The Science of the Total Environment 314 316 (2003) 513 534 Fingerprinting suspended sediment sources in a large urban river system Julie Carter, Philip N. Owens *, Desmond E. Walling, Graham J.L. Leeks a,1 a, a b a Department of Geography, University of Exeter, Exeter, Devon EX4 4RJ, UK b Centre for Ecology and Hydrology, Wallingford, Oxfordshire OX10 8BB, UK Accepted 1 January 2003 Abstract Very few studies have attempted to quantify the sources of suspended sediment transported in urban river systems. In this study, statistically verified composite fingerprints and a multivariate mixing model have been used to identify the main sources of the suspended sediment transported by the River Aire and its main tributary, the River Calder. Because of the polluted nature of the AireyCalder catchment and its effect on fingerprint property concentrations, source tracing was undertaken separately for the upper and lower reaches. The mean contributions from individual source types (i.e. surface materials from woodland, uncultivated and cultivated areas, channel bank material, road dust and solids from sewage treatment works) varied between the upper and lower reaches of the rivers, reflecting the change in land use from primarily pasture and moorland in the upper reaches to mainly urban areas (with some cultivated land) in the lower reaches. The suspended sediment in the upper reaches of the River Aire originates largely from channel bank sources (43 84%) and from uncultivated topsoil (16 57%). In the lower reaches of the AireyCalder system, local sources of cultivated topsoil contribute 20 45% of the suspended sediment load and there is a significant contribution from urban sources, such as road dust (19 22%) and solids from sewage treatment works (14 18%). In the upper reaches, the proportion of sediment derived from each of the two main geological areas corresponds broadly to the proportion of the catchment occupied by each geological area. The relative contribution from the Rivers Aire and Calder to the suspended sediment load transported below the confluence demonstrates that most of the sediment is derived from the River Calder. 2003 Elsevier Science B.V. All rights reserved. Keywords: Suspended sediment; Sediment sources; Fingerprinting; Urban sources; Mixing model 1. Introduction *Corresponding author. National Soil Resources Institute, Cranfield University, North Wyke, Okehampton, Devon EX20 2SB, UK. Tel.: q44-1837-883524; fax: q44-1837 82139. E-mail address: philip.owens@bbsrc.ac.uk (P.N. Owens). 1 Present address: Institute of Water and Environment, Cranfield University, Silsoe, Bedfordshire MK45 4DT, UK. The suspended sediment load transported by a river commonly represents a mixture of sediment derived from different locations and different sediment source types within the contributing catchment. Information on suspended sediment provenance is an important requirement in the 0048-9697/03/$ - see front matter 2003 Elsevier Science B.V. All rights reserved. doi:10.1016/s0048-9697(03)00071-8
514 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 examination of sediment routing and delivery, and in the construction of catchment sediment budgets (Dietrich and Dunne, 1978; Walling and Webb, 1983). From a management perspective, there is also a need to identify sediment sources in order to implement appropriate strategies to control sediment mobilisation and subsequent siltation of river channels and reservoirs. There is also increasing awareness of the role of suspended sediment as a vector for the transport of sediment-associated nutrients and contaminants (e.g. heavy metals, phosphorus, PCBs) in fluvial systems (Horowitz, 1991; Owens et al., 2001). The source of the sediment influences its chemical and physical properties and its contaminant loading, and as such is an important consideration in the management of polluted river systems. The precise type of information required on sediment source depends 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 different land use types, erosion of channel banks or solids from urban runoff) and spatial location (e.g. which tributary or part of the basin) is frequently required (Walling et al., 1999). Information on sediment provenance has traditionally been collected using direct monitoring techniques (cf. Loughran and Campbell, 1995). However, studies employing erosion pins and troughs to estimate soil erosion rates, or sediment load measurements to quantify the relative contribution of suspended sediment from different source areas within a catchment, face a number of problems (Peart and Walling, 1986) and fingerprinting techniques are increasingly being used as an alternative (e.g. Walling and Woodward, 1995; Collins et al., 1998; Bottrill et al., 1999). Advantages include the limited field data collection involved (i.e. collection of source material and suspended sediment samples as opposed to extensive longterm monitoring at a number of sites). The basic principle underlying the fingerprinting approach is that different potential sediment sources can be characterised, or fingerprinted, using a number of diagnostic physical and chemical (and possibly biological) properties, and that comparison of these fingerprints with equivalent information for suspended sediment samples permits the relative importance of the different potential sources to be determined (Oldfield et al., 1979; Walling and Woodward, 1995). Many different physical and chemical properties have been successfully used to discriminate potential sediment sources in drainage basins, including mineralogy (Klages and Hsieh, 1975; Johnson and Kelley, 1984), sediment chemistry (Wall and Wilding, 1976; Peart and Walling, 1986), mineral magnetism (Oldfield et al., 1979; Slattery et al., 2000) and environmental radionuclides (Walling and Woodward, 1992; He and Owens, 1995). The use of a single diagnostic property is now recognised as being inadequate to discriminate between the wide range of sediment sources found in river catchments, and composite fingerprinting procedures, based on a number of different properties, have been developed (Collins et al., 1998; Walling et al., 1999; Collins and Walling, 2002). This approach involves the use of a number of different diagnostic properties to establish statistically verified composite fingerprints, and the subsequent application of a multivariate mixing model to establish the relative importance of the different potential sources (He and Owens, 1995; Walling and Woodward, 1995; Collins et al., 1998; Walling et al., 1999). This has the advantage of permitting