited in at least two distinct phases. A first phase comprises the sediments at a depth of 4 m to 3 m, which were deposited between ka (sample GLL ) and before 1 ka ago (sample GLL ). A second phase is characterised by a significantly higher sedimentation rate, as the uppermost 3 m of sediment must have been deposited in less than 1 ka. We hypothesise that more heterogeneous resetting of the luminescence signal in the uppermost samples gives an indication for a high sedimentation speed. A high sedimentation speed lowers the probability for all grains to have their luminescence signal reset at the moment of deposition. The depositional and sedimentary history of the Korbeek- Dijle site shows formation of a basal peat layer between 8000 BC and 500 BC. It is unclear whether a second date of the top of the peat layer giving 5000 BC is correct, as this date is obtained from a bulk sample and reworking of peat cannot be Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

62 FLUVIAL HOLOCENE SEDIMENT DYNAMICS IN A BELGIAN LOESS CATCHMENT 51 excluded. A subsequent sedimentation phase ends around 1100 AD, with a major sedimentation period around 500 AD. Of the three samples yielding ages of 500 AD, the deepest one is yielding the oldest uncalibrated age, but they are overlapping on a 2s calibrated interval. The top of a second peat layer is dated by radiocarbon dating around 900 AD and by OSL dating around 1200 AD. The OSL age is considered to be more accurate: the radiocarbon date is obtained from a bulk sample, with possibly age overestimation due to reservoir effects and radiocarbon dates not obtained from bulk samples of comparable peat layers at other locations yield ages between 1000 and 1400 AD. The main sedimentation phase at Korbeek- Dijle is situated after the formation of this peat layer. This phase suffers from a low dating resolution. However, we discussed that the heterogeneity of the luminescence signal resetting gives an indication for high sedimentation rates. Additionally, when assuming that the error on the luminescence ages with rejected aliquots gives a good indication of the real age, it can be argued that sedimentation is concentrated around ka ago, with the deposition of m sediments. This implies that during the last 500 a only about 0.6 m of sediment is deposited. Catchment-wide pattern of floodplain and colluvial deposition Figure 5 plots the age and relative accumulated mass per unit depositional area for the colluvial sites, whereas Fig. 6 provides this information for alluvial sites. Based on Fig. 6 the average accumulated sediment mass for the tributaries and main trunk floodplain was calculated (Fig. 7). The calculated rates are not corrected for the dating resolution of the different sites and are therefore influenced by averaging effects. Data for the colluvial sites show the start of deposition between 4160 BC and 600 BC, while the main accumulation of sediment mass occurred for all sites in the last 3 ka (Fig. 5). During the last ca. 1 ka between 60% and 30% of the sediment mass accumulation took place. As for most alluvial sites only two or three dates are available, these curves suffer largely from an averaging effect and interpretation should consider this averaging. Within the alluvial graphs (Fig. 6) one major outlier is apparent: at the site Bonlez U almost all sedimentation took place before ca BC. After the early and mid Holocene aggradation phase the river incised the floodplain to a depth of 6 m (Notebaert, 2009). The only dating control on this incision phase is the age of the top of the floodplain (ca BC). This site has an important influence Figure 6. Age and relative accumulated sediment mass for the different dated corings in alluvial deposits in the Dijle catchment. Data for Rotspoel are obtained from De Smedt (1973). A value of 70% is used for the percentage of organic matter in the peat layers. on the average values calculated for the tributary floodplains (Fig. 7). For sites where the first Holocene floodplain accumulation is dated, this phase is situated between and 8000 BC. Mass accumulation before 500 BC is rather limited for most sites, although there is a wide scatter. Most sites show a first increase in sediment accumulation between 4000 BC and 500 BC, with again a high scatter. A second increase in sediment accumulation is observed for most sites around 1000 AD. Due to the averaging effect and the rather low resolution of the dates, it is not possible to identify short-term variations in sediment accumulation, and thus it is not clear how long the sedimentation phases lasted, or what the intensity of the phases was. The possibility of intermediate periods with less accumulation cannot be excluded. The presence in most crosssections of a peat layer within the upper clastic layer (e.g. Korbeek-Dijle site, see above) suggests a stable phase. Ages for this peat layer range between 700 and 1400 AD. The average relative sediment mass accumulation (Fig. 7) provides an overview of the catchment-wide sedimentation patterns. These averages are, however, still largely influenced by single dated cross-sections, as indicated by the influence of the Bonlez U cross-section. These averaged values indicate that Figure 5. Age and relative accumulated sediment mass for the different dated corings in colluvial deposits in the Dijle catchment. Data for the Nodebais are reported by Rommens et al. (2006) and data for the Beauvechain site by Rommens (2006). Figure 7. Age and average relative accumulated sediment mass for the different dated corings in alluvial deposits in the Dijle catchment. Average values are calculated for the main trunk valley, tributaries, tributaries excluding the dates from the Bonlez U cross-section and for all cross-sections. A value of 70% is used for the percentage of organic matter in the peat layers. Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

63 52 JOURNAL OF QUATERNARY SCIENCE deposition in the trunk valley floodplain is mainly concentrated in the last 1 2 ka. Deposition in the tributary floodplains increased earlier than in the trunk valley floodplain, with an important increase starting around 2000 BC. This pattern is, however, influenced by averaging effects caused by the dating resolution. To make a correct interpretation of the sedimentation history for the trunk valley floodplains, for each time frame only sites should be included with a sufficient temporal dating resolution for the considered period. Analysis of sedimentation rates Figure 8. Sedimentation rates (SR) and relative mass accumulation rates (MR) plotted vs. depth for the available radiocarbon dates in the Dijle catchment. (A) Sedimentation rates; (B) relative mass accumulation rates; (C) relative mass accumulation rates including the different modelled scenarios. Table 4. Model efficiencies (ME) for the different scenarios for relative mass accumulation rates compared to calculated relative mass accumulation rates based on radiocarbon ages of the Dijle catchment. Scenario: increase from ME colluvial dates ME alluvial dates ME alluvial dates excluding Bonlez U 2000 BC BC AD AD AD The plots of sedimentation rates (SR) and the relative mass accumulation rates (MR) (Fig. 8) show rather constant SR and MR values before 4000 BC, slightly increasing values between 4000 BC and 1000 BC, and then more increasing values after 1000 BC, while also the scatter increases. Model efficiency values for several sedimentation rate scenarios were compared with these plots (Table 4). For the fluvial ages ME was also calculated excluding the dates of Bonlez U (Table 4; see above). ME is highest for the model with increased colluvial sedimentation starting at 1000 BC and alluvial sedimentation at 1 AD. However, from the pattern of floodplain and colluvial deposition it is clear that such scenarios are an oversimplification of the real sedimentation history as these do not incorporate several phases of increased or decreased sedimentation. Nevertheless, the models give an indication of a time lag between the start of increased sedimentation in colluvial and alluvial deposits. An increase in sedimentation rate with a decreasing measuring period is reported for many sedimentary deposits, and this phenomenon has been linked to a time dependency in sediment accumulation rates (e.g. Sadler, 1981; Schumer and Jerolmack, 2009). This is being linked to unsteady discontinuous sedimentation. As such, the observed increase in sediment deposition cannot only be explained by a true increase due to variations in driving forces, but also by an apparent increase caused by such scaling effects. As there is no systematic sampling for shallower or deeper Holocene deposits, sampling density (in depth) can be more or less equal for the entire Holocene sequence. As such, higher sedimentation rates for shorter time spans are a logical consequence. Lower sedimentation rates, and thus more or longer periods with little or no sedimentation, are also an essential part the sediment dynamics. In order to correct for a possible measuring period dependence in sedimentation rates, dating should be performed homogeneously in time over the Holocene period, which implies for most dated places in the Dijle catchment that the sampling density in depth should largely vary. A coupling between the calculated sedimentation rates or (relative) mass accumulation rates and the fluvial architecture and site-specific sedimentation histories also indicates that time dependency is not the only explaining factor for high recent rates. Cumulative probability functions of radiocarbon ages The CPD of floodplain fines ages (Fig. 9A) shows peaks around BC, AD and AD. The CPD of colluvial ages (Fig. 9B) shows peaks around BC, BC, 900 BC 500 AD, 900 AD 1400 AD and 1450 AD to present. The charcoal ages from Bilande are considered to be reworked and redeposited, and therefore they were not included in the original analysis. However, they also indicate phases of past colluviation more upstream, which was subsequently eroded (Lang and Hönscheidt, 1999), and therefore they can also be introduced into the CPD (Fig. 9C). The resulting CPD differs slightly: there are additional peaks between 2000 and 1000 BC. All these CPDs rely on a low number of ages which cause a high sensibility to individual samples. The combined CPD for aggradation ages (Fig. 10, B and C) shows comparable peaks, with aggradation ages concentrated after 2000 BC with lower probabilities at ca BC, ca BC, BC, ca. 800 BC, ca. 400 BC and AD. The CPD off all peat ages (Fig. 10A) shows peaks at BC, BC, 8250 BC, BC, BC, BC, BC, BC, BC, BC, BC, AD, AD and AD. Compared with the aggradation ages, the stability ages show a larger scatter. Although the CPDs in this study are based on low numbers of data, which biases their interpretation, we consider that they still provide a framework for an objective analysis of the available radiocarbon ages. From the CPDs it is clear that aggradation, both colluvial and alluvial, occurred mainly during the end of the Holocene, after 2000 BC. The given data Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

64 FLUVIAL HOLOCENE SEDIMENT DYNAMICS IN A BELGIAN LOESS CATCHMENT 53 Figure 9. Cumulative probability functions (CPDs) for radiocarbon ages in the Dijle catchment. Probabilities are divided by the probability of equally spaced ages. The grey shaded area indicates where the relative probability is larger than average. (A) Floodplain fines (n ¼ 10); (B) colluvial deposits (n ¼ 21); (C) colluvial deposits including rejected charcoal ages from Bilande (n ¼ 29). Figure 10. Cumulative probability functions (CPDs) for radiocarbon ages in the Dijle catchment. Probabilities are divided by the probability of equally spaced ages. The grey shaded area indicates where the relative probability is larger than average. (A) All peat ages (n ¼ 49); (B) all aggradation ages (n ¼ 31); (C) all aggradation ages including rejected charcoal ages from Bilande (n ¼ 39). suggest that between 2000 BC and 1 AD colluvial aggradation was more important than alluvial aggradation. Peat ages, indicating stable phases, are rather well spread over the Holocene. The simultaneous occurrence of stability and aggradation peaks during the last 4 ka suggests a spatial heterogeneity within the catchment: while some parts are prone to aggradation, floodplains in other parts are stable. Such a spatial heterogeneity can point to different responses of floodplains to the same external forcings, or to heterogeneity in these external forcings. Time-differentiated sediment budget Table 5 and Fig. 11 show the results of the temporal differentiated sediment budget for the two scenarios described above. These results are based on the dating results and thus show comparable patterns of varying sedimentation rates. As colluvial valley deposition is very low for the first period (early Holocene to 2000 BC), scenario 1 yields a hillslope sediment delivery ratio (HSDR) for this first period of 96%, which is very high. These HSDR rates are considered unrealistically high by Verstraeten et al. (2009a). Using scenario 2, HSDR drops to 83%, which is at first sight a more realistic value. On the other hand, hillslope colluvial deposition is often reported to be associated with agricultural practices and the creation of lynchets (e.g. Houben, 2006, 2008). Colluvial deposition is also very rare or absent in the historical forests of the catchment, for which we can assume that these experienced only a minimal anthropogenic impact (e.g. Langohr and Sanders, 1985). Indeed, before 4000 BC no traces of colluvial deposition could be found so far in the Dijle catchment (see CPD analysis), which would mean that the HSDR is indeed near 100%. Off course, the spatial representativeness of our dataset is limited and it cannot therefore be excluded that some slopes do have limited colluvial deposits. Based on all these observations both scenario 1 and 2 can be valid (Fig. 12). Table 5. Temporal differentiated sediment budget for the Dijle catchment, using two different scenarios (see text). Hillslope erosion (Tg) Hillslope deposition (Tg) Colluvial valley deposition (Tg) Hillslope export (Tg) HSDR (%) Tributary floodplain deposition (Tg) Trunk valley floodplain deposition (Tg) Scenario 1 Period 1: 9000 BC 2000 BC Period 2: 2000 BC 1000 AD Period 3: 1000 AD present Scenario 2 Period 1: 9000 BC 2000 BC Period 2: 2000 BC 1000 AD Period 3: 1000 AD present Total Holocene Export (Tg) SDR (%) Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

65 54 JOURNAL OF QUATERNARY SCIENCE Figure 11. Time-differentiated sediment budget for the Dijle catchment, using scenario 1 for the time differentiation of the soil erosion masses. The results (Fig. 12) show that during the second period (2000 BC 1000 AD) colluvial deposition is more important than alluvial deposition. The relative importance of floodplain deposition increases again during the last period (1000 AD present). This demonstrates a lag time between colluvial storage and floodplain storage, as was demonstrated before. Figure 12. Evolution of SDR and HSDR in the Dijle catchment, using the two scenarios to construct the time-differentiated sediment budget. Discussion Several methods were used to analyse the available radiocarbon dates (Table 6). Relative sedimentation and mass accumulation rates were constructed for each individual coring site and a general pattern was extracted (Fig. 8). Construction of these diagrams, especially the mass accumulation, requires the availability of a detailed coring description (e.g. layer thickness, texture, OM and DBD). Owing to the low number of dates per coring site the interpretation suffers from a large averaging effect and a low spatial resolution. The older the sample, the larger this averaging effect will be, causing a lower temporal resolution and a decreased sensitivity (Hoffmann et al., 2009). Therefore only long-term changes in sedimentation rates can be identified. On the other hand, this method allows the reconstruction of site-specific sedimentation histories. Local deviations from the general pattern can give more insight into the spatial variability of the catchment sediment dynamics. Scenarios of deposition rates can be developed and compared with the distribution of the ages to come to general conclusions about sedimentation history. Compared to the sedimentation rate analysis of the Rhine catchment (Hoffmann et al., 2009), the methodology was Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

66 FLUVIAL HOLOCENE SEDIMENT DYNAMICS IN A BELGIAN LOESS CATCHMENT 55 Table 6. Overview of the different methods used for the analysis of radiocarbon ages and the derived sedimentation history of the Dijle catchment: output parameter, temporal resolution, derived start of the increase in colluvial sedimentation and derived start of the increase in alluvial sedimentation. Temporal resolutions are partially based on Hoffmann et al. (2009). Sedimentation history per site Mass accumulation rate Frequency distribution Output parameter (Relative) sedimentation rate (Relative) sedimentation rates Probability densities of radiocarbon ages Temporal resolution Variable, a Variable, decreasing with age a Start of increasing colluvial sedimentation First increase BC 1000 BC 4300 BC Main deposition last 1000 a Start of increasing alluvial sedimentation First increase BC 1 AD 1500 BC Second increase 1000 AD improved in three ways. This improved analysis was facilitated by the availability of a detailed coring description which was missing in the synthetic analysis of Hoffmann et al. (2009). By using relative mass accumulation rates instead of volumetric sedimentation rates, the calculated values take into account the large variations in organic material of floodplain sediments. Peat layers can be responsible for an important vertical aggradation of the floodplain, especially during the early Holocene, which does not correspond to important sediment accumulation due to the low mineral content of these layers. Additionally, the use of relative rates excludes the effect of local differences in total Holocene accumulation. Finally, the use of relative rates and the methodology used for the models results in equal total Holocene sediment accumulation amounts for all models (100%), whereas Hoffmann et al. (2009) use models which yield different total Holocene accumulation amounts (in total thickness of the Holocene deposit). As a result, our models allow a more straightforward comparison with the calculated rates. The construction of CPDs based on radiocarbon ages provides a detailed image of phases during which the different processes were active. CPDs were constructed for several groups of radiocarbon data, representing stability or aggradation of the floodplain (see also Hoffmann et al., 2009; Macklin et al., 2010). Interpretation of these CPDs suffers largely from the limited amount of available radiocarbon dates: peaks can be caused by a single radiocarbon date, which makes this method extremely sensitive for outliers. Spatial heterogeneity of processes can also play a role: while stability dominates in one part of the catchment, aggradation can dominate elsewhere. The CPD method can provide insight into short time variations (Hoffmann et al., 2009) and is thus more suited to studying the effects of short time variations of the driving forces. Additionally there are two major methodological drawbacks (Hoffmann et al., 2009). First, the probability frequencies cannot be transferred into volumes or fluxes. Where the CPD of aggradation ages shows comparable or even more dominant peaks for the period around 1000 BC compared to those for the period after 1000 AD, the other applied methods suggest that the fluxes for the last period are more important. Second, low probabilities do not necessarily suggest low activities but rather a lack of dating evidence, which can be caused by a low preservation potential or by the sampling method. Indeed, the CPD of the peat ages shows no peaks between 8000 and 6000 BC, which is caused by the sampling strategy: often only the top and the base of the basal peat layer are sampled and nothing in between, and as a result few dates are available for the middle of the period with peat formation. An increased number of available ages would improve the application of this method in the Dijle catchment. It remains unclear, however, how many dates are needed to construct CPDs (Hoffmann et al., 2008). Comparison of CPDs of the Dijle catchment with those from other European catchments is biased by the differences in methodology used, the related different objectives and the limited number of ages available in this study. For Poland the database was divided into different groups, excluding a group representing the overbank deposition (Starkel et al., 2006). Conclusions are mainly based on ages of channel facies and abundance of palaeochannels. Dates coinciding with overbank aggradation above peat layers peak mainly in the expansion period of agriculture during the late Roman period and the 11 15th centuries AD (Starkel et al., 2006). For Great Britain, phases of major flooding are identified based on a database of radiocarbon ages (Macklin and Lewin, 2003). Lewin et al. (2005) classify the dates into groups, including a group representing dated floodplain surfaces (overbank) ages. The results show that floodplain sedimentation peaks mainly during the later Holocene and alluviation rates were not started but enhanced by anthropogenic effects. The greater availability of fine-grained material by weathering, with the progressing Holocene, can also play a role. Further, also the importance of preservation potential is stressed. For Great Britain, Macklin et al. (2010) have recently reviewed the CPDs for different depositional environments. Comparable to the results of this study, an acceleration in overbank floodplain sedimentation is reported after 1 cal ka BP, for the first time recognising an unequivocal anthropogenic signal affecting the British rivers. In this study we used an approach comparable with the approach used in Germany (Hoffmann et al., 2008) and for the Rhine catchment (Hoffmann et al., 2009). Due to the size of the database we used in our study, fewer groups could be differentiated. For the Rhine catchment (Hoffmann et al., 2009), stable floodplain environments occurred during the early and Middle Holocene ( BC), followed by increased floodplain deposition in the Late Holocene (since 1500 BC). Within the last period, peaks of higher probabilities for floodplain deposition are reported at 1000 BC, 300 BC and 1100 AD, while lower probabilities occur at 700 BC, 200 AD and 1400 AD. Colluvial ages do not show a general trend in this catchment, but most important peaks of increased probabilities are observed around 7000 BC, 5500 BC, 2500 BC and since 1200 AD. The database for the whole of Germany (Hoffmann et al., 2008) overlaps largely with the database from the Rhine catchment (Hoffmann et al., 2009), resulting in the same patterns. In this study a large number of dated cross-sections and dates were used to provide a catchment-wide overview of the sedimentation history. By using a large number of crosssections and corings, the influence of local variations is minimised. When considering site-specific sedimentation rates (Fig. 6) the variation is relatively high. Comparison of the relative accumulated mass of the Bonlez U and Bonlez D cross- Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

67 56 JOURNAL OF QUATERNARY SCIENCE sections indicates that there are important differences between both cross-sections, although cross-section Bonlez U is located only 2.0 km upstream of Bonlez D. As no major tributaries join the Train valley between the study sites, the upstream catchment of both sites is only slightly different. These observations stress the importance of using dating results from different sites to obtain insight into the catchment sediment dynamics. Local sedimentation histories will be influenced by environmental variations in the upstream catchment of the studied site, and even with an almost identical upstream catchment individual coring sites can show a different sedimentation history. As a result, simple extrapolation of local sedimentation histories to a larger catchment is not justified. The different methods used to evaluate the radiocarbon dates of the Dijle catchment yield comparable sedimentation histories. Evidence for early Holocene colluvial deposition is lacking and first colluvial deposition is reported from ca BC on, coinciding with the Neolithic Period. Colluviation appears to become more important between 2000 BC and 500 BC, and a second increase occurs between 0 AD and 1300 AD. The age/depth and frequency distribution analysis of radiocarbon dates suggest that the increase in alluvial deposition occurs later than the increase in colluvial deposition. For most sites the increase in floodplain deposition starts somewhere between 4000 and 500 BC, while the major part of floodplain sedimentation occurred since ca AD. This general pattern of colluvial and alluvial deposition coincides with the general pattern of an increasing intensity in agricultural activities from the Neolithic Period on. Little evidence exists for the influence of climatic variations. When studying the sedimentation rates at the Bilande site in detail, the main sedimentation phase occurred somewhere between 1150 and 1450 AD. In other catchments sedimentation phases during the high Middle Ages are linked to a wetter climate during the 14th century and especially due to major rainfall events in 1342 AD (e.g. Bork et al., 1998), while the same period coincides with a peak in agricultural activities (e.g. Bork et al., 1998). It is unclear from the available data whether the peak in sediment deposition is caused by increased agricultural pressure, a wetter climatic phase or a combination of both. An apparent lag time is reported between the increase in colluvial and alluvial deposition, which can be created by a lag time between catchment disturbance and its influence on the floodplain deposition, or by a threshold which has to be met before a disturbance influences floodplain deposition. Given observations in other catchments (e.g. sediment cascade model; Lang and Hönscheidt, 1999) and the earlier increase in colluvial deposition compared to alluvial deposition, we hypothesise that it is rather a threshold that has to be met than a lag time between erosion and sedimentation. An apparent lag time or threshold between changes in catchment environment and changes in deposition rates is evident for other catchments (e.g. Kalicki et al., 2008; Lespez et al., 2008; Trimble, 2009). Comparable to the Dijle catchment, the sediment budget of Coon Creek, Wisconsin, also shows a downstream shift of the sediment dynamics following catchment disturbances (Trimble, 1983, 1999, 2009). Increase of sediment erosion caused a downstream shifting wave of sediment deposition, while decrease of sediment erosion caused a downstream shifting evolution of cessation of floodplain accumulation or even floodplain erosion. An important limitation on the temporal differentiated sediment budget is, however, that it does not take into account the repeated reworking and redeposition of sediments. Several periods of deposition and erosion of colluvial deposits have possibly occurred, as demonstrated, for example, by Lang and Hönscheidt (1999) and supported by the presence of reworked charcoals in the colluvial deposits at Bilande. There is no field evidence to quantify the importance of the reworking of sediments. Reworking can have an important influence on the sediment budget: due to a lower preservation potential (e.g. Lewin and Macklin, 2003) the older colluvial deposits are possibly underestimated compared to more recent deposits. Therefore the total erosion and deposition amounts will be underestimated, also resulting in errors in the sediment delivery ratio (SDR) and HSDR. The absence of large amounts of colluvial material in the historical forests of the catchment (see above) suggests, however, that reworking of colluvial deposits since the early Middle Ages of older material will be limited. Additional budgeting of old colluvial deposits in these historical forest could supply more information on the importance of reworking in sediment budgeting. A main advantage of the time-differentiated sediment budget is that it provides an integrated sediment dynamics history of the catchment. The site-specific sedimentation histories demonstrate that there is a large variation in sedimentation history between sites. Therefore a lumped approach is preferable, integrating the sedimentation history over space and time. The resulting averaged trend can be displayed with a larger confidence. A main advantage of using a time-differentiated sediment budget over site-specific sedimentation curves (Figs. 5 7) or analysis of the radiocarbon dates database (Figs. 8 10) is that it allows the quantification of the changing importance of different sinks and links in the sediment budget. Although MR calculations and sitespecific sedimentation histories allow a quantification of processes, the analysis remains qualitative and a quantification of fluxes is not possible. By using a spatial and temporal integrated approach, the results are filtered for variability in the different processes, providing a catchment-wide pattern and enabling a more robust correlation with the environmental changes, for which information is often also available on a catchment scale. By quantifying the different processes and links, variations in connectivity between the different components of the sediment dynamics can be quantified. The different methodologies applied in this work indicate an important influence of anthropogenic land use changes on floodplain sedimentation in the Dijle catchment. However, the dating resolution does not allow identifying the influence of short-term climatic variations, lasting some decades or even a few centuries, on the catchment processes. A more detailed temporal framework is needed to identify such phases. An extension of the radiocarbon database could possibly address this problem, as the identification of shorter phases becomes more reliable, depending on fewer data points (e.g. Macklin et al., 2010). Different studies have indicated the importance of climatic events on catchment and fluvial processes (e.g. Macklin and Lewin, 2003; Starkel et al., 2006). Land use changes may have made the landscape more sensitive to climatic events (see, for example, Knox, 2001), and we hypothesise that probably the interplay between land use changes and climatic events has determined the Holocene sediment dynamics of the Dijle catchment. Conclusions In this study, AMS radiocarbon and OSL dating were applied to the dating of alluvial and colluvial sediments in the mediumsized Dijle catchment. A database of 81 radiocarbon data values was developed and from this sedimentation rates and cumulative frequency distributions were analysed. The results for the Dijle catchment show that there is a large variation in sedimentation history between the different study sites and that data from a single coring cannot simply be extrapolated to an entire catchment. Dating different sites within the same catchment can provide deeper insight into catchment- Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

68 FLUVIAL HOLOCENE SEDIMENT DYNAMICS IN A BELGIAN LOESS CATCHMENT 57 wide processes. In addition, the construction of sedimentation rates and relative mass accumulation rates allows a catchmentwide insight into the sedimentation history. Comparably, the use of cumulative probability functions of radiocarbon ages provides an important methodological tool for the analysis of catchmentwide floodplain or colluvial aggradation and stability phases. Results show that net floodplain aggradation in the studied catchment is much higher in recent periods then before. The detailed sedimentation history shows the influence of anthropogenic land use on both colluvial and alluvial aggradation since Neolithic times. Changes in sediment dynamics for Neolithic times and possibly also for subsequent periods was limited, and more important changes occurred only later, with the most important sedimentation phase during the last 1 ka. An influence of climatic variation on the studied sedimentary archives could not be identified, which can be attributed to the absence of such an influence or to the dating resolution, which does not allow the detection of short duration variations. Possibly land use changes have made the catchment more sensitive to climatic events (e.g. Knox, 2001), and the interplay between both land use changes and climatic events may have determined the catchment sediment dynamics during the Holocene. A further refinement of the dating resolution or a modelling study is necessary to validate this hypothesis. The results also indicate the large variability between the different studied cores, even when they are located in the same fluvial system at a short distance. This stresses the importance of studying several profiles in order to incorporate local variations and in order to develop a catchmentwide sedimentation pattern. The time-differentiated sediment budget of the Dijle catchment indicates that there are important variations in the relative importance of colluvial and alluvial deposition as sediment sinks. We hypothesised that the resulting lag time between colluvial and alluvial sedimentation can be attributed to a threshold that has to be met in soil erosion and/or sediment transport in order to result in certain amounts of floodplain deposition. This time lag indicates the importance of internal system dynamics. Studies concerning entire catchment and sediment pathways, like sediment budget studies, can help in understanding the temporal and spatial dynamics of the sediments within those catchments (e.g. Verstraeten et al., 2009a). Modelling studies can also provide an important contribution to the understanding of past sediment dynamics, as such models would allow control of the variations of each driving force (e.g. Ward et al., 2009). Acknowledgements. This research is part of a project funded by the Research Foundation Flanders (FWO, research project G ), and by the Interuniversity Attraction Poles Programme Belgian State Belgian Science Policy, project P6/22. Their support is gratefully acknowledged. The authors would also like to thank Bjorn Dieu, Jeroen Monsieur, and the several MSc and PhD students in physical geography for their assistance during fieldwork. We would also like to thank M. Macklin and T. Hoffmann for reviewing this paper and providing useful comments. Abbreviations. AMS, accelarator mass spectrometry; a.s.l., above sea level; BLSL, blue light stimulated luminescence; CPD, cumulative probability distribution; HSDR, hillslope sediment delivery ratio; IRSL, infrared stimulated luminescence; ME, model efficiency; MR, relative mass accumulation rate; OSL, optically stimulated luminescence; SDR, sediment delivery ratio; SR, sedimentation rate. References Adamiec G, Aitken MJ Dose-rate conversion factors: update. Ancient TL 16: Aitkin MJ Thermoluminescence Dating. Academic Press, London. Beach T The fate of eroded soil: sediment sinks and sediment budgets of agrarian landscapes in southern Minnesota, Annals of the Association of American Geographers 84: Bisutti I, Hilke I, Raessler M Determination of total organic carbon: an overview of current methods. Trends in Analytical Chemistry 23: Bork H-R, Bork H, Dalchow C, et al Landschaftsentwicklung in Mitteleuropa: Wirkungen des Menschen auf Landschaften. Klett- Perthes: Gotha, Germany. Bøtter-Jensen L, Andersen CE, Duller GAT, Murray AS Developments in radiation, stimulation and observation facilities in luminescence measurements. Radiation Measurements 37: Bronk Ramsey C Development of the radiocarbon calibration program OxCal. Radiocarbon 43: Bronk Ramsey C Bayesian analysis of radiocarbon dates. Radiocarbon 51: Chiverrell RC, Foster GC, Thomas GSP, et al Robust chronologies for landform development. Earth Surface Processes and Landforms 34: de Moor J, Verstraeten G Alluvial and colluvial sediment storage in the Geul River catchment (The Netherlands): combining field and modelling data to construct a Late Holocene sediment budget. Geomorphology 95: De Smedt P Paleogeografie en kwartair-geologie van het confluentiegebied Dijle-Demer. Acta Geographica Lovaniensia 11: 141 pp. Dotterweich M The history of soil erosion and fluvial deposits in small catchments of central Europe: deciphering the long-term interaction between humans and the environment a review. Geomorphology 101: Duller GAT Distinguishing quartz and feldspar in single grain luminescence measurements. Radiation Measurements 37: Foulds S, Macklin M Holocene land-use change and its impact on river basin dynamics in Great Britain and Ireland. Progress in Physical Geography 30: Fryirs K, Brierley GJ Variability in sediment delivery and storage along river courses in Bega catchment, NSW, Australia: implications for geomorphic river recovery. Geomorphology 38: Fuchs M, Lang A OSL dating of coarse-grain fluvial quartz using single-aliquot protocols on sediments from NE Peloponnese, Greece. Quaternary Science Reviews 20: Hoffmann T, Erkens G, Cohen KM, et al Holocene floodplain sediment storage and hillslope erosion within the Rhine catchment. Holocene 17: Hoffmann T, Lang A, Dikau R Holocene river activity: analysing 14 C-dated fluvial and colluvial sediments from Germany. Quaternary Science Reviews 27: Hoffmann T, Erkens G, Gerlach R, et al Trends and controls of Holocene floodplain sedimentation in the Rhine catchment. Catena 77: Houben P A Holocene sediment budget for the Rockenberg catchment. Pre-workshop field guide. In Open Lucifs Workshop, Houben P Scale linkage and contingency effects of field-scale and hillslope-scale controls of long-term soil erosion: anthropogeomorphic sediment flux in agricultural loess watersheds of Southern Germany. Geomorphology 101: Kalicki T, Sauchyk S, Calderoni G, et al Climatic versus human impact on the Holocene sedimentation in river valleys of different order: examples from the upper Dnieper basin, Belarus. Quaternary International 189: Knox JC Agricultural influence on landscape sensitivity in the Upper Mississippi River Valley. Catena 42: Knox JC Floodplain sedimentation in the Upper Mississippi Valley: natural versus human accelerated. Geomorphology 79: Lang A Recent advances in dating and source tracing of fluvial deposits. In Sediment Dynamics in Changing Environments, Schmidt J, Cochrane T, Phillips C, et al. Publication 325. IAHS: Wallingford, Oxon, UK; Lang A, Hönscheidt S Age and source of colluvial sediments at Vaihingen-Enz, Germany. Catena 38: Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

69 58 JOURNAL OF QUATERNARY SCIENCE Langohr R, Sanders J The Belgian Loess Belt in the last years: evolution of soils and relief in the Zonien Forest. In Soils and Quaternary Landscape Evolution, Boardman J editor. Wiley: Chichester; Lespez L, Clet-Pellerin M, Limondin-Lozouet N, et al Fluvial system evolution and environmental changes during the Holocene in the Mue valley (Western France). Geomorphology 98: Lewin J, Macklin M Preservation potential for Late Quaternary river alluvium. Journal of Quaternary Science 18: Lewin J, Macklin M, Johnstone E Interpreting alluvial archives: sedimentological factors in the British Holocene fluvial record. Quaternary Science Reviews 24: Macklin MG, Lewin J River sediments, great floods and centennial-scale Holocene climate change. Journal of Quaternary Science 18: Macklin M, Benito G, Gregory K, et al Past hydrological events reflected in the Holocene fluvial record of Europe. Catena 66: Macklin M, Jones A, Lewin J River response to rapid Holocene environmental change: evidence and explanation in British catchments. Quaternary Science Reviews 29: Messerli B, Grosjean M, Hofer T, et al From nature-dominated to human-dominated environmental changes. Quaternary Science Reviews 19: Meybeck M Global analysis of river systems: from Earth system controls to Anthropogene syndromes. Philosophical Transactions of the Royal Society of London, Series V 358: Mullenders W, Gullentops F, Lorent J, et al Le Remblement de la vallée de la Nethen. Acta Geographica Lovaniensia 4: Murray AS, Wintle AG Luminescence dating of quartz using an improved single-aliquot regenerative-dose protocol. Radiation Measurements 32: Murray AS, Wintle AG The single aliquot regenerative dose protocol: potential for improvements in reliability. Radiation Measurements 37: Nash JE, Sutcliffe JV River flow forecasting through conceptual models, a discussion of principles. Journal of Hydrology 10: Notebaert B Sensitivity of river systems to human actions and climatic events across different environments: a Holocene perspective. PhD thesis. Department Earth and Environmental Sciences, KU Leuven, Belgium. Notebaert B, Verstraeten G, Rommens T, et al Establishing a Holocene sediment budget for the river Dijle. Catena 77: Page MJ, Trustrum NA, Dymond JR Sediment budget to assess the geomorphic effect of a cyclonic storm, New Zealand. Geomorphology 9: Prescott JR, Hutton JT., Cosmic ray contributions to dose rates for luminescence and ESR dating: large depths and long-term time variations. Radiation Measurements 23: Reid L, Dunne T Sediment budgets as an organizing framework in fluvial geomorphology. In Tools in Fluvial Geomorphology, Kondolf M, Piégay H editors. Wiley: New York; Reimer PJ, Baillie MGL, Bard E, et al IntCal04 Terrestrial radiocarbon age calibration,0-26 ka BP. Radiocarbon 46: Rommens T Holocene sediment dynamics in a small river catchment in central Belgium. PhD thesis. Department Geography Geology, KU Leuven, Belgium. Rommens T, Verstraeten G, Bogman P, et al Holocene alluvial sediment storage in a small river catchment in the loess area of central Belgium. Geomorphology 77: Rommens T, Verstraeten G, Peeters I, et al Reconstruction of late-holocene slope and dry valley sediment dynamics in a Belgian loess environment. Holocene 17: Sadler PM Sediment accumulation rates and the completeness of stratigraphic sections. Journal of Geology 89: Schumer R, Jerolmack DJ Real and apparent changes in sediment deposition rates through time. Journal of Geophysical research Earth Surface 114: 12 pp. [F00A06]. Slaymaker O The sediment budget as conceptual framework and management tool. Hydrobiologia 494: Starkel L, Soja R, Michczynska DJ Past hydrological events reflected in the Holocene history of Polish rivers. Catena 66: Thorndycraft V, Benito G. 2006a. Late Holocene fluvial chronology of Spain: the role of climatic variability and human impact. Catena 66: Thorndycraft V, Benito G. 2006b. The Holocene fluvial chronology of Spain: evidence from a newly compiled radiocarbon database. Quaternary Science Reviews 25: Törnqvist T, Dejong A, Oosterbaan W, et al Accurate dating of organic deposits by AMS 14 C measurement of macrofossils. Radiocarbon 34: Trimble SW A sediment budget for Coon Creek basin in the Driftless Area, Wisconsin, American Journal of Science 283: Trimble SW Decreased rates of alluvial sediment storage in the Coon Creek Basin, Wisconsin, Science 285: Trimble SW Fluvial processes, morphology and sediment budgets in the Coon Creek Basin, WI, USA, Geomorphology 108: Vandenberghe D Investigation of the optically stimulated luminescence dating method for application to young geological sediments. PhD thesis, Universiteit Gent, Belgium. Vandenberghe D, De Corte F, Buylaert J-P, et al On the internal radioactivity in quartz. Radiation Measurements 42: Verstraeten G, Poesen J Factors controlling sediment yield from small intensively cultivated catchments in a temperate humid climate. Geomorphology 40: Verstraeten G, Rommens T, Peeters I, et al. 2009a. A temporarily changing Holocene sediment budget for a loess-covered catchment (central Belgium). Geomorphology 108: Verstraeten G, Lang A, Houben P. 2009b. Human impact on sediment dynamics: quantification and timing. Catena 77: Walling D, Quine T Using Chernobyl-derived fallout radionuclides to investigate the role of downstream conveyance losses in the suspended sediment budget of the river Severn, United Kingdom. Physical Geography 14: Walling DE, Russel M, Hodgkinson R, et al Establishing sediment budgets for two small lowland agricultural catchments in the UK. Catena 47: Walling DE, Collins A, Jones P, et al Establishing fine-grained sediment budgets for the Pang and Lambourn LOCAR catchments, UK. Journal of Hydrology 33: Ward P, van Balen R, Verstraeten G, et al The impact of land use and climate change on late Holocene and future suspended sediment yield of the Meuse catchment. Geomorphology 103: Copyright ß 2011 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 26(1) (2011)

70 The Holocene 15,7 (2005) pp /1043 Soil erosion and sediment deposition in the Belgian loess belt during the Holocene: establishing a sediment budget for a small agricultural catchment Tom Rommens, 1 * Gert Verstraeten, 1,2 Jean Poesen, 1 Gerard Govers, 1 Anton Van Rompaey, 1,2 Iris Peeters 1 and Andreas Lang 3 ( 1 Physical and Regional Geography Research Group, Katholieke Universiteit Leuven, Redingenstraat 16, B-3000 Leuven, Belgium; 2 Fund for Scientific Research-Flanders, Belgium; 3 Department of Geography, University of Liverpool, Liverpool L69 7ZT, UK) Received 21 September 2004; revised manuscript accepted 4 March 2005 Abstract: A method to establish a Holocene sediment budget for a 103 ha agricultural catchment representative for the Belgian loess belt is presented. Soil erosion and sediment deposition were determined based on 185 coring locations and a large excavation in the valley bottom. Results were integrated in a GIS and interpolation techniques applied to derive spatial patterns of erosion and sedimentation. Total soil erosion, sediment deposition and sediment export were calculated and the results show that volumes are highly dependent on the interpolation technique used. Sediment delivery ratios between 20% and 42% are derived and are consistent with data reported in previous studies. This clearly shows that the majority of the sediments produced during the Holocene have been stored near their source area and have not been delivered to the downstream rivers. The spatial distribution of soil erosion and sediment deposition within the catchment is strongly dependent on slope gradient and position within the catchment, which suggests that, since human impact began, topography has been the main factor determining long-term soil erosion and sedimentation. Key words: Historical erosion, agricultural catchment, soil erosion, sediment budget, sediment delivery, loess, Belgium, Holocene. Introduction In many European landscapes, erosion features and sediments related to past land use can be widely found, for example, as deeply truncated soils on the slopes, ancient rills and gullies, colluvial deposits on the lower slopes and clastic deposits in floodplains and lakes. As a result, the evolution and distribution of contemporary soils can be understood only by taking into account impacts of the past. Over the last three decades geomorphological research has led to a significantly better understanding of present-day soil erosion and sediment redistribution processes. Whereas current processes can be observed *Author for correspondence ( Tom.Rommens@geo.kuleuven. be) and measured directly and in most cases using relatively simple techniques, the intensity of past soil erosion events can only be estimated by studying their cumulative effect, resulting in colluvial, alluvial and lake sediments. Studies addressing the long-term effects of soil erosion processes on landscape development are usually carried out to reconstruct palaeoenvironments. Such studies usually focus on alluvial and lake sediments (e.g., Page and Trustrum, 1997; Macaire et al., 1997; Foster et al., 2000; Zolitschka, 2002; Oldfield et al., 2003; Zolitschka et al., 2003). Often the main emphasis is on sediment dating to create a temporal framework for different erosion phases and to estimate the relative importance of climate and land-use change in influencing Holocene erosion rates (e.g., Lang, 2003). Although such approaches can be used to distinguish periods with higher # 2005 Edward Arnold (Publishers) Ltd / hl876ra

71 Tom Rommens et al.: A soil erosion sediment budget for the Belgian loess belt 1033 and lower sediment dynamics, they are limited for deriving quantitative information on erosion and sediment transport rates and only provide integrated information with low spatial resolution. Only in a limited number of cases, an attempt is made to assess the quantities of colluvial deposits and soil profile truncation on a catchment scale (Lewis and Lepele, 1982; Bork, 1983), or to establish sediment budgets for larger-scale river catchments (e.g., Larue, 2000; Macaire et al., 2002). A thorough quantification of historical soil erosion and sediment budgets on a broader spatial scale is necessary in order to better understand sediment dynamics on a catchment scale as a function of changing climate and land use. Soil erosion and sediment transport models that are spatially variable, operate at the catchment scale and are based on current erosion process knowledge, can aid in this respect. However, in order to validate such modelling exercises, spatially resolved data on historic soil erosion and sediment deposition are needed. The aims of this study are therefore (1) to establish and discuss a methodology to interpret soil profiles and sediment deposits in the Belgian loess belt in terms of long-term soil erosion; (2) to investigate the spatial patterns of Holocene soil erosion and sediment deposition in a small agricultural catchment in the Belgian loess belt; and (3) to establish a long-term sediment budget for this catchment. Materials and methods Study area For this study, a small catchment (103 ha) in Nodebais, 13 km south of Leuven (Figure 1), was selected. This catchment is representative of the Belgian loess belt, which is characterized by gentle rolling hill topography. The Belgian loess belt is part of the large W/E running loess region extending from northern France towards the Ukraine. In the study area the loess generally overlays Tertiary marine sands, but in the bottom of small river valleys Tertiary marine clays also occur. The loess is of variable thickness and the majority of it is of Weichselian age (Tavernier, 1948; Gullentops, 1954; Van Vliet and Langohr, 1981). In more detail, the topography at the study location can be described as a plateau that is dissected by several dry valleys. Soils in this region are mainly loess-derived Luvisols (silt loam texture) that have been recognized for their fertility since prehistoric times and are at present still intensively farmed. Today arable fields that in some cases can extend over several tens of hectares dominate the flat areas on the plateaus. Some patches of woodland or fallow land are found where the less fertile Tertiary sands outcrop. Within the Nodebais catchment, land use is mainly arable land (/90%). The main crops grown are wheat, sugar beet, potatoes, chicory and flax. Two important topographical units can be distinguished in the catchment; on the one hand an almost flat plateau area with slopes less than 3% in the south, and on the other three small confluent dry valleys with more or less SW- to NE-oriented axes in the north. The sides of the valleys are steep with maximum slope angles of 28%. Most of the slopes are smooth and show only few signs of local anthropogenic disturbance such as closed depressions, sunken lanes or earth banks (tillage steps or colluviation terraces), that otherwise can be found abundantly in the region. Height varies from m a.s.l. in the uppermost part down to 80 m a.s.l. at the outlet. Soil and sediment survey In total, 185 augerings were conducted using an Edelman hand-auger. Locations were selected to represent different topographic positions in the landscape and forming transects perpendicular to the dry valley bottom or thalweg. Augering positions were measured with a high precision RTK-GPS. Augering density was chosen higher in zones with higher topographical variation, namely the incised valleys in the northern part of the catchment (Figure 1). Augering depths varied from 0.4 m on severely eroded slopes to a maximum of 6 m in the valley bottom. As mentioned above, soils in the catchment are developed in late Pleistocene loess deposits. This loess initially contained 10/13% CaCO 3 (Goossens, 1987). After deposition, decalcification occurred and subsequently Luvisol development commenced. In the upper decalcified soil horizons clay became mobile, migrated and formed a clay-accumulation (B t ) horizon. The B t horizon, also called Argic horizon, is a key property of the Luvisol soil type (Food and Agriculture Organization, International Soil Reference and Information Centre and International Society of Soil Science, 1998). It is characterized by a higher clay content, which sometimes results in the occurrence of brown clay skins, and a compact and blocky structure, which often shows hydromorphic mottles. Erosion and sediment deposition after Luvisol development changes the typical soil profile. Soil erosion led to soil profile truncation at erosive sites and deposition of colluvium and alluvium at lower slopes and in valley bottoms. Thus, studying Figure 1 Spatial distribution of the augerings and transects (with numbers) in the study area, and location of the study area in central Belgium

