Fingerprinting suspended sediment sources in the catchment of the River Ouse, Yorkshire, UK

Transcription

1 HYDROLOGICAL PROCESSES Hydrol. Process. 13, 955±975 (1999) Fingerprinting suspended sediment sources in the catchment of the River Ouse, Yorkshire, UK Desmond E. Walling 1 *, Philip N. Owens 1 and Graham J. L. Leeks 2 1 Department of Geography, University of Exeter, Exeter, Devon, EX4 4RJ, UK 2 Institute of Hydrology, Wallingford, Oxfordshire, OX10 8BB, UK Abstract: Statistically veri ed composite ngerprints and a multivariate mixing model have been employed to establish the main sources of the suspended sediment transported through the lower, non-tidal reaches of the River Ouse and one of its main tributaries, the River Wharfe, during the period 1994±1997. In the case of the suspended sediment samples collected from the River Ouse, the load-weighted mean contributions from uncultivated topsoil, cultivated topsoil and channel bank sources were estimated to be c. 25, 38 and 37%, respectively, while for the River Wharfe these sources contributed c. 70, 4 and 23%, respectively (c. 4% was derived from woodland topsoil). Suspended sediment samples collected during higher ows evidenced a greater contribution from channel banks than samples collected during lower ows. Source materials were also di erentiated according to the three main geological source areas (Carboniferous, Permian and Triassic, and Jurassic) and their load-weighted mean contributions were estimated to be c. 24, 41 and 35% for the River Ouse and c. 91, 9 and 0% (there are no Jurassic rocks in this catchment) for the River Wharfe, respectively. When suspended sediment samples from tributary streams were used to characterize each geological source area, the equivalent results for the River Ouse were c. 30, 46 and 24%. Considering the three main tributaries that contribute to the River Ouse, the load-weighted mean contributions from the rivers Swale, Ure and Nidd were estimated to be 82, 15 and 3%, respectively. These values have been compared with estimates of the relative magnitude of the annual suspended sediment loads of these three rivers for the years 1995 and 1996 derived from continuous monitoring of discharge and turbidity. Di erences between the two sets of results are ascribed to the di erent periods of record involved and to the timing of suspended sediment sampling relative to the overall storm hydrograph, and thus the degree to which the available samples are representative of the overall suspended sediment ux. Although a number of limitations must be recognized, the ngerprinting approach to source ascription is seen as providing valuable information regarding suspended sediment sources in the study catchments. Copyright # 1999 John Wiley & Sons, Ltd. KEY WORDS suspended sediment; sediment ngerprinting; source tracing; sediment provenance; multivariate mixing model; composite ngerprints INTRODUCTION The suspended sediment transported by a river will commonly represent a mixture of sediment derived from di erent locations and di erent types of sediment source within the contributing drainage basin. Information on suspended sediment provenance is an important requirement in the examination of sediment routing and delivery and in the construction of catchment sediment budgets (cf. Dietrich and Dunne, 1978; Swanson et al., 1982; Walling and Webb, 1983). From a management perspective, there is also a need to identify * Correspondence to: Professor D. E. Walling, Department of Geography, University of Exeter, Exeter, EX4 4RJ. geography@exeter.ac.uk Contract grant sponsor: UK NERC. Contract grant number: GST/02/774. CCC 0885±6087/99/070955±21$1750 Received 1 January 1998 Copyright # 1999 John Wiley & Sons, Ltd. Revised 1 April 1998 Accepted 1 April 1998