the discrimination of a greater range of sediment sources and is likely to prove more effective in establishing source sediment linkages by reducing the possibility of spurious matches that may occur with the use of individual tracers. Although the fingerprinting approach is being increasingly used to establish sediment sources in agricultural drainage basins in contrasting environments (e.g. Collins et al., 1998, 2001; Bottrill et al., 1999; Walling et al., 1999; Owens et al., 2000; Slattery et al., 2000; Russell et al., 2001), very few studies have attempted to determine suspended sediment sources in urbanised and industrialised catchments. Notable exceptions include the work of Charlesworth et al. (2000) and Charlesworth and Lees (2001), although their studies have tended to focus on small (i.e. usually -10 km 2 ) catchments. As far as we are aware, there have been no previous attempts to determine suspended sediment sources in a large (i.e. )1000 km 2 ) highly urbanised catchment using composite fin-
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 515 gerprints. This paper reports the use of the fingerprinting approach to establish the relative importance of different potential suspended sediment sources, defined in terms of both source type (i.e. agricultural and urban sources) and spatial location (i.e. different geological zones and individual sub-basins), within the heavily polluted drainage basin of the River Aire in Yorkshire, UK. To the best of our knowledge, this study represents the first attempt to include urban sources of road dust and solids from sewage treatment works (STWs) within the composite fingerprinting approach. 2. Study basin and methods 2.1. Study basin The River Aire is a tributary of the River Ouse, which drains into the North Sea via the Humber Estuary (Fig. 1). It has a total catchment area of 2 1932 km above the tidal limit, and a long-term (1958 1998) mean annual discharge of 35.4 m 3 s at the Environment Agency (EA) gauging station at Beal (NERC, 2001). The River Calder is the main tributary of the River Aire and has a 2 catchment area of 930 km and a long-term (1988 3 1998) mean annual discharge of 19.0 m s at the EA gauging station at Methley, thus contributing over half of the discharge of the lower Aire (NERC, 2001). Both rivers rise on land dominated by pasture and rough grazing, where water quality is generally good, except for small discharges of agricultural effluent and some limited diffusesource pollution. However, in the middle and lower reaches, the AireyCalder system drains a heavily urbanised and industrialised catchment with a population of approximately 2 000 000 people. The main industries in the catchment include wool, textiles, chemicals, engineering, and food and drink production. Most of the industrial effluent is treated by sewage treatment works (STWs), although some industries have consent for trade effluent to be discharged directly to the river. Consequently, the middle and lower reaches of both rivers are heavily polluted and receive discharges of sewage effluent from STWs, combined sewer overflows and sewer dykes (CSOSD), and industrial discharges (direct and indirect), in addition to inputs from road drainage. This transition from a non-polluted upstream river to a heavily polluted system in downstream reaches provides an ideal opportunity to investigate contrasts in sediment source between rural and urban areas. The underlying geology of the upper reaches of the AireyCalder system is Carboniferous limestone and millstone grit (Fig. 1). The middle and lower reaches are underlain by Carboniferous coal measures, while below the confluence of the two rivers this gives way to Permian magnesian limestone. The soils in the upper reaches are dominated by raw oligo-fibrous peats, and stagnohumic and stagnogley soils. These give way to typical brown earths and pelo-stagnogley soils in the middle and lower reaches. The narrow band of Permian magnesian limestone at the catchment outlet is overlain by typical brown calcareous earths, which are frequently cultivated. 2.2. Field sampling Bulk suspended sediment samples were collected from a number of locations throughout the AireyCalder catchment (Fig. 1). Over 70 bulk suspended sediment samples were collected between November 1997 and January 1999. Because most of the suspended sediment load of the River Aire is transported during a few high flow events (cf. Wass and Leeks, 1999), sediment samples were collected primarily during these events. As an example, Fig. 2 shows the time of sample collection at Beal in relation to river flow at this site. It is clear from Fig. 2 that sediment samples were collected during most of the high flow events that occurred during the sampling period. Furthermore, the samples collected throughout the Aire catchment encompassed a range of suspended sediment concentrations (range ;10 400 mg l, which is typical for this river; cf. Wass and Leeks, 1999) and likely seasonal, inter- and intra-storm variations in sediment properties. The sediment samples collected are, therefore, considered to be reasonably representative of the suspended sediment load transported past each site. The sediment samples were collected from the centre of the channel, using a submersible
516 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 Fig. 1. Location map showing the study area, the suspended sediment and floodplain sampling sites and the main geological subdivisions. pump powered by a portable generator, to fill several 25-l acid-washed polyethylene containers. The suspended sediment was recovered from the bulk samples by continuous flow centrifugation and freeze-dried prior to analysis. At one site (Allerton), where no suspended sediment samples were collected, overbank floodplain deposits were used as a surrogate (cf. Bottrill et al., 1999). Such