72 1034 The Holocene 15 (2005) soil profiles and sediments can be used to determine past erosion and sediment deposition. However, such an interpretation necessitates assumptions about the initial pre-human impact reference situation. This poses a problem, since it is very likely that there are no undisturbed Holocene soils conserved in the study area. Most of the region was already deforested for agricultural purposes in the Middle Ages, so erosion and deposition processes have influenced vast areas of the soil surface. Recent studies show that even in old forests, such as the nearby Meerdaal Forest, which used to be considered as pristine woodland, past human impact on the local geomorphology is significant (Vanwalleghem et al., 2003). Despite inconclusive data, some reasonable assumptions on the early Holocene soil geography are proposed here as a working hypothesis (Figure 2a): (1) The decalcification and soil development rates in nonerosive circumstances are assumed to be independent of the slope gradient and aspect. In this view the original, uneroded Holocene soil in the catchment is considered homogeneous, which means identical soil profiles occurred on the plateaus and the slopes. (2) The depth of decalcification in a non-eroded soil in the catchment is estimated at 2.3 m. This estimate is calculated as the average of results from 14 soil profiles in flat landscape positions, where soil erosion and sediment deposition were supposed to be insignificant. Former studies in the Belgian loess belt have found more or less similar decalcification depths (Dudal, 1955; Desmet, 1986; Goossens, 1987; Vandaele, 1997), except for the study in the Zoniën-forest (Langohr, 2001), where decalcification seems to be much deeper (Table 1). (3) The depth of the upper border of the clay illuviationhorizon (B t ) of non-eroded soils in the catchment is estimated at 0.4 m and the depth of the lower border of the B t horizon at 1.5 m (Table 1). These values were also calculated as average values derived from the 14 soil profiles in flat topographic locations. These values again correspond well with other studies (Desmet, 1986; Goossens, 1987; Rampelberg and Deckers, 1995; Langohr, 2001). Based on these assumptions, the Holocene soil erosion depth and thickness of corresponding sediment deposits can be estimated. Table 2 shows a classification of the soil profile types that were observed in the Nodebais-catchment and the way they were interpreted in terms of erosion- and colluviumdepth. Interpretations are based on three criteria. First, the occurrence of a B t horizon can be identified by morphological characteristics. The higher clay content results in the occurrence of brown clay skins, a compact and blocky structure and hydromorphic mottles. Secondly, the occurrence of colluvial sediments can be recognized by the presence of charcoal, ceramics and brick fragments (and a reduced density compared with the in situ soil). Thirdly, a subdivision is made between locations in which calcareous loess was recovered and locations in which the calcareous parent material could not be recovered. Representative profiles from three different landscape positions (on the plateau, the slope and the thalweg) illustrate this methodology in Figure 2b. Calculation of sediment volumes All data were integrated in a GIS, and analysed using Surfer (version 8.00) and Idrisi32 software. Two relatively simple interpolation methods were applied to produce grid maps for erosion depth and colluviation thickness for the Nodebais catchment at 5 m resolution. The first interpolation used is the Inverse Distance to a Power (IDP) gridding method, which uses a weighted average interpolator: Z ffl j X n Z i i1 h b ij X n 1 i1 h b ij where: h ij is the distance between grid node j and neighbouring point i; Z ffl j is the interpolated value for grid node j; Z i is the value for a neighbouring point; and b is the weighting power. A weighting power of 2 was used. As erosion or deposition values vary considerably over short distances, the interpolation value for a point was based on the six nearest data points only (n/6). Secondly, a Point Kriging procedure was applied, which is a geostatistical interpolation that uses a variogram to determine weighing factors for the data points. For the variogram a simple linear model was chosen. Total erosion and sediment deposition volumes were calculated from the resulting grid maps using the GridVolumemodule in Surfer. Mathematically, the volume (V) under the interpolated grid surface (function f (x,y)) is defined by a double integral V f x max f y max f (x; y)dxdy; which is approximated in Surfer using the extended trapezoidal x min y min rule: and: V : Dy 2 [A 12A 2 2A 3 2A r1 A r ] A j : Dx 2 [G 1; j 2G 2; j 2G 3; j 2G c1; j G c; j ] where: Dx is the grid column spacing; Dy is the grid row spacing; G i,j is the grid node value in column i and row j; r is the number of grid rows; and c is the number of grid columns. Sediment volumes were also calculated for different morphometric units. The catchment was subdivided into five morphometric classes according to slope gradient and geomorphologic position: the plateau (with slope angles less than 3%); slopes between 3% and 5%, 5% and 8% and more than 8%; and finally the valley bottom and adjacent foot-slopes (with slope angles less than 8%). Average values of soil erosion and sedimentation were then calculated for every class ( Average Per Unit method : APU). Volumes were estimated by multiplying the average soil erosion or deposition by the class area. Similar methods are also used in other studies (Lewis and Lepele, 1982; Bork, 1983; Macaire et al., 2002). For validation of the volume calculations, 14 valley crosssections were manually drawn based on the augering results and field observations. These cross-sections were considered as most reliable representations of reality. The accuracy of the volume calculations was then assessed by (1) extracting the corresponding cross-sections from each of the different interpolated maps, and (2) comparing the interpolated with the manually drawn cross-sections. To further elucidate the performance of the interpolation methods eight additional crosssections were extracted in between the augering cross-sections and compared with neighbouring sections.

73 Tom Rommens et al.: A soil erosion sediment budget for the Belgian loess belt 1035 Figure 2 a. Illustration of the working hypotheses to deduce soil erosion and sediment deposition. Evolution of a slope through time: (1) situation just after loess deposition: the whole landscape is covered by calcareous loess; (2) after soil formation but before erosion was initiated: soil characteristics (Luvisol) are similar in every topographical position; (3) after soil erosion (present-day situation): soil profiles on the slopes are truncated, footslopes are covered with sediment. The broken line indicates the land surface during phases (1) and (2). b. Profile description and interpretation of three representative cores, from contrasting landscape positions

74 1036 The Holocene 15 (2005) Table 1 Depth of soil horizons. Depth to calcareous parent material and upper and lower border of the argic horizon of non-eroded soils in the Nodebais catchment. Soil profiles to be used as references were selected on plateau positions within the catchment at locations with insignificant erosion B t upper border (m) B t lower border (m) Depth to calcareous loess (m) Nodebais catchment, 14 reference profiles: Mean Standard deviation Values for other studies in the Belgian loess belt: Dudal, 1955 / / 2.0/2.5 Desmet, Goossens, 1987 / 0.9/ Rampelberg and Deckers, / / Vandaele, 1997 / / 2.5 Langohr, / / /3.5 Table 2 Classification and interpretation of soil profiles in the Nodebais catchment No erosion: Insignificantly truncated soil profiles (more or less standard profile) Moderate erosion: Moderately eroded soil profiles (B t horizon partly present) Severe erosion: Severely eroded soil profiles (B t horizon missing) No colluvium C 1 present C 1 not present a Erosion/C 1,ref /C 1,current (All reference soil horizons well developed to normal depths.) Erosion/BC ref /BC current (Well developed Bt horizon with normal thickness and normal depth.) Colluvium C 1 present Erosion/assume 0 Colluvium/B t,current /B t,ref (All reference soil horizons normally developed, but profile buried.) C 1 not recovered a Erosion/assume 0 Colluvium/B t,current /B t,ref No colluvium C 1 present Erosion/C 1,ref /C 1,current (Should more or less equal BC ref /BC current ) C 1 not present a Erosion/BC ref /BC current Colluvium C 1 present Erosion/C 1,ref /(C 1,current /B t,current ) (Should more or less equal BC ref /(BC current /B t,current )) Colluvium/B t,current No colluvium C 1 not present a Erosion/BC ref /(BC current /B t,current ) Colluvium/B t,current C 1 present C 1 not present a Erosion/C 1,ref /C 1,current Assume Erosion/BC ref b Assume Erosion/C 1,ref (Tertiary sand outcropping.) Colluvium C 1 present Erosion/C 1,ref /(C 1,current /BC current ) b C 1 not present a Colluvium/BC current b Erosion/C 1,current c Colluvium/C 1,current c Assume Erosion/BC ref b Colluvium/BC current b Assume Erosion C 1,ref Colluvium/BC current (Sandy BC horizon, transition to Tertiary sand.) Erosion/C 2,current c Colluvium/C 2,current c Erosion, depth to which soil is eroded; Colluvium, thickness of colluvial deposits; B t, depth of upper border of B t horizon; BC, depth of upper border of BC horizon; C 1, depth of decalcification front; C 2, depth at which sandy substratum is found; ref, observed in reference profiles; current, observed in core. a C 1 horizon not present because of limited thickness of initial loess cover: sandy substratum at shallow depth. Or sediment thickness larger than maximum coring depth (usually in valley bottom). b Silt loam BC horizon partly present. c Sharp boundary between colluvial sediment and underlying calcareous loess or Tertiary sand: old pit or gully.

75 Tom Rommens et al.: A soil erosion sediment budget for the Belgian loess belt 1037 The precision of the calculations was assessed assuming an observational error of 0.05 m for erosion depth and sedimentation height and applying Gaussian error propagation. Results and discussion Interpretation of the soil augerings To validate results of the soil augerings, a 65 m long and 1/4m deep trench was opened along one of the slopes into the bottom of the dry valley. The resultant outcrop showed 5/3.5 m thick silt loam sediment overlying a soil remnant. The soil remnant consists of a dark brown clayey silt loam horizon with a thickness of 9/0.5 m in the lower part and a pale silt layer of 0.3 m thickness, with dark and reddish iron and manganese mottles in the upper part. The dark horizon is intersected by cracks, which in the vertical section appeared as tongues, infilled with pale silty material. In a horizontal section, these cracks show polygonal patterns, with polygons of 0.15/0.20 m diameter. Texture analysis shows a clay content of 19% for the lower dark brown horizon, whereas the overlaying soil horizon contains only 11/12% clay. These characteristics fit well with the description of a fragipan, as found in the upper part of the B t horizon in silty soils under forest (Van Vliet and Langohr, 1981; Langohr and Sanders, 1985; Langohr, 1990, 2001). This type of original forest soil is still conserved in the lowest parts of the landscape, protected by a thick blanket of colluvial deposits. Traces of this palaeosol were detected in almost all cores studied in the valley bottoms of the Nodebais catchment, at depths that varied between 0.9 m in the valley heads and 5.9 m near the outlet of the catchment. This suggests that there has been no extreme gullying in the valley bottoms of this catchment, in contrast to other parts of the Belgian loess belt where large gully systems have been described (Arnould-De Bontridder and Paulis, 1966; Bollinne, 1976; Vanwalleghem et al., 2003). However, it indicates that the early Holocene topography was more pronounced than today. Four charcoal samples were taken from the sediments directly overlying the palaeosol. They returned calibrated radiocarbon ages of 970/400 BC (Table 3). These ages show that the colluvial sediments in the catchment were deposited in the late Holocene (late Bronze Age to Iron Age). The higher erosion rates are probably caused by more extensive forest clearing, cultivation and grazing from the late Bronze Age onwards. Indications for the intensification of soil erosion in this period are also found in many other parts of Europe (e.g., Starkel, 1992). Moreover, traces of late Bronze Age settlements were found in the proximity of the study area (Martens, 1981). In the outcrop, several 1 m to more than 2 m thick infills of old depressions were observed on the slope. The depressions Table 3 AMS-radiocarbon ages of charcoal samples from the sediments overlying the palaeosoil observed in the excavation Lab. code Sampling depth (cm) 14 C-age BP Calibrated 14 C-age a GrA /35 970/800 BC GrA /45 770/400 BC GrA /40 770/410 BC GrA /35 790/410 BC a Calibrated using OxCal v3.9 (Bronk Ramsey, 1995, 2001). Calibration curve based on atmospheric data from Stuiver et al. (1998). appear to be man-made and are similar to those described by Gillijns et al. (2005) from other parts of the Belgian loess belt. They were clearly dug into the calcareous loess, which is present already at shallow depth on the slopes. Hence, they probably served as marl pits. In historical times, calcareous loess was often used as so-called marl to neutralize the acidity of cropland (Lindemans, 1952). As the depressions are completely refilled with sediment, they cannot be recognized on the present-day land surface. The occurrence of infilled pits can serve as an explanation for a number of anomalies that were encountered by augering. In several places, on slopes as well as on the plateau, surprisingly thick sediment deposits were observed, directly overlaying calcareous loess. This could be due to old refilled gully systems or, more likely, infilled manmade pits, as the ones revealed by the excavation. GIS grid interpolations Estimations for the total eroded sediment volumes in the catchment using the original data set vary between m 3 and m 3, depending on the interpolation method (Table 4). The total volume of sediment stored in the catchment ranges between m 3 and m 3. To quantify the effect of the infilled depressions, calculations were repeated after erasing anomalous observations and replacing them with values interpolated from neighbouring points. In the Nodebais catchment, seven of these anomalous sites were detected in the soil profiles, of which only one site can still be identified as a small depression in the present-day topography. Calculations based on the revised data set resulted in soil erosion volumes that were on average m 3 lower, and volumes of deposition that were on average m 3 lower than the volumes calculated on the basis of the complete data set (Table 4). Research on historical man-made closed depressions in the Belgian loess belt (Gillijns et al., 2005) suggests that these features have an average diameter of 43 m and an average maximum depth of 3.3 m. Assuming that these characteristics are applicable also for the seven pits that were detected in the Nodebais catchment, a maximum total excavated volume of m 3 can be expected in the catchment. So, it seems that the influence of these marl pits on the total sediment budget is relatively small, and that taking into account these anomalies leads to an overestimation of the sediment volumes extracted from and stored in these features. Therefore, for further calculations, the revised data set was used and the anomalies were ignored. Total volumes of sediment that were exported out of the catchment can be calculated as the difference between total soil erosion volume and total sediment deposition volume. This sediment export ranges from m 3 to m 3. The mean dry bulk density of sediment samples from the study area is 1.53 t/m 3, which results in a sediment yield of 53/68 t/km 2 per yr for the last 2500 years. Furthermore, the sediment delivery ratio (SDR), which is the proportion of eroded sediment that is exported from the catchment (Reid and Dunne, 2003), can be estimated at 20/23% (Table 4). Figure 3 shows maps of soil erosion and sediment deposition for the Nodebais catchment. It can be seen that the catchment has undergone severe erosion, especially on the steepest slopes, where the unweathered calcareous loess is found directly under the plough layer. This suggests that on these slopes the top 2 m of the original soil have disappeared. In contrast, in the valley bottom, sediment deposits with a thickness of 2 m to /5 m are found, whereas on the plateau soils are left unchanged or only slightly eroded or covered with sediment.

76 1038 The Holocene 15 (2005) Table 4 Eroded and deposited sediment volumes and sediment delivery ratios (SDR), calculated using different interpolation methods in a GIS: APU (/volumes calculated for different geomorphologic units using average values per unit), IDP (/inverse distance to a power interpolation method) and kriging. Error margins for the calculations assume an error of 0.05 m for the estimated erosion depth and sediment thickness Plateau Slope Slope Slope Thalweg Total SDR B/3% 3/5% 5 /8% /8% B/8% (%) APU Eroded volume (m 3 ) (9/8) Deposited volume (m 3 ) Exported volume (m 3 ) Number of data Area (m 2 ) Average erosion depth (m) Average deposition height (m) IDP Eroded volume (m 3 ) (9/4) a Deposited volume (m 3 ) a Exported volume (m 3 ) Kriging Eroded volume (m 3 ) (9/3) a Deposited volume (m 3 ) a Exported volume (m 3 ) a Calculations on the basis of the complete data set, before revision. The maps clearly show the inability of a simple interpolation method to deal with isolated augerings that were not embedded in a transect. These data points cause typical bull s eyes in the interpolation pattern. This indicates that the augering density and the choice of augering locations strongly influences the interpolated patterns and calculated volumes. The soil erosion and sediment deposition maps can be used to reconstruct the original topography before soil erosion was initiated (probably in the late Bronze Age or early Iron Age) (Figure 4). Erosion depths are added to and sedimentation heights are subtracted from, the digital elevation model (DEM) of the present-day surface that was derived from the topographical map. The resulting DEM shows a much more pronounced topography, with deeper valleys and steeper slopes. Average per unit (APU) approach Average soil erosion depth and sediment deposition thickness were calculated for five different geomorphologic units that were distinguished in the Nodebais catchment (Table 4). As expected, there is a clear relationship between average soil erosion and slope gradient, showing most intense soil erosion for the steepest slope class. Slopes steeper than 8% account for an erosion volume of m 3, which is 33% of the total volume eroded ( m 3 ). Average sediment deposition is highest in the thalweg, where m 3 of sediment is stored over an area of 7.5 ha. This means that 38% of the total sediment deposited in the catchment is stored on an area that is only 7% of the total area of the catchment. The high deposition volumes on the steepest slopes are remarkable. Probably, they are caused by the fact that the steepest slope gradients in this catchment occur near the foot of the slope, where a quite sharp boundary can be found between the more or less flat valley bottom and the steep slopes. It is in this particular zone that very often eroded soil profiles are found covered by colluvium. The sediment delivery ratio, calculated using this method, is 429/8%, and the sediment yield comes out at 131 t/km 2 per yr. These values are significantly different from those obtained with GIS interpolations. This is due to the fact that, compared with the APU method, the GIS interpolation methods lead to an underestimation of eroded volumes and an overestimation of deposited volumes. This seems to be especially the case for the steepest slope class, where the interpolated erosion volumes are much lower and depositional volumes are at least twice as high as with the APU method. Transects The locations of the transects are given in Figure 1. Figure 5 illustrates how volumes of erosion and deposition along the transects are calculated. Zones of deposition, low and high intensity soil erosion can be distinguished according to the position in the landscape. The overall distribution of erosion and sedimentation corresponds with the manually drawn transects. In Table 5 the volumes of erosion and deposition are shown for all extracted cross-sections. Relative differences between the manual method and the calculation methods point to an overestimation of sediment quantities in most of the cross-sections. For the GIS interpolation methods, this could be explained by the fact that these methods tend to smooth out the zones of erosion and deposition, thereby creating large zones where sediment deposits are overlying partly eroded soil profiles. In reality, these zones are much narrower. However, it appears that the average per unit approach does not tackle this problem. Considering the predicted erosion in the cross-sections, the three calculation methods show more or less the same level of accuracy. For transect 14, a transect following the thalweg, there is a striking difference between erosion values based on the average per unit method and the GIS methods. The APU method largely overestimates erosion in the valley bottom, as a

77 Tom Rommens et al.: A soil erosion sediment budget for the Belgian loess belt 1039 Figure 3 Soil erosion and sediment deposition maps, produced with the kriging procedure, using the revised data set result of the high average erosion depth (0.57 m) for this geomorphic unit caused by partly eroded soil profiles, covered with colluvium, at the borders of the valley bottom. This kind of soil profile seems to be over-represented in the data set, which biases the calculated average. For the sediment deposits, the IDP approach seems to predict cross-sectional areas better than the two other methods. There is a remarkable underestimation for the sediment deposits stored in the thalweg transect (number 14). This underestimation is highest for the APU method. Eight cross-sections were extracted from the interpolation maps between the augering transects (1a /13a). They all seem to be very similar to the cross-sections in their vicinity. This suggests that the values predicted between the augering transects / at least in the part of the catchment where these transects are not too widely spaced / are realistic.

78 1040 The Holocene 15 (2005) Figure 4 Digital elevation models for the original soil surface (after loess deposition and soil formation, but before soil erosion) and the present-day surface. (Northern half of the Nodebais catchment. Geomorphic differences are illustrated by three cross-sections) Discussion The results illustrate that calculated sediment volumes are highly dependent on the method used. However, when focusing on the sediment delivery ratio, a good correspondence is found between the results of this study and other studies in loess areas. Long-term sediment delivery ratios range between 3% and 52% (Figure 6). Sediment delivery ratios generally decrease with increasing drainage area and the obtained sediment delivery ratios for the Nodebais catchment fit well in this negative trend (Figure 6). It is difficult to define which calculation method performs best. All of the studies mentioned in Figure 6 use a classified approach, estimating mean values for profile truncation and sediment deposition for geomorphologic classes (similar to the APU approach used here). The underlying idea is that erosion and deposition are strongly related to morphometric properties, as for instance slope angle and curvature, as has been shown in various studies on soil erosion processes (Savat and De Ploey, 1982; Govers, 1987; Poesen et al., 1998). One of the advantages is that it offers the possibility to obtain estimates for larger areas by gathering only a limited data set for each geomorphologic class. However, soil and sediment classification is not always straightforward and can certainly / as can the choice of sampling location / influence the results. For this study, the results obtained by the APU method seem not to be more reliable than those obtained using interpolation procedures. According to the average relative prediction errors for erosion and deposition in the transects, the IDP interpolation method tends to perform best. The GIS interpolation techniques are easily applicable and more objective as they do not require a geomorphic classification of the catchment, but they do not allow extrapolation to larger study areas. The biggest advantage is that the interpolation methods can be used to produce detailed maps of erosion and deposition, if sampling density is high enough. The correctness of the volume estimations is difficult to assess for several reasons. First, generalizing point data to areas and ultimately to volumes involves an error, which can hardly be quantified. An attempt was made to evaluate this error by comparing a number of calculated cross-sections with the corresponding manually drawn sections, assuming that the latter are most realistic. This comparison suggests an overestimation of both soil erosion and sediment deposition in most of the catchment, except for the thalweg, where sediment deposition seems to be severely underestimated, especially by the APU approach. The latter problem could be due to a sampling bias: only a minority of the soil augerings in the thalweg were conducted close to the outlet where sediment deposits are thickest. This leads to a calculated average deposition that is too low and results in a small volume of sediment in the thalweg area. Secondly, evaluating the volume estimations is hindered by the uncertainty on the interpretation of the cores. In fact, values for soil erosion and sediment deposition in the data set depend on the quality of the observations and the initial assumptions on which the interpretation is based. For the majority of the cores, the first factor causes no problem: most of the soil horizons, sediment packages and / in particular / the decalcification front can be delineated within a margin of

79 Tom Rommens et al.: A soil erosion sediment budget for the Belgian loess belt 1041 Table 5 Comparison between cross-section areas of soil erosion and sediment deposition derived from manually drawn transects, GIS interpolations (IDP/inverse distance to a power) and average per unit (APU) approach Transect Soil erosion Sediment deposition Manually (m 2 ) IDP (m 2 ) Kriging (m 2 ) APU (m 2 ) Manually (m 2 ) IDP (m 2 ) Kriging (m 2 ) APU (m 2 ) a a a a a a a a a Mean relative difference (%) b Standard deviation (%) a Transect 14 is following the thalweg. b Relative difference between results based on interpolation and results based on the manual approach. Given as a percentage of the results based on the manual approach. Average for transects 1 to 14. Figure 5 Spatial distribution of soil erosion and sediment deposition along a transect and based on different techniques. Example: transect 8. (IDP, Inverse Distance to a Power interpolation; APU, Average Per Unit approach)

80 1042 The Holocene 15 (2005) Figure 6 Relationship between long-term sediment delivery ratio (SDR) and catchment area for catchments characterized by loess-derived soils. Range of resulting sediment delivery ratios for the Nodebais catchment is indicated less than 5 cm. The validity of interpretation assumptions is more problematic. As a working hypothesis, the early Holocene soil in the catchment is assumed homogeneous and independent of topography, and the properties of the uneroded soil are derived from present-day soil profiles on reference positions in the catchment. As mentioned before, there is no way to check these assumptions, which means that erosion as well as deposition could be over- or underestimated. Thirdly, former human activities in the catchment, such as the excavation of calcareous loess can lead to irregularities that are hard to quantify. However, the influence of such irregularities on the total sediment volumes is relatively small and can be neglected for the study area. Establishing long-term sediment budgets for larger areas based on interpolation of data sets is not feasible. Data sets of high spatial resolution, such as the one presented here, will always be limited to relatively small areas. Such information is, however, crucial for calibration and validation of scenarios from model runs to predict long-term and large-scale erosion and deposition patterns. Conclusions We show that the results of an intensive soil and sediment survey can be used to obtain a reasonable estimate for volumes of erosion, sediment storage and sediment yield for a small agricultural catchment in the Belgian Loam Belt during the Holocene. The studied site has undergone soil erosion since at least 2500/3000 years from the Late Bronze Age and Early Iron Age onwards. Long-term soil erosion and sediment deposition are shown to be highly dependent on topography. Highest soil losses are estimated for the steepest slopes, whereas thickest sediment deposits are measured in the valley bottom near the outlet of the catchment. Underneath the sediments in the valley, traces of the early Holocene palaeosol can usually be recovered. The maximum soil erosion volume for the period considered is estimated at more or less m 3 (/309 t/km 2 per yr), whereas estimations for sediment deposition yield at least m 3 (/178 t/km 2 per yr). Volume estimations are highly dependent on the interpolation technique used. The sediment yield for the last 2500 years is estimated at 55/131 t/km 2 per yr and the sediment delivery ratios for the Nodebais catchment (103 ha) vary between 20% and 42%. The low sediment delivery ratios confirm that the majority of the sediments produced during the Holocene are still sitting in the low-order catchments and did not reach the higher-order channels. Acknowledgements We would like to thank the farmers of the study area for their interest and for allowing access to their fields, especially Mr Evrard, who gave permission to excavate the large trench from which we learnt a great deal about soil erosion on the slopes, hidden depressions and colluvial packages. AMS radiocarbon ages were determined at the Centre for Isotope Research, Groningen University, The Netherlands. The GPS equipment was granted by the Fund for Scientific Research Flanders (Belgium, grant no ). The PhD students and the technical staff of the Physical and Regional Geography Research Group are thanked for their contribution to the fieldwork and for providing background information. The work is financed through the OT programme of the KU Leuven, grant no: 3E References Arnould-De Bontridder, O. and Paulis, L. 1966: Etude du ravinement Holocene en Foret de Soignes. Acta Geographica Lovaniensia 4, 182/91. Bollinne, A. 1976: L évolution du relièf à l Holocène. Les processus actuels. In Pissart, A. and Macar, P., editors, Géomorphologie de la Belgique / Hommage au Professeur P. Macar. Université de Liège. Laboratoire de géologie et de géographie physique, 224 pp. Bork, H.-R. 1983: Bodenerosion, Holozäne und Pleistozäne Bodenentwicklung. Catena, Supplement 3, 138 pp. Bronk Ramsey, C. 1995: Radiocarbon calibration and analysis of stratigraphy: the OxCal program. Radiocarbon 37, 425 / : Development of the radiocarbon program OxCal. Radiocarbon 43, 355 /63.

81 Tom Rommens et al.: A soil erosion sediment budget for the Belgian loess belt 1043 Desmet, P. 1986: Bijdrage tot de kwantificering van de totale akkererosie op lemige gronden.unpublished M.Sc. dissertation. Katholieke Universiteit Leuven. Dudal, R. 1955: Bijdrage tot de kennis van gronden op loess-leem in Midden-België. Unpublished Ph.D. thesis. Faculty of Agricultural and Applied Biological Sciences, Katholieke Universiteit Leuven. Food and Agriculture Organization, International Soil Reference and Information Centre and International Society of Soil Science 1998: World reference base for soil resources. World Soil Resources Reports 84. Foster, I.D.L., Mighall, T.M., Wotton, C., Owens, P.N. and Walling, D.E. 2000: Evidence for Mediaeval soil erosion in the South Hams region of Devon, UK. The Holocene 10, 261/71. Gillijns, K., Poesen, J. and Deckers, J. 2005: On the characteristics and origin of closed depressions in loess-derived soils in Europe / a case study from central Belgium. Catena 60, 43/58. Goossens, D. 1987: Sedimentatiemechanismen bij natuurlijke stofdeeltjes in lucht. Unpublished Ph.D. thesis. Department of Geography, Katholiek Universiteit Leuven. Govers, G. 1987: Spatial and temporal variation in rill development processes at the Huldenberg experimental site. Catena, Supplement 8, 17/34. Gullentops, F. 1954: Contribution à la chronologie du Pleistocène et des formes du relief en Belgique. Mémoires de l Institut géologique de l Université de Louvain 18, 125 /252. Lang, A. 2003: Phases of soil erosion-caused colluviation in the loess hills of South Germany. Catena 51, 209/21. Langohr, R. 1990: The dominant soil types of the Belgian loess belt in the Early Neolithic. In Cahen, D. and Otte, M., editors, Rubané et Cardial. Liège: Université de Liège, 117 / : L anthropisation du paysage pédologique agricole de la Belgique depuis le Néolithique ancien / Apports de l archéopédologie. Etude et Gestion des Sols 8, 103/18. Langohr, R. and Sanders, J. 1985: The Belgian Loess Belt in the Last years: evolution of soils and relief in the Zonien forest. In Boardman, J., editor, Soils and Quaternary landscape evolution. Chichester: Wiley, 359/71. Larue, J.P. 2000: Small valley bottom deposits in the sandy district of the Sarthe basin (France): climatic and/or human origin. Geomorphology 45, 309 /23. Lewis, D.T. and Lepele, M.J. 1982: Quantification of soil loss and sediment produced from eroded land. Soil Science Society of America Journal 46, 369/72. Lindemans, P. 1952: Geschiedenis van de landbouw in België. Deel 1. Antwerpen: Genootschap voor geschiedenis en volkskunde. Macaire, J.-J., Bossuet, G., Choquier, A., Cocirta, C., De Luca, P., Dupis, A., Gay, I., Mathey, E. and Guenet, P. 1997: Sediment yield during the Glacial and Holocene Periods in the Lac Chambon watershed, Massif Central, France. Earth Surface Processes and Landforms 22, 473/89. Macaire, J.-J., Bellemlih, S., Di-Giovanni, C., De Luca, P., Visset, L. and Bernard, J. 2002: Sediment yield and storage in the Négron river catchment (south western Parisian basin, France) during the Holocene period. Earth Surface Processes and Landforms 27, 991 / Martens, E. 1981: Uit het verleden van de gemeente Oud-Heverlee. Neerijse, Belgium: Stroobants. Oldfield, F., Wake, R., Boyle, J., Jones, R., Nolan, S., Gibbs, Z., Appleby, P., Fisher, E. and Wolff, G. 2003: The late-holocene history of Gormire Lake (NE England) and its catchment: a multiproxy reconstruction of past human impact. The Holocene 13, 677/90. Page, M.J. and Trustrum, N.A. 1997: A late Holocene lake sediment record of the erosion response to land use change in a steepland catchment, New Zealand. Zeitschrift für Geomorphologie, Neue Folge 41, 369/92. Paulissen, E. and Desmet, P. 1988: Estimation of total erosion and colluviation since agricultural practices. In Geomorphological Processes and Soils. Excursion guide 4th Benlux Colloquium, 24 April/2 May, 1988, University of Amsterdam and University of Leuven, 101/103. Poesen, J., Vandaele, K. and van Wesemael, B. 1998: Gully erosion: importance and model implications. In: Boardman, J. and Favis- Mortlock, D., editors, Modelling soil erosion by water. NATO ASI Series, Vol Berlin: Springer, 285/311. Rampelberg, S. and Deckers, J. 1995: Bodems van Meerdaalwoud. Bespreking van twee transecten in het noorden van Meerdaalwoud. Intern publication 32, Institute for Land and Water Management, Katholieke Universiteit Leuven. Reid, L.M. and Dunne, T. 2003: Sediment budgets as an organizing framework in fluvial geomorphology. In Kondolf, G.M. and Piégay, H., editors, Tools in fluvial geomorphology. Chichester: Wiley, 463 /99. Savat, J. and De Ploey, J. 1982: Sheetwash and rill development by surface flow. In Bryan, R. and Yair, A., editors, Badland geomorphology and piping. Norwich: GeoBooks, 113 /26. Starkel, L. 1992: The feedback mechanisms between the environmental systems and the behaviour of prehistoric man. In Frenzel, B. and Reisch, L., editors, Evaluation of land surfaces cleared from forests by prehistoric man in early Neolithic times and the time of migrating Germanic tribes. ESF Project: European Climate and Man 3, 25/37. Stuiver, M., Reimer, P.J., Bard, E., Beck, J.W., Burr, G.S., Hughen, K.A., Kromer, B., McCormac, G., van der Plicht, J. and Spurk, M. 1998: INTCAL98 radiocarbon age calibration, 24000/0 cal BP Radiocarbon 40, 1041/83. Tavernier, R. 1948: Les formations quaternaries de la Belgique en rapport avec l évolution morphologique du pays. Bulletin de la Société Belge de Géologie 62, 609/41. Vandaele, K. 1997: Temporele en ruimtelijke dynamiek van bodemerosieprocessen in landelijke stroomgebieden (Midden- België) / Een terreinstudie. Unpublished Ph.D. thesis. Department of Geography, Katholieke Universiteit Leuven. Van Vliet, B. and Langohr, R. 1981: Correlation between fragipans and permafrost with special reference to silty Weichselian deposits in Belgium and Northern France. Catena 8, 137/54. Vanwalleghem, T., Van Den Eeckhaut, M., Poesen, J., Deckers, J., Nachtergaele, J., Van Oost, K. and Slenters, C. 2003: Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium). Geomorphology 56, 15 /29. Zolitschka, B. 2002: Late Quaternary sediment yield variations / natural versus human forcing. Zeitschrift für Geomorphologie, Neue Folge 128, 1/15. Zolitschka, B., Behre, K.E. and Schneider, J. 2003: Human and climatic impact on the environment as derived from colluvial, fluvial and lacustrine archives / examples from the Bronze Age to the Migration Period, Germany. Quaternary Science Reviews 22, 81/100.

82 Geomorphology 77 (2006) Holocene alluvial sediment storage in a small river catchment in the loess area of central Belgium Tom Rommens a,, Gert Verstraeten a,b, Pieter Bogman a, Iris Peeters a, Jean Poesen a, Gerard Govers a, Anton Van Rompaey a,b, Andreas Lang c a Physical and Regional Geography Research Group, K.U. Leuven, Celestijnenlaan 200 E, B-3001 Heverlee, Belgium b Fund for Scientific Research-Flanders, Belgium c Department of Geography, University of Liverpool, Liverpool, L69 7ZT, UK Received 8 July 2005; received in revised form 9 January 2006; accepted 10 January 2006 Available online 23 February 2006 Abstract Soil erosion and sediment deposition widely affect landscape development, particularly in erosion-prone areas with loessderived soils. Nevertheless, until now, few attempts were made to quantify soil losses and sediment storage over long (centennial or millennial) timescales. In this study, the Holocene alluvial sediment storage in a small river catchment (52km 2 ) of the Belgian loess belt is estimated, and a preliminary sediment budget for the catchment is presented. In the valley of the Nethen River (c. 13km long), a detailed survey of the alluvial sediment archive was conducted. Hand augerings and percussion drillings were made along cross-valley transects at 12 locations in the catchment. AMS 14 C dating of peat samples provided a temporal framework for the interpretation of the cores. Results show that the thickness of Holocene sediment deposits in the Nethen valley is 4 to 6m, which corresponds to a total clastic sediment mass of t stored in the valley bottom. Three alluvial units could be distinguished and associated with deposition phases from 9600 to 2900 B.C., 2900 B.C. to A.D and A.D to present. In contrast to the older sediments (units 1 and 2), deposits from the last 1000 year (unit 3) contain little organic matter. They are seldom intercalated with peat layers, and devoid of tufa. Unit 3 reaches a thickness of c. 2m, thereby representing 50% of the Holocene sediment mass stored in the alluvial plain. The mean sedimentation rate in the alluvial plain for this last phase is 26t ha 1 a 1, which is about tenfold larger than the sedimentation rates calculated for the older Holocene sediment units. Sediment supply towards the alluvial plain has therefore increased tremendously since Medieval times. These results are in contrast to dating results obtained for colluvial sediments in a nearby dry valley within the catchment of the Nethen, where soil erosion and sediment deposition started in the early Iron Age and was already substantial during the Roman Age. This means that there is a time lag of about one millennium between the onset of high sedimentation rates in the upstream area and high deposition rates in the alluvial plain. This is probably caused by a change in coupling (sediment connectivity) between the plateau, slopes, and rivers. As soil erosion proceeds, first the dry zero-order valleys in the catchment act as sediment traps, and only after these are filled sediment reaches the floodplains. The preliminary sediment budget for the Nethen catchment illustrates that 50% of the sediment that was eroded during the Holocene was stored in colluvial deposits, which are mainly located on footslopes and in dry valley bottoms. Another 29% of the sediment mass is stored in the alluvial plain Elsevier B.V. All rights reserved. Keywords: Holocene; Alluvial sediment storage; Sediment budget; Soil erosion; Human impact; Belgium Corresponding author. Tel.: ; fax: address: Tom.Rommens@geo.kuleuven.be (T. Rommens) X/$ - see front matter 2006 Elsevier B.V. All rights reserved. doi: /j.geomorph

83 188 T. Rommens et al. / Geomorphology 77 (2006) Introduction Soil erosion and sediment deposition are the main landscape-shaping processes in loess areas during the Holocene period. They result in the infilling of valleys and depressions and in the overall reduction of slope gradients, thus smoothing the topography. Insight into the intensities and rates of these processes is necessary to understand the past and present-day relations between climate, humans, and the environment. This is important if we aim at sustainable land and water management. Direct monitoring of sediment yields at a wide range of spatial scales has resulted in a better understanding of present-day sediment dynamics. High resolution data sets have been used to establish sediment budgets, describing the input, transport, storage and export of sediment from fluvial systems (e.g., Reid and Dunne, 2003). They provide information on the relative importance of different sediment sources and sinks in the sediment pathways over short (days or years) (e.g., Page et al., 1994; Walling et al., 2002) and longer (decades) time spans (e.g., Trimble, 1999; Fryirs and Brierly, 2001), thereby showing the impact of environmental changes on sediment redistribution in river basins. Moreover, available data sets made it possible to develop and validate soil erosion models (e.g., Flanagan and Nearing, 1995; De Roo et al., 1995; Morgan et al., 1998; Van Oost et al., 2000). In contrast, over long (centennial or millennial) timescales, estimation of eroded and deposited sediment volumes is difficult. Most studies of long-term sediment dynamics focus on the analysis of different types of sediment records, such as lake sediments (e.g., Zolitschka, 1998), alluvial sediments (e.g., Macklin et al., 1991; Lang and Nolte, 1999; Taylor et al., 2000; Grossman, 2001; Larue, 2002; Kukulak, 2003; Daniels and Knox, 2005), or colluvial deposits (e.g., Lang and Hönscheidt, 1999; Foster et al., 2000; Lang, 2003; Bertran, 2004). From these archives, periods with higher and lower sedimentation rates can be identified and linked to climate and land use changes. However, the spatial resolution of the data is usually too low to derive spatially resolved quantitative information on soil erosion and sediment transport. Attempts to establish sediment budgets for longer time periods (e.g., the Holocene), by quantifying erosion and storage, and comparing slope and river dynamics are scarce (e.g., Meade, 1982; Macaire et al., 2001; Macaire et al., 2002). Nevertheless, studies like these, which characterize the entire system of sediment sources and sinks, are essential if we want to evaluate and model the impact of land use or climate change on soil erosion and sediment transfer through river basins. The objectives of this study are to estimate the sediment volume stored in the Holocene alluvium of a representative small river in the Belgian loess belt and to establish a chronology of sediment deposition. Furthermore, estimates of Holocene soil erosion intensities and colluvial sediment storage from previous work (Rommens et al., 2005) in this catchment are supplemented with the new data to establish a preliminary sediment budget for this river catchment. 2. Materials and methods 2.1. Study area The Nethen is a small tributary of the Dijle River, in central Belgium (Fig. 1). It has a length of 13km, and a catchment area of 52km 2. Soils in the area developed on silt loam and sand. The silt loam originates from Late- Pleistocene loess deposits (e.g., Tavernier, 1948; Gullentops, 1954; Van Vliet and Langohr, 1981). These are underlain by Tertiary marine deposits, which are mainly sandy, except for the lowest parts of the Nethen basin where in places Tertiary clays outcrop. The dominant soil type in the region is the Luvisol (FAO, 1998). The topography in the Nethen catchment can be characterised as gently rolling hills alternating with flat plateaus. The majority of the slopes in the catchment are less than 5% inclined, but slope gradients up to 50% occur along the major axis of the Nethen valley. The slopes oriented to the south are usually steeper, making the Nethen valley asymmetric. This is a remnant of the Pleistocene as the NNE-oriented slopes received a thicker loess cover (Goossens, 1987). The width of the alluvial plain does not exceed 500m. Floodplains are generally poorly drained and used for pasture or forestry (poplar wood: Populus sp.), whereas the loamy soils in the rest of the catchment are intensively used as arable land. Population in the catchment is concentrated in some villages along the River Nethen. Large parts of the Nethen river channel remained more or less unregulated, whereas in places watermills and fishponds were introduced from Medieval times ( A.D.) onward. Most of these constructions have disappeared by now, but they are clearly indicated on old maps; and some traces can still be recognized in the landscape morphology. Research locations were chosen away from these sites to minimize the bias caused by local anthropogenic disturbances.