2 956 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS sediment sources in order to implement appropriate strategies to control sediment mobilization and associated o -site e ects such as the siltation of river channels and reservoir sedimentation. Furthermore, sediment source will also exert a fundamental control on the sediment-associated transport of nutrients and contaminants in river systems, since the source of the sediment will in uence its physical and chemical properties and its contaminant loading and any management strategy would need to take account of sediment provenance. The precise type of information on sediment source required will depend on the purpose in hand and the nature of any sediment-related problem. However, information on both the source type (e.g. sheet erosion on di erent land use types or erosion of channel banks) and spatial location (e.g. which tributary or part of the basin) will frequently be required. Because of the many problems associated with traditional procedures for identifying suspended sediment sources (cf. Peart and Walling, 1988), ngerprinting techniques are increasingly being used as an alternative approach to assembling such information. Advantages include the limited amount of eld data collection required (i.e. collection of source material and suspended sediment samples as opposed to expensive longterm monitoring at a wide range of sites) and the range of complementary information that can be obtained using the ngerprinting approach. The basic principle underlying the ngerprinting approach is that di erent potential sediment sources can be characterized, or ngerprinted, using a number of diagnostic physical and chemical properties and that comparison of these ngerprints with equivalent information for suspended sediment samples permits the relative importance of the di erent potential sources to be determined (cf. Old eld et al., 1979; Peart and Walling, 1986; Walling and Woodward, 1992). The selection of suitable diagnostic properties will depend on the nature of the potential sources to be distinguished and the drainage basin characteristics. Many di erent physical and chemical properties have been successfully used to discriminate potential sediment sources in drainage basins, including mineralogy (Klages and Hsieh, 1975), sediment chemistry (Wall and Wilding, 1976), mineral magnetism (Old eld et al., 1979) and environmental radionuclides (Walling and Woodward, 1992). Because of the natural spatial variability of source materials in uvial systems, the complexity of sediment routing and delivery processes in drainage basins, and the need to deal with several potential sources, use of a single diagnostic property will rarely prove adequate to discriminate a number of potential sediment sources and to establish their relative importance. The rigour and reliability of the ngerprinting approach, and the range of potential sediment sources that can be discriminated, will generally be signi cantly increased if several di erent diagnostic properties are used to establish statistically veri ed composite ngerprints and if a multivariate mixing model is subsequently used to establish the relative importance of the di erent potential sources (cf. Yu and Old eld, 1989, 1993; Walling et al., 1993; He and Owens, 1995; Walling and Woodward, 1995; Collins et al., 1996, 1997a, b). To maximize the e ectiveness of the ngerprinting approach, the diagnostic properties should be in uenced by di erent environmental controls and thus possess a substantial degree of independence, so that in combination they a ord a high degree of source discrimination (cf. Walling et al., 1993). This paper reports the use of statistically veri ed composite ngerprints in association with a multivariate numerical mixing model to establish the relative importance of di erent potential suspended sediment sources, de ned in terms of both source type (cultivated or uncultivated topsoil or channel banks) and spatial location (individual sub-basins and di erent geological zones), within the drainage basins of the River Ouse, and one of its major tributaries, the River Wharfe, in Yorkshire, UK. Suspended sediment provenance was determined by comparing the properties of suspended sediment samples collected from several primary sites (Figure 1) with those of both potential source materials (i.e. soils, channel bank material, etc.) and suspended sediment samples collected from major and minor tributaries. Use of source material ngerprints provided information on both source type and spatial provenance de ned in terms of geological zones. The suspended sediment samples used to establish the importance of di erent sub-basins and di erent geological zones were collected from the three major tributaries of the River Ouse upstream of its tidal limit (the rivers Swale, Ure and Nidd) and from minor tributaries that were used to characterize the main geological groups within the study area (see Figure 1). Initially, the suspended sediment and source material samples were analysed for a number of potential ngerprint properties, including

3 FINGERPRINTING SUSPENDED SEDIMENT SOURCES 957 Figure 1. Location map showing the study area, the suspended sediment sampling sites and the main geological subdivisions of the study area environmental radionuclides ( 137 Cs, 226 Ra and unsupported 210 Pb), mineral magnetism (susceptibility and saturation isothermal remanent magnetisation Ð SIRM), geochemical composition (Al, Ca, Cr, Cu, Fe, K, Mg, Mn, Na, Ni, Pb, Sr and Zn), phosphorus content (total, inorganic and organic) and organic matter content represented by C and N. Subsequently, a two-stage statistical procedure was used in order to establish which of these properties provided good discrimination between individual sources within a particular category (i.e. source types, geological zones and subcatchments) and which combination of those properties provided the best composite ngerprint for each category of potential sources. A multivariate mixing model

4 958 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS was then used to compare the composite ngerprints of the suspended sediment samples with those of the potential sources, in order to estimate the relative importance of those sources. STUDY AREA The River Ouse is one of two main rivers that drain into the Humber Estuary in north-east England: the other being the River Trent (Figure 1). The River Wharfe is a major tributary of the River Ouse, and joins the latter downstream of the tidal limit, south of York. The catchment areas of the rivers Ouse and Wharfe at the Environment Agency gauging stations at Skelton ( just upstream of York) and Tadcaster are 3315 and 818 km 2, respectively. The Ouse and Wharfe basins are predominantly rural and agricultural with a low population density. Geology, topography and land use within the study area are closely interrelated and exhibit a well-de ned pattern of variation running from west to east (cf. Figure 1). The topography is dominated by the Pennine Hills, which occupy the western parts of the study area and which represent a dissected plateau reaching altitudes of over 700 m, and by the western edge of the North York Moors to the east, which rise to an altitude of over 300 m within the study area. The lowland of the Vale of York lies between these two areas of higher relief. The underlying solid geology of the Pennine Hills (cf. Figure 1) is predominantly of Carboniferous age (limestone and Millstone Grit). To the east, the Carboniferous strata give way to the softer Permian (Magnesian limestone) and Triassic (New Red Sandstone) strata that underlie the Vale of York. Further to the east, the North York Moors are underlain by Jurassic sandstones, limestones, shales and clays. In many areas, particularly in the Vale of York, the solid geology is mantled by glacial drift deposits, such as boulder clay and lacustrine deposits, but the topography closely re ects the underlying solid geology. The soils are dominated by peats and stagnohumic and stagnogley soils in the uplands, and by stagnogleys, sandy gley soils and brown earths in the lowlands. In the low-lying Vale of York, much of the land is cultivated and the main crops are cereals, potatoes and sugar beet. In the upland areas in the Pennines and North York Moors, land use is dominanted by pasture, rough grazing and moorland. There is relatively little woodland in the study basin and York is the only major urban area. Mean annual precipitation varies considerably over the study area, from c. 600 mm at the tidal limit to over 1700 mm in the headwaters of the River Ure near Hawes. More detailed information on the physiography and hydrology of the study area can be found in Jarvie et al. (1997) and Law et al. (1997). METHODS The collection of suspended sediment and source material samples Bulk river water samples (between c. 100 and 250 litres) were collected from above the tidal limits of the River Ouse (at sites either just above or below York) and the River Wharfe (at Tadcaster) and from both major and minor tributaries (Figure 1) using a submersible pump powered by a portable generator. The water samples were collected during storm events when suspended sediment concentrations were high and typically within the range 100±1000 mg l 1. Over 150 bulk suspended sediment samples were collected during the period November 1994±February 1997, and these samples are considered to be reasonably representative of the suspended sediment transported past each sampling site. The sediment was recovered from the bulk water samples by continuous ow centrifugation and the resulting sediment samples were freeze-dried before analysis. In order to characterize potential source materials, over 160 source material samples (4500 g) were collected throughout the study area using a stainless steel trowel. Within each of the main geological zones, representative samples were collected from the faces of eroding channel banks and from the surface (c. top 2 cm) of woodland, uncultivated (i.e. pasture, rough grazing and moorland) and cultivated areas. These samples were dried at 40 8C and disaggregated prior to analysis. Laboratory analysis The suspended sediment and source material samples were analysed for a range of diagnostic properties (cf. Table I). All source materials were screened through a 63 mm sieve prior to analysis in order to provide a