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 517 Fig. 2. Variation in flow of the River Aire at Beal during the study period, and the timing of the collection of suspended sediment samples. deposits were collected using Astroturf mats (cf. Lambert and Walling, 1987), which were deployed prior to overbank flooding and collected soon after the floodwaters had receded. In order to characterise potential source materials, ca. 150 source material samples ()500 g) were collected throughout the study area using a stainless steel trowel. These were collected over a period of 12 months in order to take account of seasonal fluctuations in tracer properties. Source material sampling was stratified to incorporate geological and land-use variations. Within each of the main geological zones (cf. Fig. 1), representative samples were collected from the surface (top ;2 cm) of woodland, uncultivated (i.e. pasture, rough grazing and moorland) and cultivated areas, and also from the faces of eroding channel banks. Previous fingerprinting studies have not included urban areas as a potential sediment source. This was clearly an important consideration in an urban catchment (Stone and Marsalek, 1996), and therefore samples of road dust and of solids from a STW were also collected. Samples of road dust in the most built-up areas were swept into labelled polythene bags. Care was taken to sample areas located close to drains to ensure that the sediment was representative of sediment likely to enter the river. Due to logistical and safety problems associated with collecting samples of STW effluent discharged to the River Aire during high-flow conditions, and the fact that the river level is usually above the level of the STW discharge pipes, samples (ns3) of the influent to Esholt STW (a large STW that serves the city of Bradford; see Fig. 1) were collected. Such samples are likely to be representative of the material discharged into rivers, particularly during storm events, when storm bypass channels mean that sewage influent is discharged directly to the river (Owens and Walling, 2002). All source material samples were dried at 40 8C, gently disaggregated and then dry-sieved to -63 mm to facilitate direct comparison with fluvial (suspended and overbank) sediment samples (Collins et al., 1998; Walling et al., 1999). 2.3. Laboratory analysis Laboratory analysis of both source material and suspended sediment samples was undertaken for a range of potential fingerprint properties. Phosphorus (P) concentrations (total, organic and inorganic) were determined using a Pye Unicam SP6 UVy visible spectrophotometer after chemical extraction (Mehta et al., 1954). Metal concentrations were determined using a Unicam 939 atomic absorption spectrophotometer after acid (concentrated HCl and HNO 3) digestion (Allen, 1989). Carbon (C) and nitrogen (N) concentrations were measured using a Carlo Erba ANA 1400 analyser. Radionuclide concentrations wcaesium-137 ( Cs), 137 radium- 226 210 226 ( Ra) and unsupported lead-210 ( Pb)x were determined using an EG&G Ortec hyper-pure germanium well-type detector. In order to correct for further contrasts in grain size composition between source materials and suspended sediment, the specific surface area of the samples was estimated from their particle size distributions. The latter were determined using a Coulter LS130 laser diffraction granulometer, after removal of organic matter (by H O ) and chemical w(napo ) x and 2 2 3 6 ultrasonic dispersion. 2.4. The fingerprinting approach In order to exploit fully the potential of particular properties to differentiate potential source
518 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 types, firm verification is needed that a property is capable of discriminating between the source groups identified. A two-stage statistical verification procedure was used to produce multicomponent signatures for the discrimination of different source types and areas (Collins et al., 1998; Walling et al., 1999). First, either the Mann Whitney U-test (to distinguish between two potential sources) or the Kruskal Wallis H-test (to distinguish between three or more potential sources) was used to establish which properties 137 (i.e. P, Cr, Cs) exhibited significant differences between the sources (i.e. cultivated topsoil, channel bank material, street dust). Properties failing this test were removed from the subsequent analysis. Secondly, multivariate discriminant function analysis was applied to the properties selected in the first stage, in order to identify the set of properties or composite fingerprint that afforded the best discrimination between source groups. A stepwise selection algorithm, based on the minimisation of Wilks lambda, was used in this analysis. 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 successfully quantified the provenance of suspended sediment in predominantly agricultural catchments (Collins et al., 1998, 2001; Bottrill et al., 1999; Walling et al., 1999; Owens et al., 2000; Russell et al., 2001). Suspended sediment samples are generally enriched in fines compared to source materials (Walling et al., 2000), resulting in sediment samples exhibiting higher concentrations of many constituents than the source material (Horowitz, 1991). The effect of contrasts in grain size composition between source materials and suspended sediment was partly addressed by restricting analysis to the -63-mm fraction, but further correction was required to make the sediment samples directly comparable with the source materials. The values for the fingerprint properties of samples from each source were therefore corrected for differences in particle size composition compared with the suspended sediment being traced, using the ratio of the specific surface area of each suspended sediment sample to the mean value for each source (e.g. cultivated topsoil or limestone geology or Calder sub-basin). This provides a simple and convenient means of correcting all of the fingerprint property values for each source (Walling et al., 1999). Peart and Walling (1986) noted that suspended sediment is also likely to be enriched in organic matter, which may then act as a scavenger for many elements (Horowitz, 1991). Consequently, Collins et al. (1998) suggested that it was necessary to include correction for organic matter content in the model. However, the relationship between organic matter content and heavy metal content is complex and difficult to generalise (Walling et al., 1999), and such correction was not included in this study, since it may also result in the over-correction of tracer properties. Mean values of the particle size-corrected tracer properties for each source were used in the mixing model. Due to the wide range of discharges and suspended sediment concentrations when samples were collected (cf. Fig. 2), the relative contributions from each source for the individual sediment samples were weighted according to the values of discharge and suspended sediment concentration at the time of sampling (cf. Walling et al., 1999; Owens et al., 2000). This sediment load-weighting approach ensures that the importance of source contributions associated with periods of high sediment load is emphasised, and therefore 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. However, for the purposes of this study, continuous discharge and turbidity data were only available at the most downstream sites on each river, and consequently mean values for the individual samples were used at other sites. The goodness-of-fit provided by the mixing model was assessed by comparing the actual fingerprint property concentrations found in the suspended sediment samples with the corresponding values predicted by the mixing model, using the procedures described in Collins et al. (1998). The mean (average for all properties within each composite fingerprint) relative errors for the mixing