84 T. Rommens et al. / Geomorphology 77 (2006) Fig. 1. Topographic map of the Nethen catchment and studied cross-sections in the Nethen-alluvium. Central Belgium has a temperate oceanic climate with a mean annual rainfall which varies between 700 and 800 mm and is well distributed throughout the year Methods Soil and sediment successions were studied in cross-valley augering transects at eight locations along the Nethen and in three large tributaries. An Edelman hand auger was used to recover sediment and to establish the stratigraphy and texture of sediment deposits. Presence of CaCO 3 in the soils and sediments was tested with HCl (1M). Samples for texture analysis and AMS 14 C dating were taken from cores retrieved with a percussion drill. In total, 106 auger cores and 9 percussion drills were retrieved. Laboratory texture analysis was done with the sievepipette method. Carbonates were determined by digestion with HCl, and organic matter was determined by H 2 O 2 oxidizing. For calculating sediment volumes, the alluvial plain was divided into twelve homogenous zones, and each of them represented by a central transect. Based on the augering results stratigraphical cross sections were drawn for each transect. In each cross section, three

85 190 T. Rommens et al. / Geomorphology 77 (2006) different alluvial units could be distinguished (see below). The volume of each unit was calculated as (Fig. 2): V alluv:unit ¼ A alluv:unit dðmeanþ ¼A alluv:unit ðcs=wþ where: CS V alluv.unit A alluv.unit d(mean) W ð1þ is the cross-sectional area of a unit in a specific transect (m 2 ); is the volume of sediment (m 3 ) of a unit represented by cross section CS; is the area (m 2 ) of the alluvial zone represented by cross section CS; is the mean depth of sediments along the cross section; and is the width (m) of the alluvial plain at the central transect. For uniform sediment bodies, the sediment mass in an alluvial unit (M alluv.unit,t) can be calculated as M alluv:unit ¼ V alluv:unit DBD ¼ A alluv:unit ðcs=wþdbd ð2þ where DBD is the dry bulk density of the sediment (t/m 3 ). In the alluvial fill of the Nethen, peat and organic rich sediments intercalate with minerogenic sediments. For these mixed sediment bodies a different approach was used. The mass of clastic sediment in each stratum can be calculated by taking into account the percentage of organic matter (%OM), the dry bulk density of the organic matter (DBD OM ) and the dry bulk density of the clastic component (DBD ms ) (e.g., Verstraeten and Poesen, 2001a): M alluv:unit ¼ V alluv:unit %OM=100 DBD om 1 : þ 1 ð%om=100þ DBD ms ð3þ The calculation of minerogenic sediment masses thus requires several parameters to be estimated, each of which can only be quantified with a certain precision. The overall errors of sediment volumes and masses can be estimated using Gaussian error propagation if all variables are independent, and values are normally distributed. For u=f(x 1,x 2, x j ) and all x j s are independent, Su, the standard deviation of u, is given by vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ux j Su ¼ t ðau=ax i Þ 2 dðsx i Þ 2 ð4þ i¼1 Fig. 2. Calculation of alluvial sediment volume. W: width of alluvial plane at the location of a transect; A: area of the alluvial zone; S: area of the sediment fill estimated in the cross section; d(max): maximum depth of Holocene alluvium; d(mean) =S/ W.

86 T. Rommens et al. / Geomorphology 77 (2006) where Sx i is the standard deviation of x; and Δu, the standard error of u, is given by: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ux j Du ¼ t ðau=ax i Þ 2 dðdx i Þ 2 ð5þ i¼1 where Δx i is the standard error of x i (Parrat, 1961). Relative errors can be calculated by: REðuÞ ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P j i¼1 ðau=ax i Þ 2 dðreðx i Þdx i Þ 2 u ð6þ with RE(x i ) the relative error on x i. Errors on the measurement of the coring depth (d; Fig. 2), dry sediment bulk density and alluvial plain area (A alluv.unit ) on the digital soil map are independent and normally distributed. At several locations, parallel cores with hand augers and percussion corers were taken. From these, a relative error of 5% on the observed depths was estimated. A value of 1.42±0.13t/m 3 was used for DBD min.sed. of the alluvium. This is the average value and the standard deviation of 16 samples. DBD O.M. was taken from the literature as 0.35±0.15t/m 3 (e.g., Brady, 1990; Locher and de Bakker, 1994; Verstraeten and Poesen, 2001b). Digitalization of the alluvial polygons, with an accuracy of 2mm, on the 1:10,000 soil map causes an error on the area of less than 10%. Imprecision introduced by the methodology for discriminating the sediment units are difficult to account for. The delineation of these units was not always straightforward and, after all, based on a relatively limited amount of augering interpretations. Conditions for Gaussian error propagation may not be valid here. The uncertainty on the limits between sediment units was therefore estimated at ± 0.25 m, and this error margin was introduced into the volume calculations. The position of the border of the alluvial deposits on the soil map (Baeyens et al., 1957; Baeyens, 1958; Baeyens and Dudal, 1959) was evaluated during the field work and proved to be fairly accurate. Imprecision on the soil map was further not taken into account. Laboratory analyses of 48 samples showed typical contents of organic matter (%OM) in the order of 0.5% for sediments that were classified as poor in organic matter, 2.5% for Fig. 3. Sediment profile from drilling core extracted in transect d (see Fig. 4, core at c. 90m). Sediment texture according to the USDA classification (Soil Survey Staff, 2003).

87 192 T. Rommens et al. / Geomorphology 77 (2006) sediments rich in organic matter, and 20% for peaty sediments. These values were considered constant in the error calculation, because organic matter content and bulk density are correlated. Sensitivity of the results to varying values of %OM was tested by repeating the calculations using minimal and maximal values of the dataset. This yielded results which only differed 3% to 15%. 3. Results 3.1. Characteristics of the Nethen sediments The valley of the River Nethen is filled with Holocene sediments that reach a depth of 4 to 6m in the axis of the valley. Pre-Holocene sediments were not recovered or recognized in every augering. However, in most of the transects early Holocene peaty strata, Late-Pleistocene calcareous loess, or even Tertiary sands or clays were observed in some of the cores; so it was possible to approximate the depth of Holocene sediments. The sediment profile in a typical drilling core can be seen in Fig. 3. In Figs. 4 and 5, a cross section and the longitudinal section along the Nethen River are shown, respectively. The majority of the Holocene sediments consist of silt loam, loam, and sandy loam deposits (Soil Survey Staff, 2003). Sands are rarely found. Although Tertiary sand crops out at several places in the study area (especially on the steepest slopes), its contribution to the texture of the alluvial fill of the Nethen valley is rather limited. Fig. 4. Cross section d through the valley bottom of the Nethen River (see Fig. 1 for location).

88 T. Rommens et al. / Geomorphology 77 (2006) Fig. 5. Longitudinal profile of the Nethen River. Sampling locations for 14 C-dating are indicated with an asterisk (*). Note that vertical scales for topography (left) and augerings (right) are different. (Legend: see Fig. 4.) The stratigraphy of the deposits along the Nethen shows a rather complex alternation of silty and loamy strata, peaty soils, and peat of variable thickness and at variable depths. However, two important peat strata can be distinguished when analysing the augering results. Unit 1 has its top surface at a depth of about 3.5m, and unit 2 is found at approximately 2m depth (Fig. 6). In several augerings, tufa deposits are found, together with other calcareous remains (shells). These calcareous strata are generally located at depths greater than 2.75m (Fig. 6) Dating results Dating efforts were concentrated to determine when the phases of peat growth and organic matter accumulation stopped (Table 1). For example, peat samples were taken from a core in transect b (Fig. 1) at depths of about 2 and 3.70m. The calibrated radiocarbon age of peat at m depth is B.C. (2σ). A peat stratum at 1.90m depth in the same core returned a calibrated radiocarbon age of A.D (2σ). At three other locations in the Nethen valley (transects c, d and h; see Fig. 1), the ages of peat strata overlain by mineral sediments were determined. The calibrated radiocarbon age of peat from 2.30m depth is A.D (2σ), from 1.60m depth A.D (2σ) and from the deepest peats at 3.20 and 5.50m B. C. (2σ) and B.C. (2σ), respectively. In one of the tributaries of the Nethen (transect l), organic matter at 1.60 and 2.05m depth returned an age of A.D Two previously published radiocarbon dates (Mullenders et al., 1966) were used as well. The samples were taken from peat layers at 5.10 and 4.10m depth and date from B.C. to B.C., respectively Holocene sediment volumes and masses in the Nethen alluvial fill Holocene sediment deposits were divided into three strata (Fig. 6). Unit 1 mainly exists of peat and tufa, below 3 to 3.5m depth. On top of that lays unit 2, which generally ends in a peat at a depth of 2m. There begins the uppermost and youngest loamy stratum (unit 3).

89 194 T. Rommens et al. / Geomorphology 77 (2006) Fig. 6. Frequency analyses of the depth to peat strata and the minimum depth to tufa in cores from the alluvium. Calibrated age (2σ) and depth of available samples from different locations in the alluvium are shown. Sediment volumes and masses were calculated for these three alluvial units using Eqs. (1) and (3) (Table 2). A total volume of Holocene sediment in the order of (10.92±1.94) 10 6 m 3 is stored along the Nethen and its tributaries. This corresponds to a total clastic sediment mass of (13.77±6.11) 10 6 t. About half of this volume is stored in unit 3, the youngest stratum. Based on the radiocarbon dating results, a chronological framework for the phases of sediment deposition of units 1, 2, and 3 was established. Phase 1 probably starts at the beginning of the Holocene (9600 B.C.) and ends around 2900 B.C. (the most probable youngest date obtained on unit 1; Lv. 277; B.C.) at the end of the Atlantic Period. Phase 2 starts around 2900 B.C. and ends around A.D. 1000, in Medieval times. This age is chosen because it lies in the middle of the interval [A.D. 640 A.D.1430], which contains the five radiocarbon ages obtained on peaty strata directly underlying unit 3. Consequently, 400year was assigned as uncertainty from the deviation of these dating results. The average accumulation rate (Table 2; Fig. 7) in the alluvial plain for the entire Holocene is 0.4±0.1mm a 1. Before the Middle Ages, accumulation rates were low ( 0.2 to 0.3mm a 1 ) and mainly from peat growth, whereas during the last 1000year it was an order of magnitude higher (2.5±1.2mm a 1 ) and mainly reflects clastic deposition A sediment budget for the Nethen catchment In a 100ha, zero-order basin in the southern part of the Nethen-catchment (see Fig. 1), eroded and deposited sediment volumes were calculated, based on soil truncation and colluvial deposition (Rommens et al., 2005). The small catchment was divided into morphometric units according to slope gradient (0 3%, 3 5%, 5 8%, and N8%), and a fifth unit for the valley bottom, containing the footslopes (b8%) and the concave thalweg (Fig. 8A). Average soil profile truncation as well as the average depth of the colluvium in each of the morphometric units was estimated based on a dense network of soil augerings. The small basin is seen as characteristic for most of the Nethen catchment. To obtain a tentative sediment budget for the entire

90 T. Rommens et al. / Geomorphology 77 (2006) Fig. 7. Evolution of the average sediment accumulation depth in the Nethen alluvial plain during the Holocene. catchment the same morphometric units were calculated for the Nethen catchment based on the digital elevation model of the area. Erosion and deposition were calculated as: E tot ¼ E mean A unit DBD D tot ¼ D mean A unit DBD ð3þ ð4þ where E tot and D tot are total eroded and deposited sediment masses (t) for a morphometric unit; E mean and D mean are mean erosion and deposition magnitude (m) for a morphometric unit that were determined in a characteristic 100ha zero-order basin (Rommens et al., 2005); A unit is the area of the morphometric unit (m 2 ); and DBD is the mean dry bulk density of the sediment (t/m 3 ). Here as well measurement errors of 10% and 5% were assigned to areas and depths, respectively. The dry bulk density of 32 samples (18 from soil and 14 from sediment) collected in zero-order basins in the study area was 1.51 ± 0.10 (average value and standard deviation). Propagation of these errors was calculated using Eq. (5). Monte Carlo simulations showed that the uncertainty on E mean and D mean due to the limited augering density is in the order of 10% to 30%. Taking into account the possible difficulties in soil and sediment profile interpretation and the augering densities for each class (see: Rommens et al., 2005), an uncertainty of 10% was assigned to the mean erosion depths on the slopes and in the valleys, 30% for average erosion and deposition in the plateau class (slope gradient: 0 3%), and 20% for other deposition heights. Results are listed in Table 3 and depicted in Fig. 8B. Overall (47.85±9.08) 10 6 t of sediment have been eroded in the Nethen catchment during the Holocene period. Almost 50% of this mass ((23.75 ±8.54) 10 6 t) is stored in colluvial deposits on the plateau and footslopes and in dry valley bottoms. Twenty-nine percent ((13.77 ± 6.11) 10 6 t) was deposited along permanent streams of the Nethen and its tributaries. The rest of the eroded sediments (21%, or (10.33 ± 19.54) 10 6 t) left the Nethen catchment and was delivered to the Dijle River, the higher order stream. 4. Discussion 4.1. Sedimentation history of the Nethen valley Sediment survey and dating results from this study, together with results of previous research in the region (Mullenders et al., 1966) allows us to differentiate the

91 196 T. Rommens et al. / Geomorphology 77 (2006) Fig. 8. (A) Schematic representation of the geomorphic units (plateaus, hillslopes, thalwegs, alluvium) in the Nethen catchment, used for calculating the sediment budget. (B) Holocene sediment distribution over these units in the Nethen catchment ( export from each unit is calculated as the difference between estimated erosion and deposition). alluvial fill of the Nethen into three sedimentation phases. Sediment unit 1 is characterized by peat and peaty loam, intercalated with tufa and is located between 3 and 6 m depth. Its accumulation ended around 2900 B.C., and thus it represents deposits from the early Holocene biozones Preboreal (starting at 9600 B.C.), Boreal and Atlanticum. The peat stratum of the Atlanticum has already been described and dated at several locations in the alluvial plain of the Dijle River, downstream of the Nethen (e.g., Mullenders and Gullentops, 1957; De Smedt, 1973). Moreover, the study of travertine deposits and tufa in a number of small Belgian river valleys indicates that these deposits mainly formed during the Boreal and Atlanticum biozones and ceased in the second half of the Holocene (Geurts, 1976). Our data confirm these findings.

92 T. Rommens et al. / Geomorphology 77 (2006) Table 1 Radiocarbon ages for the Nethen sediments Depth (cm) Radiocarbon age (lab code) Calibrated age a Location, core Sampled material 14 C-dates obtained in this study ±30 BP (Beta ) / A.D. Beauvechain (Transect b) Peaty stratum (c. 0.40m thick) ±40 BP (Beta ) B.C. Beauvechain (Transect b) Peaty stratum (c. 0.10m thick) in between tufa ±40 BP (GrA-28419) B.C. Hamme Mille (Transect c 1 ) Peaty stratum (c. 0.60m thick), above tufa layer ±40 BP (GrA-28456) / B.C. Hamme Mille (Transect c 2 ) Wood fragment ±40 (GrA-28458) / B.C. Hamme Mille (Transect c 2 ) Peat and tufa layers (N0.40m thick) ±40 BP (Beta ) A.D. Hamme Mille (Transect d) Peaty stratum (c. 0.40m thick) ±40 BP (Beta ) A.D. Nethen (Transect h 1 ) Thick peat (N3m thick) ±25 BP (KIA-28287) / A.D. Nodebais (Transect l) Loam, rich in organic matter ±25 BP (KIA-28286) / A.D. Nodebais (Transect l) Loam, rich in organic matter 14 C-dates extracted from Mullenders and Gullentops, ±150 BP (Lv-277) B.C. Nethen (Transect h 2 ) Peat (c. 1.20m thick) ±150 BP (Lv-279) B.C. Nethen (Transect h 2 ) Peat (c. 0.60m thick) a Calibrated at the 2σ confidence level, using OxCal v3.10 (Bronk Ramsey, 1995; Bronk Ramsey, 2001). Calibration curve based on atmospheric data from (Reimer et al., 2004). Between c. 2 and c. 3.5m depth, a second unit can be defined. Unit 2 has a high content of organic material (soil and peat remnants) but is almost free of tufa. Radiocarbon dating places the formation of these sediments in the time span between 2900 B.C. and the Middle Ages (1000 A.D.). The youngest unit (phase 3) generally consists of clastic sediments that were deposited during the last 1000 year. These homogeneous sediments show few traces of old soil surfaces or peat accumulation, which indicates a fast accumulation of sediments. Delineation of the three sediment units is tentative, and radiocarbon ages so far do not allow studying the chronology of the valley infill in detail. Yet, morphological characteristics of the units differ enough to establish a reasonable temporal framework, which enables us to calculate average sedimentation and accumulation rates. These sedimentation rates probably varied over space and time. This explains why the deepest sample from the alluvial sediments (LV-279) is not the oldest one. However, average rates for each phase were similar over the whole alluvial plain and illustrate the dramatic shift in floodplain sedimentation (Table 2, Fig. 7): Over the last 1000 year, the accumulation rate has increased almost tenfold compared to earlier periods. This caused the end of peat formation and impeded the development of organic soils in most of the alluvium. Obviously, there has been a dramatic increase in sediment supply towards the Nethen River from Medieval times onwards. Possible explanations are deforestation and intensification of agriculture, which took place in large parts of Europe during that period (e.g., Aerts et al., 1985). Similar effects of vast historical Table 2 Holocene sediment volumes and masses stored in the Nethen alluvium Deposition phase Volume ( 10 6 m 3 ) Accumulation rate (mm a 1 ) Mass of mineral sediments ( 10 6 t) Sedimentation rate (t ha 1 a 1 ) Phase ± ± ± ± B.C. Phase ± ± ± ± B.C A.D. Phase ± ± ± ± A.D. present 9600 B.C. present 10.92± ± ± ±1.9 Average accumulation and sedimentation rates in the alluvial plain of the Nethen.

93 198 T. Rommens et al. / Geomorphology 77 (2006) Table 3 Tentative (Holocene) sediment budget for the Nethen catchment Geomorphic unit Erosion ( 10 6 t) Deposition ( 10 6 t) Plateau, 0 3% 9.65± ±3.03 Slope, 3 5% 9.20± ±0.50 Slope, 5 8% 9.42± ±0.09 Slope, N8% 15.61± ±1.37 Thalweg (b8%) 3.98± ±3.54 Total (plateau and dry valleys) 47.85± ±8.54 Alluvium? 13.77±6.11 Total (Nethen catchment) 47.85± ±14.65 Sediment export: 10.33± t Sediment delivery ratio: 22±41% deforestations were observed in the sediment archives of other European rivers (e.g., Larue et al., 1999; Larue, 2002; Macaire et al., 2002; Kukulak, 2003; Lespez, 2003). The extension of arable land and the cultivation of less favorable soils on steeper slopes and in the direct vicinity of permanent streams can improve coupling (sediment connectivity) between the plateau, slopes, and rivers, and facilitate sediment transfer through the fluvial system (e.g., Brunsden and Thornes, 1979; Brunsden, 2001; Harvey, 2001; Harvey, 2002). Holocene climate variability or extreme rainfall events may also have been an important driving factor (e.g., Bork and Bork, 1987; Kalis et al., 2003; Daniels and Knox, 2005). To evaluate the relative importance of human and climate related factors, however, sediment archives with a high resolution temporal framework are needed. So far, geomorphic research in central Belgium has pointed the transition from the Subboreal to the Subatlantic Period (ca BP) as the start of massive alluvial sedimentation in the river valleys (e.g., Tavernier, 1948; Mullenders et al., 1966; De Smedt, 1973). Peat strata and soil remnants within the Late Holocene sediments were given little attention. Our results show that in the alluvial plain of the Nethen large patches of peat continued to accumulate and stable phases of soil formation occurred after the Atlanticum, at least until the early Middle Ages. The majority of the silty and loamy sediments accumulated in the alluvial plain only later. In other words, the time span between the onset of the Subatlanticum and 1000 A.D. was of minor importance for the infill of river valleys Sediment budget As shown in Table 3, considerable uncertainties are involved in the calculation of a long-term sediment budget. Nevertheless, an attempt is made to give an idea of the enormous amounts of sediment that have been transported through time. High errors on the values for sediment export and sediment delivery ratio are inevitable. We should also stress that no value is available for riverbank erosion. The Nethen is a meandering river and thus riverbank erosion does occur but this is limited to a narrow meander belt. Therefore, it does not affect the entire alluvial plain, but probably b10% of it. To evaluate the overall accuracy of the sediment budget, the calculated sediment export value is compared to the output of an empirical relationship between catchment area (A, ha) and specific sediment yield (SSY, t ha 1 a 1 ), obtained for 26 cultivated catchments in the central Belgian loess belt over a period of 2 to 46 years during the 20th century (Verstraeten and Poesen, 2001a). This formula states that SSY=26A For the Nethen catchment, this results in an average contemporary annual sediment yield of 1.30t ha 1 a 1. Using this figure, the mean annual sediment export for the Nethen catchment can be calculated to be 6800t a 1. If we extrapolate this to the Holocene period ( 11,600years) the calculated sediment export becomes t. Because the formula used here was derived for catchments under intensive present-day agriculture, it is certainly not applicable for the entire Holocene period. For the last 2500year, which is the estimated period of agricultural land use in the region (Rommens et al., 2005), sediment export comes to t, whereas for the period in which sediment unit 3 was deposited (1000 A.D. present) a value of t is obtained. So, it is plausible that the real export value lies somewhere between and t, which corresponds well with the estimated export of c t obtained in the sediment budget. As expected, the majority of the sediments entering the Holocene sediment pathway originate from the steeper slopes. Some sediment is produced on the plateaus, but most of it is deposited in close proximity to its source. Overall, the sediment budget demonstrates that a substantial amount (50%) of sediment is stored in the zero-order valleys and has not, until now, been transferred to higher order streams. Moreover, dating of colluvial sediments suggests that soil erosion and sediment deposition in the zeroorder valleys at the fringes of the plateau started in the Early Iron Age (c B.C.) (Rommens et al., 2005) and was already considerable in the Roman Age (c A.D.) (Rommens, unpublished data).

94 T. Rommens et al. / Geomorphology 77 (2006) This is much earlier than the massive sediment deposition phase (unit 3) in the alluvial plain, which apparently only started at least 500 years later (during the Middle Ages, A.D.). So, deforestation of part of the catchment clearly started well before medieval times. Probably, this has triggered severe soil erosion all over the plateau and in many dry valleys, but because of insufficient coupling between slopes and permanent streams, for instance, it did not immediately induce a rise of sedimentation rates along the Nethen River. Sediment transport at a catchment scale is thus very likely to follow a nonlinear pathway (e.g., Brierly and Fryirs, 1999; Lang and Hönscheidt, 1999; Malmon et al., 2003; Lang et al., 2003), influenced by complex long-term interactions of sediment sources and sinks, which change through time. Sinks can switch into sources and vice versa, depending on human and climatic impact on the environment. 5. Conclusions In the alluvial deposits of the Nethen, three sediment units could be distinguished on the basis of sedimentological properties and dated using 14 C. The youngest unit, which started to accumulate during the Middle Ages (c A.D.), contains 50% of the total Holocene sediment mass. Mean sedimentation rates for these youngest deposits are almost 10 times as high as accretion rates earlier in the Holocene. The sediment budget for the Nethen river catchment demonstrates that most of the Holocene sediments are stored as colluvial deposits on footslopes and in dry zero-order valley bottoms. Only 29% of the eroded material was delivered to the valleys of permanent streams, where it was stored as alluvial deposits, and only 21% has been transferred through the catchment and delivered to the higher order stream. A study of colluvial sediments in the upstream area of the Nethen shows that, in the time span between the onset of the Iron Age (c B.C.) and A.D. 1000, a large volume of sediment was produced and stored in dry valleys. During that period, very little deposition took place in the alluvium. So, there exists a time lag of about one millennium between the higher erosion and sedimentation rates in the upstream area and the increase in sediment supply to the river valley. Early deforestations on the plateau, for instance, are therefore not necessarily represented in the contemporary sediment record of the alluvial plain downstream. This raises questions concerning the reliability of studies that aim to reconstruct environmental change in river or lake catchments by analysing the sediment archive at one location only. Further research to unravel the longterm routing mechanism of sediments through nonlinear sediment pathways could clarify this further. Acknowledgements The Ph.D. students and the technical staff of the Physical and Regional Geography Research Group are thanked for their contribution to the fieldwork. This project is funded by the K.U.Leuven, grant no. 3E The GPS equipment was granted by the Fund for Scientific Research Flanders (Belgium), grant References Aerts, E., Dupon, W., Van der Wee, H., De economische ontwikkeling van Europa: documenten-deel I-Middeleeuwen. Leuven University Press, Leuven. Baeyens, L., Bodemkaart van België. Kaartblad 104W Meldert. IWONL. Baeyens, L., Dudal, R., Bodemkaart van België. Kaartblad 103W Duisburg. IWONL. Baeyens, L., Tavernier, R., Scheys, G., Bodemkaart van België. Kaartblad 103E Hamme-Mille. IWONL. Bertran, P., Soil erosion in small catchments of the Quercy region (southwestern France) during the Holocene. The Holocene 14 (4), Bork, H.-R., Bork, H., Extreme jungholozäne hygrische Klimaschwankungen in Mitteleuropa und ihre Folgen. Eiszeitalter und Gegenwart 37, Brady, N.C., The Nature and Properties of Soils. McMillan, New York. Brierly, G.J., Fryirs, K., Tributary-trunk stream relations in a cutand-fill landscape: a case study from Wolumla Catchment, New South Wales, Australia. Geomorphology 28, Bronk Ramsey, C., Radiocarbon calibration and analysis of stratigraphy: the OxCal program. Radiocarbon 37 (2), Bronk Ramsey, C., Development of the radiocarbon program OxCal. Radiocarbon 43 (2A), Brunsden, D., A critical assessment of the sensitivity concept in geomorphology. Catena 42, Brunsden, D., Thornes, J.B., Landscape sensitivity and change. Transactions of the Institute of British Geographers, New Series 4, Daniels, J.M., Knox, J.C., Alluvial stratigraphic evidence for channel incision during the Mediaeval Warm Period on the central Great Plains, USA. The Holocene 15 (5), De Roo, A.P.J., Wesseling, C.G., Jetten, V.G., Offermans, R.J.E., Ritsema, C.J., LISEM, Limburg Soil Erosion Model, User Manual. Department of Physical Geography, Utrecht University, The Netherlands. De Smedt, P., Paleogeografie en kwartair-geologie van het confluentiegebied Dijle-Demer. Acta Geographica Lovaniensia 11, 130.

95 200 T. Rommens et al. / Geomorphology 77 (2006) FAO, World Reference Base for Soil Resources. World Soil Resources Reports 84 Rome. Flanagan, D.C., Nearing, M.A., USDA-Water Erosion Prediction Project: Hillslope Profile and Watershed Model Documentation. NSERL Report No. 10, West Lafayette, IN. Foster, I.D.L., Mighall, T.M., Wotton, C., Owens, P.N., Walling, D.E., Evidence for Medieval soil erosion in the South Hams region of Devon, UK. The Holocene 10 (2), Fryirs, K., Brierly, G.J., Variability in sediment delivery and storage along river courses in Bega catchment, NSW, Australia: implications for geomorphic river recovery. Geomorphology 38, Geurts, M.-A., Genèse et stratigraphie des travertins de fond de vallée en Belgique. Acta Geographica Lovaniensia 16, 66. Goossens, D., Sedimentatiemechanismen bij natuurlijke stofdeeltjes in lucht. Ph.D. dissertation, K.U. Leuven, Belgium. Grossman, M.J., Large floods and climatic change during the Holocene on the Ara River, Central Japan. Geomorphology 39, Gullentops, F., Contribution à la chronologie du Pleistocène et des formes du relief en Belgique. Mémoires de l' Institut géologique de l' Université de Louvain 18 (4), Harvey, A.M., Coupling between hillslope and channels in upland fluvial systems: implications for landscape sensitivity, illustrated from the Howgill Fells, northwest England. Catena 42, Harvey, A.M., Effective timescales of coupling within fluvial systems. Geomorphology 44, Kalis, A.J., Merkt, J., Wunderlich, J., Environmental changes during the Holocene climatic optimum in central Europe-human impact and natural causes. Quaternary Science Reviews 22, Kukulak, J., Impact of Mediaeval agriculture on the alluvium in the San River headwaters (Polish Eastern Carpathians). Catena 51, Lang, A., Phases of soil erosion-derived colluviation in the loess hills of south Germany. Catena 51, Lang, A., Hönscheidt, S., Age and source of colluvial sediments at Vaihingen-Enz, Germany. Catena 38, Lang, A., Nolte, S., The chronology of Holocene alluvial sediments from the Wetterau, Germany, provided by optical and 14 C dating. The Holocene 9 (2), Lang, A., Bork, H.-R., Mäkel, R., Preston, N., Wunderlich, J., Dikau, R., Changes in sediment flux and storage within a fluvial system: some examples from the Rhine catchment. Hydrological Processes 17, Larue, J.P., Small valley bottom deposits in the sandy districts of the Sarthe basin (France): Climate and/or human origin? Geomorphology 45, Larue, J.P., Mahoué, J.-P., Monnier, J., Erosion in cultivated soils and river morphodynamics: The example of the Tortue Basin (Sarthe, France). Geodinamica Acta 12 (2), Lespez, L., Geomorphic responses to long-term land use changes in eastern Macedonia (Greece). Catena 51, Locher, W.P., de Bakker, H., Bodemkunde van Nederland. Deel 1: Algemene bodemkunde. Malmberg, Den Bosch. Macaire, J.-J., Bellemlih, S., Cocirta, C., De Luca, P., Di-Giovanni, C., Gay-Ovejero, I., Quantification des flux et des stocks solides holocènes dans le bassin de la Loire (France). La Houille Blanche 8, Macaire, J.-J., Bellemlih, S., Di-Giovanni, C., De Luca, P., Visset, L., Bernard, J., Sediment yield and storage variations in the Négron River catchment (south western Parisian basin, France) during the Holocene period. Earth Surface Processes and Landforms 27, Macklin, M.G., Passmore, D.G., Stevenson, A.C., Cowley, D.C., Edwards, D.N., O'Brian, C.F., Holocene alluviation and land-use change on Callaly Moor, Northumberland, England. Journal of Quaternary Science 6 (3), Malmon, D.V., Dunne, T., Reneau, S.L., Stochastic theory of particle trajectories through alluvial valley floors. Journal of Geology 111, Meade, R.H., Sources, sinks, and storage of river sediments in the Atlantic drainage of the United States. Journal of Geology 90, Morgan, R.P.C., Quinton, J.N., Smith, R.E., Govers, G., Poesen, J., Auerswald, K., Chisci, G., Torri, D., Styczen, M.E., The European soil erosion model (EUROSEM): a dynamic approach for predicting sediment transport from fields and small catchments. Earth Surface Processes and Landforms 23, Mullenders, W., Gullentops, F., Palynologisch onderzoek in de alluviale vlakte van de Dijle, te Heverlee-Leuven. Agricultura 5 (1), Mullenders, W., Gullentops, F., Lorent, J., Coremans, M., Gilot, E., Le remblaiement de la vallée de la Nethen. Acta Geographica Lovaniensia 5, Page, M.J., Trustrum, N.A., Dymond, J.R., Sediment budget to assess the geomorphic effect of a cyclonic storm, New Zealand. Geomorphology 9, Parrat, L.G., Probability and Experimental Errors. Wiley, New York. Reid, L.M., Dunne, T., Sediment budgets as an organizing framework in fluvial geomorphology. In: Mathias Kondolf, G., Piégay, H. (Eds.), Tools in Fluvial Geomorphology. Wiley, New York, pp Reimer, P.J., Baillie, M.G.L., Bard, and 26 others, IntCal04 terrestrial radiocarbon age calibration, 0 26Cal Kyr BP. Radiocarbon 46 (3), Rommens, T., Verstraeten, G., Poesen, J., Govers, G., Van Rompaey, A., Peeters, I., Soil erosion and sediment deposition in the Belgian loess belt during the Holocene: establishing a sediment budget for a small agricultural catchment. The Holocene 15 (7), Soil Survey Staff, Keys to Soil Taxonomy. USDA-NRCS, Washington DC. Tavernier, R., Les formations quaternaires de la Belgique en rapport avec l'évolution morphologique du pays. Bulletin de la Société Belge de Géologie 62, Taylor, M.P., Macklin, M.G., Hudson-Edwards, K., River sedimentation and fluvial response to Holocene environmental change in the Yorkshire Ouse Basin, northern England. The Holocene 10 (2), Trimble, S.W., Decreased rates of alluvial sediment storage in the Coon Creek Basin, Wisconsin, Science 285, Van Oost, K., Govers, G., Desmet, P., Evaluating the effects of changes in landscape structure on soil erosion by water and tillage. Landscape Ecology 15, Van Vliet, B., Langohr, R., Correlation between fragipans and permafrost with special reference to silty Weichselian deposits in Belgium and northern France. Catena 8,

96 T. Rommens et al. / Geomorphology 77 (2006) Verstraeten, G., Poesen, J., 2001a. Variability of dry sediment bulk density between and within retention ponds and its impact on the calculation of sediment yields. Earth Surface Processes and Landforms 26, Verstraeten, G., Poesen, J., 2001b. Factors controlling sediment yield from small intensively cultivated catchments in a temperate humid climate. Geomorphology 40, Walling, D.E., Russel, M.A., Hodgkinson, R.A., Zhang, Y., Establishing sediment budgets for two small lowland agricultural catchments in the UK. Catena 47, Zolitschka, B., A 14,000year sediment yield record from western Germany based on annually laminated lake sediments. Geomorphology 22, 1 17.

97 The Holocene 17,6 (2007) pp Reconstruction of late-holocene slope and dry valley sediment dynamics in a Belgian loess environment Tom Rommens, 1 Gert Verstraeten, 1 * Iris Peeters, 1 Jean Poesen, 1 Gerard Govers, 1 Anton Van Rompaey, 1,2 Barbara Mauz, 3 Susan Packman 3 and Andreas Lang 3 ( 1 Physical and Regional Geography Research Group, K.U. Leuven, Celestijnenlaan 200 E, B-3001 Heverlee, Belgium; 2 Fund for Scientific Research-Flanders, Belgium; 3 Department of Geography, University of Liverpool, Liverpool L69 7ZT, UK) Received 24 July 2006; revised manuscript accepted 20 February 2007 Abstract: To unravel the evolution of a dry valley in the Belgian loess area soils and sediments along a slope catena were studied in detail. A 67 m long trench was opened from the upper slope to the centre of the valley bottom. The exposed soils and sediments showed evidence of severe soil erosion and other human disturbances that significantly changed the valley topography. The early-holocene slope gradient was up to 25%, whereas now it is less than 10%. In the thalweg, a remnant of the early-holocene soil was found underneath colluvial deposits, which were more than 3 m thick. A chronology for the valley evolution was established based on AMS 14 C dating of charcoal fragments and optical dating of colluvial sediments. The first sediment deposition occurred in the early Iron Age, with an average sedimentation rate of approximately 3.4 ± 1.3 t/ha per yr. This increased to c. 5.4 ± 2.2 t/ha per yr during the Roman Period and further to 18.0 ± 2.2 t/ha per yr in the Middle Ages. Although sediment accumulation in the valley was substantial, soil-erosion processes were mainly lowmagnitude and signs of gullying are absent in the thalweg until the last few centuries. Key words: Holocene, soil erosion, sediment dynamics, radiocarbon dating, optical dating, loess, landscape evolution, Iron Age, Roman Period, Middle Ages, Belgium. Introduction Contemporary soil erosion in the Belgian loess area is well known and intensively studied. The aeolian silts (loess) that form the parent material of the present-day soil in this region are amongst the most erosion-prone materials in the world (Bryan and De Ploey, 1983; Govers, 1985). Contemporary erosion rates are in the order of several tonnes per hectare per year and can reach more than 10 t/ha per yr (Verstraeten et al., 2006). The main agent responsible for these high erosion rates is water, which causes sheet, rill and, at some locations, gully erosion. However, soil cultivation itself also causes soil translocation, known as tillage erosion. On arable land, this last form of erosion may represent up to 50% of the total sediment transport (Van Oost et al., 2000). All these processes redistribute large amounts of sediment within the landscape and therefore have a strong impact on the *Author for correspondence (gert.verstraeten@geo.kuleuven.be) topography on longer timescales. In other European loess areas, studies have shown evidence for localized soil erosion from Neolithic times onwards, and first signs of more widespread soil erosion in the Late Bronze Age or Early Iron Age (Bollinne, 1976; Kalis et al., 2003; Zolitschka et al., 2003). In the Belgian loess belt the tradition of agricultural land use can also be traced back to Neolithic times (Lindemans, 1994). As a result, soils on the slopes were truncated and valleys were buried under colluvial and alluvial sediments. The resulting soilscape is described in detail by Gullentops and Scheys (1950), Paulissen et al. (1981) and Langohr (2001), and Holocene sediment budgets were established for several small catchments (Desmet and Govers, 1995; Vandaele and Poesen, 1995; Rommens et al., 2005). Despite these efforts, detailed time frames for the Holocene sediment dynamics are still missing. Here we establish for the first time a chronology of Holocene slope evolution and valley infill by carefully analysing soils and sediments along a slope catena and the footslope colluvium in a 2007 SAGE Publications /

98 778 The Holocene 17,6 (2007) Figure 1 Location and view of the outcrop in the Nodebais catchment (central Belgium). The uppermost photograph shows the soils and sediments at the valley bottom, including a palaeosol overlain by colluvium dry valley. The chronology is established by combining AMS 14 C dating and OSL dating results. Based on this chronology, minimum sedimentation rates for various time periods are determined. The analyses were carried out in the Nodebais catchment, for which a Holocene sediment budget was already established (Rommens et al., 2005). Combining the information enables the calculation of sediment fluxes for the first time. Materials and methods Study area A small dry valley in Nodebais was selected as a representative site for the Belgian loess Belt (Figure 1). Here, Pleistocene loess deposits of varying thickness overlie marine sands and clays of Tertiary age. Holocene soils developed mainly in Weichselian loess. Older loess deposits are rare because of intense erosion and denudation during interglacial periods (Bolt et al., 1980). Reference Holocene soil profiles can only be found at stable locations on topographic plateaus that are free from soil erosion and sediment deposition (Rommens et al., 2005). They show that the initially calcareous loess is decalcified to a mean depth of 230 cm. Along with the decalcification, clay formation and mobilization led to the development of an illuviation horizon at between 0.40 m and 1.50 m depth. On the slopes and in the valleys, soils were truncated or buried, because of water and tillage erosion (Rommens et al., 2005). Today most of the area is under intense arable use for the cultivation of wheat and barley, sugar beets and chicory roots. The landscape has an open field character, with patches of woodland or pasture in areas where less fertile Tertiary sands crop out, and in the low-lying wetlands of the river valleys. Archaeological and historical research has elucidated the long occupation and cultivation history of the region. Earliest evidence of human presence goes back to Neolithic times (Martens, 1981). Traces of settlements and burial places from the Bronze Age (c cal. BC) and the Iron Age (c cal. BC) have been studied (Vincent and Vincent, 1909; Olyslager, 1960). From the Roman Age (c. AD cal.), remains of several houses are reported. One of these Roman villas was located close to (800 m NNE) the slope investigated here. Many of the farms that still exist today originate from at least Mediaeval times. Moreover, several old Christian monasteries owned large farming estates since early Mediaeval times (Tarlier and Wauters, 1873; Schayes, 1990; Bertrand, 2003). Even the so-called pristine forests in the area show evidence of anthropogenic disturbances and periods of deforestation (Vanwalleghem et al., 2004, 2005). Field observations and analyses Based on a survey of intense augering, a 67 m long and 2 m wide trench was opened on a WNW-facing slope (Figures 1 and 2). The trench ran perpendicular to the slope in WNW direction from the upper slope (~100 m a.s.l.) to the centre of the valley bottom (~94 m a.s.l.). While excavating the trench, care was taken to ensure that the trench penetrated all of the Holocene soils and sediments. Thus, the depth of the trench varied between 1 m in the upper part and 4 m in the centre of the valley.