5 FINGERPRINTING SUSPENDED SEDIMENT SOURCES 959 more direct comparison with suspended sediment samples (additional corrections for contrasts in particle size composition were also applied at a later stage). Radionuclide concentrations [caesium-137 ( 137 Cs), radium-226 ( 226 Ra) and unsupported lead-210 ( 210 Pb)] were determined using an EG&G Ortec hyper-pure germanium well-type detector. Mineral magnetic properties were measured at Coventry University using a Bartington MS2B dual frequency sensor for the susceptibility (w) parameters and a Molspin pulse magnetiser and Molspin minispin uxgate magnetometer for SIRM (IRM at 0.8 T). Metal concentrations were measured using a Unicam 939 atomic absorption spectrophotometer after acid (concentrated HCl and HNO 3 ) digestion (cf. Allen, 1989). Organic carbon (C) and nitrogen (N) concentrations were determined using a Carlo Erba ANA 1400 automatic nitrogen analyser, and phosphorus (P) concentrations (total, inorganic and organic) were measured using a Pye Unicam SP6 UV/visible spectrophotometer after chemical extraction (cf. Mehta et al., 1954). In order to correct for further contrasts in grain size composition between source materials and suspended sediment or between suspended sediment from tributary streams and the main river sites, the speci c surface area of the samples was estimated from their particle size distributions. These were determined using a Coulter LS130 laser di raction granulometer, after removal of organic matter (by H 2 O 2 ) and chemical [(NaPO 3 ) 6 ] and ultrasonic dispersion. Statistical and numerical methods A two-stage statistical procedure was used to identify optimum sets of source material and sediment properties for use as composite ngerprints (cf. Collins et al., 1996, 1997b). First, the Kruskall±Wallis H-test was used to establish which properties exhibited signi cant di erences between the individual source groups within a particular category of sources. Secondly, multivariate discriminant function analysis was applied to the properties selected in the rst stage in order to identify the set of properties or composite ngerprint that a orded optimum discrimination between source groups. A stepwise selection algorithm, based on minimization of Wilks' lambda or U-statistic, was used in this analysis. It is important to note that although it is possible to use ratios (such as C/N) to discriminate source groups statistically, ratio data are not suitable for use in a numerical linear mixing model and, therefore, have not been used in this study. A multivariate mixing model was subsequently used to estimate the relative contribution of the potential sediment sources to a particular suspended sediment sample. This mixing model is similar in principle to those used in other investigations that have quanti ed the provenance of recent uvial and lacustrine sediments (e.g. Yu and Old eld, 1989, 1993; Walling et al., 1993; He and Owens, 1995; Slattery et al., 1995; Walling and Woodward, 1995; Collins et al., 1996, 1997a, b). For each of the tracer properties i in the composite ngerprint, a linear equation is constructed that relates the concentration of property i in the suspended sediment sample to that in the mixture representing the sum of the contributions from the di erent source groups. Thus, the composite ngerprint is represented by a set of linear equations (one for each of the properties in the composite ngerprint). Instead of solving the set of linear equations directly, the least-squares method was used, and the proportions derived from the individual sources s are established by minimizing the sum of the squares of the residuals (R es ) for the n tracer properties involved, where R es ˆ Xn iˆ1 C ssi C si P s 2 1 C ssi where C ssi is the concentration of tracer property i in the suspended sediment sample, C si is the mean concentration of tracer property i in source group s and P s is the relative proportion from source group s. The model must satisfy two linear constraints, namely. (a) the contribution from each source must lie within the range 0 to 1, i.e. 0 4 P s 4 1; and 2