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 519 Table 1 The results of using the Mann Whitney test to assess the ability of each tracer property to discriminate between source materials from areas underlain by limestone and millstone grit in the AireyCalder catchment Tracer property N 0.083 C 0.071 As 0.000 * Pb 0.025 * Zn 0.000 * Cr 0.441 K 0.002 * Ca 0.006 * Mn 0.003 * Mg 0.869 Na 0.005 * Cu 0.001 * Al 0.036 * Fe 0.091 Total P 0.007 * Inorganic P 0.024 * Organic P 0.018 * 137Cs 0.135 226Ra 0.001 * 210 Unsupported Pb 0.005 * * Significant at Ps0.05. model calculations typically ranged between 8 and 20%. This indicates that the mixing model provides reasonably accurate predictions of the concentrations of fingerprint properties associated with individual suspended sediment samples in the Rivers Aire and Calder. It is estimated that the errors associated with the source tracing results are of the order of approximately "15%, and this should be borne in mind when interpreting the results presented below. The validity of the composite fingerprinting technique depends heavily on the assumption that the concentrations of properties measured in suspended sediment can be directly compared with those of the same properties in a variety of source materials. However, many substances, such as organic constituents and a number of heavy metals, may be introduced into a river from point sources as it flows downstream. Several studies (Neal et al., 1999; Carton et al., 2000; Owens et al., 2001; Owens and Walling, 2002) have shown that the Rivers Aire and Calder are enriched in a number P of sediment-associated nutrients (P) and contaminants (certain heavy metals) relative to most other rivers in the UK. This affects the effective use of the composite fingerprint, as some properties may no longer directly reflect the original source of the sediment. For this reason, different composite fingerprints have been employed in the upstream and downstream reaches of the study rivers. 3. Results and discussion 3.1. Sediment source ascription in the upstream reaches 3.1.1. Source area (geology) The Mann Whitney test was used to assess the ability of tracer properties to discriminate between the two geological areas that occur in the upper reaches of the Aire catchment (cf. Fig. 1), namely Carboniferous limestone and millstone grit (Table 1). The majority of tracer parameters exhibit P- values well below the significance value of 0.05, indicating that they can strongly discriminate between the two geological areas. Six parameters 137 (N, C, Cr, Mg, Fe and Cs) were found to be not significant in making the discrimination, and were therefore removed at this stage. Multivariate discriminant function analysis was subsequently used to identify the best composite fingerprint incorporating a number of these properties, and the results are shown in Table 2. The optimum multicomponent fingerprint, comprising K, Cu, As, Mn, Na and total P, was able to distinguish Table 2 Results of using stepwise discriminant function analysis to identify which combination of tracer properties provides the best composite fingerprint for discriminating source materials on the basis of geology (i.e. limestone or millstone grit) Tracer property K 69.8 Cu 74.6 As 87.3 Mn 92.1 Na 88.9 Total P 96.1 Cumulative geology samples classified correctly (%)
520 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 Table 3 Mean contributions of each geological area to the suspended sediment samples collected from the upper reaches of the River Aire during the period November 1997 January 1999 River Site Number of Geological area contribution (%) sediment samples Limestone Millstone grit Otterburn Beck Bell Busk 2 100 (100) 0 (0) Aire Bell Busk 5 100 (100) 0 (0) Eller Beck Skipton 3 100 (100) 0 (0) Aire Kildwick 18 56 (59) 44 (41) Aire Apperley 12 13 (38) 87 (62) The approximate surface areas underlain by each geological area are listed in parentheses. correctly 96% of the source area samples. The addition of further tracer properties to the composite fingerprint does not increase the success of the classification. The numerical mixing model was used to establish the relative contribution of each source to the individual suspended sediment samples collected from the upstream sites on the River Aire. The mean results for each site are presented in Table 3. In the catchment of the River Aire above Bell Busk and the Eller Beck at Skipton, there are no millstone grit outcrops, and consequently the area underlain by limestone is the only source of the suspended sediment loads at these sites. At Kildwick, 56% of the sediment originates from the limestone area. This is in close agreement with the area occupied by this geology (approx. 59%). At Apperley there is a decrease in the proportion of sediment derived from the limestone area. This reflects the increase in the proportion of land underlain by millstone grit above this sampling site. The fact that the millstone grit at Apperley supplies 87% of the suspended sediment load while only occupying approximately 62% of the area may reflect higher rates of erosion and sediment supply associated with the millstone grit. However, the apparent importance of the millstone grit rocks may simply reflect the timing of suspended sediment sampling in relation to the storm hydrograph. The considerable variation in the relative contributions from the different geological areas associated with the individual suspended sediment samples collected at Kildwick and Apperley is shown in Fig. 3. The proportion of sediment supplied by the limestone area reached 80% at Kildwick and 40% at Apperley. The results highlight the importance of the timing of suspended sediment sampling, since the routing of sediment contributions from different parts of the catchment causes these contributions to pass the sampling Fig. 3. Inter-storm variability in the relative contribution of areas underlain by limestone and millstone grit to suspended sediment samples collected from Kildwick and Apperley.