99 Tom Rommens et al.: Slope and dry valley sediment dynamics in a Belgian loess landscape 779 Figure 2 (a) Overview of sediments and soils along the trench; present-day topography and estimated original topography. (b) A more detailed stratigraphy of the sediments and soils at the midslope section (for location see Figure 2a)

100 780 The Holocene 17,6 (2007) Table 1 Characteristics of the colluvial sediments in the thalweg, see Figure 3a Label Colour Description Samples a 10YR 4/4 (brown) Ploughing layer; silt loam. T1.1 b 10YR 5/6 (yellowish brown) Silt loam; many roots, worm holes and other T1.2; T1.3; T1.4; macro-pores (10 15/100 cm²); some pale LV76; LV77 mottles;, homogeneous sediment. c 10YR 6/6 (bright Yellowish brown almost horizontal layer, T1.5 yellowish brown) c. 7 cm thick; sandy loam; diffuse boundaries; silt loam pockets mixed in. d 10YR 5/6 (yellowish brown) Silt loam; very few macro-pores (< 5/100 cm²); T1.6; T1.7; LV75 some small pockets of sandy loam in homogeneous silty matrix. e 10YR 5/6 (yellowish brown) Silt loam; fuzzy boundaries; c. 10 cm T1.8; C6 thick; yellowish orange (10YR 8/4)) mottles; very small charcoal fragments. f 10YR 5/6 (yellowish brown) Silt loam; bright yellowish brown T1.9; T1.10; T1.11; (10YR 7/6) mottles in homogeneous matrix; T1.12; C1; C2; C4; few charcoal fragments, except for lowest C5 LV72; LV 73 ; LV74 sediments; some small sandstone fragments; irregular lower boundary. g 10YR 7/4 (dull yellowish Pale silt loam; oximorphic properties: iron T1.13; orange) depletion, bright reddish brown (5YR 5/6) C3; iron mottles, brownish black (10YR 2/3) LV71 Mn-oxides; high concentration of charcoal fragments. h 10YR 3/4 (dark brown) Clayey silt loam; whitish tongues (cracks or T1.14 root casts), 5 cm thick, cm long, filled with material from overlying horizon; large amounts of Mn-oxides; oxidized iron along the tongues; blocky structure; compact. i 10YR 4/6 (brown) Loam, sandy loam and loamy sand; thick layers and lenses; little clay; pale mottles and layers; few oxides; less compact than horizon h. j 5Y 8/4 (yellow) Marine sand (Tertiary) capped by gravel (sandstone and flint cobbles). C, charcoal sample; LV, OSL sample; T, sample for texture analysis. Table 2 Results of AMS 14 C dating of charcoal sampled in the thalweg sediment deposits (see Figure 3a for sampling location) Sample ID Lab code Depth below surface (m) Radiocarbon age a (yr BP) Calibrated radiocarbon age (yr BC/AD) a,b Sampling context 1 GrA ± / cal. BC Thalweg 2 GrA ± cal. BC Thalweg 3 GrA ± cal. BC Thalweg 4 GrA ± cal. BC Footslope 5 GrA ± 35 cal. AD Thalweg 6 GrA ± 35 cal. AD Thalweg 7 GrA ± 35 cal. AD Thalweg, gully fill 8 KIA ± cal. BC Pit fill a Given at 2σ confidence level. b Calibrated using Oxcal v.3.10 (2005) (Bronk Ramsey, 2001). Calibration curve based on atmospheric data from (Reimer et al., 2004) Soils and sediments in the NNE-facing exposure were described and sketched at 1:20 scale. Colours were defined with a Munsell colour chart, texture was estimated in the field and samples for laboratory grain size analysis (sieve-pipette method) were taken from the main soil horizons and sediment units (Table 1). The CaCO 3 -content of the loess was tested with HCl (1M). Supplementary information on soil and sediment stratigraphy was obtained from auger cores in the bottom of the trench and in its immediate surroundings. Laboratory analyses Phosphorus determination The total phosphorus content was measured by inductively coupled plasma mass spectrometry (ICP AES) on samples that were taken from a percussion core in a non-eroded reference soil over a depth of 1 m, and also in the thalweg colluvium next to the trench over a depth of 2 m. Phosphorus added through fertilizers is quickly immobilized and fixed to soil particles. Thus, at locations where no soil erosion or sediment deposition occurs, P is

101 Tom Rommens et al.: Slope and dry valley sediment dynamics in a Belgian loess landscape 781 Figure 3 (a) Exposure and stratigraphy of the thalweg sediments (for location of this section and legend see Figure 2a). (b) Clay content and median grain size for samples from the colluvial sediments (sample locations T2 and T1 in Figure 3a). Note the changing texture (median grain sizes) in T1 at ~1.40 m and ~2 m. The changes correspond to layers c and e stored only in the plough layer. Ploughing results in homogeneous distribution of P over the upper 25 to 30 cm of the soil, whereas the rest of the soil profile does not receive extra P. Leaching of P is only possible in case of P saturation, but because of the high P-holding capacity of the silt loam this is not likely (Steegen et al., 2000). P-rich sediments derived from the soil surface are mobilized by soil erosion and accumulate in sedimentation zones (Halbfass and

102 782 The Holocene 17,6 (2007) Table 3 Analytical data and OSL dating results: sample code and depth, water content (, in % of dry sample weight), U, Th and K concentrations, effective dose rate (dd/dt effective ), equivalent dose (D e ), number of aliquots used for D e calculation (n) and OSL ages (see Figure 3a for sampling location) Lab code Sampling (%) U (µg/g) Th (µg/g) K (%) dd/dt effective D e (Gy) n OSL OSL age depth (m) (Gy/ka) age (ka) a (calendar year) b LV ± ± ± ± ± ± BC LV ± ± ± ± ± ± BC LV ± ± ± ± ± ± 0.08 AD LV ± ± ± ± ± ± BC LV ± ± ± ± ± ± BC LV ± ± ± ± ± ± BC LV ± ± ± ± ± ± 0.03 AD LV ± ± ± ± ± ± BC a OSL ages were converted to calendar years by subtracting 2005 (the year of measurement) from the OSL age and rounding to the next decade. b Given at 1σ confidence level. Grunewald, 2003; Heckrath et al., 2005). In colluvial deposits one can thus expect enrichment in P to depths greater than the normal ploughing depth. Data on the Belgian production and consumption of P fertilizers indicate that larger quantities of phosphorus were applied to the fields only after AD (Steegen et al., 2000). The depth to which higher P concentrations occur can thus be correlated to ~ AD 1920 and used to calculate an accumulation rate since then (Hofman and Verloo, 1989; Swinnen and Tollens, 1989; Steegen et al., 2000). AMS 14 C chronometry Eight charcoal fragments were collected for AMS 14 C dating (Tables 1 and 2; Figures 2b and 3a). In the field, samples were placed in polyethylene sampling bags, and in the laboratory charcoal remains were cleaned using distilled water and dried. Seven samples were sent for analysis to the Groningen Centre for Isotope Research (the Netherlands). One sample was analysed in the Royal Institute for the Study and Conservation of Belgium s Artistic Heritage (Belgium). OSL chronometry Optically stimulated luminescence (OSL) dating of soil-erosion derived colluvium has been used in the past (Lang, 2003). Compared with other techniques, OSL dating offers the possibility of determining the time when sediment grains were last exposed to daylight often the time of sediment deposition. So far, the majority of these OSL studies of loess have utilized the luminescence signal from feldspars, and employed multiple-aliquot additive-dose procedures. Here we use a single-aliquot regenerative-dose (SAR) protocol (Murray and Wintle, 2000) on fine-grained quartz to (1) exclude anomalous fading often associated with feldspar luminescence; (2) benefit from the high precision obtainable with SAR and (3) exclude underestimations that have so far limited the use of SAR on feldspar (Banerjee et al., 2001). In the central part of the valley-fill, sediment samples for OSL dating (sample code LV71 LV77) were taken at regularly spaced intervals of about 0.5 m in vertical succession (Table 1 and 3; Figure 3a). Another sample (LV78) was taken from sediments further upslope (Figure 2b). Samples for OSL-dating were collected by hammering steel cylinders into the sediment. Sample preparation procedures applied are described in detail by Mauz et al. (2002). In brief, sample preparation involved removal of organics with hydrogen peroxide and calcium carbonate with hydrochloric acid. Grains of µm were extracted by settling in cylinders following Stokes Law. A 20% hydrofluoric (HF) acid digestion for 40 minutes was employed to concentrate the quartz grains, followed by feldspar contamination tests (Mauz and Lang, 2004a) to ensure the purity of the quartz extracts. The quartz grains were then suspended in acetone to separate grains of the size 8 15 µm and deposited onto aluminium discs to provide 2 mg of sample on each disc. Two Risø DA-15 TL/OSL readers were used for luminescence measurements, both equipped with EMI 9235QB photomultiplier tubes, blue LEDs providing 30 mw/cm 2 stimulation, and Hoya U-340 filters for detection (transmission nm). IR stimulation used in feldspar contamination tests (Mauz and Lang, 2004a; Shen et al., 2007) was either by a 1 W laser diode providing ~400 mw/cm 2 to the sample (peak emission 830 nm) or LEDs providing 110 mw/cm 2 (peak emission 870 nm). Blue LED stimulation was conducted at 125 C for 40 s, and IR stimulation at room temperature for 100 s. Different beta-irradiation units were used, all of which are calibrated for irradiation of siltsized quartz (Mauz and Lang, 2004b): dose rate was ~4.70 Gy/min or ~6.00 Gy/min for all samples except LV77 for which ~0.55 Gy/min was used. For D e estimation, a SAR protocol (Murray and Wintle, 2000) was applied. For age calculation D e -values were only used when (i) recycling ratios fell within 1.0±0.1; (ii) thermal transfer and recuperation were insignificant (< 5% of the natural OSL); (iii) the natural dose was bracketed by at least two regenerated dose points; and (iv) the dose response curve was linear. LM OSL (linear modulated OSL) was measured for all samples. An ultrafast OSL component (Jain et al., 2003) is present in samples LV75 and LV77. After a preheat test had indicated no dependence of D e with temperature a 200 C/10 s preheat and cutheat to 200 C was used. This reduces the impact of the ultrafast component on D e determination (Packman et al., 2007). An α efficiency (a-value) of 0.031±0.002 was determined on sample LV 78 following the procedures of Mauz et al. (2006). For age determination, a-values of 0.030±0.003 were used. Annual dose rates were calculated from radionuclide concentrations determined using high-resolution, low-level γ spectrometry, and the burial depth of the sample. The beta dose was corrected using the attenuation factors given by Mejdahl (1979). Further details about the techniques applied can be found in Mauz et al. (2002). Gaussian error propagation at the 1σ level was applied including all random errors and all quantifiable systematic errors, including uncertainties involved in source and detector calibration. Results Chronometry Radiocarbon dating results are presented in Table 2; OSL dating results are given in Table 3. No significant radioactive

103 Tom Rommens et al.: Slope and dry valley sediment dynamics in a Belgian loess landscape 783 Figure 4 Calibrated radiocarbon ages and OSL ages plotted versus depth of sampling. Accumulation rates for different phases are indicated. For an explanation of the stratigraphic units, see Table 1 disequilibrium was detected in any of the OSL samples. Reproducibility of the luminescence measurements was generally good and aliquots responded well to the SAR procedure (80% or more of the aliquots passed all rejection tests). LV77 was an exception: five aliquots (out of 24) showed poor recycling ratios and were excluded from further analyses. We attribute this low reproducibility to the presence of the ultrafast component (Packman et al., 2007). Slope An overview of the soils and sediments recovered along the trench is shown in Figure 2. On the slope, an eroded soil profile was found that is typical for much of the Central Belgian loess area (Figure 2a). Below 0.5 m depth, bright yellowish brown calcareous loess is found that is overlain by a thin horizon of decalcified silt loam, and in turn by a c. 0.3 m deep ploughing horizon. On the plateau, some 20 m upslope from the end of the trench, the decalcification front was found much deeper, ie, 2.1 m. The entire loess cover can be up to several metres thick and is deposited on top of Tertiary sand. A gravel layer with sandstone and flint pebbles marks the contact between Tertiary (sand) and Quaternary (loess) deposits. Figure 2a shows that the surface of the underlying sand is not parallel to the present-day surface. Two different loess deposits were found, which have been previously described by Haesaerts (1984) and Frechen et al. (2001): the lower unit (~67 ka to ~40 ka; Hesbayan loess ) has a coarser texture and a more yellowish colour (10YR 6/8), shows fine layering of coarser and finer silt grains, and is sometimes devoid of carbonates. The upper loess has a bright colour (10YR 6/6) and a uniform fine silt texture. In Belgium this unit, which originates from ~25 ka to ~20 ka, is termed Brabantian loess. The exposure also revealed three large structures cutting into the calcareous loess. The same structures were obvious in both exposures on the two sides of the trench. They have distinct borders, flat horizontal bottoms and steep walls. The width of the largest one is 15 m, and the depth varies between 1.5 m and 2.5 m below the present-day surface (Figure 2b). The compact infill is yellowish brown (10YR 5/6) and has a uniform silt loam texture. Some parts of the exposure show fine horizontal lamination of white silt and dark brown clay. Some irregularly shaped calcareous sediment lumps are found in the lower colluvium. No ceramics or other artefacts were found, and charcoal fragments were very scarce. One charcoal, sampled at a depth of 1.3 m in the largest sediment body, returned a radiocarbon age of cal. BC (Sample 8, Table 2). Sediment sampled from laminated deposits at a similar depth c. 6 m upslope gave an OSL age of BC (LV78; Table 3). Thalweg deposits Soils and sediments recovered towards the lower end of the trench in the central part of the dry valley are shown in Figure 3a. Sediment characteristics are listed in Table 1. The marine sand of Tertiary age (j in Figure 3a) was found by augering. It is unconformably overlain by decalcified silt (i in Fig 3a). A stone line of flint pebbles and sandstones marks this unconformity. Similar stone lines can be found in the overlaying unit where decalcified loam and sandy loam are found in a succession of darker and brighter layers (i in Figure 3a). Above the layered silts a more homogeneous decalcified silt is present in which a compact clay-illuviation horizon (argic horizon (Food and Agriculture Organization (FAO), 1998)) developed (h in Figure 3a). It has a blocky structure and its most compact upper part is c. 0.6 m thick. The clay content of this material is c. 20% (Figure 3b), which is significantly higher than that of the overlying sediments. A dull yellow orange (10YR 7/4) horizon with a thickness of 0.20 to 0.30 m lies on top of this horizon (g in Figure 3a). It consists of silt loam with much bright reddish brown (5YR 5/6) and brownish black (10YR 2/3) mottles, as a result of Fe- and Mn-oxide concretions. Narrow cracks and tongues, filled with this material penetrate into the underlying clay horizon. The buried palaeosol has the typical characteristics of an Albeluvisol (FAO, 1998), and the typical compact clayey horizon, which causes water stagnation, is a so-called fragipan (Langohr and Sanders, 1985). Currently, albeluvisols can still be found under forest in the loess areas of central Belgium (Langohr, 2001). The palaeosol can be traced in the exposure over a length of c. 10 m, rising to a depth of c. 1.5 m. At the edge of the

104 784 The Holocene 17,6 (2007) stratigraphically inconsistent. The OSL age from the upper part of layer b is AD (LV77). A charcoal fragment sampled from the infill of unit β returned a calibrated 14 C age of AD (Sample 7, Table 2). Discussion Figure 5 Holocene slope dynamics in Nodebais flat valley bottom, the transitions between unaltered parent material, soil horizon and colluvial sediments become indistinct. The sediments stored in the thalweg have a thickness of up to 3.2 m (Figure 3a; Table 1). The texture of the colluvial deposits is homogeneous silt loam, except for a 0.07 m thick sandy layer (c in Figure 3a) at c m and a 0.10 m thick fine silt layer at c. 2 m depth (e in Figure 3b) at c. 2 m depth (Figure 3b). The clay content in the upper half of the infill is slightly higher than in the lower half of the sediments, and the organic matter content is rather low (<1%). Wormholes and plant roots are found within the upper 1.25 m. Close to the top of the colluvial sediments further sedimentary features can be observed: the infills of two small gullies. Gully (α) is c. 0.8 m wide, shows a parabolic cross-section, and cut 1.3 m deep into the colluvium. It is filled with pale (10YR 7/4) sandy loam and loam, alternated with brown (10YR 4/6) sand. Reddish, black and white mottles occur, resulting from Fe- and Mn-oxides. Gully (β) is filled with a mixture of yellowish brown and brown silt loam on top of a compact thin layer with flint pebbles, brick and tile fragments of post-mediaeval age. All chronometric information available for this section is plotted in Figure 4 against depth of sampling. All age estimates below layer e and above the middle of layer b (Figure 3a) are consistent with the stratigraphic order of the sediments. In between, however, age reversals occur. The oldest OSL ages are obtained for the palaeosol underlying the colluvial sediments and the sediment in layer e. An age of BC (LV71) is obtained for the palaeosol. Radiocarbon ages of charcoal sampled from the old surface just above the palaeosol fall within BC (Samples 1 4, Table 2), the late Bronze Age. The lower parts of the colluvial sediment (layer f) returned OSL ages of BC (LV72) and AD (LV73), the Iron Age/Roman Period. OSL and radiocarbon ages from layers e to c and the lower part of b are Assuming that the albeluvisol-type soil initially formed along the entire slope, and taking the average soil profile on the plateau as a reference (Rommens et al., 2005), it is possible to reconstruct the early-holocene topography using the stratigraphical and chronological information obtained in this study (Figure 5). On the steepest parts of the slope, an erosion depth of more than 2 m is found. In the thalweg, before sediment deposition started, the land surface was 3.15 m lower than today. This implies that the slope gradient before soil erosion was up to 25%, much steeper than the current slope (maximum of c. 10%). The slope gradient of the valley scoured in the Tertiary sands before Pleistocene loess deposition was even higher (up to 50%). The major uncertainty associated with the above approach is to what extent soil development on the slopes was similar to that on the plateau. More intense overland flow on the slope may reduce the amount of water percolating vertically in the soil and thus the decalcification front on the slopes may be shallower than on the plateau. This leads to an overestimation of the height of the presoil erosion surface and to an overestimation of the amount of soil erosion. The contrary may also be true because of additional water delivered to the slope resulting from interflow. On the other hand, at locations where today s erosion surface has caught up with the decalcification front and calcareous loess outcrops, only minimal values for pre-erosion soil depth can be estimated. Clear evidence for the very steep original slope profile is found where the old argic horizon is still recognizable in the exposure (Figure 3a). Assuming the early Holocene surface was parallel to the upper border of this horizon, the slope angle was at least 20%. The three large structures cutting into the calcareous loess found on the upper and mid slope are interpreted as man-made excavations, which were filled with colluvium after abandonment. Their clear-cut straight and vertical sidewalls and flat horizontal bottom, as well as the fact that these pits do not drain into the valley, suggest that they do not originate from gullying. At present, no traces of these excavations can be recognized at the surface. Additional augering in the proximity of the trench in order to trace the colluvia in the two largest pits (Figure 6), indicates that the pits had a diameter of m, and a depth of at least 1.5 m to 2 m. Gillijns et al. (2005) and Vanwalleghem et al. (2007) studied a number of closed depressions in several locations close to Nodebais. They reconstructed pits that originally had similar characteristics to the ones observed here, and explain these closed depressions as partly filled remnants of so-called marl pits from Roman and Mediaeval times. Calcareous loess was extracted from these quarries and used as an acid-neutralizing fertilizer on the fields. Most probably the pits found in the present study have a similar origin. Dating results obtained here suggest that the pits are much older than the marl pits described by Gillijns et al. (2005) and Vanwalleghem et al. (2007). The OSL age of BC (LV78) for the pit-fill and a calibrated 14 C age of a charcoal of cal. BC clearly pre-date the Roman period. The OSL age has to be taken with caution, as it may be an overestimate because of insufficient light exposure of the Quartz grains (bleaching) prior to deposition. However, the sample was taken from finely layered sediment, suggesting the deposition of loose and well-sorted grains and enabling maximum bleaching during transport. Because no sand is present in these sediments OSL ages were

105 Tom Rommens et al.: Slope and dry valley sediment dynamics in a Belgian loess landscape 785 Figure 6 Map of the depth to calcareous loess on the slope in the surroundings of the trench (white lines in the centre). Augering locations are shown as black points. Inverse distance to a power (2) interpolation was used. Greatest depths occur in the central area and correspond to the largest of the pits recovered (see Figure 2b) obtained on silt and techniques developed to check for poor bleaching of OSL cannot be applied. The 14 C-age, however, seems reliable and would link the pit to the Bronze Age. This 14 C-age was derived from a charcoal fragment thus reworking cannot be ruled out and the age should only be viewed as a maximum age of deposition. Other evidence, however, also points to rather early activities: the oldest colluvia on the foot slope seem to originate from the Bronze Age, and in the nearby Meerdaal forest colluvia from the Bronze Age, Iron Age and Roman Age were found in closed depressions and old gully systems (Vanwalleghem et al., 2003; Gillijns et al., 2005). Moreover, historical sources from Roman literature mention that local farmers already used dug-up soil to fertilize the fields before the Roman occupation in the first century BC (Lindemans, 1994). The colluvial fill in the valley bottom shows very little textural variation (Figure 3b). Nevertheless, it is clear that the younger sediments have higher clay contents than the older sediments. This illustrates how soils in the area were successively eroded: the older sediments originate from the original upper eluvial soil horizons, which delivered silty material devoid of clay, whereas later, the erosion of the argic soil horizons resulted in clayenriched colluvia. The layers with coarser texture at 1.50 m and finer texture at 2 m depth point to temporary changes in the erosion processes and sediment source. These sediments are derived from higher magnitude erosion events, probably gullies on the slopes upstream. From the chronometric results the following time frame for the deposition of the lower slope colluvium can be established. (1) Soil formation stopped some time after BC the OSL age obtained on the soil horizon (layer g; LV71). This OSL age is significantly younger than the age of formation of the Brabantian loess deposits. During soil formation, bleached grains from the surface are mixed in by bioturbation or anthropogenic disturbance, which may lead to sufficient bleaching of the OSL signal (eg, Lang and Hönscheidt, 1999). The OSL age may thus represent the last soil formation/ cultivation. If the mixing processes at the soil surface were insufficient to completely bleach all grains then the OSL age will be an overestimate. (2) Colluviation started some time during or after the Late Bronze Age (~ cal. BC). The abundance of charcoal fragments just above the old argic horizon, all of which returned similar 14 C ages, is probably a legacy of the old slash and burn system that was used to clear arable fields in prehistory. Similar abundance of charcoal at this stratigraphic level was found in numerous auger holes in the surroundings. If we assume the average age of the charcoal samples 1 4 to represent the onset of agriculture and associated soil erosion, a long-term sediment accumulation rate of 1.2 mm/yr is derived for the last ~ years. (3) By BC (LV 72) there was already a significant layer of colluvium deposited, and by AD (LV 73) the

106 786 The Holocene 17,6 (2007) Table 4 Accumulation and deposition rates in the thalweg for different time periods Phase Accumulation rate (mm/yr) Width of valley bottom deposits (m) Weight factor a Deposition rate (t/ha per yr) b 600 BC 160 BC 0.80 ± (Avg.: 6) ± BC AD ± (Avg.: 10.5) ± 2.2 AD 320 AD ± (Avg.: 16) ± 2.2 AD 1440 present 1.27 ± (Avg.: 21.5) ± 1.9 a Sediment accumulation height is weighed against sediment accommodation space. Weight factor = former average width of valley bottom/present average width of valley bottom. b Local sediment deposition rate, per ha of current depositional area. (Deposition rate = weighted accumulation rate dry bulk density). Dry bulk density of the sediment: 1.51 ± 0.10 t/m 3. This value was measured for 32 soil samples in the Nodebais catchment (Rommens, 2006). Figure 7 Phosphorus concentrations obtained from colluvial sediments (Colluvium P) and an undisturbed Luvisol (Reference P) colluvium thickness amounted to c m. This clearly shows strong soil erosion during the Iron Age as was found in similar areas close by (Vanwalleghem et al., 2003; Gillijns et al., 2005) and in other European loess areas (eg, Lang and Hönscheidt, 1999). The depositional style during that period had a rather low-magnitude high-frequency character. (4) After ~ AD 320 a change in sediment mobilization/deposition occurred that led to age estimates that are inconsistent with stratigraphy. Poor bleaching is most probably responsible for the OSL age overestimates (LV 75 and LV 76) and reworking of older charcoal most probably led to 14 C ages that are older than the OSL ages at corresponding depths. This disturbance probably reflects changes in the sediment delivery pathway, which could be due to higher magnitude erosion events (gullying) during which sediment is transported as aggregates or anthropogenic disturbance. As further upslope there are no signs of ancient gullies present but clear evidence of quarrying, most likely increased activity at the marl-pit caused these effects. The excavation probably led to major sediment mobilization that over time delivered older and older material, which is reflected in the OSL age overestimates increasing with time. This would imply increased human activity on the slope during the Mediaeval period between c. AD 400 and AD (5) Colluvial sedimentation since AD (LV 77) returned to a more continuous pattern but was interrupted by phases of gully erosion that cut into the colluvium. The infill of gully β returned a calibrated 14 C age of AD The fill of gully α contained modern materials such as post- Mediaeval potsherds and bricks (De Bie, personal communication, 2004). This suggests that gullying, at least in this part of the catchment, is a rather recent phenomenon. The dimensions of the gullies suggest that these were formed at the end of the winter or early spring (Poesen and Govers, 1990; Nachtergaele et al., 2001). So, probably the more intensive farming that was introduced during the last few centuries provoked the formation of these ephemeral gullies. The new crop rotation schemes were characterized by reduced fallow periods and a limited use of green manure, which resulted in large areas of uncovered arable land in winter. The chronometric results can also be used to calculate average accumulation rates (Table 4 and Figure 4). Standard errors on these accumulation rates were determined using Gaussian error propagation techniques assuming an error of ±3 cm on the depth of the dated material, and 1σ on the sediment age. From the early Iron Age to the Roman Period, an accumulation rate of 0.80 ± 0.50 mm/yr is obtained. This rate did not change significantly during the Roman Period. In the Middle Ages, however, the average accumulation rate is much higher: 1.61 ± 0.17 mm/yr. After the Middle Ages, there is again a decrease in sediment accumulation rate to an average value of 1.27 ± 0.09 mm/yr.

107 Tom Rommens et al.: Slope and dry valley sediment dynamics in a Belgian loess landscape 787 It is not straightforward, however, to compare the sediment accumulation rates for different time periods. A correction has to be made to these sedimentation rates since accommodation space increased through time. The valley width increased from about 3 m during the onset of deposition to c. 23 m today. Thus, 1 mm/yr at the onset of deposition corresponds only with 4.5 kg of deposits per unit length ( m accumulation times 3 m valley width times 1510 kg/m 3 sediment bulk density times 1 m unit length along the thalweg axis), whereas an identical sediment rate at present corresponds with 34.7 kg per unit of thalweg length. Table 4 therefore also shows the converted sedimentation rates expressed per unit of depositional area. Sediment deposition almost doubled from the Iron Age to the Roman Age, and increased fivefold from the Iron Age to the Middle Ages. However, interpreting these accumulation rates in terms of soil erosion rates in the catchment, which are generally expressed in the same units (i.e. t/ha per yr) may be misleading. Rommens et al. (2005) calculated a mean soil erosion rate of 3.09 t/ha per yr since the Iron Age for the Nodebais catchment in which the studied slope is situated. Local sedimentation rates (Table 4) are thus higher than catchment-wide erosion rates, simply because the areas with intense sediment deposition are very limited in spatial extent (Rommens et al., 2005). For the last century, average accumulation rates can be calculated on the basis of the P-profile (Figure 7). Comparing the P enrichment in the reference soil and the colluvium yields a difference of 15 cm ( 1) to 25 cm ( 2), which results in an average sediment accumulation of 1.7 mm/yrto 2.9 mm/yr since AD 1920, or a sedimentation rate of 26 to 44 t/ha per yr in the valley bottom. This suggests a recent increase in sediment supply. The intensification and mechanization of agriculture following World War II is probably responsible for the accelerated infill of the valley. Even though sediment redistribution by water may have been more or less constant, intense tillage erosion has caused considerable sediment accumulation in concave landscape positions (Van Muysen, 2001; Van Oost et al., 2003). Conclusions The soil- and sediment-scape in central Belgium testify to a long history of land use and related sediment dynamics. The morphological study of a typical slope catena in the Belgian loess areas illustrates that before the onset of agriculture, slopes were much steeper (slope angles up to 25%) and valleys were several metres deeper. Besides soil erosion, quarrying of carbonate-rich sediments also leads to significant changes in the landscape: on the slope, several man-made pits were discovered that are entirely filled with colluvial sediments today. The exact age of these features remains uncertain but most probably their origin goes back to the Iron Age. The colluvial infill of the dry valley started in the early Iron Age. At that time sedimentation in the valley bottom was still limited (~ 3.4 t/ha per yr). Sediment delivery to the valley bottom increased to ~ 5.4 t/ha per yr during the Roman Period, and boosted to ~ 18.0 t/ha per yr during the Middle Ages. Afterwards, sedimentation in the valley slightly decreased, but because of mechanization and intensification of agriculture during the twentieth century there are indications for a recent increase to a value of t/ha per yr. For the last few centuries, there is also more evidence for gullying, whereas before, low-magnitude processes, resulting in deposits with little textural variations, dominated erosion and deposition. Acknowledgements We are grateful to Mr Evrard, who gave us the permission to dig a great trench in his field. We appreciate his cooperation and interest in our work. We hereby also thank the colleagues of the Physical and Regional Geography Research Group for their contribution to the fieldwork. This project is funded by the K.U. Leuven, grant no. 3E References Banerjee, D., Murray, A.S., Bøtter-Jensen, L. and Lang, A. 2001: Equivalent dose estimation using a single aliquot of polymineral fine grains. Radiation Measurements 33, Bertrand, T. 2003: Valduc, abbaye de moniales cisterciennes à Hamme-Mille. Le Bulletin du Centre culturel de la vallée de la Néthen, Bollinne, A. 1976: L évolution du relief à l Holocène. Les processus actuels. In Pissard, A., editor, Géomorphologie de la Belgique Hommage au Professeur P. Macar. Université de Liège, Bolt, A.J.J., Mücher, H.J., Sevink, J. and Verstraten, J.M. 1980: A study on loess-derived colluvia in southern Limbourg (the Netherlands). Netherlands Journal of Agricultural Science 28, Bronk Ramsey, C. 2001: Development of the radiocarbon program OxCal. Radiocarbon 43, Bryan, R.B. and De Ploey, J. 1983: Comparability of soil erosion measurements with different laboratory rainfall simulators. Catena 4, Desmet, P.J.J. and Govers, G. 1995: GIS-based simulation of erosion and deposition patterns in an agricultural landscape: a comparison of model results with soil map information. Catena 25, Food and Agriculture Organization 1998: World reference base for soil resources. World Soil Resources Reports 84, FAO. Frechen, M., Van Vliet-Lanoë, B. and van den Haute, P. 2001: The Upper Pleistocene loess record at Harmignies/Belgium high resolution terrestrial archive of climate forcing. Palaeogeography, Palaeoclimatology, Palaeoecology 173, Gillijns, K., Poesen, J. and Deckers, J. 2005: On the characteristics and origin of closed depressions in loess-derived soils in Europe a case study from central Belgium. Catena 60, Govers, G. 1985: Selectivity and transport capacity of thin layer flows in relation to rill generation. Catena 12, Gullentops, F. and Scheys, G. 1950: Premiers résultats de la cartographie des sols en Hesbaye Septentrionale. Bulletin de la Société Belge d Etudes Géographiques 19, Haesaerts, P. 1984: Aspects de l évolution du paysage et de l environnement en Belgique au Quaternaire. In Cahen, D., Haesaerts, P. and Watteyne, D., editors, Peuples chasseurs de la Belgique préhistorique dans leur cadre naturel. Patrimoine de l Institut royal des sciences naturelles de Belgique, Halbfass, S. and Grunewald, K. 2003: Spatial variability of Phosphorus contents in topsoils of two small catchments under agricultural use. Journal of Plant Nutrition and Soil Science Zeitschrift für Pflanzenernährung und Bodenkunde 166, Heckrath, G., Djurhuus, J., Quine, T. A., Van Oost, K., Govers, G. and Zhang, Y. 2005: Tillage erosion and its effects on soil properties and crop yield in Denmark. Journal of Environmental Quality 34, Hofman, G. and Verloo, M. 1989: Minerale meststoffen. In Milieuzorg in de landbouw. Monografieën Stichting Leefmilieu. Pelckmans, Jain, M., Murray, A.S. and Bøtter-Jensen, L. 2003: Characterisation of blue-light stimulated luminescence components in different quartz samples: implications for dose measurement. Radiation Measurements 37, Kalis, A.J., Merkt, J. and Wunderlich, J. 2003: Environmental changes during the Holocene climatic optimum in central Europe human impact and natural causes. Quaternary Science Reviews 22, Lang, A. 2003: Phases of soil erosion-derived colluviation in the loess hills of South Germany. Catena 51, Lang, A. and Hönscheidt, S. 1999: Age and source of soil erosion derived colluvial sediments at Vaihingen-Enz, Germany. Catena 38, Langohr, R. 2001: L anthropisation du paysage pédologique agricole de la Belgique depuis le Néolithique ancien Apports de l archéopédologie. Etude et Gestion des Sols 8,

108 788 The Holocene 17,6 (2007) Langohr, R. and Sanders, J. 1985: The Belgian loess belt in the last years: evolution of soils and relief in the Zonien Forest. In Boardman, J., editor, Soils and Quaternary landscape evolution. John Wiley and Sons Ltd, Lindemans, P. 1994: Geschiedenis van de landbouw in België. Genootschap voor geschiedenis en volkskunde Antwerpen. Martens, E. 1981: Uit het verleden van de gemeente Oud-Heverlee. Drukkerij Stroobants Neerijse. Mauz, B. and Lang, A. 2004a: Removal of the feldspar-derived luminescence component from the polymineral fine silt samples for optical dating applications: evaluation of chemical treatment protocols and quality control procedures. Ancient TL 22, b: The dose rate of beta sources for optical dating applications: a comparison between fine silt and fine sand. Ancient TL 22, Mauz, B., Bode, T., Mainz, E., Blanchard, H., Hilger, W., Dikau, R. and Zöller, L. 2002: The luminescence dating laboratory at the University of Bonn: equipment and procedures. Ancient TL 20, Mauz, B., Packman, S. and Lang, A. 2006: The alpha effectiveness of silt-sized quartz: new data obtained by single aliquot regeneration dose protocols. Ancient TL 24, Mejdahl, V. 1979: Luminescence dating beta dose attenuation in quartz grains. Archaeometry 21, Murray, A.S. and Wintle, A.G. 2000: Luminescence dating of quartz using an improved single-aliquot regenerative-dose protocol. Radiation Measurements 32, Nachtergaele, J., Poesen, J., Steegen, A., Takken, I., Beuselinck, L., Vandekerckhove, L. and Govers, G. 2001: The value of a physically based model versus an empirical approach in the prediction of ephemeral gully erosion for loess-derived soils. Geomorphology 40, Olyslager, W. 1960: De tumuli van Heverleebos en Meerdaalwoud. In Oostbrabantse werkgemeenschap, editor, Oost-Brabant.2: Heemkundig handboek voor Leuven en omliggende gemeenten. Renova, Packman, S.C., Mauz, B., Rousseau, D.-D., Rossignol, J., Antoine, P. and Lang, A. 2007: Implications of broad dose distributions obtained with the single-aliquot regenerative-dose method on quartz fine grains from loess. Quaternary Geochronology 2, Paulissen, E., Gullentops, F., Vermeersch, P.M., Geurts, M.-A., Gilot, E., Van Neer, W., Van Vooren, E. and Wagemans, E. 1981: Evolution Holocène d un flanc de vallée sur substrat perméable (Hesbaye sèche), Belgique. Mémoires de l Institut géologique de l Université de Louvain 31, Poesen, J. and Govers, G. 1990: Gully erosion in the loam belt of Belgium: typology and control measures. In Boardman, J., Foster, I.D.L. and Dearing, J.A., editors, Soil erosion on agricultural land. Wiley, Reimer, P.J., Baillie, M.G.L., Bard, E., Bayliss, A., Beck, J.W., Bertrand, C.J.H., Blackwell, P.G., Buck, C.E., Burr, G.S., Cutler, K.B., Damon, P.E., Edwards, R.L., Fairbanks, R.G., Friedrich, M., Guilderson, T.P., Hogg, A.G., Hughen, K.A., Kromer, B., McCormac, G., Manning, S., Ramsey, C.B., Reimer, R.W., Remmele, S., Southon, J.R., Stuiver, M., Talamo, S., Taylor, F.W., van der Plicht, J. and Weyhenmeyer, C.E. 2004: IntCal04 terrestrial radiocarbon age calibration, 0 26 cal kyr BP. Radiocarbon 46, Rommens, T. 2006: Holocene sediment dynamics in a small river catchment. Unpublished PhD thesis, K.U.Leuven, Faculty of Sciences, Geography. Rommens, T., Verstraeten, G., Poesen, J., Govers, G., Van Rompaey, A.J.J., Peeters, I. and Lang, A. 2005: Soil erosion and sediment deposition in the Belgian loess belt during the Holocene: establishing a sediment budget for a small agricultural catchment. The Holocene 15, Schayes, J. 1990: Les sentiers de l histoire à Beauvechain et environs. Beauvechain. Shen, Z., Mauz, B., Lang, A., Bloemendal, J. and Dearing, J. 2007: Optical dating of Holocene lake sediments: elimination of the feldspar component in fine silt quartz samples. Quaternary Geochronology 2, Steegen, A., Govers, G., Beuselinck, L., Van Oost, K., Quine, T.A. and Rombaut, A. 2000: The use of phosphorus as a tracer in erosion/sedimentation studies. In Stone, M., editor, The role of erosion and sediment transport in nutrient and contamination transfer (Proceedings of a symposium held at Waterloo, Canada, July 2000). IAHS Publ. no. 263, Swinnen, J. and Tollens, E. 1989: Technische ontwikkelingen. In De Coster, M., editor, Milieuzorg in de landbouw. Monografieën Stichting Leefmilieu. Pelckmans, Tarlier, J. and Wauters, A. 1873: La Belgique ancienne et moderne. Géographie et histoire des communes belges. Province de Brabant, Canton de Jodoigne. Van Muysen, W. 2001: Tillage translocation and tillage erosion: an experimental approach. Unpublished PhD-thesis, K.U.Leuven, Faculty of Sciences, Geography. Van Oost, K., Govers, G. and Desmet, P. 2000: Evaluating the effects of changes in landscape structure on soil erosion by water and tillage. Landscape Ecology 15, Van Oost, K., Van Muysen, W., Govers, G., Heckrath, G., Quine, T.A. and Poesen, J. 2003: Simulation of the redistribution of soil by tillage on complex topographies. European Journal of Soil Science 54, Vandaele, K. and Poesen, J. 1995: Spatial and temporal patterns of soil erosion rates in an agricultural catchment, central Belgium. Catena 25, Vanwalleghem, T., Van Den Eeckhaut, M., Poesen, J., Deckers, J., Nachtergaele, J., Van Oost, K. and Slenters, C. 2003: Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium). Geomorphology 56, Vanwalleghem, T., Verheyen, K., Hermy, M., Poesen, J. and Deckers, J. 2004: Legacies of Roman land-use in the present-day vegetation in Meerdaal forest (Belgium)? Belgian Journal of Botany 137, Vanwalleghem, T., Poesen, J., Van Den Eeckhaut, M., Nachtergaele, J. and Deckers, J. 2005: Reconstructing rainfall and landuse leading to the development of old gullies. The Holocene 15, Vanwalleghem, T., Poesen, J., Vitse, I., Bork, H.R., Dotterweich, M., Schmidtchen, G., Deckers, J., Lang, A. and Mauz, B. 2007: Origin and evolution of closed depressions in central Belgium, European loess belt. Earth Surface Processes and Landforms 32, Verstraeten, G., Poesen, J., Goossens, D., Gillijns, K., Bielders, C., Gabriels, D., Ruysschaert, G., Van Den Eeckhaut, M., Vanwalleghem, T. and Govers, G. 2006: 1.30 Belgium. In Boardman, J. and Poesen, J., editors, Soil erosion in Europe. John Wiley and Sons, Vincent, A. and Vincent, G. 1909: Les tumuli de la Forêt de Meerdael. Sont-ils rangés le long d une route Romaine? Annales de la Société d Archéologie 23, Zolitschka, B., Behre, K.-E. and Schneider, J. 2003: Human and climatic impact on the environment as derived from colluvial, fluvial and lacustrine archives examples from the Bronze Age to the Migration period, Germany. Quaternary Science Reviews 22,

109 The Holocene 15,3 (2005) pp. 378/386 Reconstructing rainfall and land-use conditions leading to the development of old gullies T. Vanwalleghem, 1,2 * J. Poesen, 1 M. Van Den Eeckhaut, 1,2 J. Nachtergaele 1 and J. Deckers 3 ( 1 Laboratory for Experimental Geomorphology, K.U. Leuven, Redingenstraat 16, 3000 Leuven, Belgium; 2 Fund for Scientific Research, Flanders, Belgium; 3 Institute for Land and Water Management, K.U. Leuven, Vital Decosterstraat 102, 3000 Leuven, Belgium) Received 20 August 2003; revised manuscript accepted 10 April 2004 Abstract: Knowledge of past erosion events and their controlling factors is an important key to understanding the impacts of environmental change (climate/land use) on the landscape. In this study, knowledge about erosion processes on the development of present-day ephemeral gullies is used for reconstructing conditions leading to the formation of old, permanent gullies. Empirical relations between flow hydraulics and channel geometry have been recently established for gullies. Hence, using measured bottom width W bottom of old gullies as input, peak flow discharges (Q p ) of these gullies can be estimated. In two forested areas in central Belgium, 52 old gullies were mapped. The old gullies had an average W bottom ranging between 1.1 and 1.5 m. Corresponding calculated Q p values ranged between 0.04 and 0.07 m 3 /s. Rainfall intensities (I) were also deduced from Q p using the rational formula. By simulating various land use scenarios and thus various runoff coefficient (C) values, I and concentration time (T c ) could be calculated for each land-use class. Using I, T c and intensity/duration/frequency tables for the study area, the recurrence interval (RI) of the rain events, needed to erode the observed gully channels was assessed. Although analysis of historical documents indicates that both areas have probably been under forest since the Middle Ages, it is unlikely that the old gullies originated under forest vegetation or even degraded forest vegetation, since RI/200 years were obtained for these land-use scenarios. Cropland is the only land use that provides acceptable values of RI (11/128 years). Key words: Historical gully erosion, gully width, (peak) flow discharge, rainfall intensity, land use, Belgian loess belt, late Holocene. Introduction In many parts of northwest Europe old, permanent gullies under forest can be observed (e.g., Bollinne, 1976; Gullentops, 1992; Semmel, 1995; Bork et al., 1998; Stankoviansky, 2003). These old gullies are also present under cropland, but often no longer visible, since they are completely filled with sediments (Lang et al., 2003; Vanwalleghem et al., in press). During the last decade, interest in historical gully erosion has increased significantly (Poesen et al., 2003) because these geomorphic features provide important insights into the impacts of environmental change (climate/land use) on erosion processes acting at timescales that are not measurable in laboratory or *Author for correspondence: ( tom.vanwalleghem@geo. kuleuven.ac.be) field studies. Considerable efforts have been made to quantify historical erosion volumes and to date important past erosion phases (Bork et al., 1998; Lang, 2003; Dotterweich et al., 2003). The most important factors causing past gullying phases are, however, not well-known. Bork et al. (1998) attributed many large gullies in Germany to a combination of intensive land use and extreme rainfall. Stankoviansky (2003) came to the same conclusions for a large study area in Slovakia. Vogt (1953) concluded on the basis of his research in northwestern France, that land use was the main driving factor for gully erosion. In this respect, it is important to document past land use. Based on historical documents, Schmitt et al. (2003) reconstructed a detailed land-use history in the catchment of a large gully in Bavaria. Gábris et al. (2003) used historical maps for monitoring the extension of a large gully system in Hungary. If these historical sources are absent, it is very hard # 2005 Edward Arnold (Publishers) Ltd / hl807rp

110 T. Vanwalleghem et al.: Reconstructing rainfall and land-use conditions during gully formation 379 to reconstruct past land use. At present, no tools are readily available for reconstructing the environmental factors leading to phases of past gully incision. In contrast to this absence of data about past gully formation, there is extensive knowledge about actual gully formation processes, gathered in recent decades (Poesen et al., 2003). The application of this presentday process knowledge to old gullies could open valuable perspectives for the analysis of factors controlling the development of these gullies. For example, in a completely infilled historical gully under cropland in central Belgium, Vanwalleghem et al. (in press) used contemporary measurements of sediment deposition rates to reconstruct historical gully infilling. In this study an approach is proposed to reconstruct land use and rainfall conditions in the runoff-contributing area of old gullies at the time of incision. Theoretical background For rivers and rills a good correlation exists between channel width and the peak flow discharge measured in the field or during laboratory experiments (Leopold and Maddock, 1953; Lane and Foster, 1980; Gilley et al., 1990; Haan et al., 1994). Nachtergaele et al. (2002a) showed that such a relation is also valid for ephemeral winter gullies in the Belgian loess belt. They proposed the following power relation, based on data collected for rills, ephemeral gullies and small rivers (n/99): W 6:56Q 0:56 p R 2 0:95 (1) where W is channel width (m) and Q p is peak flow discharge (m 3 /s). Boundary conditions for the use of this equation are 0.1 mb/wb/22.3 m and m 3 /sb/q p B/6.5 m 3 /s. The W of the old, permanent gullies, studied here, falls between the data range for rills and ephemeral gullies on the one hand and small rivers on the other. Typical channel widths for rills and ephemeral gullies range from 0.08 to 0.7 m and the associated runoff discharges are between m 3 /s and 0.02 m 3 /s. For small rivers, typical channel widths are between 2.4 m and 22.3 m and the corresponding discharges range from 0.1 m 3 /s to 6 m 3 /s. Therefore, a regression equation based on both groups together was used. Using a Q p /W relation, based on only one of these data sets would lead to, respectively, an over- or underestimation of Q p. Using equation (1) and the measured mean channel width at the bottom (W bottom ) of the studied old gullies under forest, allows us to estimate the peak flow discharges that must have caused gully incision. In order to derive Q p from W bottom, however, a regression equation has to be used that treats the independent and dependent variables symmetrically. Therefore, a new linear orthogonal regression equation was calculated based on the original, log-transformed data collected by Nachtergaele et al. (2002a): Q p 0:0344W 1:77 R 2 0:95 (2) In theory, a relation between W and Q p is only valid for soils with a uniform erosion resistance with depth (Knighton, 1974). This is generally not the case in the Belgian loess belt, where a more resistant clay illuviation horizon or fragipan is found at about 35/45 cm below the soil surface (Nachtergaele et al., 2002a). However, this constraint has only a negligible effect because results from Nachtergaele et al. (2002a) indicate that a Q p /W relation is applicable to most gullies formed in the Belgian loess belt. In combination with a runoff model, Q p can then be used to estimate the rainfall intensities (I, mm/h) that are required to cause gullies with the morphological characteristics measured in the field. For this, the rational formula (e.g., Chow et al., 1988) was used, i.e., Q p 0:00278 CIA (3) where C is the runoff coefficient and A is the runoffcontributing area (ha), Q p is peak flow discharge (m 3 /s) and I is rainfall intensity (mm/h) The whole catchment must contribute to the runoff discharge at the outlet before peak flow discharges are reached (Chow et al., 1988). Therefore equation (3) is only applicable for rainfall events where the calculated I has a duration greater than the concentration time (T c, min). T c was determined according to the nomogram given by Rantz (1971), suited for small agricultural basins. T c is calculated taking into account the overland flow travel distance, the average slope of the runoff-contributing area and the value of the runoff coefficient. Although valid criticisms have been raised about the adequacy of the rational method, it is still widely used because it is a simple model that requires few input parameters. In spite of this simplicity the method seems to give consistent results (Viessman et al., 1989; Titmarsh et al., 1995). As is generally the case with palaeoenvironmental studies, detailed and accurate input data are often lacking, which makes the use of a simple model very appropriate. Land use is one of the main factors determining the C value. The C value further depends on the type and condition of the soil, the rainfall intensity, proximity of the water-table, degree of soil compaction, subsoil porosity, slope of the soil surface and depression storage (Chow et al., 1988). Because the former land use and more specifically the former C value is not known, calculations were made for a range of C values (from 0.05 to 0.7). Rather then determining one C value per land-use type, a range of C values was accorded to every land-use type. In this study, three possible land-use types were distinguished and attributed a C value based on data extracted from the literature: forest (0B/C5/0.3), disturbed forest (0.35/C5/0.5) and cropland (0.45/C5/0.7). An additional problem with the runoff coefficient is that it varies with the recurrence interval (RI, years) (Eliasson, 2002). To incorporate this effect, the C value is often multiplied with a frequency factor varying from 1.0 for RI5/10 years to 1.25 for RI/100 years (Atlanta Regional Commission, 2001). Also, more complex formulae exist to relate C with RI (e.g., Rossmiller, 1981). In this study the following regression equation between RI and C was used, based on data given by Chow et al. (1988), because it allows extrapolation for RI up to 500 years and because a continuous relation between C and RI can be established: C 0:81 C 10 RI 0:09 (4) where C 10 is the the runoff coefficient valid for RI5/10 years. In a first step, I and T c are calculated using the default value of C 10, valid for RI5/10 years, as indicated in the literature. Based on the assumptions of the rational formula, the duration of the rainfall event (D) required to erode the observed gully channels can be set equal to T c. Subsequently, the calculated D and I can be used to estimate the RI of the rainfall event causing the old gullies for each I /D combination. This RI can be extracted from published Intensity/Duration/ Frequency (IDF) curves for the nearby climatic station of Brussels (Ukkel), based on a 100-year precipitation data set.