6 960 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS (b) the sum of the contributions from all sources is 1, i.e. X n sˆ1 P s ˆ 1 3 Because of the well-documented relationship between the speci c surface area of a sediment sample and the concentration of a given determinand in that sediment, shown by many elements (cf. Horowitz, 1991), there is a need to take account of di erences in particle size composition between suspended sediment samples collected from di erent (main and tributary) sampling sites and between suspended sediment and source materials. The e ects of contrasts in grain size composition between source materials (or tributary suspended sediment) and downstream suspended sediment were partly addressed by restricting analysis to the 563 mm fraction, but further correction was required for both source material and tributary suspended sediment samples in order to take account of di erences in the grain size composition of this fraction. The values for the ngerprint properties of samples from each source group were therefore corrected for di erences in particle size composition compared with the suspended sediment sample whose source was being traced, by using the ratio of the speci c surface area of each suspended sediment sample to the mean value for an individual source group. Since the precise relationship between speci c surface area and element concentration (cf. He and Walling, 1996) was not determined for each tracer property, use of the ratio described above provided a simple and convenient means of correcting all of the ngerprint property values for each source group. No corrections were made for di erences in organic matter content between source materials or tributary suspended sediment and downstream suspended sediment (cf. Collins et al., 1996, 1997b) because the relationship between organic matter content and element concentration is complex and di cult to generalize. Also, the simultaneous correction of raw source material or tributary suspended sediment data for di erences in both particle size composition and organic matter content di erences may result in overcorrection of ngerprint property values (Collins et al., 1997b). Furthermore, the source material (or tributary suspended sediment) and downstream suspended sediment data for each of the tracer properties in the composite ngerprint were standardized so that the values for each of the tracer properties ranged between zero and one. Mean values of the particle size corrected, standardized tracer properties for each source group were used in the mixing model. The goodness-of- t provided by the mixing model was assessed by comparing the actual ngerprint property concentrations for a representative selection of the suspended sediment samples collected from the downstream reaches of the rivers Ouse and Wharfe, with the corresponding values predicted by the mixing model based on the estimates for the percentage contributions from each of the sources within each source category (cf. Collins et al., 1997a, b). The mean (average for all properties within each composite ngerprint) relative errors for the mixing model calculations typically ranged between + 8% and + 15%, and suggest that the mixing model is able to provide an acceptable prediction of the concentrations of the ngerprint properties associated with individual suspended sediment samples from the study rivers. Because the suspended sediment samples collected from each sampling site covered a range of discharge conditions and suspended sediment concentrations (e.g. 30 samples were collected from the River Ouse), the mean values for the contribution of the various sources for individual sampling sites have been weighted according to the suspended sediment load at the time of sampling (derived as the product of the instantaneous values of discharge and suspended sediment concentration), namely P sw ˆ Xn P sx xˆ1 where P sw is the load-weighted relative contribution from source group s, L x (kg s 1 ) is the instantaneous suspended sediment load for suspended sediment sample x, L t (kg s 1 ) is the sum of the instantaneous loads L x L t 4

7 FINGERPRINTING SUSPENDED SEDIMENT SOURCES 961 (L x ) for all sediment samples from that sampling site and P sx is the relative contribution from source group s for sediment sample x. This load-weighted approach provides a more realistic estimate of the proportion of the total suspended sediment load at a sampling site contributed by individual sources than a simple average of the percentage contribution values associated with individual suspended sediment samples. Further information on the monitoring of river discharge and turbidity in the study rivers is provided by Evans et al. (1997), Law et al. (1997), Wass et al. (1997) and Wass and Leeks (1999). RESULTS Sediment source ascription using source materials (a) Source type. Table I presents the results of the Kruskall±Wallis H-test, which was used to assess the ability of the tracer properties to discriminate the four source types, namely, surface soil from woodland, uncultivated and cultivated areas and channel bank material. Because the Wharfe catchment is adjacent to the Ouse catchment and has similar characteristics, the source materials for both catchments were pooled (tests showed that the values for virtually all tracer properties were not signi cantly di erent between the two catchments at the 95% level of con dence). The Kruskall±Wallis test indicated that most (20 out of 24) of the tracer properties provided a clear discrimination between woodland topsoil, uncultivated topsoil, cultivated topsoil and channel bank material. Multivariate discriminant function analysis was subsequently used to Table I. The results of using the Kruskall±Wallis test to assess the ability of each tracer property to discriminate between surface (c. top 2 cm) soil from woodland, uncultivated and cultivated areas and channel bank material from the Ouse and Wharfe basins (pooled data) Tracer property H value* P value 137 Cs (mbq g 1 ) { 226 Ra (mbq g 1 ) { Unsupported 210 Pb (mbq g 1 ) { N (%) { C (%) { Total P (mg g 1 ) { Inorganic P (mg g 1 ) { Organic P (mgg 1 ) { w lf (mm 3 kg 1 ) { w fd (nm 3 kg 1 ) { SIRM (mam 2 kg 1 ) { Al (mgg 1 ) { Ca (mgg 1 ) Cr (mgg 1 ) { Cu (mg g 1 ) Fe (mg g 1 ) { K(mg g 1 ) { Mg (mgg 1 ) { Mn (mgg 1 ) { Na (mg g 1 ) Ni (mg g 1 ) { Pb (mgg 1 ) Sr (mg g 1 ) { Zn (mgg 1 ) { * Critical H value ˆ { Signi cant at p ˆ 005.