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 521 Fig. 4. Storm-period variation in the relative contribution of areas underlain by limestone and millstone grit to suspended sediment samples collected from Kildwick. site at different times. Fig. 4 shows the intra-storm variations in sediment source areas for samples collected from Kildwick during two storm events. During these events, the proportion of sediment derived from the limestone area peaks on the rising limb or at the peak, and then decreases during the falling limb. This reflects the close proximity of the site to the area underlain by limestone. Fig. 5 shows selected intra-storm variation in sediment source at Apperley. During these events, it appears that the proportion of sediment derived from the limestone area increases during the latter stages of an event, as sediment from more distant parts of the catchment underlain by limestone takes longer to reach the site. 3.1.2. Source type (land use) The source materials collected from the upstream areas were also classified according to source type. Previous authors (Collins et al., 1998; Walling et al., 1999; Owens et al., 2000) have divided sources into four categories (i.e. surface material from woodland, uncultivated and cultivated areas, and material from channel banks). There is very little cultivated land in the upper reaches of the AireyCalder catchment, and because it is unlikely that it would have an influence on the supply of sediment to the river, this source type was removed from the analysis at this stage. The results of the Kruskal Wallis H-test are shown in Table 4. A total of 14 out of 20 tracer properties provided clear discrimination between surface material from woodland and uncultivated (moorlandypasture) areas and channel bank material. A multicomponent signature containing organic P, 137 Cs, Mg, K, Mn and Fe was selected as the optimum fingerprint capable of classifying 87% of the source material samples correctly (Table 5). Summary results for the use of the mixing model are presented in Table 6. For the suspended sediment samples collected from the Otterburn Beck at Bell Busk and the River Aire at Kildwick and Apperley, the contribution from surface material from woodland areas is zero. This finding is consistent with that obtained for other Yorkshire rivers (e.g. Walling et al., 1999) and reflects the Fig. 5. Storm-period variation in the relative contribution of areas underlain by limestone and millstone grit to suspended sediment samples collected from Apperley.
522 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 Table 4 Results of applying the Kruskal Wallis test to assess the ability of each tracer property to discriminate between surface material from uncultivated and woodland areas, and channel bank material collected from the AireyCalder catchment Tracer property H P N 39.723 * 0.000 C 32.567 * 0.000 As 2.579 0.275 Pb 9.704 * 0.008 Zn 8.647 * 0.013 Cr 0.921 0.631 K 23.729 * 0.000 Ca 3.279 0.194 Mn 0.361 0.835 Mg 14.828 * 0.001 Na 0.189 0.910 Cu 0.237 0.888 Al 9.150 * 0.010 Fe 8.966 * 0.011 Total P 18.639 * 0.000 Inorganic P 17.709 * 0.000 Organic P 22.054 * 0.000 137Cs 37.570 * 0.000 226Ra 16.761 * 0.000 210 Unsupported Pb 28.267 * 0.000 * Significant at Hs5.99. limited spatial extent of woodland in the catchment and the lack of surface erosion from such areas. For the River Aire at Bell Busk and the Eller Beck at Skipton, small contributions from woodland sources are identified, and this is probably due to the existence of areas of woodland in close proximity to the sampling sites. The predominant sediment source at Bell Busk, for both the Otterburn Beck and the River Aire, and at Kildwick is Table 5 Results of using stepwise discriminant function analysis to identify which combination of tracer properties provides the best composite fingerprint for discriminating between surface materials from woodland and uncultivated areas, and channel bank material Tracer property Organic P 6400 137Cs 7350 Mg 7960 K 8250 Mn 8510 Fe 8720 Cumulative source type samples classified correctly (%) material from eroding channel banks. The contribution from channel banks is higher than that reported previously for studies in the UK (Collins et al., 1998; Walling et al., 1999; Owens et al., 2000), for which values have typically ranged between ;10 and 40%. The high contribution of sediment from channel banks in the upper reaches of the River Aire can be partly explained by the absence of cultivated land. This means that the actual amount of sediment supplied by channel banks may not be substantially greater than in other catchments, but because there is a lack of cultivated land as a sediment source, the proportion contributed by eroding channel banks is increased. However, there is some evidence of severe bank erosion in the upper reaches of the catchment. In addition, the greater than average rainfall during the study period may have contributed to an above average incidence of bank col- Table 6 Mean contributions of each source type to the suspended sediment samples collected from the upper reaches of the River Aire during the period November 1997 January 1999 River Site Number of Source type contribution (%) sediment samples Channel Uncultivated Woodland bank topsoil a topsoil Otterburn Beck Bell Busk 2 84 16 0 Aire Bell Busk 5 50 48 2 Eller Beck Skipton 3 45 47 8 Aire Kildwick 18 55 45 0 Aire Apperley 13 43 57 0 a Mainly moorland and pasture.