111 380 The Holocene 15 (2005) For the C values where the calculated I and D result in RI/10 years, the default C 10 value is adapted according to equation (4) and the critical rainfall intensity and concentration time are re-calculated. This calculation is repeated for the whole range of C values (and thus for the entire range of land-use scenarios). Finally, the RI values obtained yield information on the most likely land use in the gully catchment at the time of gully incision. The larger the calculated RI, the less likely is the corresponding land-use. Study area The study area that was selected for this research encompasses two forest areas in the Belgian loess belt where large gullies were found (Figure 1): Meerdaal Forest (1329 ha) and Tersaart Forest (20 ha). Altitudes range from 35 to 103 m a.s.l., with a typical rolling topography, characteristic of the Belgian loess belt. The two forests are only about 4 km apart and have a similar geomorphologic setting. The loess cover, overlying Tertiary marine sands, has a thickness ranging from a few decimeters up to about 10 m (Poesen et al., 1993). In both forests, mainly Albeluvisols or Luvisols are found. Locally, on eroded hilltops, the Tertiary substratum outcrops and Podzols can be found. No historical documents indicating agricultural activity were found for the study area. Both the Meerdaal and Tersaart Forests were already forested on the first available maps of the region. Historical maps and documents indicate that from the fourteenth century, no agricultural activities took place in the Meerdaal Forest. For the Tersaart Forest, land-use history is less well documented; the first maps date from the sixteenth century. Before that time however, several archaeological sites from Bronze, Iron and Roman Age (Martens, 1981) indicate that it is very possible that certain areas within both Meerdaal and Tersaart Forests were cleared and used for cropland. This was also concluded by Vanwalleghem et al. (2003), based on a study of the morphological and topographical characteristics of the gullies in Meerdaal Forest. Figure 1 Location map of the two study areas in central Belgium and mapped gullies in Tersaart Forest (1) and mapped gullies in Meerdaal Forest (2)

112 T. Vanwalleghem et al.: Reconstructing rainfall and land-use conditions during gully formation 381 Methods In previous studies all old, permanent gullies in Meerdaal Forest and Tersaart Forest were mapped (Poesen et al., 2000; Vanwalleghem et al., 2003). Detailed measurements were made of their morphological and topographical characteristics. Each gully was subdivided into sections with comparable morphology. Bottom width (W bottom ), top width (W top ), depth (GD) and length of every section was measured with a measuring tape. Total gully length (GL) was obtained by summing the length of all the sections. W bottom, W top and D of every gully were averaged with the length of every section as a weighting factor. Gully volume (GV) was obtained by summing the different segment volumes. Since gully heads may have retreated upslope since their formation, the slope of the soil surface at the gully head (S gully ) was defined as the steepest slope of the soil surface along the gully trajectory and measured with a clinometer (Suunto, error m/m). The runoff-contributing area of each gully (A) was delimited with markers in the field and measured with a measuring tape. The position was also recorded with GPS (Trimble Pro-XL). Because of multipath effects, large errors (up to several metres) remained. Length (L A ) and height difference (H A ) from every runoff-contributing area were obtained from the GPS measurements. Gully characteristics were tested for significant differences at a confidence level of P/0.05. Results and discussion In Meerdaal Forest, Vanwalleghem et al. (2003) mapped 43 large gullies in an area of 1329 ha (Figure 1). In Tersaart Forest, with a surface area of 20 ha, nine large gullies were mapped by Poesen et al. (2000). Minimum depth for all mapped incisions is 0.3 m. Gullies that have a clear origin as a road were excluded. A typical gully under forest is shown in Figure 2B. For comparison, a typical gully under cropland is shown in Figure 2A. Detailed morphological and topographical characteristics of these gullies are summarized in Table 1. Although the difference in absolute values is small (0.4 m), the W bottom of the mapped gully channels in Meerdaal and Tersaart Forest differs significantly. For most parameters shown in Table 1, however, the gullies in the two forested areas are quite comparable and show no significant differences. Only the runoff-contributing area and the associated L A differ significantly between the Meerdaal and Tersaart Forest gullies. For Meerdaal Forest, A is only half of the corresponding value of the gullies in Tersaart Forest. W bottom for individual gullies varies between 0.4 and 2.8 m. Gullies with such large dimensions are rather rare, however. In total, only four of the 52 gullies had a bottom width/2 m. The W-value used in this study is based on the mean W bottom of all the mapped gullies for each forest. Based on the mean W bottom of 1.1 m for Meerdaal Forest (n/43) and 1.5 m in Tersaart Forest (n/9) it can be calculated, using equation (2), that concentrated flow discharges of about 0.04 m 3 /s and 0.07 m 3 /s, respectively, were needed to erode channels with the observed W bottom in the two forests. As a result of partial infilling of the old forest gullies (e.g., Figure 2B), it is difficult to delineate W bottom precisely in the field. Figure 3 gives an indication of how W bottom was determined in the field. However, field measurements of W bottom of an old gully will most probably never be equal to the W bottom of that gully at the time of formation. After gully incision, infilling will begin, through a combination of soil fall Figure 2 A. Recent gully under cropland (Bertem, Belgium, January 1999). Note soil fall processes acting on gully wall. B. Old gully under forest (Meerdaal Forest, Belgium, December 2001) (see Figure 2A) and fluvial deposition processes (Vanwalleghem et al., in press). This is schematically represented in Figure 3. This infilling is, however, not necessarily problematic, as it probably results in an underestimation of W bottom. As a consequence, Q p and I will be underestimated, resulting in an underestimation of the RI. The effect of this underestimation will be the final acceptance of more land uses as being probable (i.e., also smaller C values will lead to acceptable RI). Another problem is that the final cross-sectional shape of an old gully is the result of several rainfall events, possibly even extending over several years. Nevertheless, available evidence points to the fact that the most important shaping of a gully channel occurs during the very beginning of a gully s life. During the greater part of a gully s lifetime, it is morphologically near-stable (Sidorchuk, 1999). In Figure 4, data on the evolution of gully dimensions as a function of gully life (GT) are compiled from various sources. Data from Kosov et al. (1978), Nachtergaele et al. (2002b) and Vanwalleghem et al. (in press) show the evolution of gully volume (GV). Data from Rutherford et al. (1997), Graf (1977), Nachtergaele et al. (2002b) and Vanwalleghem et al. (in press) indicate the evolution of gully length (GL). From Figure 4 it can be seen that the different data sets correspond very well. All studies report a degressive exponential increase of GV or GL with time: GV 96:5 (1e 0:068GT ) (5) GL98:9 (1e 0:043GT ) (6)

113 382 The Holocene 15 (2005) Table 1 Morphological and topographical characteristics of the mapped gullies and characteristics of the runoff contributing area of the gullies in Meerdaal Forest (n/43) and Tersaart Forest (n/9). Parameter Meerdaal Forest Tersaart Forest Sig. diff. Average SD n Average SD n W bottom (m) / W top (m) / GD (m) / GL (m) / GV (m 3 ) / S gully (mm) / A (ha) / L a (m) / S a (mm) / Values are averages for all the mapped gullies. SD, standard deviation; W bottom, average gully bottom width, W top, average gully top width; GD, average gully depth; GL, total gully length; GV, total gully volume; S gully, slope of soil surface at gully head; A, gully runoff-contributing area; L A, length of gully runoff-contributing area; S a, average slope of gully runoff contributing area. The last column indicates significant differences between the gullies in Meerdaal and Tersaart Forest (/ /significant difference; /, no significant differences, P/0.05). Poesen et al. (2000); Vanwalleghem et al. (2003). Overall, half of the GV is already eroded during the first moments after gully initiation (10% of total gully life). This indicates that the main erosion phase of a gully occurs during the first major runoff event causing gully initiation and during the first subsequent runoff events. If the gully volume or area increases further after this initial stage, it is not due to an increase of W bottom, but mainly to an increase in length (see Figure 4; about 30% at GT/10%). This is supported by other studies (e.g., Nachtergaele et al., 2002b). Therefore, no large differences in channel width evolution exist between old permanent gullies under forest, that possibly developed over many years, and ephemeral gullies, erased by tillage after every cropping season, since both type of gullies erode mostly during the beginning of their total lifetime. Using equation (3) and the average runoff-contributing area of the gullies in the two forests (0.15 ha for Meerdaal Forest and 0.30 ha for Tersaart Forest), mean critical rainfall Figure 3 Schematic representation of the evolution of the gully cross-sectional area through time and the effect on measured gully bottom width (W bottom ). The cross-section of the original gully incision and the cross-section of the gully channel after infilling are based on field observations of contemporary gullies under cropland (see Figure 2A) and of old gullies mapped under forest (e.g., Figure 2B), respectively intensities needed to generate these discharges were then calculated for a range of different C values. As discussed above, three different land-use types were distinguished: forest degraded forest and cropland. Each of these were attributed a range of C values based on data reported in the literature and summarized in Table 2. Traditionally, forest, woodland and cropland are distinguished by most authors. Since T c also depends on C, this was also calculated for the entire range of C values with the nomogram given by Rantz (1971). For forest and woodland, C values between 0.05 and 0.41 were found (Table 2). It is clear that present-day forests produce no, or almost no, runoff, apart from runoff that concentrates locally on compacted roads (Elliot et al., 1999). Therefore a lower limit of 0.05 was set. As an upper limit, 0.3 was taken. This might seem high for forest areas, but from various historical documents it is clear that ancient forests were not at all comparable with present-day forests (Tack and Hermy 1993). During the Middle Ages, and probably even earlier, forests were used very intensively for grazing, charcoal burning, gathering of firewood and collection of dead leaves for manuring. Therefore, runoff conditions were probably closer to those reported for woodland (0.305/C5/0.41). Several historical sources indicate that, during certain periods, human impact in forests was even more intense than described above. Therefore, a second class of degraded forest was distinguished. While the first class implies only a minor disturbance of the original forest vegetation and soil surface, this class implies a forest with a very degraded and compacted soil surface, mainly due to trampling and overgrazing of cattle. One source (Van Uytven, 1975) even describes that, during the eighteenth century, the Meerdaal Forest harboured so many boars, bred for hunting purposes, that these animals destroyed neighbouring cropland in search of food. According to Butler (1995) such animals definitely have a far from negligible impact on hillslope processes. In the literature no values are reported for degraded forest. Therefore, the boundaries for degraded forest were based on the upper C values reported for forest and the lower C values reported for cropland: i.e., 0.3/0.5. For cropland, most C values in the literature vary between 0.4 and 0.7, although in some studies values as low as 0.21 are given (McCuen, 1989). An upper limit of 0.7 was set because for agricultural and even urban basins it is very rare to find a higher value.

114 T. Vanwalleghem et al.: Reconstructing rainfall and land-use conditions during gully formation 383 Figure 4 Evolution of gully volume (GV, in % of total) or length (GL, in % of total) in relation to gully life time (GT, in % of total gully life (GT total )) based on data from several studies Calculated I and T c were compared with IDF curves from the nearby station of Ukkel, near Brussels (KMI, 2001) to obtain the RI for each I/T c combination. For the C values where the calculated I and T c resulted in RI/10 years, the default C value was adapted according to equation (4) and the critical rainfall intensity and concentration time were recalculated again. Finally, for the range of C values between 0.05 and 0.7, critical rainfall intensities needed to incise the old gullies between 117 and 1455 mm/h and between 106 and 1273 mm/h were obtained for Meerdaal Forest and Tersaart Forest, respectively (Figure 5A). Corresponding T c varied between 5 and 25 min for both areas. Although in Meerdaal Forest the average W bottom and consequently also the calculated peak flow discharge of the 43 gullies was smaller than that of the gullies in Tersaart Forest, the rainfall intensities needed to form the gullies are larger in Meerdaal Forest because there the average runoff-contributing area is only half of that in Tersaart Forest. In Figure 5B the calculated RI are shown on the left axis and the associated exceedence probabilities (P) on the right axis. The exceedence probability is the inverse of the RI and is thus simply interpreted as the probability of the occurrence of events larger than a given magnitude (Dunne and Leopold, 1978). The probability that the gullies incised with a certain land use (having a given C value) in the catchment is directly related to this exceedence probability. Since the duration of every land-use phase is not known, it is impossible to derive the exact probability of gully incision for every land-use type. From Figure 5B, it can be seen that rainfall events with a very high RI and thus a very low probability of occurrence (i.e., B/0.005) are needed to initiate gullies for 0B/C5/0.4 in the Meerdaal and the Tersaart Forest. From this analysis it becomes clear that, for both areas, it is virtually impossible that the gullies formed while a forest cover persisted in their catchments. This is also shown in Figure 6, where the data points for Meerdaal and Tersaart Forest are indicated in relation to the IDF curve that was used to determine the RI. Since these IDF Table 2 Runoff coefficients (C) extracted from the literature for forest, woodland and cropland Runoff coefficient (C) Author Criteria a Original source Forest Woodland Cropland ST TO RI (years) n.a Chow et al. (1988) N Y B/ 10 City of Austin, Texas n.a Chow (1964) Y N B/ 10 Bernard, s.d. n.a Dunne and Leopold (1978) N N n.a. American Society of Civil Engineers (1969); Rantz (1971) n.a Haan et al. (1994) Y Y 5/10 Schwab et al. (1971) 0.08/0.15 n.a. n.a. Kang et al. (2001) N N n.a. Own measurement in Chinese loess belt 0.14/0.18 n.a. 0.21/0.28 McCuen (1989) N N B/ 25 American Society of Civil Engineers (1969) 0.05/0.2 n.a. n.a. Mutreja (1995) N N 5/10 n.a. 0.1/0.2 n.a. 0.4/0.6 Raghunath (1988) N Y 5/10 Richards, s.d. 0.2 n.a. 0.4/0.6 Rodda et al. (1976) Y Y 5/10 Hempsall (1962) a, The author takes into account soil type (ST), topography (TO) and for which recurrence interval (RI) these are valid; N, the author does not take these into account. When relevant, the indicated values are for conditions in Meerdaal Forest (i.e., loamy soils and average slope of 8%). n.a, not available. s.d, no date given in reference cited in column 4.

115 384 The Holocene 15 (2005) Figure 5 A. Calculated rainfall intensities (I) needed to incise the observed gullies in Meerdaal Forest and Tersaart Forest for a range of land-use scenarios, expressed by the runoff coefficient (C). Calculated I values for the lowest runoff coefficient (C/0.05) are not shown in the graph for practical reasons. B. Calculated recurrence interval (RI) and associated exceedence probability (P) of the rainfall event (with intensity (I) and duration (D)) necessary to create the observed gullies in Meerdaal Forest and Tersaart Forest, for a range of land-use scenarios (0.055/C5/0.7) in the runoff-contributing area of old gullies curves are based on a 100-years measuring series only, the calculated upper limit for RI is 200 years. Determining IDF curves for RI/200 years is statistically not relevant. The data points that were calculated under the assumption of forest in the gully catchments are far from the 200-year RI curve. Since IDF curves for RI/200 years are statistically not relevant, it is impossible to determine the exact value of RI. Nevertheless, it is clear that C valuesb/0.3 yield virtually impossible solutions. In the range of degraded forest (0.35/C5/0.5), the results are very much the same as for forest. Degraded forest does not seem to be a very likely land use under which the gullies could have formed. Only at the very upper limit (C/0.5), does degraded forest yield a solution that falls within the limits of the IDF curves. The obtained RI of the rainfall event causing the gullies for this scenario would be minimum 128 years (P/0.01) for Meerdaal and 56 years (P/0.02) for Tersaart Forest. In Figure 6 it can also be seen that the upper limit (C/0.5) of the degraded forest data points falls within the zone of the IDF curve with RIB/200 years. For C/0.6, the obtained RI is 19 years (P/0.05) for Meerdaal Forest and 16 years (P/0.06) for Tersaart Forest. For C/0.7, the RI even lowers to 13 (P/0.07) and 11 years (P/0. 09), respectively. These results imply that the most probable land use at the time of gully initiation is cropland (0.45/C5/0.7). Forest as land use in the catchment at the time of gully incision is virtually impossible. A degraded forest cover would still be very improbable, since only the upper limit (C/0.5) yields acceptable solutions. Thus although W bottom and A of the gullies found in the two areas differ, the conclusions are the same for both forests. These conclusions corroborate the findings by Vanwalleghem et al. (2003), who stated that the old gullies in Meerdaal Forest were probably caused by local, human-induced clearings of the forest cover on the plateau positions. This means that after the deforestation of these plateau positions, intense rainfall caused the development of several large gullies. Since the gullies are still clearly visible nowadays, the deforested plateaus must have been abandoned quite rapidly after this

116 T. Vanwalleghem et al.: Reconstructing rainfall and land-use conditions during gully formation 385 Figure 6 Intensity/duration/frequency graph (KMI, 2001) with indication of rainfall conditions (intensity (I), duration (D), and recurrence interval (RI)) needed to incise the observed gullies in Meerdaal Forest and Tersaart Forest under different land use scenarios (0.15/C5/0.7). The data points corresponding with the lowest runoff coefficient (C/0.05) were not shown on the graph for practical reasons incision phase. After this incision phase, and under the climatic conditions prevailing in the study area, natural vegetation could recover quickly, preventing important runoff generation. If the cultivation phase had continued for long timespans, the gullies would have filled in completely, as was observed under cropland by several authors (Lang et al., 2003; Vanwalleghem et al., in press). However, it is not necessarily the case that all the gullies formed at the same time. It is very possible that they developed over a long time period and thus also as a consequence of different rainfall events. In the pre-roman slash and burn culture (Lindemans, 1952), farmers often moved to a new location, depleting the soil s nutrient supply in a certain area. Hence, gullies that developed in the consecutively cultivated areas would date from several consecutive time periods. The timing of gully development is an aspect that was not investigated in this study, since only one mean value was used for the gully parameters (e.g., W bottom and A) of the Meerdaal and Tersaart Forests. Only dating of the colluvium of several different gullies would allow a more definite answer to this question. Conclusions The method presented in this study allows us to infer past rainfall and especially past land-use conditions leading to the formation of old gullies under forest. Far from predicting exact rainfall and runoff conditions, however, the method presented here provides an extra tool to reconstruct the historical land use in the gully runoff-contributing area. The obtained results indicate that, in both Meerdaal and Tersaart Forests, the gullies that are nowadays preserved under forest, formed originally under cropland. Previous research has already shown that parts of Meerdaal Forest were deforested and used for agriculture. These results demonstrate the important effect of land use and land-use changes on gully erosion. Gullies are important links for sediment export through the catchment and contribute to an important increase in soil loss. Moreover, the formation of large gullies in farmers fields of ancient times might well have been an important reason to give up the affected field plot and move to a new location. Acknowledgements The authors would like to thank Ir B. Meuleman of the Flemish government, department Bos & Groen, for permission to conduct research in the Meerdaal Forest. Also the help of Dr G. Demarrée (KMI) in providing valuable background information on rainfall characteristics is greatly appreciated. We would also like to thank Prof. Dr M. Dotterweich, Dr G. Schmidtchen and Prof. Dr M. Stankoviansky for their critical review of a previous draft. References American Society of Civil Engineers 1969: Design and construction of sanitary and storm sewers. Manuals and reports on engineering practices 37, 1/332. Atlanta Regional Commission 2001: Georgia stormwater management manual. Vol. 2. Atlanta GA: Atlanta Regional Commission. Bollinne, A. 1976: L évolution du relief à l holocene, les processus actuels. In Pissart, A. and Macart, P., editors, Geomorphologie de la Belgique. Université de Liège, 159 /68. Bork, H.-R., Bork, H., Dalchow, C., Faust, B., Piorr, H.-R. and Schatz, T. 1998: Landschaftsentwicklung in Mitteleuropa. Stuttgart: Klett-Pertes. Butler, D.R. 1995: Zoogeomorphology: animals as geomorphic agents. Cambridge: Cambridge University Press. Chow, V.T. 1964: Handbook of applied hydrology. New York: McGraw-Hill. Chow, V.T., Maidment, D.R. and Mays, L.W. 1988: Applied hydrology. New York: McGraw-Hill. Dotterweich, M., Schmitt, A., Schmidtchen, G. and Bork, H.-R. 2003: Quantifying historical gully erosion in northern Bavaria. Catena 50, 135 /50. Dunne, T. and Leopold, L.B. 1978: Water in environmental planning. San Francisco CA: W.H. Freeman and Company.

117 386 The Holocene 15 (2005) Elliasson, J. 2002: The rational formula as a linear element in computer runoff models. European Geophysical Conference, 22/26 April, Nice, France. Elliot, W.J., Hall, D.E. and Graves, S.R. 1999: Predicting sedimentation from forest roads. Journal of Forestry 97, 23/29. Gábris, G., Kertész, A. and Zámbó, L. 2003: Land use change and gully formation over the last 200 years in a hilly catchment. Catena 50, 151/64. Gilley, J.E. Kottwitz, E.R. and Simanton, J.R. 1990: Hydraulic characteristics of rills. Transactions of the ASAE 33, 1900 /906. Graf, W.L. 1977: The rate law in fluvial geomorphology. American Journal of Science 277, 178 /91. Gullentops, F. 1992: Holocene soil erosion in the loess belt of Belgium. In Van der Haegen, H. and Van Hecke, E., editors, Liber Amicorum, Prof. Dr. M. Goossens. Acta Geographica Lovaniensa 33, 671/84. Haan, C.T., Barfield, B.J. and Hayes, J.C. 1994: Design hydrology and sedimentology for small catchments. New York: Academic Press. Hempsall, A. 1962: The assessment of flood risk in the UK. Journal of the Chartered Insurance Institute 59, 115/37. Kang, S.Z., Zhang, L., Song, X.Y., Zhang, S.H., Liu X.Z., Liang, Y.L. and Zheng, S.Q. 2001: Runoff and sediment loss responses to rainfall and land use in two agricultural catchments on the Loess Plateau of China. Hydrological Processes 15, 977 /88. KMI 2001: Extreme precipitation in Flanders, new IDF-curves based on long-term precipitation measurements. Brussels: Ministerie van de Vlaamse Gemeenschap (in Dutch). Knighton, A.D. 1974: Variation in width/discharge relation and some implications for hydraulic geometry. Geological Society of America Bulletin 85,1069/76. Kosov, B.F., Nikol skaya, I.I. and Zorina, Ye.F. 1978: Eksperimental nyye issledovaniya ovragoobrazovaniya. In Makkaveev, N.I., editor, Eksperimental naya geomorfologiya. Vol. 3. Moskva: Izd Moskva Univ., 113/40 (in Russian). Lane, L.J. and Foster, G.R. 1980: Concentrated flow relationships. In Knisel, W.G., editor, CREAMS / a field scale model for chemicals, runoff and erosion from agricultural management systems. USDA Conservation Research Report no. 26, United States Department of Agriculture (USDA), Washington DC: US Governmental Printing Office, 474/85. Lang, A. 2003: Phases of soil erosion-derived colluviation in the loess hills of South Germany. Catena 51, 209/21. Lang, A., Bork, H.R., Mäckel, R., Preston, N., Wunderlich, J. and Dikau, R. 2003: Changes in sediment flux and storage within a fluvial system-some examples from the Rhine catchment. Hydrological Processes 17, 3321/34. Leopold, L.B. and Maddock, T. JR. 1953: The hydraulic geometry of stream channels and some physiographic implications. U.S. Geological Survey Professional Paper 252. Lindemans, P. 1952: History of agriculture in Belgium. Vol. 1. Antwerp: De Sikkel (in Dutch). Martens, E. 1981: From the past of the community of Oud-Heverlee. Neerijse: Stroobants (in Dutch). McCuen, R.H. 1989: Hydrologic analysis and design. Englewood Cliffs NJ: Prentice Hall. Mutreja, K.N. 1995: Applied hydrology. New Delhi: Mc Graw-Hill. Nachtergaele, J., Poesen, J., Sidorchuk, A. and Torri, D. 2002a: Prediction of concentrated flow width in ephemeral gully channels. Hydrological Processes 16, 1935/53. Nachtergaele, J., Poesen, J., Oostwoud Wijdenes, D. and Vandekerckhove, L. 2002b: Medium-term evolution of a gully developed in a loess-derived soil. Geomorphology 46, 223 /39 Poesen, J., van Wesemael, B. and Cammeraat, E. 1993: Gully erosion in the loess belt: typology and control measures (Kinderveld site, Korbeek-Dijle). In Goossens, D., editor, Geomorphological processes in the Belgian loessbelt. Excursion guide. International Symposium on Experimental Geomorphology and Landscape Ecosystem Changes. Laboratory for Experimental Geomorphology, K.U. Leuven, Belgium, 16/28. Poesen, J., Nachtergaele, J. and Deckers, J. 2000: Gullies in the Tersaart Forest (Huldenberg): climatic or anthropogenic cause? In Verstraeten, G., edito, Gully erosion processes in the Belgian loess belt: causes and consequences. Excursion guide. International Symposium on Gully Erosion under Global Change, Laboratory for Experimental Geomorphology, K.U. Leuven, Belgium, 15/25. Poesen, J., Nachtergaele, J., Verstraeten, G. and Valentin, C. 2003: Gully erosion and environmental change: importance and research needs. Catena 50, 91 /133. Raghunath, H.M. 1988: Hydrology: principles, analysis and design. New Delhi: Wiley Eastern. Rantz, S.E. 1971: Suggested criteria for hydrologic design of stormdrainage facilities in the San Francisco Bay region, California. Menlo Park, CA: US Geological Survey Open File Report. Rodda J.C., Downing, R.A. and Law, F.M. 1976: Systematic hydrology. London: Butterworth. Rossmiller, R.L. 1981: The runoff coefficient in the rational equation. Ames IA: Engineering Research Institute, Iowa State University. Rutherford, I.D., Prosser, I.D. and Davis, J. 1997: Simple approaches to predicting rates and extent of gully development. In Wang, S.S.Y., Langendoen, E.J. and Shields, F.D., editors, Proceedings of the conference on management of landscapes disturbed by channel incision. Oxford MI: Univerdity of Mississippi 1125/30. Schmitt, A., Dotterweich, M., Schmidtchen, G. and Bork, H.R. 2003: Vineyards, hopgardens and recent afforestation: effects of late Holocene land use change on soil erosion in northern Bavaria, Germany. Catena 51, 241/54. Schwab G.O., Frevert, R., Edminster, T.W. and Barnes, K.K. 1971: Elementary soil and water conservation engineering. New York: Wiley. Semmel, A. 1995: Development of gullies under forest cover in the Taunus and Crystalline Odenwald Mountains, Germany. Zeitschnift für Geomorphologie Neuro Funde Supplement Board. 100, 115 /27. Sidorchuk, A. 1999: Dynamic and static models of gully erosion. Catena 37, 401 /14. Stankoviansky, M. 2003: Historical evolution of permanent gullies in the Myjava Hill Land, Slovakia. Catena 51, 223/39. Tack, G. and Hermy, M. 1993: Woodlands of Flanders. A historical ecology. Leuven: Davidsfonds (in Dutch). Titmarsh, G.W., Cordery, I. and Pilgrim, D.H. 1995: Calibration procedures for the rational and USSCS design flood methods. Journal of Hydraulic Engineering 121, 61 /70. Van Uytven, R. 1975: Map of Meerdaal, Mollendaal and Heverlee (1769); History (in Dutch). In: Jaarbulletin van de vrienden van Heverleebos en Meerdaalwoud, Heverlee, 25/36. Vanwalleghem, T., Van Den Eeckhaut, M., Poesen, J., Deckers, J., Nachtergaele, J., Van Oost, K. and Slenters, C. 2003: Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium). Geomorphology 56, 15 /29. Vanwalleghem, T., Bork, H.R., Poesen, J., Schmidtchen, G., Dotterweich, M., Nachtergaele, J., Bork, H., Deckers, J., Brüsch, B., Bungeneers, J. and De Bie, M. in press: Rapid development and infilling of a historical gully under cropland, central Belgium. Catena. Viessman, W.J., Knapp, J.W. and Lewis, G.L. 1989: Introduction to hydrology. New York: Harper and Row. Vogt, J. 1953: Erosion des sols et techniques de culture en climat tempere maritime de transition (France et Allemagne). Revue de Geomorphologie Dynamique 4, 157/83.

118

119 Geomorphology 56 (2003) Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium) T. Vanwalleghem a,b, *, M. Van Den Eeckhaut b, J. Poesen b, J. Deckers c, J. Nachtergaele b, K. Van Oost b, C. Slenters b a Fund for Scientific Research-Flanders, Belgium b Laboratory for Experimental Geomorphology, K.U. Leuven, Belgium c Institute for Land and Water Management, K.U. Leuven, Belgium Received 29 August 2002; received in revised form 20 December 2002; accepted 23 December 2002 Abstract In many forests of Northwestern Europe old gullies can be found, but few studies have reported their genesis and characteristics. This study investigates these old gullies under forest in the large case-study area of Meerdaal Forest, in the Central Belgian loess belt. The objectives are (1) to determine the spatial distribution of these gullies, (2) to measure their morphological and topographical characteristics and (3) to reconstruct the factors that led to their development. In the 1329-ha study area, 252 channel-like incisions were mapped. Different types of incisions could be distinguished. Besides small and large gullies, many incisions were sunken lanes or road gullies. These road gullies are aligned along north south oriented lines, whereas the concentration of old gullies is strongly related to the distribution of archaeological sites. Out of the 252 mapped incisions, 43 large gullies and 21 representative road gullies were selected for detailed morphological and topographical measurements. The characteristics of these two types of incisions were compared with ephemeral gullies formed under nearby cropland. Significant differences in morphology between the three types could be demonstrated. Ephemeral gullies under cropland and large gullies under forest differ significantly in all measured parameters, except bottom width. Both the old gullies and road gullies under forest have a significantly larger cross section and total eroded volume compared with the ephemeral gullies observed under cropland. This indicates that once formed, the old gullies were not ploughed in nor were they filled by sediment originating in their drainage areas, because of limited sediment production. Comparing topographical characteristics (i.e. slope at the gully head and runoff contributing area) of forest gullies and ephemeral gullies that formed under cropland yields important indications about their formation. The larger sedimentation slope of forest gullies, compared with ephemeral gullies and road gullies, suggests that the forest gullies incised on vegetated slopes as a consequence of runoff from the adjacent plateau, where the forest cover was disturbed. For the old gullies under forest, no relation between slope at the gully head and runoff contributing area is observed, probably because most gullies occur on very steep slopes. When simulating arable land-use in the study area, zones where ephemeral gullies are expected to develop can be predicted using published topographical threshold relationships. Comparing the zones where ephemeral gullies are predicted with the position of old gullies under forest leads to the conclusion that gully incision was most probably not triggered by extreme rainfall events * Corresponding author. Laboratory for Experimental Geomorphology, K.U. Leuven, Redingenstraat 16, B-3000 Louvain, Belgium. Tel.: ; fax: address: tom.vanwalleghem@geo.kuleuven.ac.be (T. Vanwalleghem) X/$ - see front matter D 2003 Elsevier Science B.V. All rights reserved. doi: /s x(03)

120 16 T. Vanwalleghem et al. / Geomorphology 56 (2003) and that they are not of periglacial origin. The observed gully pattern can best be explained by local, anthropogenically determined land-use changes. D 2003 Elsevier Science B.V. All rights reserved. Keywords: Gully erosion under forest; Central Belgium; Topographical thresholds; Human-induced land-use changes 1. Introduction The problem of gully erosion in the West European loess area is well known and documented (e.g. Poesen and Govers, 1990; Papy and Douyer, 1991; Auzet et al., 1995; Ludwig et al., 1995; Nachtergaele et al., 2002). Most studies, however, focus on ephemeral gully development under cropland. With respect to permanent gullies, as observed under forest in the Belgian loess belt, relatively few studies have been conducted (e.g. Arnould-De Bontridder and Paulis, 1966; Langohr and Sanders, 1985; Gullentops, 1992; Poesen et al., 2000, 2003). Currently, little is known about their spatial distribution, about their morphological and topographical characteristics or about their genesis (Poesen et al., 2003). With the current climatic conditions in Northwestern Europe, no gully erosion is expected to occur under undisturbed forest vegetation, mainly because of the high infiltration capacity of these forest soils, which makes Hortonian runoff almost impossible. Therefore, these forest gullies must be old landscape features. Different hypotheses can explain the formation of these gullies. A first is that these gullies developed under forest as a consequence of high magnitude low frequency rainfall events during which significant runoff was generated, even under a protective forest cover. Comparable with this is the hypothesis of Gullentops (1992), who, based on the study of peat sequences, stated that many large gullies formed under forest during the wetter Atlanticum ( BP), a period characterised by higher rainfall intensities then today. A second hypothesis is that runoff was generated locally in areas with a disturbed forest cover due to humaninduced land-use changes (e.g. cropland, intensive cattle grazing in forests, forest logging), and that reforestation occurred afterwards. A third hypothesis is a combination of human-induced land-use change and extreme rainfall. Bork et al. (1998) showed that this mechanism was the driving factor for the incision of many gullies presently under forest in Germany. A final hypothesis was proposed by Langohr and Sanders (1985) on the basis of research in the Zonien forest (Central Belgium), who suggested these old gullies could be periglacial features, cut into the loess cover long before a protective plant cover was established during the Holocene. In order to elucidate the conditions leading to the development of gullies presently found under forest, representative old gullies in forested areas are studied in detail. More specifically, the objectives of this study are (1) to determine the spatial distribution of these old gullies under forest in a representative large study area, (2) to measure morphological and topographical characteristics of old gullies under forest and to compare them with the characteristics of ephemeral gullies formed under cropland and (3) to reconstruct the environmental conditions that triggered gully formation in a large forest area. Therefore, gully position is predicted based on the assumption that the entire area was used as cropland in the past. The predicted ephemeral gully pattern is compared with the distribution pattern of the mapped old gullies. Together with the distribution pattern and characteristics, these elements should indicate whether gully formation was triggered by extreme rainfall events, human-induced land-use changes, periglacial processes, or a combination of processes. 2. Study area The focus of this study is on the Belgian loess belt and the area selected for this research is one of the last original forests in Central Belgium: i.e. the Meerdaal Forest (Fig. 1). As far as is known by historians, this forest was never deforested or used for intensive agriculture on a large scale. The Meerdaal Forest covers almost 1700 ha. Part of the forest was inaccessible because of military

121 T. Vanwalleghem et al. / Geomorphology 56 (2003) Fig. 1. Location of the Meerdaal Forest within Belgium and digital elevation model of the study area. installations, therefore the study area is somewhat smaller (i.e hectares). Altitudes range between 35 and 103 m a.s.l., with a typical rolling topography characteristic of the Belgian loess belt. However, steep slopes, even >25% do occur. Mean annual rainfall is f 800 mm (Baeyens et al., 1957). The main soil types are Luvisols and Albeluvisols. Only locally, on eroded hilltops, Tertiary sands that underlie the loess deposits outcrop and there Podzols can be found (Baeyens et al., 1957). 3. Materials and methods After conducting an intensive field survey to locate all channel-like incisions in the study area, their position was mapped with GPS (Trimble Pro-XL). Channel-like incisions were defined as linear incisions with a minimum length of about 15 m and a minimum depth of 0.2 m. Furrows and drainage ditches were excluded. Mainly because of multi-path effects, large errors remained (maximum error up to several tens of metres), even after differential correction. Where possible, these errors were corrected manually. In total, 252 channellike incisions were mapped. Three different types of incisions were recognised: small gullies (< 0.3 m depth, n = 39), large gullies (>0.3 m deep, n = 43) and road gullies (n = 170). The distinction between large gullies and road gullies will be discussed in detail later. A group of anthropogenic features was also mapped. These are structures that are clearly man-made. Some are large levees up to 3 m in height, but most of these structures are deep sunken lane -like features (maximum depth around 5 m), extending over several hundreds of metres, even on flat terrain. Although their original function is not fully understood, it is generally assumed that these features served as roads (Dens, 1908; Vincent, 1925; Martens, 1981). In a second phase, the 43 large gullies and 21 randomly selected road gullies were characterised in more detail. The 39 small gullies were omitted from further detailed analysis, because they are not clearly visible in the landscape and this made it difficult to objectively determine gully morphology. Only the clearly visible, large (depth >0.3 m) gullies were selected for detailed measurement. Gullies were divided into representative segments and for each segment length, top and bottom width were measured with a measuring tape. Depth was measured several times along each individual cross section in order to calculate cross-sectional area and gully volume. For each gully, only the maximum depth per segment was used to calculate mean gully depth. The respective mean value of each parameter for the whole gully was obtained by summing up the different segments. Each segment was weighted by its length. Gully volume

122 18 T. Vanwalleghem et al. / Geomorphology 56 (2003) was also obtained by summing up the different segment volumes. Width depth ratio (WDR) was calculated as the ratio of mean gully top width and depth. Slope of the soil surface at the gully head (S g ) and slope of the soil surface where the gully ends by sediment deposition (i.e. sedimentation slope, S s ) were determined with a clinometer (type Suunto, error m/m). Since gully heads may have retreated upslope since their formation (Nachtergaele et al., 2001), S g was taken as the steepest slope along the gully trajectory. The runoff contributing area (A) was delimited with markers and measured in the field with a measuring tape. In a third phase, the mapped gully location was compared with the zones where ephemeral gullies would be expected to form if the area was cleared for cropland. Only the large gullies that were measured in detail (n = 43) are included in this analysis. The small gullies and road gullies are excluded. To predict gully location, mainly known equations were used. For the Belgian loess belt, several studies have fairly accurately predicted the position of ephemeral gullies under cropland, using only topographical parameters. Desmet et al. (1999) and Vandaele et al. (1996) worked in a catchment that is comparable to the Meerdaal Forest in terms of topography and soils. Therefore, the topographical threshold equations they proposed may be extrapolated to the Meerdaal Forest area. These equations are of the form: S g > aa b ð1þ with S g the slope of the soil surface at the gully head (m/m), A the runoff contributing area (ha) and a, b are coefficients. These topographical attributes are derived from a grid-based DEM. The DEM (pixel size 5 5 m) was constructed by digitising and interpolating the 1: topographical map (NGI, 1972). Slopes were calculated with the SLOPE-module in IDRISI32 and the contributing area for each pixel is based on the multiple flow algorithm of Desmet and Govers (1996). Desmet et al. (1999) showed that Eq. (1) has to be written as: S g > ava bv s where A s ¼ the unit runoff contributing area ðm 2 =mþ ð2þ When A s is derived from a grid-based DEM, the zones where ephemeral gully erosion can be expected to occur if the land-use was cropland, can be obtained using these threshold equations. Depending on the value of the coefficients, two types of threshold equations can be distinguished: the first type predicts gully trajectories, whereas the second type predicts gully initiation points (Desmet et al., 1999). Apart from the published threshold equation coefficients, several new coefficients were used in the predictions in order to obtain a larger percentage of predicted pixels or higher efficiency. Prediction efficiency is defined as the ratio of the total number of correctly predicted gully pixels and the total number of predicted pixels. 4. Results and discussion 4.1. Spatial distribution of gullies in Meerdaal Forest In total, 252 incisions were mapped, of which 82 were classified as small or large gullies and 170 as road gullies. Fig. 2 shows all mapped channel-like incisions within the study area. Of the 82 gullies, 39 were classified as small gullies and 43 as large gullies. As mentioned before, the small gullies were not analysed further because they are only poorly visible in the landscape. Nevertheless, their spatial distribution is very similar to that of the large gullies (Fig. 2). This distribution pattern can also be observed in fields that are under cultivation. During an erosive rain event, many rills can form, but only a few will develop into ephemeral gullies, because they capture more runoff than the other rills. Thus, the presence of these small gullies near large gullies indicates runoff production in their catchment. Besides the small and large gullies, a very large group of incisions (n = 170) could be identified as old sunken lanes or road gullies. In the field, several criteria were used to distinguish between large gullies and road gullies (Fig. 3). Road gullies are typically found on gentle slopes and have no or almost no drainage area. Furthermore, most road gullies are often found in groups, with an interval of 5 15 m separating them from each other. The distribution of the road gullies shows a linear pattern: most of them are aligned along straight,

123 Fig. 2. Spatial distribution of the mapped incisions within the study area, with indication of the different identified road systems and the location of known archaeological sites. Several tumuli or burial mounds that are located immediately next to each other are indicated with only one symbol on the map. T. Vanwalleghem et al. / Geomorphology 56 (2003)

124 20 T. Vanwalleghem et al. / Geomorphology 56 (2003) Fig. 3. Illustration of two types of incisions in Meerdaal Forest: (A) large gully and (B) road gully. Note the typical V-shaped and trapezoidal cross-sectional profile of the large gully and the road gully, respectively. north south oriented directions (Fig. 2). The observed road gully pattern under forest is related to the location of the anthropogenic features, described earlier. They also show the same linear pattern as the road gullies. Together with the road gullies they are indicated in Fig. 2 as road systems (dashed lines) through the Meerdaal Forest. The gullies, on the other hand, are more widely distributed, with the largest concentration in the southern part of the forest. To explain this distribution pattern, the relation with slope gradient, aspect, lithology and soil type was examined. Although the steepest slopes occur in the southern part of the forest, many steep slopes did not show traces of incision. Also with slope aspect, lithology or soil type, no clear spatial correlation could be found. The only correlation that could be established is when comparing the distribution of gullies with the pattern

125 T. Vanwalleghem et al. / Geomorphology 56 (2003) of known archaeological sites in the Meerdaal Forest (based on Martens, 1981), which are also located mainly in the south of the forest. In Fig. 2, the location of these sites in the study area is indicated. Four main periods of human occupation can be distinguished. Stone Age artefacts were found on several locations in the Meerdaal Forest but most of the sites are Iron Age or Roman tumuli or burial mounds. Apart from these burial mounds, no archaeological evidence exists that would indicate the presence of human settlements, large-scale deforestation or agriculture. Only at one location (Fig. 2) do some brick fragments suggest the possible presence of a Roman villa. In the 14th century, the forest obtained a special status of Vrijwoud, meaning that it could only be used for hunting purposes by the local nobility. Since then, it is certain that no large clearings have taken place. Nevertheless, the correspondence between the spatial distribution of gullies and that of archaeological sites yields a first indication that the distribution of the gullies is strongly related to very local anthropogenic disturbances of the forest cover, probably between Iron Age and Roman times Morphological and topographical characteristics The morphological and topographical characteristics of the gullies and road gullies in the Meerdaal Forest are summarised in Table 1. The characteristics of typical ephemeral gullies formed under cropland (EG) in the Belgian loess belt, are also indicated for comparison (Nachtergaele, 2001). The large gullies under forest differ significantly from the EG in all measured parameters, except bottom width. Also, most characteristics of the road gullies and large gullies under forest differ significantly from each other, except for the parameters length and bottom width (a = 5%). The length of EG is significantly (a = 5%) larger than both types of incisions under forest, although the standard deviation is large. The longest gully under forest is 214 m long and the shortest only 19 m. The longest ephemeral gully measured was almost 437 m long. These differences can be explained by the fact that forest gullies are often located on steep, short slopes. This indicates that the forest gullies are not always located in the same landscape positions compared to ephemeral gullies in cropland, which often occupy valley-bottom positions (Nachtergaele and Poesen, 1999). Gully-bottom width of EG does not differ significantly from the incisions under forest. However, gully depth (GD) and gully-top width (GW t ) of both gullies and road gullies is significantly larger than that of EG. The larger GD and GW t of both types of incisions under forest compared with the EG results in a significantly larger cross section. This larger cross section compensates for the shorter length, so that mean gully volume (GV) is up to 30 times larger for the gullies under forest. The maximum individual cross-sectional area measured for Table 1 Morphological and topographical characteristics of the large gullies and road gullies in Meerdaal Forest and ephemeral gullies under cropland in Central Belgium All incisions under forest Gullies forest Road gullies forest Ephemeral gullies cropland (Nachtergaele, 2001) n Mean SD n Mean SD n Mean SD n Mean SD GL (m) GW b (m) GW t (m) GD (m) WDR GA (m 2 ) GV (m 3 ) 64 69E 1 16E E 1 19E E 1 32E S g (m/m) S s (m/m) A (ha) Data for ephemeral gullies were collected by Nachtergaele (2001). GL = gully length; GW b = gully width at bottom; GW t = gully width at top; GD = gully depth; WDR = width depth ratio of gully ( = GW t /GD); GV = gully volume; GA= gully cross section; S g = slope of soil surface at gully head; S s = sedimentation slope.