8 962 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS Table II. The results of using stepwise discriminant function analysis to identify which combination of tracer properties provides the best composite ngerprint for discriminating source materials on the basis of source type (i.e. surface material from woodland, uncultivated and cultivated areas and material from channel banks) Tracer property Cumulative % samples classi ed correctly N Total P Sr Ni Zn Ra Cs Unsupported 210 Pb Fe Al identify the best composite ngerprint incorporating a number of these tracer properties, and the results are shown in Table II. Table II indicates that a composite ngerprint comprising 10 tracer properties is capable of classifying almost 95% of the source material samples correctly (the addition of more tracer properties to the composite ngerprint does not increase the level of success of the classi cation). The composite ngerprint includes tracer properties from several di erent property groups (i.e. radionuclides, heavy metals, phosphorus, etc.) and con rms the need to use properties with di erent environmental controls in order to obtain a composite ngerprint that a ords a high degree of discrimination (cf. Walling et al., 1993). The fact that not all of the samples were classi ed correctly into the appropriate source groups is not unexpected, because some of the source material sampling sites that were thought to be uncultivated are likely to have been ploughed at some time and then returned to pasture. Samples collected from such elds could therefore be expected to possess ngerprints re ecting characteristics of both uncultivated and cultivated soils and the discriminant function analysis con rmed that there was some overlap between the samples collected to represent uncultivated and cultivated source groups. Having identi ed the optimum composite ngerprint for distinguishing between the di erent source groups in this category, the numerical mixing model was used to establish the relative contribution of each source to the individual suspended sediment samples collected from the primary sampling sites on the rivers Ouse and Wharfe, and the load-weighted results for each basin are presented in Table III. For the Table III. Load-weighted mean contributions of each source type to the suspended sediment samples collected from the main downstream sampling sites on the rivers Swale, Ure, Nidd, Ouse and Wharfe during the period November 1994± February See Figure 1 for location of sampling sites River Number of sediment samples Woodland topsoil Source type contributions (%) Uncultivated topsoil Cultivated topsoil Channel bank material Swale Ure Nidd Ouse Wharfe

9 FINGERPRINTING SUSPENDED SEDIMENT SOURCES suspended sediment samples collected from the River Ouse during the period November 1994 to February 1997, the contribution from topsoil in woodland areas is zero (or so low that it is not recognized by the mixing model). This nding is consistent with that obtained for other drainage basins in the UK (e.g. Collins et al., 1997b) and re ects the limited areal extent of woodland in the study area and, perhaps more importantly, the lack of surface erosion and related sediment mobilization associated with woodland areas (cf. Morgan, 1986). The relative contributions from the three remaining source groups are of similar magnitude, but surface sources, and particularly those associated with cultivated areas, are dominant. Approximately 38% of the suspended sediment is derived from the surface of cultivated elds, which mainly occur in the low-lying Vale of York. Although the area occupied by cultivated land is smaller than that occupied by uncultivated grazing and moorland, the dominance of the former as a sediment source re ects the higher rates of soil loss commonly associated with cultivated soils (cf. Morgan, 1986), the close proximity of this land to the channel network, since much of the cultivated land occurs in the valley bottoms, and its location in the lower reaches of the river basin, which minimizes conveyance losses such as those associated with overbank oodplain deposition. The contribution from uncultivated source areas is, nevertheless, substantial and re ects the dominance of pasture, rough grazing and moorland land use over much of the catchment, particularly in the upland areas of the Pennines and North York Moors, where slope gradients and precipitation amounts are greatest. Channel banks are also a signi cant sediment source and this in part re ects the well-developed river banks (often 42 m in height; see also Lawler et al., 1999) found along the main river channels in the Vale of York, which frequently show evidence of erosion. Although this loadweighted contribution from channel banks of c. 37% could be seen as rather high, it is in reasonable agreement with the ndings of other studies undertaken in agricultural catchments in the UK (e.g. Walling et al., 1993; Ashbridge, 1995; He and Owens, 1995; Walling and Woodward, 1995), which have reported bank contributions to the overall suspended sediment load of 12±21%. It is also consistent with independent evidence of the importance of bank erosion generated by other studies within the LOIS programme (cf. Lawler et al., 1999). It is worthy of note that the mean contribution from channel banks for the six samples collected from the River Ouse during periods with the highest instantaneous suspended sediment loads (450 kg s 1 ) was 46%, while for the three samples collected during periods with the smallest loads (55 kgs 1 ) the contribution from channel banks was only 13%. This suggests that the detachment and entrainment of channel bank material is signi cantly greater during high ows than during lower ow conditions. Information concerning the relative importance of the contributions from the individual source types to the suspended sediment loads of the three main subcatchments that drain into the River Ouse (Swale, Ure and Nidd) is also presented in Table III. The results are again broadly in accordance with the land use and general characteristics of each basin. There are signi cant contributions from channel bank sources in all three catchments and these range between 15 and 37%. The contributions from topsoil in woodland areas are very low. In the case of the samples collected from the River Swale, uncultivated topsoil, which characterizes much of the headwater area in the Pennines, represents the dominant source. However, there is also a substantial contribution from cultivated topsoil sources and this re ects the considerable amount of cultivated land found in the downstream portion of the Swale catchment. The results for the samples collected from the River Ure are broadly similar to those for the River Swale and this re ects the similar characteristics of these two adjacent catchments. However, the reduced contribution from cultivated topsoil evidenced by the Ure catchment is consistent with its smaller proportion of cultivated land. Sediment from uncultivated topsoil sources represents the primary source of the suspended sediment load transported by the River Nidd, and this again re ects the dominance of pasture and moorland land use in this catchment. The relative contribution of cultivated topsoil to the suspended sediment load of the River Nidd is arguably lower than might be expected on the basis of the proportion of the catchment occupied by cultivated land, and this may re ect lower sediment delivery ratios for cultivated areas in this catchment. The source information provided by the seven suspended sediment samples collected from the River Wharfe is also presented in Table III. The results are broadly similar to those obtained for the main