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 523 Fig. 6. Inter-storm variability in the relative contribution of surface material from uncultivated and woodland areas, and channel bank material to suspended sediment collected from the upper reaches of the River Aire (at Bell Busk, Kildwick and Apperley) and two upstream tributaries (Otterburn Beck and Eller Beck). lapse and channel scour by increasing the erosive potential of the river during periods of high discharge. The contribution of surface material from uncultivated areas to the suspended sediment load is substantial and reflects the large amount of pasture and moorland in the upper catchment. The contribution from channel banks to the suspended sediment samples collected at Apperley is lower than at the other sites, as many channels below Kildwick are protected and bank erosion is not as common. The eroded bank material from upstream reaches is, therefore, diluted by substantial inputs of surface materials from uncultivated areas in the vicinity of Apperley. The mean values reported in Table 6 conceal significant inter- and intra-storm variations in source type, which are highlighted in Fig. 6. Appreciable temporal variation in the relative importance of sediment sources was also noted by He and Owens (1995), Walling and Woodward (1995) and Walling et al. (1999). Such variations reflect antecedent conditions and changes in landuse and land-cover between events, exhaustion of sources as an event proceeds, and the timing of sampling in relation to the hydrograph peak. For the River Aire at Bell Busk, Kildwick and Apperley, there is evidence to suggest that events during the summer months contribute smaller proportions
524 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 Fig. 7. Storm-period variation in the relative contribution of surface material from woodland and uncultivated areas, and channel bank material to suspended sediment samples collected from Kildwick. of bank material. This may relate to the absence of freeze thaw processes, which prepare sediment for entrainment during the winter months. Alternatively, it may reflect the fact that discharges are lower in the summer months, and thus have less potential for entraining sediment from large areas of riverbank. However, insufficient samples were collected during the summer to statistically verify Fig. 8. Storm-period variation in the relative contribution of surface material from woodland and uncultivated areas, and channel bank material to suspended sediment samples collected from Apperley.
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 525 this trend. The intra-storm variability in source type contributions at Kildwick and Apperley is shown in Figs. 7 and 8, respectively. At Kildwick, for the events shown, there is a slight increase in the proportion of bank material supplied during the latter stages of the event. This is consistent with the findings of Walling et al. (1999) and Owens et al. (2000), who suggested that bank material is entrained at high discharges and that higher amounts of bank material can thus be expected at the discharge peak or shortly afterwards, depending on the distance from the source of the material to the sampling site. The delayed bank input could also reflect bank collapse as the water levels recede. At Apperley on the 10 December 1997, there was also an increase in the proportion of bank material supplied during the falling limb. This also coincided with an increase in the proportion of sediment derived from limestone areas (cf. Fig. 5) and again reflects the importance of upstream sources within the limestone area in the supply of channel bank material. On the 3 January 1998, the proportion of bank material was low and this also corresponds with a low proportion of sediment derived from the limestone area (cf. Fig. 3). The event occurring on the 16 October 1998 exhibited the highest proportions of bank material and also the highest proportion of sediment from limestone areas, thus highlighting the consistency between the two sets of sediment source investigations. 3.2. Sediment source ascription in the lower reaches Fingerprinting of sediment sources at the catchment outlet is complicated by the well-documented fact that the fluvial sediment in the AireyCalder catchment is contaminated by a variety of sediment-associated nutrients and heavy metals (Neal et al., 1999; Carton et al., 2000; Owens et al., 2001; Owens and Walling, 2002). Consequently, it is not appropriate to use the composite fingerprints that were used for the upstream reaches, which were derived using source materials collected from agricultural areas only. Instead, it is necessary to also incorporate likely urban sources, such as road dust and solids from STWs. Furthermore, because there is the potential problem that some tracer properties may be discharged from point sources to rivers in solution and subsequently sorb onto existing suspended sediment in the river (Owens and Walling, 2002), thereby elevating the property concentration of the suspended sediment, it is necessary to exclude properties that show an elevated concentration in suspended sediment relative to those for the various potential sources before the fingerprinting exercise is carried out. 3.2.1. Source area (geology) In order to ascribe sources to the suspended sediment collected in the lower reaches of the river, all source materials were classified according to whether they were sampled from limestone, millstone grit, coal measures or magnesian limestone. The concentrations of tracer properties in the source materials were then corrected for particle size effects by multiplying the concentration by the ratio of the mean specific surface area of the suspended sediment to that for the source material. When the mean property concentrations for the suspended sediment load were compared to the particle size-corrected concentrations in the source materials, the concentrations of eight of the 20 properties fell outside the range of values represented by the source materials. Consequently, these properties were deemed unsuitable for fingerprinting and they were excluded at this stage. The ability of the remaining properties to discriminate between the four geological areas was tested using the Kruskal Wallis H-test. Six properties were shown to discriminate between the four geological areas. However, multivariate discriminant function analysis showed that -70% of the samples could be correctly classified using these properties. For this reason it was decided that it was not viable to attempt to use these fingerprint properties to establish source area contributions in the downstream reaches of the study catchment. These problems highlight the difficulties of source ascription in a contaminated catchment. 3.2.2. Source type (land use) In order to establish the source type contributions to the suspended sediment collected at the catchment outlet, source materials were classified