126 22 T. Vanwalleghem et al. / Geomorphology 56 (2003) gullies under forest was 97.1 m 2. The maximum gully depth of 7.80 m and also the largest top width of m were measured in this cross section. Ten of the 43 forest gullies have volumes >1000 m 3, the largest eroded gully volume measured is 99E 2 m 3. Only one road gully had a volume >1000 m 3. This difference in GW t and GD between the different types of incisions is also reflected in the width depth ratio, which is almost the same for gullies and road gullies under forest, but significantly higher for EG under cropland. The reason for the larger cross section and larger eroded volume for the incisions under forest is probably that EG are formed during one or two seasons with erosive rain events, after which they are erased by ploughing, whereas incisions under forest could develop during many years. Nachtergaele et al. (2002) studied an ephemeral gully that was not erased by ploughing after its formation and which developed into a permanent gully. They observed an increase of gully size during the first few years, then gully volume started to decrease, because sediment was captured by vegetation growing in the gully bottom. The gullies under forest have probably never reached this last phase of infilling or experienced only a limited infilling, because runoff and sediment production in the catchment area was cut off by a change in the land-use (e.g. cropland to forest). Another important difference between old incisions under forest and recent ephemeral gullies under cropland is the cross-sectional shape. In recent gullies under cropland, gully walls are vertical or even overhanging. Even in gullies that had formed 10 years before, walls were still vertical (Nachtergaele et al., 2002). This results in a rectangular shaped cross-sectional area for ephemeral gullies. Gullies under forest, however, have typically V-shaped cross-sectional areas (Fig. 3a). Mean gully wall gradients of m/m were measured. The maximum observed gradient was only m/m, which is still far from vertical. Road gullies typically have a trapezoidal cross section, but their wall gradients were not quantified. Besides the fact that in the thalweg of the large forest gullies and road gullies, sometimes trees of over 100 years old can be found, their cross-sectional shape is an important indication suggesting these incisions cannot be recent. As a conclusion, it can be stated that a morphological differentiation between gullies and road gullies under forest is possible, but not always very pronounced, while both types of incisions under forest differ significantly from EG, which form under cropland conditions. When comparing the slope of the soil surface at the gully head (S g ), which approximates the slope at which the gully initiated, for the three types of incision, it can be seen that forest gullies are clearly located on steeper slopes compared to the EG and the road gullies, that have almost identical S g values. The difference in S g between the forest gullies and the EG is significant at the 5% level, while no significant difference could be demonstrated between S g of road gullies under forest and EG. A similar conclusion can be drawn for S s. The forest gullies typically end at a mean soil surface slope of 0.10 m/m, which is even more than the mean slope at which road gullies and EG start to incise (0.090 and m/m, respectively). Although the difference in S s between road gullies under forest and EG is small in absolute terms (0.010 m/m), in this case however the differences between the three types are significant (a = 5%). This can be seen more clearly in Fig. 4, which shows the frequency distribution of the sedimentation slopes of the gullies under forest, the road gullies under forest and the ephemeral gullies under cropland. Besides topography, the slope at which sedimentation occurs is influenced by rock content (Poesen et al., 2002) and vegetation (Beuselinck et al. 2000). Since topsoil rock content is very low for soils in the Meerdaal Forest, the different sedimentation slopes must be controlled by either topography or vegetation. The fact that gullies under forest have steeper sedimentation slopes could thus indicate that sedimentation was not purely topographically controlled, but was promoted by the presence of vegetation on the eroding slopes. This is another important indication that these gullies were caused by human disturbances of the forest vegetation in the gully catchment. If cultivation had occurred in Meerdaal Forest, it is easy to imagine that while locally the plateau positions were deforested, the adjacent steep slopes would have remained vegetated because soils are less fertile and difficult to cultivate. In this way, Hortonian runoff generated on the plateau could then easily erode the gullies on the forested slopes. Since these slopes were vegetated, sedimentation occurred at steeper slopes compared with bare cropland slopes. Sedimentation slope of road gullies on the other hand, which have a completely bare surface due to trampling of humans and animals, is

127 T. Vanwalleghem et al. / Geomorphology 56 (2003) Fig. 4. Frequency distribution of sedimentation slope (S s ) of gullies and road gullies in Meerdaal Forest and ephemeral gullies under cropland in Central Belgium. Data for ephemeral gullies were collected by Nachtergaele et al. (2001). more comparable to the situation of ephemeral gullies under cropland, where vegetation cover is minimal and sediment deposition is essentially controlled by local topography. The runoff contributing area (A) of the forest road gullies is very small (A= ha), which again points to their anthropogenic genesis. Also A of the forest gullies is almost one order of magnitude smaller compared with A for EG. Fig. 5, showing the relation between S g and A for the forest gullies, the road gullies under forest and the ephemeral gullies under cropland, yields more insight into their formation. The main water erosion processes on arable land in the Belgian loess belt are driven by Hortonian overland flow. Studies (e.g. Patton and Schumm, 1975; Montgomery and Dietrich, 1994; Nachtergaele et al., 2001) demonstrated that under Hortonian overland flow a negative power relation between S g and A is expected (Eq. (2)). For this dataset, it could however be statistically demonstrated that no correlation exists between S g and A. According to Montgomery and Dietrich (1994), this could indicate that gully formation is affected by other processes, such as saturation overland flow or landsliding. However, no evidence for the presence of exfiltration processes was observed in the study area, although field mapping occurred in a very wet year. Also, the physical conditions for landsliding are absent. There is no soil horizon or geological layer that could function as a potential shear plane. The only explanation that can be given for the absence of correlation between S g and A is that the forest gullies are located on very steep slopes. Thus, it is reasonable to assume that drainage area plays only a minor role and the relative importance of slope in gully erosion is much larger. Therefore, for a given slope gradient, gullies can be formed within a range of runoff contributing areas and these can even be quite small compared with EG formed under cropland. When comparing the data points corresponding to the road gully heads with those corresponding to the forest gullies, it can be seen that they form two clearly distinct groups. As discussed above, the road gullies are generally on more gentle slopes and have little runoff contributing area. Since the road gullies are anthropogenically determined, and not purely topographically, it is logical that there is no correlation between S g and A. The gullies under forest also form a separate group when compared with ephemeral gullies formed under cropland conditions (Nachtergaele et al., 2001). Fig. 5 indicates that the forest gullies are not always located in the same landscape positions compared to the ephemeral gullies in cropland. This discrepancy is most probably linked to a difference in position of runoff-producing areas: i.e. under forest, locally cleared plateau positions draining to steep slopes where the forest gullies developed, whereas in the case of cropland, completely cleared zero-order catchments, leading to the development of ephemeral gullies in valley-bottom positions.

128 24 T. Vanwalleghem et al. / Geomorphology 56 (2003) Fig. 5. Relation between slope of the soil surface at the gully head (S g ) and runoff contributing area (A) for the gullies and road gullies in Meerdaal Forest and comparison with ephemeral gullies under cropland in Central Belgium. The critical topographical threshold lines that are indicated in Fig. 5 were established for ephemeral gullies under cropland in Central Belgium on the basis of field data collected by other authors. Topographic positions plotting below these threshold lines are stable and no gully incision is expected to occur, whereas gully erosion can be expected in landscape positions plotting above these lines. It has to be mentioned that for comparative purposes these lines have been extended in the graph outside the range of data points on the basis of which they were originally drawn. Nevertheless, although these thresholds are established for cropland, most points corresponding to the forest gullies are located above these lines. When considering, for example, the threshold line established by Nachtergaele et al. (2001), 31 of the 38 forest gully points are located above this threshold. Thus, although S g and A for the two gully types are different, the combination of S g and A seems to compensate for this difference and the forest gullies do generally not occur below the topographical threshold line established for ephemeral gullies under cropland Prediction of gully location Under the hypothesis that the whole of Meerdaal Forest was cleared and used as cropland in the past, Fig. 6 shows the zones (in grey) where gully trajectories are expected to occur, using the topographic thresholds for ephemeral gully development established by Desmet et al. (1999). The mapped gullies are shown in black. In the southern part of the forest, where most gullies are concentrated, mapped gully position coincides fairly well with the zones where gullies are expected to occur. However, it can be seen that the overall prediction is not so good because in a large part of the study area, especially in the north, gully erosion is predicted to occur, but no gullies are present. This results in a low prediction efficiency. This is an observation that is valid for all the different threshold equations that were used (Table 2). The

129 Fig. 6. Prediction of (ephemeral) gully trajectories based on the topographic thresholds proposed by Desmet et al. (1999) for the catchment of Hammeveld (Central Belgium), with similar topographical and lithological characteristics as the Meerdaal Forest. S g >ava s bv with S g = slope of soil surface at gully head (m/m); A s = unit runoff contributing area (m 2 /m); av= 33 and bv= 1. T. Vanwalleghem et al. / Geomorphology 56 (2003)

130 26 Table 2 Prediction of ephemeral gully trajectories or ephemeral gully initiation positions based on topographic thresholds established by several authors (S g >ava s bv ) with S g = slope of soil surface at gully head (m/m); A s = unit runoff contributing area (m 2 /m) and av, bvare coefficients Pixel size 5 5m Total area (ha) 1329 Total number of pixels corresponding to study area Total number of mapped gullies 43 Total number of gully pixels 764 Source Coefficients Total number of predicted pixels Number of gully pixels correctly predicted % of total study area predicted % of gully pixels predicted Prediction of Desmet et al. (1999) av= 52.5; bv= gully trajectory Kinderveld Desmet et al. (1999) av= 33; bv= Hammeveld Desmet et al. (1999) av= 18; bv= adapted b,c Desmet et al. (1999) av= 15; bv= adapted c Prediction of gully Vandaele et al. (1996) av= 0.486; bv= initiation location Vandaele et al. (1996) adapted d av= 0.55; bv= Vandaele et al. (1996) av= 0.7; bv= adapted d Vandaele et al. (1996) av= 0.874; bv= adapted d,e Desmet et al. (1999) DEM av= 1.15; bv= a Efficiency of prediction = total number of gully pixels correctly predicted/total number of predicted pixels100. b This relation equals part of the threshold equation proposed by Moore et al. (1988). c Adaptation of coefficient avin threshold equation Desmet et al. (1999) in order to improve number of gully pixels correctly predicted. d Adaptation of coefficient avin threshold equation Vandaele et al. (1996) in order to improve efficiency of prediction. e This relation equals the threshold equation proposed by Nachtergaele et al. (2001). Efficiency of prediction a (%) T. Vanwalleghem et al. / Geomorphology 56 (2003) 15 29

131 Fig. 7. Comparison of predicted zones where ephemeral gullies are expected to develop under cropland with mapped gullies and road gullies in the Meerdaal Forest. Prediction of gully initiation locations is based on the topographic threshold proposed by Vandaele et al. (1996): S g >ava s bv with S g = slope of soil surface at gully head (m/m); A s = unit runoff contributing area (m 2 /m); av= and bv= 0.4. T. Vanwalleghem et al. / Geomorphology 56 (2003)

132 28 T. Vanwalleghem et al. / Geomorphology 56 (2003) highest proportion of gully pixels that could be correctly predicted is somewhat more than 80%, but 27% of the total study area has to be predicted as prone to gully erosion (threshold equation by Vandaele et al., 1996). Thus, this relatively low prediction efficiency (overall < 1%) is not necessarily because the mapped gullies are not predicted correctly, but more because all the simulations predict many large zones where no gullies are observed. It is not clear why many zones are predicted as being susceptible to gullying, whereas no gullies are observed. The best explanation is probably that predicted zones where no gullies are found were never deforested. Forest zones where gullies were mapped can be predicted by the same topographic threshold relations that are valid for prediction of ephemeral gullies under cropland. Therefore, it is reasonable to assume that these zones were locally cleared and used for cropland, charcoal production, animal grazing, or any land-use that disturbed the permeable forest floor. Subsequently, ephemeral gullies could form and evolve to larger, permanent gullies after which the area was reforested. Future research, focussed on some case studies, needs to indicate the timing and nature of the land-use disturbance that initiated gully erosion. The possibility that extreme rainfall events caused severe gully erosion seems invalid. Firstly, most gullies are concentrated in the southern part of the forest, and this corresponds very well with the distribution pattern of archaeological sites, as mentioned earlier. Also, the prediction of many zones prone to gullying where no gullies are observed indicates that an extreme event did not cause gully incision. Similar arguments can also be applied to explain why it is difficult to assume that the gullies formed under periglacial conditions. Moreover, if the gullies are not of anthropogenic origin, but the result of natural periglacial erosion processes, one would expect a spatial distribution that reflects systematic patterns or a clear correlation with slope gradient, aspect or lithology. As shown above, this relation could not be found. When comparing the zones in the forest where ephemeral gully erosion can be expected with the position of the road gullies, another interesting observation can be made. This is illustrated in Fig. 7 with the threshold relation of Vandaele et al. (1996) which predicts the highest percentage of gully pixels correctly (i.e. 80.5%), but also predicts almost one third of the total study area (i.e. 27.0%) as being susceptible to incision by ephemeral gullies (Table 2). However, practically all the road gullies are located outside these predicted zones. This indicates that the location of the road pattern was not random. Since man chose the easiest trajectory (and thus with a low gully erosion risk because of gentle slopes) to cross the hills, the road gullies are located outside the predicted zones where gullies are expected to occur. This method also provides extra evidence to distinguish between gullies and road gullies. 5. Conclusions and implications The analysis of the spatial distribution, the morphological and the topographical characteristics of the large forest gullies and the modelling results of the location of ephemeral gullies in a cleared forest, indicate that the forest gullies were caused by local disturbances of the forest cover on the plateau positions. These plateau positions were very suitable for use as cropland, whereas the slopes were left vegetated because cultivation is difficult and, in some cases, the fertile loam cover overlying the Tertiary sands is very shallow. Nevertheless, large gullies could develop on these forested slopes. These gullies did not immediately hinder further cultivation of the plateau positions and several runoff-producing events could deepen and widen them. However, before the gully could fill in naturally by sediment deposition, the forest on the plateau was re-established, thus cutting off runoff and sediment production in the drainage area. The hypothesis that extreme rainfall events caused runoff and erosion under a protective forest cover proved highly unlikely. Also, the hypothesis that the gullies were formed before the establishment of a forest vegetation in a periglacial environment could not explain their spatial distribution pattern. This study shows that modelling gully location based on knowledge of ephemeral gully erosion under cropland is a useful tool for assessing the controlling factors of past gully development. This study also indicates that even in forested areas, which were thought to represent natural conditions, human impact has been important in the past. The traces (i.e. the forest gullies) of these old human-induced land-use changes are still conserved under the protective forest

133 T. Vanwalleghem et al. / Geomorphology 56 (2003) cover and therefore these areas are of unique value. More research is however needed to date the main land-use changes and erosion phases. Acknowledgements This study was supported by the Fund for Scientific Research-Flanders. The authors would like to thank the Forestry Department of Leuven for the permission to conduct this research in the Meerdaal Forest and for their support and advice in collecting data, especially Ir. B. Meuleman, H. Nackaerts and C. Vandenbempt. References Arnould-De Bontridder, O., Paulis, L., Étude du ravinement Holocène en Forêt de Soignes. Acta Geographica Lovaniensia 4, Auzet, A.V., Boiffin, J., Ludwig, D., Concentrated flow erosion in cultivated catchments: influence of soil surface state. Earth Surface Processes and Landforms 20, Baeyens, L., Tavernier, R., Scheys, G., Belgian Soil Map, Sheet 103E (Hamme-Mille). IWONL, Belgium. Beuselinck, L., Steegen, A., Govers, G., Nachtergaele, J., Takken, I., Poesen, J., Characteristics of sediment deposits formed by intense rainfall events in small catchments in the Belgian Loam Belt. Geomorphology 32, Bork, H.-R., Bork, H., Dalchow, C., Faust, B., Piorr, H.-R., Schatz, T., Landschaftsentwicklung in Mitteleuropa. Klett-Pertes, Stuttgart, Germany. Dens, C., Fouilles à Meerdaal. Annales de la Société d Archéologie de Bruxelles, Tome 22, Desmet, P.J.J., Govers, G., Comparison of routing algorithms for digital elevation models and their implications for predicting ephemeral gullies. International Journal of Geographical Information Systems 10, Desmet, P.J.J., Poesen, J., Govers, G., Vandaele, K., Importance of slope gradient and contributing area for optimal prediction of the initiation and trajectory of ephemeral gullies. Catena 37, Gullentops, F., Holocene soil erosion in the loess belt of Belgium. In: Van der Haegen, H., Van Hecke, E. (Eds.), Liber Amicorum, Prof. Dr. M. Goossens. Acta Geographica Lovaniensa 33, Langohr, R., Sanders, J., The Belgium Loess belt in the last 20,000 years: evolution of soils and relief in the Zonien Forest. In: Boardman, J. (Ed.), Soils and Quaternary Landscape Evolution. Wiley, Chichester, UK, pp Ludwig, B., Boiffin, J., Chadoeuf, J., Auzet, A.V., Hydrological structure and erosion damage caused by concentrated flow in cultivated catchments. Catena 25, Martens, E., Uit het verleden van de gemeente Oud-Heverlee. Stroobants, Neerijse, Belgium. Montgomery, D.R., Dietrich, W.E., Landscape dissection and drainage area slope thresholds. In: Kirkby, M.J. (Ed.), Process Models and Theoretical Geomorphology. Wiley, Chichester, UK, pp Moore, I.D., Burch, G.J., Mackenzie, D.H., Topographic effects of the distribution of surface soil water and the location of ephemeral gullies. Transactions of the American Society of Agricultural Engineers 31, Nachtergaele, J., A spatial and temporal analysis of the characteristics, importance and prediction of ephemeral gully erosion. Unpublished PhD thesis. Department of Geography, University of Leuven. Nachtergaele, J., Poesen, J., Assessment of soil losses by ephemeral gully erosion using high-altitude (stereo) aerial photographs. Earth Surface Processes and Landforms 24, Nachtergaele, J., Poesen, J., Steegen, A., Takken, I., Beuselinck, L., Vandekerckhove, L., Govers, G., The value of a physically based model versus an empirical approach in the prediction of ephemeral gully erosion for loess-derived soils. Geomorphology 40, Nachtergaele, J., Poesen, J., Oostwoud Wijdenes, D., Vandekerkhove, L., Medium-term evolution of a gully developed in a loess-derived soil. Geomorphology 46, NGI, Topographical Map of Belgium. Sheet 32/6. National Geographical Institute, Brussels. Papy, F., Douyer, C., Influence des états de surface du territoire agricole sur le déclenchement des inondations catastrophiques. Agronomie 11, Patton, C., Schumm, S.A., Gully erosion, Northwestern Colorado: a threshold phenomenon. Geology 3, Poesen, J., Govers, G., Gully erosion in the Belgian loess belt of Belgium: typology and control measures. In: Boardman, J., Foster, I.D.L., Dearing, J.A. (Eds.), Soil Erosion on Agricultural Land. Wiley, Chichester, pp Poesen, J., Nachtergaele, J., Deckers, J., Gullies in the Tersaart Forest (Huldenberg): climatic or anthropogenic cause? In: Verstraeten, G. (Ed.), Gully erosion processes in the Belgian loess belt: causes and consequences. Excursion guide, International Symposium on Gully Erosion under Global Change, Laboratory for Experimental Geomorphology, K.U. Leuven, Belgium, pp Poesen, J., Vandekerckhove, L., Nachtergaele, J., Oostwoud Wijdenes, D., Verstraeten, G., Van Wesemael, B., Gully Erosion in Dryland Environments. In: Bull, L.J., Kirkby, M.J. (Eds.), Dryland Rivers: Hydrology and Geomorphology of Semi-Arid Channels. Wiley, Chichester, UK, pp Poesen, J., Nachtergaele, J., Verstraeten, G., Valentin, C., Gully erosion and environmental change: importance and research needs. Catena 50, Vandaele, K., Poesen, J., Govers, G., Van Wesemael, B., Geomorphic threshold conditions for ephemeral gully incision. Geomorphology 16, Vincent, M., Etude d une classe de travaux de terre préhistoriques: la forêt de Meerdael et le bois d Heverlee. Bulletin de la Sociétée de Bruxelles 9,

134 Earth Surface Processes and Landforms 574 Earth Surf. Process. Landforms 32, (2007) T. Vanwalleghem et al. Published online 8 September 2006 in Wiley InterScience ( DOI: /esp.1416 Origin and evolution of closed depressions in central Belgium, European loess belt T. Vanwalleghem, 1,2 J. Poesen, 2 * I. Vitse, 2 H. R. Bork, 3 M. Dotterweich, 3 G. Schmidtchen, 3 J. Deckers, 4 A. Lang 5 and B. Mauz 5 1 K. U. Leuven Research Fund, Belgium 2 Physical and Regional Geography Research Group, K. U. Leuven, Belgium 3 Ecology-Centre, Christian Albrechts University Kiel, Germany 4 Institute for Land and Water Management, K. U. Leuven, Belgium 5 Department of Geography, University of Liverpool, UK *Correspondence to: J. Poesen, Physical and Regional Geography Research Group, K. U. Leuven, Redingenstract 16, B-3000 Leuven, Belgium. jean.poesen@geo.kuleuven.be Received 25 October 2005; Revised 2 June 2006; Accepted 28 June 2006 Abstract Closed depressions (CDs) are lower lying areas where the sediment eroded from the surrounding soil surfaces draining towards the CD is trapped in the system. CDs have been reported in several regions of the European loess belt and are attributed either to natural processes (e.g. dissolution of subsurface horizons) or to human intervention (e.g. quarrying). Previous studies focussed mainly on cropland areas where, however, only few and largely filled in CDs remain. The objectives of this study were to i) assess the spatial distribution of CDs under forest and cropland, ii) to determine and compare the morphology of CDs under forest and under cropland, and iii) to determine the origin and age of these CDs under forest. In a study area located partly in ancient forest (13 km 2 ) and partly in cropland (29 km 2 ), a systematic survey revealed the presence of 71 CDs under forest (5 3 CD.km 2 ) and 30 CDs under cropland (1 CD.km 2 ). Comparison of their morphology showed that CDs under forest were significantly deeper, with steeper sidewalls and a smaller surface area because of the erosion and deposition processes acting on the CDs under cropland. By comparing CDs that had been under cropland for different time intervals, the rate of this morphological evolution could be reconstructed. Analysis of the soil stratigraphy of two representative CDs in the ancient forest area confirmed their origin as quarries. Most probably, calcareous loess was excavated since this soil horizon, about two to five meters thick, was completely absent within the CDs. Dating of the infilling of one CD by optically stimulated luminescence (OSL) shows that the CD filled in between the first century BC and the fourth century AD. This dating corresponds to the dating of sediment deposits in nearby, human-induced gullies that were attributed to an agricultural land use phase between the 18th century BC and the third century AD. Copyright 2006 John Wiley & Sons, Ltd. Keywords: closed depression; quarries; ancient forest; loess; land use Introduction The present-day landscape in the European loess belt is the result of a complex interaction between natural and anthropogenically driven processes. Phases of landscape stability and soil formation alternated with phases of land use and soil erosion. The present-day topography and soilscape can therefore be understood only from the past (Lang and Bork, 2006). This study focuses on the poorly understood microtopographic features of closed depressions (CDs; Figure 1). A closed depression is a lower lying area where the sediment eroded from the surrounding soil surfaces draining towards the CD is trapped in the system (Norton, 1986). These circular to elliptic landforms have a surface area between 0 01 and 10 hectares and are between 0 2 and 10 m deep. In several regions in northwestern Europe, CDs are recognized as important microtopographic features. In the framework of an archaeological inventory, Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

135 Origin of closed depressions 575 Figure 1. (a) A closed depression under ancient forest (Meerdaal Forest) with a surface area (A) of 0 22 ha and sidewall slopes (S) of 0 27 m m 1. Note person for scale. The border of the closed depression is indicated with a dashed line. (b) A closed depression under cropland near the border of the Meerdaal Forest, recently deforested (between 96 and 145 years ago; A = 0 33 ha; S = m m 1 ). Baumewerd-Schmidt and Gerlach (2001) and Gerlach (2001) found many CDs in the loess landscape of western Germany. Pissart (1958) reported almost 1000 CDs in the Paris region of northern France. Prince (1961) found an impressive number of CDs around Norfolk, UK. Also in the Belgian loess belt, CDs were reported in several studies (Meeuwis, 1948; Manil and Pecrot, 1950; Gullentops, 1952; Bollinne, 1977; Bollinne et al., 1980; Dudal, 1955). Although their presence has been reported regularly, little systematic research on their characteristics and origin exists. This can probably explain why a wide range of hypotheses exists on their origin. Possible causes that were put forward include both natural (e.g. dead ice pits or morainic kettle holes) and anthropogenic (e.g. excavations of calcareous loess, clayey or sandy material) processes. In a 43-km 2 study area in the Belgian loess belt, Gillijns et al. (2005) recently made a study of the characteristics and origin of CDs under cropland. They concluded that most CDs were quarries of calcareous loess. However, given that most CDs were located under cropland, this study gives a Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

136 576 T. Vanwalleghem et al. limited view. As already indicated by Gillijns et al. (2005), many CDs under cropland have been filled in due to centuries of water and tillage erosion. The morphology of the CDs that are still visible in the present-day landscape may have been altered significantly. However, the extent of this degradation is not known. Because of this, it remains difficult to attribute a possible origin of these features based on the spatial distribution of the CDs or to use their current morphological characteristics, such as depth, as an indication for their origin. In order to better understand the origin of the CDs, this study analyses CDs in an area with minimum disturbance. An ancient forest is an ideal archive area for this study since the landscape is preserved since the time of the last deforestation. Ancient forests are generally defined as forests that have existed continuously since at least a specified date (Hermy et al., 1999), which for Belgium is 1775 (the publication date of the Ferraris map). However, although ancient forests were long believed to be undisturbed, a number of recent studies have clearly illustrated that this is not necessarily the case (Verheyen et al., 1999; Dupouey et al., 2002; Vanwalleghem et al., 2003, 2005b). Based on the study of plant composition, soil properties or geomorphologic features, it was shown that several ancient forests were characterized by an agricultural land use phase before the first maps were drawn. The objectives of this study are therefore: (i) to determine the spatial distribution of closed depressions (CDs) within an ancient forest area and under cropland; (ii) to determine and compare the morphological characteristics of CDs under forest and under cropland; and (iii) to determine the origin and age of a representative CD under forest. Study Area The study area selected for this research is located about 11 km south of Leuven and coincides with a large part of the catchment of the river Néthen (Figure 2). The total area is 42 km 2, of which the northern part (13 km 2 ) is forested. The Meerdaal Forest is one of the largest ancient forests in central Belgium. It was traditionally considered as a primary forest. However, recent studies (Vanwalleghem et al., 2003, 2005b) of permanent gullies found in the forest showed evidence of an intensive land use phase between the 18th century BC and the third century AD (Middle Bronze Age to Roman period). The study area is part of the Belgian loess belt (Figure 2a). This west to east orientated region is characterized by Quaternary (1 8 million years ago) loess deposits that overlie a Tertiary sand substratum of Middle to Late Eocene Age ( million years ago). It is bordered to the north by lowland plains consisting of Tertiary sands and clays, which are covered by late Quaternary aeolian sands. The northern border of the study area is located in the transition zone between the loess belt and these northern plains. This implies that the loess layer in the northern part of the study area is thin to absent. Also, within the loess belt itself, strong variations in the thickness of the loess cover occur. The loess cover is absent on some sandy hilltops, but can reach a depth of up to several metres in other places (Tavernier, 1954). Moreover, Goossens (1988) showed that the loess layer is thicker on north to east oriented slopes compared with the south to west oriented slopes due to the prevailing wind direction at the time of deposition. The loess deposits were originally calcareous but the top m (Goossens, 1988) are decalcified. In places with an undisturbed loess cover, Albeluvisols (ISSS, 1998) formed in this decalcified loess layer. Cultivated areas are usually characterized by Luvisols. On the sandy hilltops, Podzols or Arenosols can locally be found. Materials and Methods The CDs were located through a systematic field survey of the entire study area since existing topographic maps do not show these microtopographic features. For every CD, the deepest point and the corresponding catchment area were measured in the field using GPS (Trimble Pro-XL, horizontal accuracy 1 m under cropland). Next, the morphological characteristics of the CDs were measured in the field. For the CDs under cropland, both surface area (A) and diameter (D) were derived from the GPS measurements. Taking into account the difficulty of accurately delineating the border of the closed depression under cropland, the estimated uncertainty on D is c. 1 2 m. For the CDs under forest, A and D were measured directly in the field with a measuring tape since the accuracy of the GPS measurements was insufficient under forest. Under forest, this implies that the error in D is much less than under cropland (c. 0 1 m). The mean slope of the depression sidewalls (S) was determined with a clinometer and the maximum depth (d max ) of all CDs was then derived from S and D. In order to investigate the origin of the CDs, a detailed soil stratigraphic analysis was performed on two representative CDs. Both CDs were selected on the basis of their morphology and location in the central part of the forest, to avoid possible disturbances due to forest clearance and agriculture in historical times. This soil stratigraphical analysis was done at two different levels of detail. Firstly, the soil stratigraphy of one CD (CD1) was investigated by means of soil Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

137 Origin of closed depressions 577 Figure 2. (a) Location of the study area within the loess belt. The extent of the loess-derived soils is based on Zagwijn and Paepe (1968). (b) Location of all the closed depressions under forest (Meerdaal Forest) and under cropland (southern part of study area) within the study area. The closed depressions selected for stratigraphical analysis are indicated by black squares. augerings (Figure 3). In these augerings, the main soil horizons could be distinguished but a detailed differentiation of the colluvial deposits was not possible. Augerings were stopped at the level of the Tertiary sand substratum. In total, 17 augerings up to 7 35 m deep were made on a north to south transect (transect 1) and an east to west orientated transect (transect 2). Detailed topography along each transect was measured with a theodolite. Secondly, the infilling of a second CD (CD2) was analysed in more detail by digging large trenches. In order to limit the size of the excavation, a smaller CD was selected for this detailed analysis. Two perpendicular trenches (see Figure 3; transect 3 = 10 m 1 2 m 3m and transect 4 = 7m 1 2 m 2 m) were excavated from the centre of the CD towards the border of the CD. During the field survey, a preliminary stratigraphy with relative ages was made and detailed drawings of the vertical walls of the trench were made. Sediment units were characterized in the field (based on colour, field texture and structure) and these observations were corroborated by additional laboratory analysis (texture, dry bulk density). Texture was analysed after decalcification. Four soil samples were taken from this CD and Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

138 578 T. Vanwalleghem et al. Figure 3. Indication of position of the transects for soil stratigraphical analysis of two typical closed depressions: (a) closed depression 1 (CD1) analysed by soil augerings; (b) closed depression 2 (CD2) analysed in detail with trenches. were analysed by optically stimulated luminescence (OSL) by the Dating Laboratory of the University of Liverpool, in order to date the infilling of the CD. The OSL preparation procedures are reported in detail by Lang et al. (1996). In this study the sedimentary structures observed in the infill of the east-facing wall (= headwall) and the south-facing wall (= sidewall) of transect 3 and of the north-facing (= headwall) and the east-facing wall (sidewall) of transect 4 are presented. Spatial distribution and morphology Results and Discussion The location of the CDs found in the study area is shown in Figure 2b. Systematic field survey revealed the presence of 101 CDs in an area of 42 km 2. From Figure 2b, it can be seen that the CDs are not evenly distributed over the study area. Most CDs (n = 71) are located under forest, while only 30 CDs were found under cropland. The density of the CDs mapped under forest (5 3 km 2 ) is about five times larger than that of the CDs under cropland (1 km 2 ). The latter is comparable with the results of other studies on CDs in the region (Pissart, 1958; Bollinne et al., 1980; Gillijns et al., 2005), which focused on cropland and reported densities ranging between 0 6 and 1 CD km 2. Also, within Meerdaal Forest the CDs are not randomly distributed: the majority of the CDs is found in the southern part. Of the CDs under cropland, nine are located within 150 m of the southern forest border. Historical maps (e.g. Joris and Cardon, 1769, Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

139 Origin of closed depressions 579 Table I. Morphological characteristics of the closed depressions under forest and under cropland in the study area All (n = 101) Forest (n = 71) Cropland (n = 30) Mean SD Mean SD Mean SD P-value D (m) <0 001 A (m 2 ) <0 001 d max (m) <0 001 S (m m 1 ) <0 001 D, diameter; A, surface area; d max, maximum depth; S, average sidewall slope; SD, standard deviation. unpublished) confirm their link with the CDs under forest as this border zone was part of the Meerdaal Forest less than three centuries ago. A second concentration of 13 CDs can be found at the southwestern border of the cropland area. The most important morphological characteristics of the CDs under forest and under cropland are summarized in Table I. The average diameter (D) and average surface area (A) of the CDs under forest are much smaller than those of the CDs under cropland. The situation for the maximum depth (d max ), on the other hand, is inversed since the average d max for forest is nearly twice as large as the average d max for cropland. Because the CDs are larger and less deep under cropland, the average sidewall slope (S) of these CDs was less steep compared to those under forest. Most of the CDs under cropland (73 per cent) have S < m m 1. All differences between forest and cropland are statistically significant with a Student s t-test. Although literature on the morphology of CDs is rare and is based mainly on observations of CDs under cropland, the few existing studies report dimensions consistent with those of the CDs under cropland in this study (Table I). The diameter of the depressions observed by Bollinne et al. (1980) in the Belgian loess belt was mostly smaller than 100 m. Pissart (1958) and Prince (1961) both found large differences in the diameter of CDs in the Paris basin and Norfolk region, with values in the range m and m, respectively. According to Bollinne et al. (1980), the CDs in the region of Gembloux have d max of 1 3 m. Pissart (1958) observed that the CDs in the Paris region had a d max of m and those in Norfolk (Prince, 1961) had a d max of 2 30 m. The higher density found for CDs under forest can be attributed either to the fact that originally there were more depressions formed or excavated, or else that more CDs have been conserved. Gillijns et al. (2005) hypothesized that the CDs under forest represent the original situation and that the morphology of the CDs under cropland is the result of centuries of water and tillage erosion processes acting on the soil surfaces draining towards the CD and of sediment deposition within the CD. The marked differences found between the morphology of the CDs under forest and under cropland, collected in this study, clearly supports this hypothesis. The rate of the morphological change of the CDs was reconstructed in Figure 4 by comparing CDs in areas with a different land use history. The period of cultivation of these areas is derived from different historical maps and therefore refers only to the period since the first maps (Joris and Cardon, 1769, unpublished). Erosion of the CD sidewalls resulted in a lower S and a larger A and D, while the eroded soil was deposited within the CD and decreased the d max. D and A increase steadily with increasing time of cultivation while S decreases gradually. On the other hand, d max remains constant for a long time ( years) before dropping drastically. This is possibly due to the trapezoidal cross-section of the CDs. Most of the sediment eroded from the sidewalls will be deposited directly at the foot of the sidewall. The centre of the depression, where the deepest point was measured, probably remains relatively unaffected for some time. This might explain the time lag between the start of cultivation of the CD and the moment when d max starts decreasing. This morphological evolution is further illustrated in Figure 1a and b showing a CD under forest and a CD that has been under cultivation for about a century. Stratigraphy of a closed depression Figure 5 shows the stratigraphy of the first closed depression (CD1), analysed by soil augerings. In transect 1, perpendicular to the slope of the present soil surface, the loess thickness outside the CD is more or less constant and varies around 6 m (Figure 5a). Transect 2 (Figure 5b), parallel to the present slope direction, shows a large variation in loess thickness. To the west, the loess layer is almost absent while it increases sharply to 7 25 m at the eastern side of transect 2. Transects 1 and 2 clearly illustrate that the shape of the CD is not present in the Tertiary topography. Since irregular loess deposition over such a short distance is highly unlikely, the only possible explanation for this is that the loess inside the depression has been excavated. Moreover, none of the augerings made inside the CD Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

140 580 T. Vanwalleghem et al. Figure 4. Evolution of the morphological characteristics (diameter, area, maximum depth and slope of the soil surface) of closed depressions according to the historical period under cultivation. Error bars encompass 2σ; n represents the number of CDs that was under cultivation for each given time frame. indicates the presence of a Luvisol. Instead, the CD is partly filled in with colluvial deposits between 1 10 and 1 60 m thick. Figure 6 shows the detailed stratigraphical analysis of the second closed depression (CD2). As in CD1, a large part of the CD is filled in with colluvial sediments up to 2 06 m thick in transect 3 and 1 85 m in transect 4. This implies that the original CD was excavated more than 3 m deep. In the centre of the CD (transects 3 and 4), the base level consists of in-situ Tertiary sands (unit a). Inside the Tertiary sand, some soil formation processes took place before the Pleistocene loess deposit. In transect 3, reworked material (unit b) can be found at the contact between the Quaternary loess and the Tertiary sands. In one such package, clear clay illuviation bands were present. On top of this sand deposit, in-situ calcareous loess deposits (unit c) can be found in transect 3, towards the southwestern border. Here, the transition between the depression infill and the in-situ soil consists of decalcified loess with clay illuviation inside (unit d). This is probably the bottom part of the original Bt, which is normally about m below the soil surface (Goossens, 1988). In transect 4 on the other hand, the Tertiary sand (unit a) is covered directly by colluvial deposits. In a large part of the depression infill of transect 3, soil formation and bioturbation have erased all layering or structures (unit e). It is, however, clear that this layering must have existed before because some isolated, clearly definable packages with fine layering can be identified in transect 3 within this homogeneous sediment unit e. Also a piece of pottery (V) was found in sediment unit e in transect 3 at a depth of about 1 4 m below the soil surface. The pottery fragment (1 cm 1 cm 0 5 cm) was too small to be dated by archaeologists. The same homogeneous sediment unit e also occurs in transect 4, although it is less important there. Next, the infill in both transects consists for a large part of very fine-layered sediments (unit f). These layered sediments are built up of several sediment sequences, consisting of alternating light (silt; several millimetres) and dark (clay or loam; <1 mm) coloured layers. A sediment sequence is defined as a continuous deposit of coarse to fine sediments, most probably deposited by sheet flow. Other sediment units found in the infill of the CD are much more homogeneous (unit g) and their shape and characteristics suggest that these sediments slid in through soil flow. In transect 3 they are clearly intercalated with the layered deposits of sediment unit f. In these fine-layered sediment units, several structures were identified that suggest that they were disturbed by animals or humans. Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

141 Origin of closed depressions 581 Figure 5. Generalized stratigraphy of the first closed depression (CD1) along (a) the north south axis (transect 1) and (b) the east west axis (transect 2). The position of the soil augerings is indicated by a dotted line. In sediment units f and g, several vertical cracks could be observed in both transects, although they are especially pronounced in transect 4. These are desiccation cracks, which indicate the surface level at several stages (several dry periods) during the infilling. In transect 4, there is another characteristic feature, namely a dark band of iron-manganese (Fe-Mn) mottles over more than half of the transect. This Fe-Mn band has the form of a pool with a diameter of >5m and indicates water ponding. Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

142 582 T. Vanwalleghem et al. Figure 6. Generalized stratigraphy of the second closed depression (CD2) along (a) the northeast to southwest axis (transect 3) and (b) southeast to northwest axis (transect 4). While for the sediment units deposited by soil flow, like unit g, it is clear that they originate from the sidewalls of the CD itself, the origin of the fine-layered sediment units is not clear. Small variations in texture of the different sediment units suggest that it is a mixture of eroded sidewall material and material eroded from the soils in the catchment of the CD, that range from Podzols to Luvisols. Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