10 964 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS tributaries of the River Ouse, re ecting the similar character of the basins, but there are some di erences. The dominant source of the suspended sediment collected at the tidal limit at Tadcaster is uncultivated topsoil (c. 70%). However, the relative contributions from cultivated and channel bank sources are generally lower than for the other tributaries, and there is a small but signi cant contribution from surface soil from woodland. These di erences re ect the limited amount of cultivated land and increased woodland in the contributing catchment, and reduced incidence of bank erosion along the River Wharfe. As with the results from the River Ouse, the two samples associated with the highest instantaneous suspended sediment loads (48 kgs 1 ) showed greatest relative contribution from channel banks (c. 30%), con rming the importance of high ow events in mobilizing channel bank material. The values for the relative contribution of di erent source types to the suspended sediment loads of the study rivers presented in Table III represent load-weighted mean values. These mean values conceal considerable inter- and intra-storm variability in the contribution of the individual source groups and Figure 2 provides examples of this variability for a representative selection of individual suspended sediment samples collected from the lower reaches of the River Ouse. The variation shown in Figure 2 re ects a number of controls including inter-storm variations in the magnitude and spatial distribution of precipitation events and in antecedent conditions, seasonal variations in land use and land cover (cf. Walling and Woodward, 1995; Collins et al., 1997b) and the timing of sample collection in relation to the timing of di erent source inputs during storm events (cf. Walling and Woodward, 1992; He and Owens, 1995; Slattery et al., 1995; Collins et al., 1997b). (b) Spatial location (geology). In order to provide information on the spatial location of the sediment sources contributing to the suspended sediment load at the sampling sites, source materials were classi ed according to the three main geological groups found in the study area (i.e. Carboniferous, Permian and Triassic, and Jurassic). These three geological groups broadly correspond to the three main topographic zones in the basin, namely, the Pennines (Carboniferous), Vale of York (Permian and Triassic) and North York Moors (Jurassic). Although in places the underlying solid geology is covered by glacial drift deposits, most of these are located in the Vale of York. Thus the study area can still be divided into three main zones (here termed geological zones) on the basis of underlying geology (and drift) and topography, and it is expected that the source materials in each zone are likely to exhibit distinct property signatures. Figure 2. The relative contribution of cultivated and uncultivated topsoil and channel bank sources to a selection of suspended sediment samples collected from the River Ouse. The mean values represent the load-weighted averages for all samples collected from this river

11 FINGERPRINTING SUSPENDED SEDIMENT SOURCES 965 Table IV. The results of applying the Kruskall±Wallis test to assess the ability of each tracer property to discriminate between the three main geological groupings of source materials within the Ouse and Wharfe basins (i.e. Carboniferous, Permian and Triassic, and Jurassic). See Table I for the units of each property Tracer property H value* P value 137 Cs { 226 Ra Unsupported 210 Pb { N { C { Total P Inorganic P Organic P w lf { w fd { SIRM Al { Ca { Cr Cu Fe { K { Mg { Mn { Na { Ni { Pb { Sr { Zn * Critical H value ˆ { Signi cant at p ˆ 005. The results of applying the Kruskall±Wallis test to these three source groups are presented in Table IV. As with source type, most (16 out of 24) of the tracer properties are able to discriminate successfully between the three main geological groups. The optimum composite ngerprint selected by stepwise discriminant function analysis (cf. Table V) was composed of eight tracer properties, which in combination were capable of correctly classifying 80% of the source materials. Summary results for the use of the mixing model for the study rivers are presented in Table VI. Table VI indicates that the contributions from the three main geological source groups to the suspended sediment load transported by the River Ouse are broadly similar. Based on the load-weighted mean values, the contribution from the Permian and Triassic rocks (41%) can be seen to be dominant, and this is in good agreement with the relative proportion of the Ouse basin occupied by these rocks (c. 40%). Furthermore, there is good agreement between these results and those for source type (see Table III), because the Permian and Triassic rocks outcrop mainly in the low-lying Vale of York (Figure 1), where much of the cultivated land in the basin is located. Table III indicates that cultivated land supplies c. 38% of the suspended sediment transported by the River Ouse in its lower reaches. Based on the area of the basin occupied by each of the geological groups (see Table VI), one might expect a somewhat larger proportion of the suspended sediment load transported by the River Ouse to be contributed from the Carboniferous rocks in the western part of the catchment and a smaller contribution to be provided by the Jurassic rocks to the east, than indicated by the results in Table VI. This suggests that rates of erosion and sediment supply from the Jurassic rocks of the North York Moors are greater than from the Carboniferous