526 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 Table 7 Comparisons between the concentration of tracer properties in source materials (corrected for particle size differences) and suspended sediment Property Source type Suspended sediment Channel Topsoil STW Road Beal Methley bank solids dust Cultivated Uncultivated Woodland N (%) 0.38 0.46 0.85 1.62 4.3 0.51 1.16 1.02 C (%) 7.26 6.53 11.41 28.86 62.4 9.79 13.58 13.20 As (mg g ) 10.23 10.73 11.38 11.12 14.40 14.60 14.00 Pb (mg g ) 171.29 87.84 104.77 221.66 337.63 615.46 259.00 246.60 Zn (mg g ) 355.76 142.87 157.89 199.03 1059.70 849.23 627.70 554.70 Cr (mg g ) 108.01 42.60 34.29 35.58 90.81 195.66 229.40 266.40 K (mg g ) 641.84 1259.61 1009.88 1019.32 3031.21 1168.28 922.20 813.70 Ca (mg g ) 23485.79 3730.69 2728.65 20679.04 8060.06 102647.23 3420.30 1274.40 Mn (mg g ) 1074.10 1323.56 969.21 1127.69 551.27 1827.79 1988.90 2042.30 Mg (mg g ) 3248.37 5222.89 1918.86 7615.05 5075.00 18452.85 3795.90 3416.90 Na (mg g ) 220.14 209.11 147.39 230.08 15027.22 884.29 351.90 246.60 Cu (mg g ) 92.86 165.52 67.06 103.43 490.76 1020.28 179.30 162.50 Al (mg g ) 10362.62 8702.60 7883.66 7423.33 32622.08 23700.93 10429.50 10420.70 Fe (mg g ) 29100.23 25084.93 25888.46 20556.56 21025.60 32814.46 36185.50 40988.10 Total P (mg g ) 1710.30 1426.59 1682.37 1576.28 22204.16 1750.11 8275.90 6644.00 Inorganic P (mg g ) 1259.24 931.22 972.91 693.15 17498.49 1453.23 5798.90 5353.40 Organic P (mg g ) 451.00 495.37 709.52 883.20 4705.67 296.89 1476.90 1290.80 137 Cs (mbq g ) 24.62 16.40 55.63 200.59 27.90 19.34 29.70 25.40 226 Ra (mbq g ) 86.99 61.54 64.25 71.94 91.31 113.80 84.10 84.10 Unsupported 15.45 13.12 51.36 144.92 223.59 153.02 101.70 83.60 210 Pb (mbq g ) Values in bold italics represent concentrations in suspended sediment that lie outside the range associated with source materials. according to whether they were surface materials from woodland, uncultivated (pastureymoorland) or cultivated areas, channel bank material (i.e. agricultural sources), solids from STWs or road dust (i.e. urban sources). The source material properties were then corrected for particle size differences (Table 7). Average concentrations for suspended sediment for 16 out of 20 properties fell within the range for the source materials and were consequently subjected to the Kruskal Wallis H-test (Table 8). All of the 16 properties were able to distinguish between the six source types, and subsequent analysis produced a multicomponent signature containing Zn, C, N, unsupported 210 137 Pb, Cs and total P. This fingerprint classified 76% of the source type samples correctly (Table 9). The mean load-weighted results provided by the mixing model for the River Aire at Beal and the River Calder at Methley are presented in Table 10. For the River Aire at Beal, the dominant sediment source is from channel banks (approx. 33%). This reflects the importance of the erosion of channel banks in the downstream reaches of the river, where banks are often )2 m in height. It also reflects the downstream location of the sampling sites, and thus the distal location of many topsoil sources, particularly pastureymoorland and woodland areas, which are mainly located in upstream areas. Due to their distal location, the opportunity for conveyance losses is greater. These findings are consistent with those documented by Walling et al. (1999) and Owens et al. (2000) for downstream reaches of the Rivers Ouse and Tweed, UK, for which the contributions from channel bank sources were approximately 37 and 39%, respectively. The increased importance of cultivated (approx. 20%) over uncultivated (pastureymoorland) topsoil sources (ca. 7%) reflects the existence of large areas of intensively cultivated land in close proximity to the sampling sites and the high rates of soil loss commonly associated with cultivated soils (cf. Morgan, 1986). The location
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 527 Table 8 Results of using the Kruskal Wallis test to assess the ability of each tracer property to discriminate between surface material from uncultivated, cultivated and woodland areas, channel bank material, solids from STWs and road dust collected from the AireyCalder catchment Tracer property H P N 85.07 0.000 C 77.29 0.000 Pb 37.52 0.000 Zn 47.89 0.000 K 49.37 0.000 Ca 38.19 0.000 Mg 25.18 0.000 Na 35.52 0.000 Cu 39.36 0.000 Al 44.78 0.000 Total P 25.43 0.000 Inorganic P 31.90 0.000 Organic P 39.69 0.000 137Cs 68.60 0.000 226Ra 47.41 0.000 210 Unsupported Pb 59.21 0.000 H significant at 7.82. of the main area of cultivated land in the lower reaches of the catchment minimises conveyance losses, such as those associated with floodplain deposition. There was insufficient sediment supplied by woodland sources for its contribution to be detected by the mixing model. This reflects both the limited extent of woodland in the catchment and the lack of erosion from such sources. Perhaps the most significant finding listed in Table 10 is the relative importance of urban sources to the suspended sediment samples collected at Beal. It is estimated that approximately 40% of the suspended sediment that is transported in the River Aire at Beal is derived from roads and STWs within the urbanised part of the catchment. The Table 9 Results of using stepwise discriminant function analysis to identify which combination of tracer properties provides the best composite fingerprint for discriminating source materials on the basis of source type (i.e. surface material from cultivated, uncultivated and woodland areas, channel bank material, solids from STWs and road dust) Tracer property Zn 36.1 C 47.6 N 61.2 210 Unsupported Pb 66.9 137Cs 69.2 Total P 75.9 Cumulative source type samples classified correctly (%) results for the River Calder at Methley are broadly similar to those for the River Aire at Beal, with the main difference being an increase in contributions of surface material from cultivated areas (45%) and a decrease in contributions from channel banks (18%) compared to Beal. As with Beal, a significant amount (approx. 