143 Origin of closed depressions 583 The upper sediment units, h and i, show a lower bulk density (respectively 0 56 and 0 53 g cm 3 ) compared to the underlying sediment unit g (0 69 g cm 3 ). This is probably the result of more recent bioturbation by roots or burrowing animals. Sediment unit j is the humic surface horizon (Ah) of a typical forest soil. Finally, the timing of infilling could be dated with OSL. The results are shown in Table II. The deepest of the four samples taken (OSL I) was not bleached sufficiently at deposition, so that no stable equivalent-dose plateau could be reached. This means that the quoted OSL age for this sample is a maximum estimate only. The remaining three samples (OSL II IV) all lie within the Roman period, between AD 90 and 350. Since the standard deviations of all these samples overlap, they were all deposited during the same period. The stratigraphy of both CDs suggests that calcareous loess was excavated, as the bottom in both cases coincides with the lower border of the loess deposits. Additional evidence for the origin of the CDs as loess quarries can be found in their spatial distribution. The soils in the northern part of the study area are too sandy. Only in the southern part of the Meerdaal Forest do the loess deposits become sufficiently thick, and calcareous loess can be found in the subsoil. The spatial distribution of the CDs under forest clearly reflects this soil pattern, as can be observed in Figure 2b. Further support for their anthropogenic origin is that reported archaeological sites in Meerdaal Forest (Neolithic to Roman period) are also mostly located in the southern part of the forest (Martens, 1981). A majority of the permanent gullies, which Vanwalleghem et al. (2003, 2005b) attributed to past agricultural activity, are also located in the southern part of the forest. There are thus several indications that the early farmers preferred the loessderived soils for establishing their cropland. These detailed stratigraphical analysis confirms the results of Gillijns et al. (2005) in a comparable study area under cropland. Whether or not all the soil material was extracted from the CD remains speculative. No direct evidence in the infill was found, indicating that soil material was thrown in again manually. Also, nowhere around the CD could levees or dumps of material be observed. The clay illuviation horizon (Bt) from the Albeluvisol, formed above the calcareous loess, was useful for brick-making. The calcareous loess was probably used as fertilizer. Although loessderived soils are traditionally considered as fertile soils, Langohr (1990) stated that the original soils, which the first farmers encountered, had in fact a very low chemical and physical fertility and were not at all suitable for cereal cropping. This is also the case for the present-day undisturbed soils under forest, Albeluvisols. These soils have a high intrinsic soil fertility (high CEC), but uncultivated, they have a very low ph (ph(h 2 O) <4) and base saturation (<50 per cent) and are poor in available nitrogen (high C:N ratio, i.e. >10) (IUSS Working Group WRB, 2006). In order to obtain acceptable crop yields, application of mineral or organic fertilizer to the soil in natural condition is indispensable. Until the 19th century, calcareous loess was the only readily available fertilizer to increase or maintain the soil ph in many rural regions (Lindemans, 1994). The extracted calcareous loess was most probably used on nearby cropland, since the nature of the infilling of the CD indicates that cropland was present in the immediate vicinity of CD2. In total, 270 sediment sequences were counted in all consecutive fine-layered sediment units. Although only a rough approximation, this number gives an indication of the number of runoff events that deposited the layered sediments in the CD. Such a large number of runoff events could only have occurred on cleared land. Observations of sediment filling in a recent and a historical gully in a nearby area under cropland (Vanwalleghem et al., 2005a) show that such deposits can form in a period of several years to a few decades. More evidence for this fast infilling can be found in the three sediment samples taken from the infill of CD2, where no significant differences exist between the OSL ages. All were dated to the Roman period and range between AD 90 and 350. Another argument in favour of fast infilling is that the pedogenesis is not interrupted by a period of stability and forest regrowth. Nowhere in the soil profile was a humic horizon observed. Since the CD is a place of deposition, and not of erosion, traces of such a horizon would still be present if it formed. Importance While it is clear now that these CDs are not natural landforms, their importance in terms of soil displacement is still not known. Such a quantitative assessment can be made by combining the morphological measurements with the data about the infilling of the CDs. If the volume of a CD is approximated as a truncated cone, the actual total volume of all the CDs in the Meerdaal Forest amounts to m 3. If it is assumed that the colluvium inside the CDs originates entirely from the sidewalls of the CDs, this volume equals the volume that was originally excavated. On the other hand, if the colluvium inside the CD originated entirely from the catchment, the excavated volume would be considerably higher. Adding an average colluvium depth of 1 62 ± 0 42 m to the actual depth, the total volume excavated would then be approximately m 3. Since both sediment sources probably contributed to the colluvial infilling of the CDs, the volume originally excavated will be somewhere between these two figures. The area in which the CDs are concentrated in Meerdaal Forest was manually delineated and is about 500 ha in size (59 of the 71 CDs occur in this area). However, probably only the potentially most fertile soils, i.e. Luvisols Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

144 584 T. Vanwalleghem et al. Table II. Characteristics and OSL age of samples I to IV from transect 3 (CD2). The moisture factor for all samples was taken as 1 15 ± 0 05 Sample Depth below Sediment Grain size code surface (m) unit (µm) U (µg g 1 ) Th (µg g 1 ) K (%) a-value* dd/dt cosm (Gy ka 1 ) dd/dteffective (Gy ka 1 ) De (Gy) Age (ka) OSL I 1 96 f ± ± ± ± ± ± 0 28 <36 8 ± 2 4 <7 05 ± 0 59 OSL II 1 20 g ± ± ± ± ± ± ± 0 11 OSL III 0 98 g ± ± ± ± ± ± ± ± 0 17 OSL IV 0 47 i ± ± ± ± ± ± ± ± 0 17 * α-efficiency factor. Cosmic dose rate. Effective dose rate. Equivalent dose. Maximum age only. Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

145 Origin of closed depressions 585 or Albeluvisols, were used for cropland. Taking into account only these two soil types limits the limed area to only 343 ha. Consequently, for an area between 343 and 500 ha, the calculated soil displacement volume would be between 174 and 317 m 3 ha 1. The volume of calcareous loess excavated is about 40 per cent of this value or between 70 and 127 m 3 ha 1, which is equivalent to a total surface raising of mm. Although this figure is difficult to compare with erosion rates, because the exact time period over which this surface raising occurred is not known, it is clear that this value is not negligible for this region. Conclusions A systematic analysis of 101 closed depressions (CDs) showed important differences in spatial density and morphology between the CDs under forest (n = 71) and those under cropland (n = 30). Comparison of the densities in the cropland (1 km 2 ) and forest (5 3 km 2 ) areas suggests that about 80 per cent of the CDs under cropland were filled in up to the point that they are no longer visible in the present landscape. The morphological characteristics further show that the CDs under forest are very close to the original situation while the CDs under cropland are severely degraded due to centuries of water and tillage erosion or anthropogenic soil displacement (e.g. manual filling of CDs). The rate of this morphological change could also be reconstructed. The analysis of the soil stratigraphy of two CDs under forest shows that the CDs in Meerdaal Forest are of anthropogenic origin. The main argument for this is that the shape of the CDs is not present in the Tertiary sand and that the original loess layer is absent within the investigated CDs. The main objective for their excavation was therefore probably the extraction of calcareous loess. This study therefore contributed to the knowledge of Bronze Age to Roman farming techniques in the loess belt. The use of calcareous loess as fertilizer could have been an important tool in the cultivation of the acid forest soils that these early farmers were confronted with. The dating of the colluvial infill of one CD shows that it filled in between AD 90 and 350. Although dating of more CDs would be required, this suggests that the closed depressions are related to archaeological traces found in Meerdaal Forest, which include mainly Bronze ( BC), Iron ( BC) and Roman age (57 BC to AD 402) tumuli and Roman villas. It also confirms that the CDs are related to the permanent gullies that were dated in the same area (Vanwalleghem et al., 2006). These gullies probably filled in between the 18th century BC and third century AD. Nevertheless, the number of dated gullies (n = 2) and closed depressions (n = 1) is very limited. With the available data it can only be cautiously concluded that all these phenomena developed during the same land use phase that started around the Middle Bronze Age ( BC) and lasted until the end of Roman times (57 BC to AD 402). This study demonstrates the importance of ancient forests for the conservation of microtopographic features resulting from past land use changes. It was shown that CDs are an important phenomenon in the Belgian loess belt, not only in terms of frequency of occurrence, but also in absolute soil volumes displaced. Over a period of a few decades to several hundreds of years, between 174 and 317 m 3 ha 1 of soil was displaced by this human activity. Compared with the water erosion rates reported by Verstraeten et al. (2006) for small catchments in Belgium, which range between 1 7 and 8 2 m 3 ha 1 a 1, it can be seen that the excavation of such CDs is an anthropogenic process of significant importance, not only in this study area but possibly also elsewhere in the European loess belt. Anthropogenic CDs have been reported in regions like southwestern Germany (Baumewerd-Schmidt and Gerlech, 2001) and the UK (Prince, 1961), although no further information on their age or morphology is available. Detailed studies are therefore needed in order to determine whether the conclusions drawn from this study can be extrapolated to other regions. References Baumewerd-Schmidt H, Gerlach R Von Restfundstellen und Scheinfundstellen-Ergebnisse einer Grabenbetreuung in der Lösslandschaft. In Reimer H, Kindermann K, Lange M (eds), Archäologische informationen 24(1): (in German). Bollinne A La vitesse de l érosion sous culture en région limoneuse. Pédologie 27(2): (in French). Bollinne A, Pissart A, Bastin B, Juvigne E Etude d une dépression fermée près de Gembloux; vitesse de l érosion des terres cultivées de Hesbaye. Annales de la Société Géologique de Belgique 103: (in French). Dudal R Bijdrage tot de kennis van gronden op loess-leem in Midden-België. Unpubl. Ph.D. thesis: Faculty of Applied Biological Sciences; K.U. Leuven, Belgium; 244. Dupouey JL, Dambrine E, Laffite JD, Moares C Irreversible impact of past land use on forest soils and biodiversity. Ecology 83(11): Gerlach R Keinesfalls Ausnahmen: Materialentnahmegruben als Befundzerstörer. In Reimer H, Kindermann K, Lange M (eds), Archäologische informationen 24(1): (in German). Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

146 586 T. Vanwalleghem et al. Gillijns K, Poesen J, Deckers J On the characteristics and origin of closed depressions in loess-derived soils in Europe a case study from central Belgium. Catena 60: Goossens D Scale model simulations on the deposition of loess in hilly terrain. Earth Surface Processes and Landforms 13: Gullentops F Phénomènes subkarstiques près de Leefdael Brabant. Bulletin de la Société Belge de Géologie, de Paléontologie et d Hydrologie 61: (in French). Hermy M, Honnay O, Firbank L, Grashof C, Lawesson J Ecological comparison between ancient forest plant species of Europe and the implications for forest conservation. Biological Conservation 91: ISSS Working Group RB World Reference Base for Soil Resources: Introduction. Acco: Leuven. IUSS Working Group WRB World Reference Base for Soil Resources World Soil Resources Reports No FAO: Rome; 128. Lang A, Bork HR Past soil erosion in Europe. In Soil Erosion in Europe, Boardman J, Poesen J (eds). John Wiley & Sons: Chichester (in press). Lang A, Lindauer S, Kuhn R, Wagner GA Procedures used for optically and infrared stimulated luminescence dating of sediments in Heidelberg. Ancient TL 14: Langohr R The dominant soil types of the Belgian loess belt in the early Neolithic. In Cahen D, Otte M (eds) Rubané et Cardial, E.R.A.U.L. 39: Lindemans P Geschiedenis van de landbouw in België. Genootschap voor geschiedenis en volkskunde: Antwerpen (in Dutch). Manil G, Pecrot A La cartographie pédologique de la région de Gembloux. In IWONL (eds) Comptes rendus de recherches travaux du Comité pour l établissement de la carte des sols et de la végétation de la Belgique. IWONL: Brussels; (in French). Martens E Uit het verleden van de gemeente Oud-Heverlee. Stroobants: Neerijse (in Dutch). Meeuwis A La représentation cartographique des dépressions sans écoulement. Bulletin de la Société Royale Belge de Géographie 72: (in French). Norton LD Erosion sedimentation in a closed drainage basin in Northwest Indiana. Soil Science Society of American Journal 50: Pissart A Les dépressions fermées de la région Parisienne, le problème de leur origine. Revue de Géomorphologie Dynamique 9: (in French). Prince HC Some reflections on the origin of hollows in Norfolk compared with those in the Paris region. Revue de Géomorphologie Dynamique 12: Tavernier R Le Quaternaire. In Prodrôme d une description géologique de la Belgique, Fourmarier P (ed.). Société géologique de Belgique Place; Liège; Vanwalleghem T, Van Den Eeckhaut M, Poesen J, Deckers J, Nachtergaele J, Van Oost K, Slenters C Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium). Geomorphology 56(1 2): Vanwalleghem T, Bork HR, Poesen J, Schmidtchen G, Dotterweich M, Nachtergaele J, Bork H, Deckers J, Brüsch B, Bungeneers J, De Bie M. 2005a. Rapid development and infilling of a historical gully under cropland, central Belgium. Catena 63: Vanwalleghem T, Poesen J, Van Den Eeckhaut M, Nachtergaele J, Deckers J. 2005b. Reconstructing rainfall and land use conditions leading to the development of old gullies The Holocene 15(3): Vanwalleghem T, Bork HR, Poesen J, Dotterweich M, Schmidtchen G, Deckers J, Scheers, S, Martens M Prehistoric and Roman gullying in the European loess belt, case-study Central Belgium. The Holocene 16(3): Verheyen K, Bossuyt B, Hermy M, Tack G The land use history ( ) of a mixed hardwood forest in western Belgium and its relationship with chemical soil characteristics. Journal of Biogeography 26: Verstraeten G, Poesen J, Govers G, Gillijns K, Bielders C, Goossens D, Ruysschaert G, Van Den Eeckhaut M, Vanwalleghem T Soil erosion in Belgium. In Soil Erosion in Europe, Boardman J, Poesen J (eds). John Wiley & Sons: Chichester (in press). Zagwijn W, Paepe R Die stratigraphie der weichselzeitlichen ablagerungen der Niederlande und Belgiens. Eiszeitalter und Gegenwart 19: Copyright 2006 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 32, (2007) DOI: /esp

147 The Holocene 16,3 (2006) pp. 393/401 Prehistoric and Roman gullying in the European loess belt: a case study from central Belgium T. Vanwalleghem, 1,2 H.R. Bork, 3 J. Poesen, 2* M. Dotterweich, 3 G. Schmidtchen, 3 J. Deckers, 4 S. Scheers 5 and M. Martens 6 ( 1 K.U. Leuven Research Fund, Leuven, Belgium; 2 Physical and Regional Geography Research Group, K.U. Leuven, Celestijnenlaan 200E, B-3001 Heverlee, Belgium; 3 Ecology-Centre, Christian Albrechts University Kiel, Germany; 4 Institute for Land and Water Management, K.U. Leuven, Belgium; 5 Department of Archaeology, Fine Arts and Musicology, K.U. Leuven, Belgium; 6 Vlaams Instituut voor het Onroerend Erfgoed (VIOE), Belgium) Received 14 March 2005; revised manuscript accepted 19 October 2005 Abstract: In contrast with the understanding of present-day soil erosion processes, knowledge on past soil erosion phenomena is still rather limited. Although some studies report on severe gully erosion phases during the fourteenth and eighteenth centuries, almost no evidence is available that documents earlier gully erosion phases. This study investigates the development and age of two old, permanent gullies that are conserved in the ancient Meerdaal forest in central Belgium. The development history of both gullies is very similar. In the first gully, archaeological evidence was found indicating an erosion phase during Roman times, followed by a partial infilling of the gully. In the second gully, radiocarbon dating provided evidence of the same Roman activity phase (cal. yr 46 BC/AD 78), but also of an earlier incision phase during the Middle Bronze Age (cal. yr 1743/1602, 1568/1533 BC). Also here, the erosion phase was followed by a partial infilling. This limited infilling indicates that the catchment of the gullies was reforested quite rapidly, hereby cutting off all runoff and sediment production. This has led to a unique situation in the Meerdaal forest, with the conservation of about 43 similar, large gullies in an area of about 17 km 2. This area has a high geovalue, as the studied gullies are among the oldest and best conserved gullies in northwestern Europe. Key words: Gully erosion, ancient forest, land use, landscape evolution, Prehistoric, Roman, Belgian loess belt. Introduction In recent decades, public awareness and concern about the irreversible effect of certain human-induced environmental changes on the ecosystem has grown rapidly. Soil degradation is one of the most important irreversible changes to the nonbiotic part of the ecosystem. Soil erosion is generally recognized as a key soil degradation process. Various studies have already demonstrated that gully erosion is not only the most visible and obvious soil erosion process, but that it is also very important in a wide range of environments (Poesen et al., 2003). Not surprisingly, in many European countries, detailed studies investigating different aspects of gully erosion were set up over the last decades (Poesen et al., 2003). *Author for correspondence ( jean.poesen@geo.kuleuven.be) In contrast with this extensive assessment and relatively good understanding of actual gully erosion phenomena, documentation of earlier gully erosion phases is very limited. Figure 1 and Table 1 give an overview of reported gully erosion phases in central and northwestern Europe. Time steps of 100 years were chosen for Figure 1, given the often coarse dating that is reported for historical gully erosion phases. Although Figure 1 is only indicative, since many studies that were published in local journals could not be accessed, it illustrates clearly that most gullies studied formed during the last century. Regarding older gully erosion phases, several authors seem to agree that two phases with increased erosion activity occurred during the historical period: a first phase during the fourteenth century and a second phase during the eighteenth century (Table 1). However, evidence for gullies forming before the fourteenth century is quite rare. Only a few cases of such gullies have been documented in the literature. In Germany, # 2006 Edward Arnold (Publishers) Ltd / hl935rp

148 394 The Holocene 16 (2006) The objectives of this study are therefore to reconstruct the erosion and infilling phases of old, permanent gullies in an ancient forest and to provide a time frame for these phases. Materials and methods Figure 1 Reported gully erosion phases in northwest and central Europe. Each case study can report on one or more gullies. Details are given in Table 1 Schmidtchen and Bork (2003) dated charcoal at the bottom of an infilled gully in Biesdorfer Kehlen in the Final Neolithic and Dotterweich (2005) found evidence of gully initiation between the eighth and tenth centuries AD. Also in Germany, Semmel (1995) described a Neolithic fossilized gully, however without chronometric dating. Schmitt et al. (2003b) described a small Bronze Age gully in Poland. Harvey (1996) investigated a fossilized gully system in the Howgill Fells, NW England, which / on the basis of several radiocarbon dates / he situated in the tenth century AD. Larue (2005) reported ravine-like incisions in the dry valleys of the Pays de Thelle (Paris Basin, France), most probably formed around 500 BC. Whether or not this implies that farmers were less confronted with the problem of gully erosion before historical times is far from certain. Every study on past soil erosion phenomena is limited by the absence of adequate written archives (Bell, 1992). Additionally, the soil archive is less well-conserved for these older erosion phases because the probability is higher that it has been destroyed by subsequent erosion phases. In this respect, the ancient Meerdaal forest offers a unique setting to study older gully erosion phases. In this 17 km 2, ancient forest area in the central Belgian loess belt (Figure 2a), Gullentops (1992) first reported the presence of fossilized badlands. A recent, systematic survey by Vanwalleghem et al. (2003) described the presence of 43 large gullies in this forest. From historical documents it is clear that this forest has been continuously forested since the fourteenth century, yet Vanwalleghem et al. (2003, 2005a) found the spatial distribution and morphology of these gullies to be indicative for small, local clearings of the forest cover. Such gullies forming under cropland are, however, characterized by a rapid cycle of incision and infilling and, if their catchment remains under cropland, they are usually completely filled in (Vanwalleghem et al., 2005b). This renders the situation in Meerdaal forest even more unique since reforestation of the gully catchment area stabilized the old gully channels. It is, however, not known to what extent these gullies filled in and conserved their original shape. In the Meerdaal forest, there is thus a unique geomorphic archive of old gullies, which are most probably human-induced. To date, however, no dating of the gully activity phases has been done. Given that the forest has existed already for more than 700 years, such a dating could possibly rank these gullies among the oldest gullies known in Europe so far. Moreover, dating of the associated gully sediment bodies would confirm or deny the arguments for their anthropogenic origin. The Meerdaal forest is located in the Belgian loess belt (Figure 2a). For this study, two gullies, representative in terms of dimension and location, were selected from the 43 permanent gullies mapped in the Meerdaal forest (Figure 2b, Vanwalleghem et al., 2003). Detailed information on this study area can be found in Vanwalleghem et al. (2003, 2005a). The selected gullies are located in the central part of the forest as some areas at the forest edges were cultivated during historical times, which could possibly disturb the old gully channels and associated sediments (Bossuyt, 2001). The two gullies selected were also chosen because they have a clearly visible, undisturbed colluvial fan. This increases the odds of finding datable material because, for the dating of the gully activity phases, it is only possible to date the deposition of the associated sediments. The position of the two selected gullies (gully G14 and G19) is indicated in Figure 2b. In each gully, a first trench was dug at the transition zone between the downslope end of the present-day gully channel and the beginning of the colluvial fan. A sketch of the first gully (G14) and its colluvial fan is shown in Figure 2c. In addition to the first trench (transect G14-TR1), a second, smaller pit was excavated more upslope, in the gully thalweg (transect G14-TR2). Transect G14-TR1 is about 10 m long, 1.2 m wide and 3 m deep. The second transect, G14-TR2 is 1.25 m long, 0.75 m wide and 1.80 m deep. Figure 2d shows a sketch of the second gully (G19), its colluvial fan and the location of the third excavation, G19-TR1. Here, no second trench upslope was dug since soil augerings indicated no important sediment deposits upslope of the first trench. During the field survey, a preliminary stratigraphy with relative ages was made and detailed drawings of the vertical walls of each trench were made. Sediment units were characterized in the field (based on colour, field texture and structure) and by additional laboratory analysis on some selected samples (texture, dry bulk density, ph, carbon, nitrogen and phosphorus content). Radiocarbon dates were established with the AMS method by the Leibniz Labor für Altersbestimmung und Isotopenforschung, Kiel. Large artefacts were dated by archaeologists (see Acknowledgements for details). Results Both selected gullies, G14 and G19, are located only a few hundred metres apart and dissect the southeast-oriented slope of one of the main valleys in the Meerdaal forest (Figure 2c and d). The present-day gully channel G14 has a length of 60 m, a mean top width of 7.96 m, a mean depth of 1.21 m and a volume of 388 m 3. Gully G19 is 68 m long, m wide, 3.39 m deep and has a volume of 2257 m 3. An overview of the first transect through gully G14 (G14-TR1) is shown in Figure 3a. The detailed stratigraphy of transect G14-TR1 is shown in Figure 4a. In this study, only the sedimentary structures observed in the infill of the southeast-facing (head wall) are presented. Transect G14-TR1 shows that the original incision at this point was about 1.6 m deeper than the present-day soil surface. The gully incised down to the top of the Tertiary sands (Figure 4a, 1), most probably because these are covered by an erosionresistant, stony layer (2). The gully channel formed thus

149 T. Vanwalleghem et al.: Prehistoric and Roman gullying in Belgian loess 395 Table 1 Reported gully erosion phases in northwest and central Europe Country Study area Timing of gully erosion phase(s) Dating method Reference(s) Germany Biesdorfer Kehlen (1) After 2857/2495 BC; C Schmidtchen and Bork (2003) (2) AD 1600/1800 Poland Kazimierz Dolny (1) 2000/1750 BC; A, C Schmitt et al. (2003b) (2) AD 1700/1800 Germany Taunus and crystalline (1) Before 540 BC H, C Semmel (1995) Odenwald mountains (Neolithic?); (2) From AD 1600 on France Paris Basin Before 540 BC A, C Larue (2005) Germany Hainbach (1) AD 800/1100; (2) AD 1300/1350; (3) AD 1500/1900 C Dotterweich et al. (2003); Dotterweich (2005) UK Howgill Fells AD 900/1000 C Harvey et al. (1996), Harvey et al. (1981) Germany Rüdershausen (1) AD 1200/1450; A, C Bork et al. (1998) (2) AD 1650/1800 Germany Wolfsschlucht (1) AD 1300/1400; (2) AD 1700/1800 A, C Bork et al. (1998); Schmitt et al. (2003a) Germany Coppengrave AD 1300/1350 A, C Bork et al. (1998) Germany Drudevenshusen (1) AD 1300/1350; A, C Bork et al. (1998) (2) AD 1400/1500 Germany Lützellinden-Zechbach AD 1300/1600 A, C Bork et al. (1998) Germany Neuenkirchen AD 1200/1500 A, C Bork et al. (1998) Germany Antreff AD 1200/1500 A, C Bork et al. (1998) Germany Sälgebach AD 1200/1500 A, C Bork et al. (1998) Germany Thiershausen AD 1200/1500 A, C Bork et al. (1998) Germany Welschbach AD 1200/1500 A, C Bork et al. (1998) Germany Mörsbach AD 1500/1600 A, C Bork et al. (1998) Germany Adelshofen AD 1500/1800 A, C Bork et al. (1998) Germany Bottenbach AD 1500/1800 A, C Bork et al. (1998) Germany Hinterreit AD 1600/1900 A, C Bork et al. (1998) Germany / AD 1600/1800 H Hempel (1976) Slovakia Myjava hill land (1) AD 1550/1730; H Stankoviansky (2003) (2) AD 1780/1850 Germany Nienwohlde (1) AD 1700/1800; A, C Bork et al. (1998) (2) AD 1974/1988 Belgium Kinderveld AD 1700/1900 A, C Vanwalleghem et al. (2005b) France, E France, SW Germany AD 1700/1800 H Vogt (1953) Germany Germany SW Germany AD 1750/1850 H Hard (1976) Hungary Szekszárd Hill country AD 1780/1860 H Zámbó (1972) Hungary Csérhat hill country c. AD 1820 H Zámbó and Gábris (1977) Hungary NE Hungary c. AD 1850 H Gábris et al. (2003) Belgium Central Belgium c. AD 1900 / Grégoire and Halet (1906) Belgium Central Belgium AD 1970/2004 F De Ploey (1990); Poesen and Govers (1990); Vandaele et al. (1996); Steegen et al. (2000); Nachtergaele et al. (2001), Vanwalleghem et al. (2005b) France Northern France; Normandy AD 1970/2004 F Auzet et al. (1995); Cerdan et al. (2002) Norway Southern Norway AD 1970/2004 F Oygarden (2003) Romania Moldavia AD 1970/2004 F Radoane et al. (1995) Russia Kursk region; Stavropol region AD 1970/2004 F Roshkov et al. (1993); Belyaev et al. (2004) Sweden Southern Sweden AD 1970/2004 F Alstrom and Akerman et al. (1992) UK South Downs; Howgill Fells; Scotland; Oxfordshire and Berkshire AD 1970/2004 F Evans and Cook (1987); Boardman (1992); Harvey (1992); Grieve et al. (1995); Boardman et al. (1996) A, archaeological dating of artefact; C, chronometric dating ( 14 C, OSL); H, dating based on manuscripts or maps; F, field observations. entirely in the Quaternary loess deposits. Part of the original, calcareous loess (3) can still be found in the northeastern part of the transect. On top is decalcified loess (4), in which a clay illuviation horizon (or Bt horizon) formed. In the gully infill, five different sediment units could be distinguished (5, 6, 7, 8, 9). The largest part of the gully infill (5) consists of silty to sandy loam, with some fine layering and intercalated with several bands of coarser material (6, 7, 8). Weakly developed clay illuviation bands run through the entire gully infill. At several depths in sediment unit 5, charcoal, Roman brick fragments (roof tiles or tegulae) and late Roman pottery were found (Table 2). On top of the previous layers and over the

150 396 The Holocene 16 (2006) Figure 2 (a) Location of the study area in Belgium. (b) Map of Meerdaal forest (adapted from AWZ / Afdeling Waterbouwkundig Laboratorium and AMINAL / Afdeling Water, 2004) with the position of the two permanent gullies G14 and G19 studied. The position of archaeological sites (Martens, 1981) is also indicated. (c) Schematic representation of gully G14 with its colluvial fan and the location of the two transects G14-TR1 and G14-TR2. (d) Schematic representation of gully G19 with its colluvial fan and the location of transect G19-TR1 entire width of the transect, a sandy loamy colluvium layer (9) is found. This sediment unit is much less dense than the rest of the profile. It is not clear whether this is the result of the homogenizing action of roots and faunal activity or if it is younger colluvium that was deposited over the entire footslope. An overview of the topographical location of the second transect in gully G14 (G14-TR2) is shown in Figure 3b. For this transect, the head wall (perpendicular to the flow direction in the gully) and the side wall (parallel to the flow direction, from north to south) are shown in Figure 4b. Comparable with the situation in the downslope transect G14-TR1, the original gully incised until the level of the Tertiary sands (Figure 4b, 1), after which it filled in partly with about 1.45/1.75 m of sediment. In this second transect through gully G14, eight different sediment units were distinguished in the infill. The first sediment unit (2) above the Tertiary sand is reworked gully floor material. On top of (2) lies a thick, homogeneous sediment unit (3), in which most of the artefacts were found. Two sediment units (4 and 5) are intercalated between (3). Sediment unit (4) is quite homogeneous, while (5) is characterized by a clear, fine layering typical for water-deposited sediments. In between the layers, a small brick fragment was found (not included in Table 2). The upper sediment units (6 and 7) are homogeneous. On top of these, a thick humus horizon has formed (8). These upper three sediment units are, as in the first transect, less compact compared with the sediment units below. An overview of the location of transect G19-TR1 is shown in Figure 2d. In Figure 4c, the detailed stratigraphy of the southsoutheast facing wall of G19-TR1 is shown. As in the previous gully, the original incision of gully 19 was considerably deeper (1.80 m) than the present surface (Figure 4c). The original soil profile is still visible at both extremities (ie, on the west-southwest and east-northeast borders) of the transect. There, the Tertiary sands (Figure 4c, 1) are covered by reworked Tertiary sand (2) and a layer with very coarse debris (3). In the colluvial gully infill, evidence was found of minimally five cut and fill cycles, each of which characterized by the deposition of a coarse gravel layer (5, 7, 9, 10, 12) and a more sandy layer (6, 8, 11, 15). At the bottom of the first channel a piece of pottery was found, which was, however, too small to be dated. Other sediment units, probably resulting from gully wall failure since they are quite similar to the material in which the gully incised, fill up the remainder of the gully (4, 13, 14). A small, isolated sand lens (16) on top of (14) is probably part of sediment unit (15). A charcoal fragment (g in Figure 4c) from sediment unit (14) (1.20 m below the soil surface) was dated to the Middle Bronze Age (Table 2: 1743/1602 cal. yr. BC 1568 /1533 cal. yr. BC). Another charcoal fragment (h in Figure 4c) from sediment unit (15) (0.88 m below the soil surface) was significantly younger, and was dated to Roman times (Table 2: 46 cal. yr BC / AD 78). Other charcoal fragments (see Figure 4c) were unfortunately destroyed during the dating procedure.

151 T. Vanwalleghem et al.: Prehistoric and Roman gullying in Belgian loess 397 Discussion The development history of both gullies is to a large extent similar and is therefore presented together. Figure 5 gives a schematic representation of the development of a typical gully under forest, based on the detailed stratigraphy observed in the gully transect G19-TR1. Phase A: soil formation After deposition of glauconite-rich sands by the Tertiary seas, these sands were covered by a sandy layer that appears to be a solifluxion layer during subsequent cold periods. During the Pleistocene, most of the area was covered with loess, which is very thin or sometimes even absent on south-facing slopes (Goossens, 1988), such as the slope on which gully G19 is located. On this slope, the loess layer is absent at the gully end but is a few metres thick on the plateau near the gully head. At the site where gully G14 incised, the loess thickness varies between 1.5 and 2.5 m. After deposition, the upper layer of the originally calcareous loess was decalcified, after which clay could migrate to form a clay illuviation horizon (Bt). In G14, this Bt was continuous, while in G19, where the parent material is more sandy, the Bt was formed in bands. Although there is still discussion in the literature about the exact timing of the main phase of clay illuviation either being Lateglacial or Holocene, most authors agree that by the time of the first gully incision, at the earliest during the Middle Bronze Age, this Bt horizon was already fully developed (Kühn, 2003). Figure 3 (a) Overview of transect G14-TR1 through the lower end of permanent gully channel G14 (see Figure 2c), with indication of the approximate gully wall and bottom. Measuring stick is 2 m high. The original incision was about 1.60 m deeper than the actual gully channel with a depth of 1.21 m. (b) Overview of transect G14-TR2 (see Figure 2c) in the thalweg of the same gully. At this location, the depth of the gully infill amounts to 1.75 m. The location of transect G14-TR1, further downslope, is also indicated. (c) Overview of transect G19-TR1 (see Figure 2d) through permanent gully G19. The actual gully channel, visible in the background, has a depth of 3.4 m. At the location of the transect, the gully is filled in with about 1.7 m of sediments These sediment units are overlain by more silty sediment units (17, 18 and 19), in which a Luvisol profile has formed. Charcoal found at the west-southwest side of the transect shows that these three soil horizons are formed in colluvium and not in in situ loess. In the vicinity of the main gully (about 10 m from transect G19-TR1, at the southern border of the colluvial fan, at a depth of 0.10 m below the soil surface), a Bronze coin was found that was dated to the first half of the first century AD (Table 2). Phase B /C: gully incision and partial infilling From the transects through both gullies, several indications were found that support the hypothesis by Vanwalleghem et al. (2003, 2005a) that these gullies were caused by a human-induced land use change from forest to cropland in the gully catchment area.the fact that no undisturbed, loessderived soil profile (eg, Albeluvisol or Haplic Luvisol) is found on top of the Tertiary sands in the thalweg of the gully is the first evidence that this gully is not of periglacial origin. Moreover, the shape of the present gully is not visible in the Tertiary sands, but is cut entirely in the loess cover (Figure 4a and c). Further evidence for the anthropogenic origin of the gully is given by the various artefacts found in the transects (Table 2) and the dating of the charcoal fragments. The C-14 datings in gully G19 originate from two different periods, which could possibly be associated with two phases of gully activity and agriculture in the catchment: a first one during the Bronze Age and a second one at the end of the Roman period. Silty layers slid in from the gully walls just after the first incision of the gully, when these walls were still very unstable. The second incision cleared out part of these sediments again. Based on the artefacts found in the infill of gully G14 (Table 2), indications were found only for a Roman gully incision phase. In both gullies, the infilling must have been a rather rapid process since no indications of a stable phase with humic horizon development within the sediment deposited was found. This is further supported by the relatively steep gully walls that can be seen in both infills (around 0.54 m/m in G14 and 3.4m/m to overhanging in G19). Field observations at the nearby Kinderveld site (Vanwalleghem et al., 2005b) confirm that the infilling of a gully of several metres deep can occur within a few decades when its catchment is continuously under cropland.

152 398 The Holocene 16 (2006) Figure 4 (a) Detailed stratigraphy of trench wall G14-TR1 (see Figure 2c). For a description and approximate age of the artefacts found in this trench that could be dated (indicated by letters), see Table 2. Most of the datable material was found in the opposite trench wall, and their position in the drawn transect wall is therefore only indicative. (b) Detailed stratigraphy of trench wall G14-TR2 (see Figure 2c). For a description and approximate age of the artefacts found within this transect, see Table 2. No sample codes are indicated here since these artefacts were not found directly in the trench walls. (c) Detailed stratigraphy of transect G19-TR1 (see Figure 2d). For a description and age of the artefacts and charcoal fragments (indicated by letters) found in this transect, see Table 2 Lang and Hönscheidt (1999) point to the fact that charcoal fragments they found in colluvium were reworked from an upslope sink. Moreover, it is possible that artefacts and charcoal fragments were at the soil surface for a long time before the gullies incised. Therefore, the age of the datable material found in the gully infills, strictly speaking, only provides a maximum estimate of the age of the gully (terminus post quem). Nevertheless, the land-use history of the Meerdaal forest suggests that the gullies are not younger. Although archaeological data are scarce and the existing sites cannot be linked directly to the land use and erosion phase that is suggested by the dating of the gully infills, all archaeological sites in Meerdaal forest are from the Roman period or earlier (see Figure 2b and Vanwalleghem et al. 2003, 2005a). At a horizontal distance of only 500 m to the south of gully G14, on the plateau of Saint Nicaise, the oldest known traces of human

153 T. Vanwalleghem et al.: Prehistoric and Roman gullying in Belgian loess 399 Table 2 Relevant datable material (archaeological artefacts and charcoal fragments) found in transects G14-TR1, G14-TR2 and G19-TR1 Excavation code Sample code Depth below soil surface (m) Type of material Dating method Laboratory code Age and/or detailed description G14-TR1 a 1.26 brick fragment ARCH / Piece of Roman tile b 1.22 brick fragment ARCH / Piece of Roman tile c 1.19 pottery ARCH / Not dated d 1.32 brick fragment ARCH / Piece of Roman tile e 0.70 brick fragment ARCH / Piece of Roman tile f 1.35 brick fragment ARCH / Piece of Roman tile G14-TR2 / 0.50 iron artefact ARCH / Clamp, probably Roman / 0.98 pottery ARCH / Pottery with metal glaze, from the Argonne, wine beaker, showing evidence of secondary burning, AD 210/270 / 0.52 ceramic, large piece ARCH / Fragment of mortarium, from Tienen, Roman / 0 ceramic ARCH / Fragment of Roman dolium, AD 50/300 / 0.80 ceramic ARCH / Fragment of Roman dolium, AD 50/300 / 1.27 ceramic ARCH / Pink ceramic with motive (glazed), Tiens smoked pottery, from beaker, AD 160/170 / 0.48 brick fragment, large ARCH / Piece of Roman tile / 0.29 brick fragment, large ARCH / Piece of Roman tile / 0.48 brick fragment ARCH / Piece of Roman tile / 0.73 intact brick tile ARCH / Roman tile / 0.89 brick fragment ARCH / Piece of Roman tile G19-TR1 g 1.2 charcoal C-14 KIA15410 BP 3380 /// 22 (cal. 1743/1602; 1568/1533 BC)* fragment h 0.88 charcoal C/14 KIA15409 BP 1988///30 (cal. 46 BC/cal. AD 78)* fragment / 0.1 metal coin ARCH / AD 0 /50, showing image of emperor Claudius For location of sample codes, see Figure 4a, b and c. ARCH, dating by archaeologists; C-14, radiocarbon dating. *two s-range. occupation in Meerdaal forest are found, with several Iron Age or Bronze Age tumuli and an earthen wall (see Figure 2b). Three Roman tumuli are at a distance of only 500 m to the west. Gully G19 is even located somewhat closer (350 m) to the three Roman tumuli. Although in situ traces have not been found so far, brick fragments found at the soil surface suggest the presence of three Roman villa sites in Meerdaal forest, of which one is located in the drainage area of gully G14 (Figure 2b). Younger archaeological sites are not present in the forest and definitely after the fourteenth century, when the Meerdaal forest received a special protective status, it can be expected that written documents would exist of any agricultural land use phase. Phase D: deposition of colluvium Next, a silty colluvial layer was deposited over the entire length of the transect in both gullies. This layer was most probably deposited simultaneously or shortly after deposition of the previous sediment units since in gully G14, artefacts of Roman Age were found inside this layer. Phase E: soil formation The archaeological evidence in Meerdaal forest and its surroundings suggest that the area was abandoned after the Roman period, most probably after the invasion of German and Frank tribes at the end of the third century (Faider- Feytmans, 1977). Cropland and/or settlements in the Meerdaal forest were also abandoned probably somewhere around this period and no more runoff or sediments could be produced in the drainage area of the gullies. The infilling of these gullies stopped abruptly after the forest recovered and the gullies remained stable for almost two millennia. During this period, further soil formation took place. In gully G19, a clay illuviation horizon in bands (band Bt horizon) developed through the entire profile and a micropodzol developed in the upper centimetres of the soil, which further stress this stability phase. Also in gully G14, such Bt bands developed in the gully infill. Additionally, in this gully, decalcification continued after the gully incision (see Figure 4a). Assuming that the original decalcification border was orientated horizontally, such as between 10 B/XB/11 m, it can be seen in Figure 4a that the current decalcification border between 8 m B/XB/10 m is orientated parallel to the gully wall and about 0.2 m lower. This is consistent with values given by Vormezeele (1999) and Aerts (2003), who found a post-roman decalcification between 0.2 and 0.5 m, associated with a grave pit and a sunken lane, respectively, near Tienen (about 25 km east of Meerdaal forest). Conclusions Previous studies have shown that gully erosion phases in past times were sometimes even more catastrophic compared with present-day gully erosion, for example during the first half of the fourteenth century (Bork et al., 1998). However, while information on gully erosion during the historical period is already scarce, information on gully erosion before this period is nearly absent. This study has shown that gully formation is definitely not a new problem and that already during Bronze Age and Roman times, farmers were confronted with the development of large gullies. The gullies that were studied are among the oldest gullies that have been dated in Europe. More important than these two studied gullies, however, is that the entire study area of Meerdaal forest encompasses a total of 43 similar gullies, mapped during a recent study (Vanwalleghem et al., 2003). Although only two of these gullies have now been analysed in detail, it is reasonable to assume that the development history of the other gullies is similar given their

154 400 The Holocene 16 (2006) Dhr T. Deweerdt for providing useful background on the local archaeology. The authors also express their thanks to the Houtvesterij Leuven, especially Ir. B. Meuleman, and Ir. D. Galou of the Cantonement de Namur for the authorization to conduct this research in the Meerdaal forest. References Figure 5 Schematic development history of a typical gully in Meerdaal forest location. This makes the Meerdaal forest a unique area in Europe, with a well-conserved late-roman topography. When looking at the landscape in this ancient forest, it becomes clear that gully erosion must have been a significant process for landscape development. Outside the forest border, under cropland, similar gullies were probably also present but centuries of water and tillage erosion have erased them from the present-day surface. This stresses the importance of the Meerdaal forest as a unique area and calls for its protection as a geosite. Acknowledgements The authors would like to thank Mrs Doris Kramer for the drawing of Figure 5. We would especially like to thank Professor Dr Johan Van Heesch and Dr Marc de Bie of the K.U. Leuven for help with the dating of the Roman coin and Aerts, K. 2003: Erosie- en colluviatiegeschiedenis van de archeologische site van Tienen-Grijpen. Masters Thesis. Faculty of Science, K.U. Leuven (in Dutch). Alstrom, K. and Akerman, B. 1992: Contemporary soil-erosion rates on arable land in southern Sweden. Geografiska Annaler Series A / Physical Geography 74, 101/108. Auzet, A.V., Boiffin, J. and Ludwig, B. 1995: Concentrated flow erosion in cultivated catchments: influence of soil surface state. Earth Surface Processes and Landforms 20, 759/67. AWZ / Afdeling Waterbouwkundig Laboratorium and AMINAL / Afdeling Water 2004: Digitaal Hoogtemodel Vlaanderen. Map sheet 32/6. Bell, M. 1992: The prehistory of soil erosion. In Bell, M. and Boardman, J., editors, Past and present soil erosion. Oxbow Monograph 22, Oxbow books, 21/35. Belyaev, V.R., Wallbrink, P.J., Golosov, V.N., Murray, A.S. and Sidorchuk, A.Y. 2004: Reconstructing the development of a gully in the Upper Kalaus basin, Stavropol Region (southern Russia). Earth Surface Processes and Landforms 29, 323/41. Boardman, J. 1992: Current erosion on the south downs, implications for the past. In Bell, M. and Boardman, J., editors, Past and present soil erosion. Oxbow Monograph 22, Oxbow books, 9 /19. Boardman, J., Burt, T.P., Evans, R. and Slattery, M.C. 1996: Soil erosion and flooding as a result of a summer thunderstorm in Oxfordshire and Berkshire, May Applied Geography 16, 21/ 34. Bork, H.-R., Bork, H., Dalchow, C., Faust, B., Piorr, H.-R. and Schatz, T. 1998: Landschaftsentwicklung in Mitteleuropa. Klett- Pertes (in German). Bossuyt, B. 2001: Plant species and soil dynamics across ancient/ recent forest ecotones: consequences for ecological restoration. Ph.D. Thesis. Faculty for Applied Biological Sciences, K.U. Leuven. Cerdan, O., Le Bissonnais, Y., Couturier, A., Bourennane, H. and Souchere, V. 2002: Rill erosion on cultivated hillslopes during two extreme rainfall events in Normandy, France. Soil and Tillage Research 67, 99 /108. De Ploey, J. 1990: Threshold conditions for thalweg gullying with special reference to loess areas. In Bryan, R., editor, Soil erosion, experiments and models. Catena Supplement 17, 147 /51. Dotterweich, M. 2005: High resolution chronology of a 1300 year old gully system in Northern Bavaria, Germany. Modelling longterm human-induced landscape evolution. The Holocene 15, 994 / Dotterweich, M., Schmitt, A., Schmidtchen, G. and Bork, H.-R. 2003: Quantifying historical gully erosion in northern Bavaria. Catena 50, 135 /50. Evans, R. and Cook, S. 1987: Soil erosion in Britain. Seesoil 3, 28/ 59. Faider-Feytmans, G. 1977: La Belgique à l époque romaine. Les dossiers de l archéologie 21, 8/10. Gábris, G., Kertész, A. and Zámbó, L. 2003: Land use change and gully formation over the last 200 years in a hilly catchment. Catena 50, 151/64. Goossens, D. 1988: Loess. Oude en nieuwe opvattingen, recente ontwikkelingen. De Aardrijkskunde 3, 169/224 (in Dutch). Grégoire, A. and Halet, F. 1906: Etude agrologique d un domaine. Bulletin de l Agriculture 12, 611 /51. Grieve, I.C., Davidson, D.A. and Gordon, J.E. 1995: Nature, extent and severity of soil-erosion in upland Scotland. Land Degradation and Rehabilitation 6, 41/55.