12 966 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS Table V. The results of using stepwise discriminant function analysis to identify which combination of tracer properties provides the best composite ngerprint for di erentiating source materials on the basis of geology (Carboniferous, Permian and Triassic, and Jurassic) source groups Tracer property Cumulative % samples classi ed correctly Mn N Mg w lf w fd K Sr Ni Table VI. Load-weighted mean contributions of each geological source group to the suspended sediment samples collected from the main downstream sampling sites of the study rivers during the period November 1994±February 1997, based on representative source materials collected from each of the three geological source groups. The approximate areas underlain by each geological group in each catchment are listed in parentheses River Geological source group contributions (%) Carboniferous Permian and Triassic Jurassic Swale 28.6 (39) 36.6 (41) 34.8 (20) Ure 55.2 (79) 44.8 (21) 0 (0) Nidd 75.9 (67) 24.1 (33) 0 (0) Ouse 23.8 (50) 41.4 (40) 34.8 (10) Wharfe 90.6 (90) 9.4 (10) 0 (0) rocks of the Pennines. The apparent importance of the Jurassic rocks may, however, also re ect the timing of suspended sediment sampling relative to the overall storm hydrograph, and this is discussed later. Information concerning the relative importance of the contributions from the three main geological source groups to the suspended sediment loads of the rivers Swale, Ure and Nidd is also presented in Table VI. There is close agreement between these results and the relative areas occupied by the individual source groups in each catchment. In the case of the River Swale, the three geologies contribute approximately equally to the suspended sediment samples collected from the downstream sampling site. However, the Permian and Triassic rock type is the main geological source (c. 37%) and this is consistent with the close location of this rock type relative to the sampling site. There is also a signi cant contribution (c. 35%) from the Jurassic rocks in the east of this catchment, and again this partly re ects the closeness of this source to the sampling site. In the Ure and Nidd catchments there are no Jurassic rocks, and Carboniferous geology is the main source of the suspended sediment load in these rivers. This is in accordance with the dominance of Carboniferous rocks in these catchments, which underlie c. 79% and 67% of the catchments, respectively. There is also good agreement with the results of the geological source groups described above and those for source type presented in Table III. In the case of the River Wharfe, the results presented in Table VI relate to only two geological groups, since there are no Jurassic rocks outcropping in this river basin (Figure 1). In this basin, the area underlain

13 FINGERPRINTING SUSPENDED SEDIMENT SOURCES 967 Figure 3. The relative contribution of the three main geological zones (Carboniferous, Permain and Triassic, and Jurassic) in the study basin to a selection of suspended sediment samples collected from the River Ouse. The mean values represent the load-weighted averages for all samples collected from this river by Carboniferous rocks (c. 90%) is the dominant sediment source, and consequently most of the suspended sediment load is derived from this source, with c. 9% being derived from Permian and Triassic rocks. As with the contributions from the di erent source type groupings (Figure 2), the relative contributions from the individual geological groups exhibit considerable variation between the individual suspended sediment samples. Figure 3 presents the mixing model results for a selection of suspended sediment samples collected from the River Ouse, which clearly demonstrate this variability. Variations in the relative importance of the sediment contributions from the di erent geological groups could re ect inter-event variability in the spatial pattern of precipitation inputs and therefore in storm runo and sediment mobilization over the catchment. However, the timing of sample collection within a storm runo event will also be important, since the routing of sediment contributions from di erent parts of the catchment will cause these contributions to pass the sampling site at di erent times. By virtue of its closer proximity to the catchment outlet, the contribution from the zone underlain by Permian and Triassic rocks is likely to predominate during the early part of a ood event, whereas the relative contribution from areas underlain by Carboniferous rocks is likely to increase during the latter part of an event because of the greater distance of these areas from the catchment outlet. Because the three main geological groupings approximately coincide with the three main physiographic zones of the study area (Pennines, Vale of York and North York Moors), which frequently receive di ering amounts of precipitation during storm events, it is possible to compare the results of the geological source tracing for the Ouse basin with the precipitation records for representative gauges in each of the three regions. The latter information will a ord an independent guide as to whether a particular geological group is likely to be of increased importance as a sediment source, by virtue of receiving a greater precipitation input. Figure 4 illustrates typical examples where rainfall in the 72 hours preceding sample collection was either concentrated in the Pennines or more evenly distributed over the whole basin. The two sediment samples collected on 25 November 1995 and 22 January 1995 (Figure 4A and B) provide examples of the former case and these are dominated (455%) by sediment derived from the Pennines. For the samples collected on 13 February 1996 and 20 December 1996 (Figure 4C and D) the precipitation input was more uniformly distributed over the three geological zones and the suspended sediment at the time of sampling contained substantial contributions from all three geological groups. In all four cases, therefore, the relative contributions of sediment from the three geological source groups is consistent with the relative magnitude of the precipitation amounts recorded for each source group.