33%) of the sediment load transported in the downstream reaches of the River Calder is derived from urban sources. The mean values shown in Table 10 again conceal many inter- and intra-storm variations in sediment source. Fig. 9 shows the source contributions for all the sediment samples collected. Insufficient samples were collected during the summer months to permit investigation of seasonal variations in sediment source. However, it is clear that considerable variation exists between different samples. Such variations are most likely to reflect the timing of sampling, as sediment is delivered from different parts of the catchment at different stages of the storm hydrograph. This is further Table 10 Load-weighted mean contributions of each source type to the suspended sediment samples collected from Beal and Methley during the period November 1997 January 1999 Site Number Source type contribution (%) of samples Channel Topsoil STW Road bank Uncultivated Cultivated Woodland solids dust Beal 18 33 7 20 0 18 22 Methley 5 18 4 45 0 14 19
528 J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 Fig. 9. Inter-storm variability in the relative contribution of surface material from uncultivated and cultivated areas, channel bank material, road dust, and solids from STWs to suspended sediment collected from the lower reaches of the Rivers Aire and Calder. Fig. 10. Variation in the relative contribution of surface material from uncultivated and cultivated areas, channel bank material, road dust, and solids from STWs to suspended sediment samples collected from Beal during a storm event during 3 4 March 1998.
J. Carter et al. / The Science of the Total Environment 314 316 (2003) 513 534 529 illustrated in Fig. 10, which shows the variation in sediment sources during a storm event at Beal. During the rising limb of the hydrograph, the relative contribution of channel bank material dominates, reflecting the entrainment of exposed bank material. As the discharge increases, the relative contribution of bank sources decreases, probably due to dilution with other sources, and the contribution of topsoil from cultivated areas and road dust increases. The increase in contributions from cultivated topsoil reflects the high erosion rates and sediment delivery associated with this land use and its close proximity to the sampling site (relative to pastureymoorland areas). The increase in the relative contribution of road dust reflects the increase in the connectivity of the road network to the channel system as the storm progresses. The decrease in the relative contributions of solids from STWs during the hydrograph may reflect the dilution of sediment derived from such sources by that from more distal parts of the catchment, including pasture topsoil, the contribution of which only becomes significant during the falling limb of the hydrograph. Fig. 10 demonstrates that variations in the relative contributions of the main sources are controlled primarily by the location of the sources within the catchment relative to the sampling site, with channel banks and STWs contributing most sediment at the start of the event, and road dust and the surface material from pastureymoorland areas contributing more towards the middle and end of the event. Table 11 Results of using the Mann Whitney test to assess the ability of each tracer property to discriminate between sediment collected from Methley and Allerton Tracer property N 0.035 * C 0.001 * As 0.690 Pb 0.095 Zn 1.000 Cr 0.008 * K 0.548 Ca 0.008 * Mn 1.151 Mg 1.000 Na 0.222 Cu 0.320 Al 0.095 Fe 0.008 * Total P 0.393 Inorganic P 0.786 Organic P 0.036 * 137Cs 0.030 * 226Ra 0.019 * 210 Unsupported Pb 0.662 * Significant at Ps0.05. 3.2.3. Spatial location (sub-basins) It is possible to compare suspended sediment samples collected from the River Aire at Beal with those collected from sites upstream of the Aire Calder confluence in order to obtain information on the relative contribution of the Rivers Aire and Calder to the suspended sediment load transported downstream of the confluence. The main advantage of such an approach (i.e. comparing suspended sediment properties from different sites) is that the complications introduced by differences in particle size composition and organic matter content between source materials and sediments are reduced (cf. Walling et al., 1999). However, because no suspended sediment samples were collected from the River Aire immediately upstream of the confluence, floodplain deposits were used in place of suspended sediment for the River Aire at Allerton. Because such deposits in essence represent suspended sediment deposited during overbank events, the property concentrations associated with overbank deposits should be representative of those of suspended sediment at the same location, once particle size effects have been taken into account. The floodplain deposits from Allerton were characterised by a similar particle size to the suspended sediment from Methley, but were, nonetheless, corrected for particle size differences in the same way as for the source materials. The Mann Whitney U-test was used to assess the ability of tracer properties to discriminate between the two rivers (Table 11). Eight tracer parameters exhibit P-values below the significance value of 0.05, indicating that they afford strong discrimination between the two rivers. A multicomponent signature containing Ca, C, Cr, Fe and organic P was subsequently identified as a fingerprint capable of classifying 100% of the samples correctly P