155 T. Vanwalleghem et al.: Prehistoric and Roman gullying in Belgian loess 401 Gullentops, F. 1992: Holocene soil erosion in the loess belt of Belgium. Acta Geographica Lovaniensa 33, 671 /84. Hard, G. 1976: Excessive Bodenerosion um und nach In Richter, G. and Sperling, W., editors, Bodenerosion in Mitteleuropa. Wege der Forschung Bd Darmstadt, 195 /239 (in German). Harvey, A.M. 1992: Process interactions, temporal scales and the development of hillslope gully systems / Howgill Fells, northwest England. Geomorphology 5, 323/ : Holocene hillslope gully systems in the Howgill Fells, Cumbria. In Brooks, S.M. and Anderson, M.G., editors, Advances in hillslope processes, vol. 2. Wiley, 247/70. Harvey, A.M., Oldfield, F., Baron, A.F. and Pearson, G.W. 1981: Dating of post-glacial landforms in the central Howgills. Earth Surface Processes and Landforms 6, 401/12. Hempel, L. 1976: Flurzerstörungen durch bodenerosion in früheren Jarhhunderten. In Richter, G. and Sperling, W., editors, Bodenerosion in Mitteleuropa. Wege der Forschung Bd Darmstadt, 181 /95 (in German). Kühn, P. 2003: Micromorphology and Late Glacial/Holocene genesis of Luvisols in Mecklenburg-Vorpommern (NE- Germany). Catena 54, 537/55. Lang, A. and Hönscheidt, S. 1999: Age and source of colluvial sediments at Vaihingen-Enz, Germany. Catena 38, 89/107. Larue, J.P. 2005: The status of ravine-like incision in the dry valleys of the Pays de Thelle (Paris basin, France). Geomorphology 68, 242/56. Martens, E. 1981: Uit het verleden van de gemeente Oud-Heverlee. Stroobants. Neerijse (in Dutch). Nachtergaele, J., Poesen, J., Steegen, A., Takken, I., Beuselinck, L., Vandekerckhove, L. and Govers, G. 2001: The value of a physically based model versus an empirical approach in the prediction of ephemeral gully erosion for loess-derived soils. Geomorphology 40, 237 /52. Oygarden, L. 2003: Rill and gully development during an extreme winter runoff event in Norway. Catena 50, 217 /42. Poesen, J. and Govers, G. 1990: Gully erosion in the loam belt of Belgium: typology and control measures. In Boardman, J., Foster, I.D.L. and Dearing, J.A., editors, Soil erosion on agricultural land. John Wiley & Sons, 513/30. Poesen, J., Nachtergaele, J., Verstraeten, G. and Valentin, C. 2003: Gully erosion and environmental change: importance and research needs. Catena 50, 91 /133. Radoane, M., Ichim, I. and Radoane, N. 1995: Gully distribution and development in Moldava, Romania. Catena 24, 127/46. Roshkov, A.G., Bakhirev, G.I. and Gorin, V.B. 1993: Rapid development of gullies in the Central Chernozem Zone. Eurasian Soil Science 25, 76/83. Schmidtchen, G. and Bork, H.R. 2003: Changing human impact during the period of agriculture in Central Europe. The case study Biesdorfer Kehlen, Brandenburg, Germany. In Lang, A., Hennrich, K. and Dikau, R., editors, Long term hillslope and fluvial system modelling / concepts and case studies from the Rhine river catchment. Lecture Notes in Earth Sciences 101, Springer- Verlag, 183/200. Schmitt, A., Dotterweich, M., Schmidtchen, G. and Bork, H.R. 2003a: Vineyards, hopgardens and recent afforestation: effects of late Holocene land use change on soil erosion. Catena 51, 241 /54. Schmitt, A., Schmidtchen, G., Rodzik, J., Zglobicki, W., Dotterweich, M., Amhöfer, S. and Bork, H.R. 2003b: Historical gully erosion in southeast Poland, an example from the loess area of the Lublin Upland. In Li, Y. and Poesen, J., editors, Gully erosion under global change. Sichuan Science and Technology Press, 223/30. Semmel, A. 1995: Development of gullies under forest cover in the Taunus and Crystalline Odenwald Mountains, Germany. Zeitschrift für Geomorphologie Neues Funde Supplementen Band 100, 115/27. Stankoviansky, M. 2003: Historical evolution of permanent gullies in the Myjava Hill Land, Slovakia. Catena 51, 223/39. Steegen, A., Govers, G., Nachtergaele, J., Takken, I., Beuselinck, L. and Poesen, J. 2000: Sediment export by water from an agricultural catchment in the Loam Belt of central Belgium. Geomorphology 33, 25 /36. Vandaele, K., Poesen J., Govers, G. and Van Wesemael, B. 1996: Geomorphic threshold conditions for ephemeral gully erosion. Geomorphology 16, 161 /73. Vanwalleghem, T., Van Den Eeckhaut, M., Poesen, J., Deckers, J., Nachtergaele, J., Van Oost, K. and Slenters, C. 2003: Characteristics and controlling factors of old gullies under forest in a temperate humid climate: a case study from the Meerdaal Forest (Central Belgium). Geomorphology 56, 15 /29. Vanwalleghem, T., Poesen, J., Van Den Eeckhaut, M., Nachtergaele, J. and Deckers, J. 2005a: Reconstructing rainfall and land use conditions leading to the development of old gullies. The Holocene 15, 378 /86. Vanwalleghem, T., Bork, H.R., Poesen, J., Schmidtchen, G., Dotterweich, M., Nachtergaele, J., Bork, H., Deckers, J., Brüsch, B., Bungeneers, J. and De Bie, M. 2005b: Rapid development and infilling of a historical gully under cropland, central Belgium. Catena 63, 221 /43. Vogt, J. 1953: Erosion des sols et techniques de culture en climat tempéré maritime de transition. Revue de Géomorphologie dynamique 4, 157/83 (in French). Vormezeele, J. 1999: Reliëfconstructies op archeologische sites: een case studie te Tienen-Grijpen. K.U. Leuven, Faculteit Wetenschappen (in Dutch). Zámbó, L. 1972: Study of some development factors of anthropogenically preformed loess gullies. Acta geographica debrecina 10, 113 /16. Zámbó, L. and Gábris, G. 1977: Examination of relationships between vegetation, slope conditions and gully formation on the model of a hill-country catchment area. Annales universitatis budapestinensis, geographica 6/12, 37/58.

156

157 Geomorphology 108 (2009) Contents lists available at ScienceDirect Geomorphology journal homepage: www. elsevie r. com/ locate/ geomor ph A temporarily changing Holocene sediment budget for a loess-covered catchment (central Belgium) Gert Verstraeten a,, Tom Rommens a, Iris Peeters a, Jean Poesen a, Gerard Govers a, Andreas Lang b a Physical and Regional Geography Research Group, Katholieke Universiteit Leuven, Celestijnenlaan 200E, box 2409, B-3001 Leuven, Belgium b Department of Geography, University of Liverpool, Liverpool, L69 7ZT, UK a r t i c l e i n f o abstract Article history: Accepted 12 March 2007 Available online 13 February 2009 Keywords: Sediment budget Holocene Loess Sediment dynamics Human impact This study presents a Holocene sediment budget for the Nethen catchment, a typical river catchment (55 km 2 ) in the Belgian loess belt. Soil erosion and hillslope sediment storage are quantified by extrapolating detailed data obtained from soil profile truncation studies in three representative zero-order sub catchments. Floodplain sediment storage is estimated by augerings along several transects across the main river and some of its tributaries. The sediment budget shows that ca. 38% of the soil eroded during the Holocene is redeposited as colluvium on hillslopes and in dry valley bottoms. Another 23% of the eroded sediment is now stored as alluvium in the floodplain. The remaining 39% or Mg is exported from the catchment. Dating of both colluvial and alluvial sediment deposits reveals that sediment dynamics between the hillslopes, the dry valley bottoms and the floodplain behave highly non-linearly. Before ~500 BC, sediment delivery from the hillslopes to the river channels was near maximum. However, since the onset of significant agriculture in the Late Bronze Age Early Iron Age, increased rates of soil erosion are only reflected in the colluvium, but not in the floodplain, resulting in very low hillslope sediment delivery ratios. From the Medieval period onwards, soil erosion increased even further, mainly as a result of a further increase in agricultural land use, but now also accelerated floodplain sedimentation took place due to an improved slope-channel coupling and the management of floodplains Elsevier B.V. All rights reserved. 1. Introduction Corresponding author. Tel.: ; fax: address: gert.verstraeten@ees.kuleuven.be (G. Verstraeten). Human impact has caused worldwide increased rates of soil erosion and sediment deposition, for instance in floodplain environments, but has also changed river channel dynamics and morphology (e.g., Trimble, 1983, 1999; Zolitschka et al., 2003; Liébault et al., 2005; Foulds and Macklin, 2006). During the Holocene, natural controls on sediment fluxes in a landscape have been replaced steadily by anthropogenic controls (e.g. Meybeck, 2003; Syvitski et al., 2005; Houben, 2006). Many studies focus only on one aspect of river basin sediment response to human impact, whether it be soil loss, colluvial sediment storage, alluvial sediment storage, sediment yield or channel behavior. Often, the impact of changing human pressure on the landscape is studied by a detailed analysis of a single sediment core in a lake or floodplain environment. Such studies do not provide sufficient spatial information to assess the actual impact of humans on the river basin as a whole as the response of a river basin to disturbance can be complex and highly non-linear (e.g., Schumm, 1977; Trimble, 1983, 1999; Richards, 2002; Lang et al., 2003). A better understanding of human environment interactions in fluvial systems requires a holistic, integrated approach. Recently, Foulds and Macklin (2006) review land-use change and its impact on river basins in Ireland and Great-Britain. They conclude that after decades of research on this topic, a greater appreciation of catchment sensitivity (including slope-channel coupling characteristics, valley floor width and slope), which controls the downstream propagation of eroded sediments (in response to environmental change), is critical to understand the role of land-use change on catchment stability. Foulds and Macklin (2006) continue stating sediment budgeting is perhaps the most fruitful method to achieve this end through identification of reach-scale zones of sediment transfer and storage. The importance of sediment budget studies in tackling the complex sediment history in river basins has long been recognized (e.g. Swanson et al., 1982). Sediment budgeting, i.e. an accounting of the sources and disposition of sediment as it travels from its point of origin to its eventual exit from a catchment (Reid and Dunne, 2003), is a widely used concept with several applications. The available data are often presented in sediment flow diagrams representing a whole range of spatial scales, from rivers and estuaries (Reid and Dunne, 2003), to coastal zones and even ocean troughs (Sommerfield and Nittrouer, 1999; Hsu et al., 2006). Sediment budgets provide information on the relative importance of sediment sources and sinks. Time spans investigated are usually X/$ see front matter 2009 Elsevier B.V. All rights reserved. doi: /j.geomorph

158 G. Verstraeten et al. / Geomorphology 108 (2009) short (days or years) (e.g., Page et al., 1994; Walling et al., 2002) and rarely expand beyond decades (e.g., Trimble, 1999; Fryirs and Brierley, 2001). Emphasis is usually on the impacts of climatic events or environmental changes on sediment redistribution in fluvial systems. The majority of established sediment budgets are limited to short time spans and/or small study areas, for which data collection is easier. This is of fundamental importance as the time scales over which rivers respond to human or natural disturbances will often be longer than the time period studied so that long-term evolution cannot be detected. Fundamental tendencies of change can only be highlighted when a sufficiently long time span is considered. Attempts to establish sediment budgets for longer time periods (e.g., the Holocene) by quantifying erosion and sediment storage, including both hillslope and river channel dynamics (e.g., Meade, 1982; Macaire et al., 2001, 2002; Schrott et al., 2003) are scarce but essential to (1) characterize the entire system of sediment sources and sinks, and (2) to fully understand the impact of land-use or climate change on soil erosion and sediment transfer through river basins. In this study, we present a Holocene sediment budget for the Nethen River catchment, which is typical for the European loess belt. We build on Holocene soil erosion and colluvial sediment storage for a small zero-order catchment established by Rommens et al. (2005) and data on floodplain storage for the Nethen river reported in Rommens et al. (2006). We report new data from two other zero-order catchments, establish an improved chronology and refine the tentative sediment budget of Rommens et al. (2006). This allows us for the first time to determine the changing role of the various sediment stores through time. 2. Materials and methods 2.1. Study area This study was carried out in the Nethen catchment, which is a small river catchment (55 km 2 ) situated in the central Belgian loess belt (Fig. 1). Soils in this region developed in aeolian calcareous silt loam of Late Pleistocene age, which was deposited on top of Tertiary marine sands and clays. Flat plateaus alternate with rolling hills and valleys. Along the course of the Nethen River a floodplain developed, which has a width up to 500 m. The majority of the slopes in the catchment are not steeper than 5% although locally slope gradients up to 50% occur near the major axis of the Nethen valley. Floodplains are often poorly drained and used for pasture or forest, whereas the silt loam soils in the rest of the catchment are cultivated for crop production. Luvisols (FAO, 2006) cover most of the area, particularly in the south-eastern and eastern part of the Nethen basin. The Luvisol cover is, however, not continuous. The north-western and western parts of the Nethen river catchment Fig. 1. Generalized topographic map of the Nethen catchment.

159 26 G. Verstraeten et al. / Geomorphology 108 (2009) show a rather complex soilscape because of a thinner loess blanket, and soils also developed in outcrops of sandy and clayey Tertiary deposits (Podzols, Cambisols, Regosols). Central Belgium has a temperate humid climate with a mean annual rainfall between 700 and 800 mm, which is well distributed throughout the year A sediment budget for the Nethen In order to construct a long-term sediment budget for the Nethen catchment, estimates of past soil erosion as well as sediment storage volumes are needed. The major sediment stores are the floodplain, the dry valleys linking the loess plateau with the floodplain, and the footslopes outside concentrated flow areas, whereas the major sediment source is hillslope soil erosion by water. Quantification of sediment stores and soil losses was primarily based on augering surveys, whereby a somewhat different approach was followed for the hillslopes than for the floodplains. Rommens et al. (2005) first quantified the volume of soil that was eroded and stored for Nodebais, a zero-order catchment (103 ha) in the Nethen river catchment (Fig. 1). These volumes were obtained by comparing present-day soil profiles with standard reference soil profiles of a typical Luvisol. Truncated soil profiles are the result of soil erosion processes, whereas buried soil profiles can be used to quantify sediment deposition. Similar approaches to quantify historic soil erosion volumes have been applied before (Trimble, 1975; Lewis and Lepele, 1982; Clemens and Stahr, 1994). For a detailed description of this method, reference is made to Rommens et al. (2005). The major problem with this method is that an accurate description of a typical, non-eroded soil profile has to be made. A comparison of soil profiles on non-eroded sites at various places in central Belgium, revealed a spatial variability in Holocene soil profile development so that the representativeness of the Nodebais catchment can be questioned. Therefore, the findings of the Nodebais catchment were compared with data from two other zero-order catchments in the Nethen catchment (Beauvechain and Hamme-Mille; Fig. 1). In addition to the 185 augerings made in Nodebais, 164 and 92 augerings were made in the Beauvechain (67 ha) and Hamme-Mille (64 ha) zero-order catchments, respectively. In total 36 reference soil profiles were identified, and data from these profiles were used to calculate average Holocene erosion depths and sediment deposition heights as a function of five different morphometric classes: slopes b3%, slopes 3 5%, slopes 5 8%, slopes N8%, and the thalweg (Rommens, 2006; Table 1). On the basis of these average values, the Table 1 A detailed Holocene sediment budget for the Nethen catchment. Morphometric unit Area (km 2 ) Average depth (m) Volume (10 6 m 3 ) Mass (10 6 Mg) Soil erosion Plateau, slope b3% ± ±4.73 Slope 3 5% ± ±2.90 Slope 5 8% ± ±2.64 Slope N8% ± ±3.80 Thalweg, slope b8% ± ±0.79 Alluvium 2.93 ( ) Total ±14.86 Sediment deposition Plateau, slope b3% ± ±1.26 Slope 3 5% ± ±0.81 Slope 5 8% ± ±0.44 Slope N8% ± ±0.82 Thalweg, slope b8% ± ±3.84 Alluvium ±3.32 ( ) Total ±10.50 Sediment export 21.20± Mg Sediment delivery ratio 39 ±35% ( ) Erosion in the floodplain is not considered (Rommens et al., 2006). total volume of eroded soil and colluvial sediments for the whole Nethen catchment was calculated as follows: First, a DEM with 5 m resolution was obtained from digitizing contours on a 1:10,000 topographic map (contour interval of 2.5 m) and applying the TIN (Triangulated Irregular Network) interpolation method of IDRISI; Next, a slope map was calculated in IDRISI and the morphometric classes defined. The thalweg class consists of the concave valley bottom where slopes are less than 8%. The total area of each morphometric class in the Nethen catchment was extracted from this slope map (Fig. 2); The average erosion rate for every morphometric class as obtained in the three zero-order catchments (Rommens, 2006, Table 1) was multiplied with their corresponding areas in the entire Nethen catchment; Finally, the obtained total volumes were multiplied with the dry bulk density of the soil and the sediment in order to obtain a total eroded soil and deposited sediment mass (t) for every morphometric unit. The dry bulk density of 32 samples (18 from soil and 14 from sediment) collected in zero-order basins in the study area was 1510±100 kg m 3. Total Holocene sediment storage in the floodplain of the Nethen and its main tributary river floodplains was estimated by Rommens et al. (2006). Based on 115 cores alongside eight transects the average depth of the Holocene sediment unit and the total mass of mineral sediment was determined. We slightly revised the data of Rommens et al. (2006), based on additional data for the dry sediment bulk density and the organic matter content. Finally, the total mass of sediment exported from the 55 km 2 Nethen catchment during the considered time period was estimated by subtracting all sediment storage components from the soil erosion component of the sediment budget. The quantification of soil loss and sediment deposition requires several measurements, assumptions and calculations. Each of which is of limited precision only. Here we quantified all random errors and assumed values for systematic errors. Errors were propagated using Gaussian error propagation technique. Measurement errors of 10% and 5% were assigned to areas and depths, respectively, whereas the interpolation uncertainty due to the limited augering density was estimated to be between 10 and 30% (Rommens et al., 2006) A timeframe for sediment dynamics In order to analyze the Holocene sediment dynamics in more depth, it is important to know when particular sediment stores become more important compared to others. The sediment budget of Coon Creek, Wisconsin (Trimble, 1983, 1999), for instance, clearly showed that there is a major downstream shift in sediment dynamics following human impact in the catchment. In order to test whether this is also the case for the Nethen on a much longer time scale, both the colluvial and the alluvial sediments were dated by AMS radiocarbon on wood and charcoal fragments incorporated in the deposits, and by optically stimulated luminescence (OSL) dating of the sediments themselves. A chronology of colluvial sediments was established both for the Nodebais and for the Beauvechain catchment. In the main thalweg of the Nodebais catchment, a trench was dug, revealing 4 m of late- Holocene deposits (Rommens et al., 2007). From the exposed colluvial deposits, seven charcoal samples for AMS were taken, as well as seven sediment samples for OSL (for details about the technique applied see Rommens et al., 2007). The obtained dates indicated that no sediment was deposited in the Nodebais thalweg prior to the Late Bronze Age Early Iron Age (~ cal BC), and that a significant increase in colluvial sedimentation rate only took place after ~320 cal AD

160 G. Verstraeten et al. / Geomorphology 108 (2009) Fig. 2. Morphometric units in the Nethen catchment as derived from a digital terrain model. This map was used for the calculation of total erosion and sediment deposition amounts on the slopes and dry valleys (1: Nodebais; 2: Beauvechain; 3: Hamme-Mille). (Rommens et al., 2007). In order to test whether slopes in other parts of the Nethen evolved in a similar way as in Nodebais, several detailed and undisturbed percussion cores were recovered from the thalweg of the Beauvechain catchment (Fig. 3). From these cores, six charcoal samples and four silty sediment samples were collected and dated with AMS 14 C and OSL, respectively. Based on the chronometric dates for Nodebais and Beauvechain, chronologies for colluviation were established and sedimentation rates calculated. Consequently, rates of total sediment accumulation were calculated taking into account the increase in accommodation space as sedimentation proceeded. Despite their different characteristics, all three investigated subcatchments show largely similar slope evolution histories. We therefore assumed that the other, non-sampled zero-order basins evolved in a similar way and scaled results up for the whole Nethen catchment by using the percentages of sediment deposited during a specific time period in the Nodebais and Beauvechain catchment multiplied with the total colluvial sediment stored in the Nethen catchment (see Section 2.2). Fig. 3. Transect through the colluvial sediments in the lower part of the Beauvechain zero-order catchment.

161 28 G. Verstraeten et al. / Geomorphology 108 (2009) Rommens et al. (2006) also provided a temporal framework of alluvial sediment deposition based on 11 radiocarbon dates of organic rich deposits. They distinguish three major periods of sediment deposition. The results from this study were used to calculate average sedimentation rates, which were again expressed as a sediment mass per unit of depositional area (t ha 1 a 1 ) in order to compare these rates with those obtained for colluvial sediment deposition. For further comparison with the colluvial sediment store, we also made a cumulative assessment of floodplain sediment storage through time. All sedimentation rates are also subject to some inaccuracies, which are related to (1) inaccuracies in the determination of the sediment depth, and (2) inaccuracies in the sediment age. The accuracy of these accumulation rates was determined using Gaussian error propagation techniques assuming an error of ±3 cm on the depth and the reported analytic uncertainty (1σ) on the sediment age. 3. Results and discussion 3.1. A Holocene sediment budget for the Nethen river catchment Eroded and stored sediment volumes and masses for the Nethen River catchment are listed in Table 1 and a sediment budget for the Holocene is shown in Fig. 4. During the Holocene, ca Mg of sediment have been eroded in the Nethen catchment. It is clear that the steep slopes acted as the main sediment source: almost one third of the eroded volume originates from the steepest slope class (N8%) while this class occupies only 13% of the total area. Ca. 38% of the eroded sediment mass is stored in colluvial deposits on the plateau, on footslopes and in dry valley bottoms, whereas ca. 23% was deposited in the floodplains of the Nethen River and its tributaries. The remaining part of the eroded sediments, i.e. ca. 39%, or Mg, was exported from the Nethen catchment and delivered to the Dijle River. The Holocene sediment budget for the Nethen catchment gives an idea of the sediment masses that have been transported in this basin through time. As expected, the majority of the sediments entering the Holocene sediment pathway originates from the steepest slopes but the soil losses from the plateau areas also contribute significantly to the budget. Overall, the budget demonstrates that a significant part of the eroded soil during the Holocene is nowadays stored in the zero-order valleys and on the slopes, and until now, has not been transferred to higher order streams. As already mentioned, we have to deal with a considerable level of uncertainty in the calculations. Part of this uncertainty can be treated as a random Gaussian error (e.g. measurement errors), but other factors must be estimated (e.g. accuracy of field observations). Apart from the factors already discussed, the remaining uncertainty on the sediment budget is to a large extent due to the variability in reference soil profiles in the three zero-order catchments. The presence of rather large measurement errors and uncertainties implies that the eroded, deposited and exported sediment masses reported above should be considered as best possible estimates, with which a relatively large uncertainty is associated. Based on the error ranges discussed above, Rommens (2006) estimated the total mass of eroded sediment to vary between 48 and Mg, and the total mass of deposited sediment between 26 and Mg. As a consequence, sediment delivery ratios obtained by the different extrapolations range as well, namely between 24% and 59%. Extreme values resulted when extrapolating data obtained in a single zeroorder catchment to the entire Nethen. It is clear from these results that one has to be careful when extrapolating observations from so-called representative test sites. The sediment budget presented here does not include erosion from sunken lanes, nor does it consider sediment storage in lynchets, nor the role of the river channel itself. We digitized all the sunken lanes in 9 km 2 areas around the Nodebais and Beauvechain zero-order catchments: this yielded sunken track densities of 0.75 km km 2 and 0.91 km km 2, respectively. Assuming a fairly large mean crosssection of 32 m 2 (=8 4 m), the erosion from the sunken lanes results in an additional soil loss of ~ m 3 km 2 or ca Mg. This value is far smaller than the total volume that has been eroded from the hillslope. Although lynchets are mainly the result of tillage operations, sediment deposition by overland flow upslope from lynchets can also be important (e.g. Macaire et al., 2002). This sediment sink was not quantified in this study. Furthermore, we did not separate the effect of tillage redistribution on hillslopes and sediment fluxes caused by water erosion. Although at present, tillage redistribution rates in central Belgium are at least of the same order of magnitude as water erosion rates, this is only true for the last century. For the entire Holocene period, the observed soil profile truncation and sediment deposition depths are mainly the result of water erosion (Van Oost et al., 2005). Part of the eroded soil is caused by root and tuber crop harvesting over the last 200 years in the study area (Poesen et al., 2001). Average soil losses by root and tuber crop harvesting (SLCH) for the Belgian Fig. 4. Holocene sediment budget for the River Nethen (new data and updated data from Rommens et al., 2006). More detailed information on hillslope erosion and sediment storage on hillslopes can be found in Table 1.

162 G. Verstraeten et al. / Geomorphology 108 (2009) Loess Belt are in the order of 1.8 Mg ha 1 a 1 for the late 20th century (Verstraeten et al., 2006). This would correspond to a maximum soil loss of Mg for the Nethen catchment and most of that soil has been transported with the crop by humans out of the catchment. However, we currently lack sufficient historic agricultural census data to calculate a more accurate value, and SLCH are therefore not further considered in the sediment budget. Finally, riverbank erosion is not considered. The Nethen is a meandering river and hence riverbank erosion does occur, but this erosion is limited to a narrow meander belt as could be observed in the nearby Dijle River (Vandaele et al., 2002). In none of the augerings in the floodplain, deposits of former river channels could be identified. There is also no indication of significant widening or narrowing of the river channel itself. Besides secondary fresh water carbonates and peat, almost the entire floodplain is made up of silt loam sediments suggesting that coarse grained bedload transport that could have influenced channel dynamics has not been important. Thus, we believe that riverbank erosion and sediment deposition in the inner meander bends of the Nethen have balanced each other and that the net role of the river channel itself in the total sediment budget is very limited Sediment dynamics through time Fig. 5 compares the sedimentation rates for the floodplain, as well as for the thalweg in the Nodebais and Beauvechain zero-order catchments. More information about the sediment dates obtained in Beauvechain (Fig. 3) can be found in Tables 2 and 3. Data for Nodebais are extensively discussed in Rommens et al. (2007). Minerogenic sediment deposition in the floodplain remained low (ca 0.2 mm a 1 ) for most of the Holocene period. This allowed the accumulation of peat and organic matter in most places. Loam and silt loam deposits on top of these organic layers only date from the last ~600 to ~1400 years, which are characterized by high sedimentation rates (ca 2.5 mm a 1 ). However, in dry valleys connecting the plateau with the floodplain, substantial sediment Table 2 Results of AMS 14 C dating of charcoal sampled in the thalwegs sediment deposits in the Beauvechain catchment (for location of samples see Fig. 3). ID Lab code Depth (m) Conventional 14 C age (a BP±1σ) Cal. (2σ) age (cal. BC AD) a B2.1 GrA (±45) AD / B3.2 GrA (±35) AD B3.3 GrA (±35) BC/120 BC AD 70 B3.5 GrA (±40) / / BC B4.1 Beta N39,400 B4.4 GrA (±35) / BC Radiocarbon dead charcoal: the age could not be defined. a Calibrated using OxCal v 3.10 (2005) (Bronk Ramsey, 2001). Calibration curve based on atmospheric data from Reimer et al. (2004). deposition probably started in the Late Bronze Age (c cal BC) or the Early Iron Age (c cal BC). Sedimentation rates in the dry valleys quickly rose from almost zero to values higher than those occurring in the alluvial plain during the same period (3 6 Mg ha 1 a 1 )(Fig. 5). A comparison of these sedimentation rates suggests that sediment transfer from the hillslopes to the river channels is typical of that described as the sediment cascade system (e.g. Lang, 2003): in a first period with limited land disturbance, soil erosion only impacts hillslopes locally and sediment deposition is mainly limited to colluvium. Only when these sediment stores are full and/or when a more efficient linkage is created between the hillslopes and the river channels, accelerated sedimentation takes place on floodplains. These observations made in the dry valleys of Beauvechain and Nodebais are similar to those reported in other studies from Western and Central Europe, which place the onset of significant colluvial sedimentation in the Bronze or Iron Age (Macklin et al., 1991; Niller, 1998; Lang and Hönscheidt, 1999; Lang and Nolte, 1999; Taylor et al., 2000; Lang, 2003; Bertran, 2004). Estimated sediment accumulation rates for these studies (measured in mm a 1 ) are also similar. Lang and Hönscheidt (1999) reported colluvial accretion rates of c. 0.6 mm a 1 in the Iron Age to 1.2 mm a 1 in recent times. Bertran (2004) measured sediment accumulation in karstic depressions and found rates from 0.4 to 1.1 mm a 1 between the Early Iron Age and 1840 AD. However, the picture becomes more complex when cumulated sediment deposition masses in the floodplain and the thalweg are considered (Fig. 6). Despite the relatively low sedimentation rates in the floodplain before the Middle Ages, the total amount of sediment deposited on the floodplain is much higher than that in the dry valleys. 46% of the total mass of Holocene sediment deposited in the Nethen alluvial plain was deposited before significant colluviation started in Nodebais or Beauvechain. For a correct interpretation of the sedimentation values, some caution is needed. The average rates in fact stand for a largely simplified model of reality. The sedimentation rates are averaged over time, which implies that potentially important fluctuations are inevitably masked A temporarily changing sediment budget for the Nethen Fig. 5. Average mineral sediment deposition rates in the alluvial plain of the Nethen (adapted from Rommens et al., 2006) compared to the corresponding average rates for the zero-order valley bottoms in Nodebais (Rommens et al., 2007) and Beauvechain (this study). Ranges (min. to max.) for the sedimentation rates in the different sections of the Nethen alluvium are indicated as well. Because of the rather poor temporal resolution of the data, the exact onset of rising or decreasing sedimentation rates in both colluvial and alluvial environments cannot be determined. Nonetheless, a temporally resolved sediment budget is suggested by introducing three periods, which are characterized by a different sedimentation rate in either the floodplain or the dry valley bottoms (Table 4). The first period runs from the beginning of the Holocene until the onset of significant colluvial deposits in the Nodebais and Beauvechain thalweg. Although the range in dates for these first colluvial deposits extends from 1090 cal BC to 450 cal BC, we selected

163 30 G. Verstraeten et al. / Geomorphology 108 (2009) Table 3 Analytical data and OSL dating results for silt samples (4 15 μm) taken from the Beauvechain thalwegs sediment deposits (for location of samples see Fig. 3): sample code and depth, water content (Δ, in % of dry weight), U, Th and K concentration, effective dose rate (dd/dt effective), equivalent dose rate (D e ) and OSL ages. Sample ID Lab code Sampling depth (m) Δ (%) U (µg g 1 ) Th (µg g 1 ) K (%) dd/dt effective (Gy ka 1 ) D e (Gy) OSL age (ka) a OSL age (calendar year) a B I ±5 3.32± ± ± ± ± ±0.02 AD B II ±5 3.28± ± ± ± ± ±0.07 AD B III ±5 3.49± ± ± ± ± ± BC AD 140 B IV ±5 3.44± ± ± ± ± ± BC a OSL ages are given at 1σ confidence level and were converted to calendar years by subtracting 2005 (the year of measurement) from the OSL age and rounding to the next decade. the end of period 1 at 500 BC. The second period runs from 500 BC to 1000 AD: the latter age corresponds to the average date of various samples taken from floodplain sediments at the transition from sedimentation unit 2 to 3 (Rommens et al., 2006). This age also corresponds to the onset of a sharp increase in floodplain sedimentation rates. The third period runs from 1000 AD to the present (2005 AD). The sediment masses in Table 4 were calculated by multiplying the average sedimentation rate (Fig. 5) with the total depositional area (Table 1) and the time period considered. For the thalweg colluvium, the average value of the Beauvechain and Nodebais data were extrapolated to the whole Nethen catchment. Unfortunately, no sediment dating for the hillslope deposits outside the dry valleys is available, nor can we date the erosion history of the slopes. Nevertheless, we attempted to calculate a complete sediment budget for the three time periods considered (Table 4; Fig. 7). Therefore, we had to make the following assumptions: 1. Sediment export follows the same trend as floodplain storage. We are aware that other studies found that sediment yield is not a reliable indicator of the internal sediment dynamics in a catchment (e.g. Trimble, 1983, 1999). In the Nethen, however, sediments are mainly cohesive silt loam and coarser grained sediments, as in Trimble's study in Coon Creek that shows clear non-linear relations between floodplain storage and sediment yield, are rare. Also, the transport capacity of a stream like the Nethen for silt loam is very large. We suggest that floodplain sediment deposition in the Nethen is primarily controlled by inundation frequency (i.e. flow discharge) and sediment concentration. Both factors also control sediment yield. Thus, periods with higher floodplain sedimentation Fig. 6. Cumulative sedimentation volumes in the floodplain of the Nethen, as well as in the dry valleys based on extrapolation of the sedimentation rates in Beauvechain and Nodebais. are probably also periods during which a lot of sediment is exported out of the catchment. 2. For each time period, mass is preserved. Thus, the erosion rate for each of the three periods equals the sum of the sediment mass exported and the sediment mass stored in the floodplain, the dry valleys and the slopes. The first three are discussed above, thus only the sediment storage on the slopes itself remains unknown. 3. The temporal dynamics of sediment deposition on the slopes was estimated in two different ways (hence, there are also two estimates for soil erosion in Table 4). First, it is assumed that sediment deposition on the slopes underwent a similar evolution as the sedimentation in the dry valleys. However, this implies that during the first time period, when almost no sedimentation is observed in the dry valleys, nearly all sediment eroded off the hillslopes was directly transported to the channels. In other words, the hillslope sediment delivery ratio would approximate 100%. This is an unrealistically high value. Therefore, we also calculated sediment storage on the slopes assuming that it follows the same trend as the erosion on the slopes. This latter value is depicted in Fig. 7. The tentative sediment budgets for the three different time periods clearly show that the relative importance of the various components of the sediment budget change significantly through time. As a consequence, the sediment delivery ratios (SDR) also vary through time. Fig. 8 shows how SDR changes, whereby a distinction is made between hillslope SDR and catchment SDR: the first is the ratio between the mass of sediment exported to the river channel divided by the eroded mass, whereas the second is the sediment export out of the catchment divided by the eroded mass. The difference in SDRvalues 1 and 2 reflects different assumptions on how to calculate soil erosion estimates on the hillslopes (see above). Although the absolute number of the various sediment budget components varies depending on the way the erosion rate is calculated (see Table 4), the resulting SDR-values of both methods are quite comparable. Hence, irrespective of the (contrasting) assumptions we made with respect to the evolution of slope erosion through time, the internal sediment dynamics of the Nethen catchment remain the same. The SDR-values are highest for the first time period, for which we can safely state that the human impact was lowest. The high value for the hillslope SDR can be explained as follows: under (semi-)natural conditions with a negligible human impact, vegetation cover is very good. The limited production of sediment on the slopes can be transported efficiently to the river channels as the transport capacity of overland flow is not reached. Especially in the dry valleys where runoff concentrates, the sediment transport is maximal. Based on extensive augerings in the Nodebais catchment (Rommens et al., 2005), the original topography prior to soil erosion and sediment deposition could be reconstructed. This shows that the slope of the dry valley bottom was much steeper than it is today, thus the sediment transport capacity also must have been larger. During the Bronze Age Iron Age the human impact became more important. This is evidenced by the abundance of charcoal from this period in the sediment deposits in Beauvechain and Nodebais, and also from gullying and pit excavations dated to this period in the Meerdaal forest, which is

164 G. Verstraeten et al. / Geomorphology 108 (2009) Table 4 A tentative sediment budget for the Nethen cachment for three different periods. Period Floodplain storage a Colluvial storage in thalweg b Sediment export c Soil erosion estimate 1 d Soil erosion estimate 2 e Colluvial sediment storage on slopes 1 f Colluvial sediment storage on slopes 2 g 10 6 Mg % 10 6 Mg % 10 6 Mg % 10 6 Mg % 10 6 Mg % 10 6 Mg % 10 6 Mg % b500 BC BC 1000 AD AD present Total a Updated from Rommens et al. (2006). b Based on sedimentation rates in the thalwegs of Nodebais and Beauvechain. c Based on floodplain storage % for each period. d,e Assume that for a given period the mass balance must be kept, thus erosion equals the sum of all storage components plus the sediment export. f Assumes that colluvial sediment storage on the slopes is proportional to the sediment storage in the thalweg. g Assumes that sediment storage on the slopes is relative to the erosion rate (in this case, b and f are determined iteratively). For explanations see text. situated in the NW-part of the Nethen catchment (Vanwalleghem et al., 2006, 2007). Furthermore, pollen analysis of sediments from the Nethen floodplain (Mullenders et al., 1966) and the Dijle floodplain less than 10 km from the confluence of the Nethen with the Dijle river (De Smedt, 1973), shows that although some cereal pollen is observed in the Atlantic period ( BP), it is not until the end of the Subboreal period (~ BP) when it occurred in significant concentrations and is abundant from the Subatlantic period. These Fig. 7. Changing sediment dynamics through time in the Nethen catchment as illustrated by a tentative sediment budget for three different time periods.

165 32 G. Verstraeten et al. / Geomorphology 108 (2009) Table 6 Oldest references to villages and watermills in the Nethen catchment (after Tarlier and Wauters, 1873; Olyslager, 1949; Delcorps et al., 2005). Village Year (AD) Nodebais 990 Hamme-Mille 1139 Tourinnes-la-Grosse 1159 Beauvechain 1018 Nethen 1147 Bossut 1092 Watermill location and owner Tourinnes-l.-G. Monastery of Aulne Before 13th century a Tourinnes-l.-G. Monastery of Averbode AD 1233 a Hamme-Mille Monastery of Valduc Before AD 1431 Hamme-Mille Lord of Bierbeek, Monastery of Valduc Before AD 1367 St.-Joris-Weert Monastery of St. Jean Baptiste, Liège Before AD 1367 Nethen Monastery of St. Jean Baptiste, Liège Before AD 1367 a a Demolished in 19th or 20th century. Fig. 8. Sediment delivery ratios (SDR) for the Nethen for three different periods and for the entire Holocene period. Numbers 1 and 2 refer to erosion estimates 1 and 2, respectively, in Table 4. See text for more information on how the SDR-values were calculated. show the increasing importance of arable land and as a result, soil erosion intensified and the sediment fluxes from the slopes increased as well. We suggest that sediment transport capacity in the dry valleys was now exceeded and important sediment deposition took place. Most likely, the deeply incised dry valleys were not cleared and the vegetation cover that remained facilitated sediment deposition. From the Middle Ages onwards, the system again reacted differently. Archeological research in the Dijle River basin suggests that prehistoric and Roman settlements are mainly found at some distance from the rivers, and that river valleys only became occupied from Medieval times onwards (Table 5; Van Hove et al., 2005). The first villages as we know them today appeared after 1000 AD in the main valley of the Nethen (Table 6). The change in settlement pattern most probably led to a concurrent change in land-use patterns and arable land was no longer confined to the plateau areas and the upper slopes. Instead, the steeper slopes closer to the valley were also cultivated, which facilitated the transport of sediment to the floodplain. Moreover, especially from the 12th and 13th century AD onwards, the establishment of large farms, which specialized in the extensive production of wheat, resulted in the removal of fences and hedges from the agricultural land and the typical open-field landscape developed (Verhulst, 1990; Lindemans, 1994). Consequently, connectivity in the landscape improved, which facilitated sediment and water transfer through the fluvial system (Van Oost et al., 2000). As the percentage of arable land increased through time, the runoff discharges also increased, thereby increasing the floodplain inundation frequencies. Furthermore, changes in the way the floodplain was managed with drainage works and the establishment of fishponds and watermills from the 12th to 14th century AD onwards, enhanced sedimentation in the floodplain (Table 5). Many of these changes in medieval agriculture and technology were introduced under the influence of upcoming local and regional monasteries. Similar findings were reported in central Germany (Houben, 2006). As discussed above, the changing spatial patterns of land occupation, as well as the introduction of new land management types (both on the hillslopes with arable land and in the floodplain), probably were at least for a large part responsible for the change in sediment dynamics in the Nethen river catchment. These two factors are more important than population density as such. Fig. 9 shows the evolution of the number of households (numbers of individuals are not known) for the village of Nodebais. Population density remained stable during the Middle Ages, and only started to increase during the 19th century. By contrast, we see that sedimentation rates in the alluvium have been high during the last 1000 years, and that, in the Nodebais zero-order catchment sedimentation already reached a high value during the Middle Ages. This clearly indicates that the evolution of population density as such cannot be used as a proxy for soil erosion rates. 4. Conclusions This study on the Nethen river catchment is one of the first that quantifies Holocene soil erosion and sediment deposition in the upland area as well as in the alluvial plain. Our results show that large masses of sediment, mainly originating from the slopes in zeroorder basins, are deposited and stored on footslopes and in dry valley bottoms, near their source area (~38% of the total sediment Table 5 Distribution of archaeological artefacts and sites with respect to the distance to the nearest river channel for various cultural periods (after Van Hove et al., 2005). Period Distance to the nearest river channel b100 m m m N1000 m Prehistoric 10% 35% 37% 18% Roman period 10% 30% 20% 40% Middle ages 31% 48% 13% 8% Post-middle ages 58% 31% 11% 0% Fig. 9. Evolution of the number of households in the village of Nodebais from the 14th century AD to the 19th century AD (data from Tarlier and Wauters, 1873).