14 968 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS Figure 4. The relative contribution of the three main geological areas to individual suspended samples collected during ood events characterized by precipitation falling primarily over the Pennines (A and B) and by precipitation falling across the entire catchment (C and D). The 72-hour precipitation totals represent the sum of the daily precipitation for the day on which the samples were collected and for the two preceding days and relate to a representative rain gauge located within each of the three geological areas, namely Carboniferous (NGR SD970724), Permian and Triassic (SE521684), and Jurassic (SE706873) Sediment source ascription using suspended sediment from tributary rivers in the Ouse basin In addition to comparing suspended sediment samples with source materials, it is possible to compare suspended sediment samples collected at downstream monitoring sites with suspended sediment samples collected from upstream tributaries, in order to obtain information on the spatial location of their source. The main advantages of comparing the properties of suspended sediment collected from the downstream (non-tidal) limit of a river with equivalent information regarding the properties of suspended sediment samples collected from tributary sites is that problems associated with di erences in particle size composition and organic matter content are reduced, and potential problems associated with tracer property transformations during transport are likely to be signi cantly less than those associated with comparing suspended sediment with source materials. Here, suspended sediment samples collected from the primary suspended sediment sampling sites on the three main tributaries upstream of the tidal limit of the River Ouse (Swale, Ure and Nidd) are used to determine the relative contributions from each subcatchment to sediment samples collected from the River Ouse near York. Similarly, suspended sediment samples collected from minor tributaries within the Ouse Basin representative of the three di erent geological zones are used to determine the relative contribution from these three zones to the suspended sediment load of the River Ouse (see Figure 1).

15 FINGERPRINTING SUSPENDED SEDIMENT SOURCES 969 Table VII. The results of applying the Kruskal±Wallis test to assess the ability of each ngerprint property to distinquish between the three main subcatchments of the River Ouse and minor tributaries representing the three main geologies (Carboniferous, Permian and Triassic, and Jurassic). See Table I for property units Tracer property Main tributaries Minor tributaries H value* P value H value* P value 137 Cs { { 226 Ra { Unsupported 210 Pb { { N { C { { Total P { Inorganic P { { Organic P { w lf { { w fd { SIRM { Al { { Ca { Cr { Cu Fe { K { { Mg { { Mn { { Na { Ni Pb { Sr { { Zn { { * Critical H value ˆ 599. { Signi cant at p ˆ 005. (a) Main subcatchments. The results of applying the Kruskall±Wallis test and discriminant function analysis to the suspended sediment samples collected from the primary sampling sites on the three main tributaries are presented in Tables VII and VIII, respectively. Of the 24 tracer properties tested 18 were able to discriminate suspended sediment samples collected from the downstream reaches of the three main tributaries. The composite ngerprint selected by stepwise discriminant function analysis contained ve Table VIII. The results of using stepwise discriminant function analysis to identify which set of tracer properties provides the best composite ngerprint to discriminate suspended sediment samples from major and minor tributaries in the Ouse basin. See Figure 1 for sampling locations Main tributaries Minor tributaries Tracer property Cumulative % samples classi ed correctly Tracer property Cumulative % samples classi ed correctly C Inorganic P Mg C Mn Al Cs w lf Inorganic P Mg Cr Cs 98.45

16 970 D. E. WALLING, P. N. OWENS AND G. J. L. LEEKS Figure 5. The relative contribution of the three main subcatchments (Swale, Ure and Nidd) to a selection of suspended sediment samples collected from the River Ouse. The mean values represent load-weighted averages for all samples collected from this river tracer properties and this was able to classify 100% of the tributary sediment samples correctly. Some of the results of applying the mixing model are illustrated in Figure 5. These results emphasize the importance of the catchment of the River Swale as a sediment source and the load-weighted mean values indicate that nearly 82% of the suspended sediment collected from the River Ouse near York originated from the River Swale, with the rivers Ure and Nidd contributing c. 15 and 3%, respectively. These gures can be compared with the estimates of the annual suspended sediment loads of the three tributaries for the years 1995 and 1996, based on continuous monitoring of discharge and turbidity (cf. Wass et al., 1997; Wass and Leeks, 1999), which are presented in Table IX. The average annual sediment load for the River Swale represents about 54% of the total load of all three tributaries, while the equivalent values for the rivers Ure and Nidd are c. 37% and 10%, respectively. Although these proportions give the same rank order for the contributions of the three main tributaries as provided by the mixing model results, the magnitudes of the relative contributions di er. These di erences may partly re ect the di erent periods of record involved, since the sediment samples were collected during the period November 1994±February 1997, whereas the sediment load data relate to the years 1995 and However, the di erences are more likely to re ect the di erent nature of the two sets of results. Those based on the annual load data refer to the total mass of sediment transported by the rivers, whereas those based on the ngerprinting study refer only to the sediment associated with the samples collected from the sampling sites on the River Ouse. These were generally collected at or soon after the hydrograph peak when sediment from the River Swale, the most important subcatchment in terms of both discharge and sediment load, is Table IX. Estimated annual suspended sediment loads for the study rivers for the years 1995 and 1996 based on continuous monitoring of discharge and turbidity (P. Wass, personal communication) Wharfe Ouse Swale Ure Nidd Area (km 2 )* Sediment load 1995 (t) Sediment load 1996 (t) Average load (t year 1 ) Average speci c suspended sediment yield (t km 2 year 1 ) * Catchment area to river gauging station.