The scale problem in tackling diffuse water pollution from agriculture: Insights from the Avon Demonstration Test Catchment programme in England

Mitigation of diffuse water pollution from agriculture is of concern in the United Kingdom, so that freshwater quality can be improved in line with environmental objectives. Targeted on‐farm mitigation is necessary for controlling sources of pollution to rivers; a positive impact must also be delivered at the subcatchment and catchment scales before good ecological status can be achieved. A farm on the River Sem in the Hampshire Avon Demonstration Test Catchment was selected for monitoring due to its degraded farmyard, track, and drainage ditch, which was targeted by the Demonstration Test Catchment programme for improvement using a treatment train of interventions. The river was monitored before and after, upstream and downstream, of the potential sources of pollution and subsequent mitigation, both locally at farm scale, and downstream at the subcatchment scale. Sediment was obtained from the riverbed using a conventional disturbance technique, and source samples were collected from across the subcatchment. Samples were analysed for geochemistry, mineral magnetism, and environmental radionuclide activity using the <63‐μm fraction, before sediment source fingerprinting was conducted to apportion sources. Source tracing revealed that, although the degraded farm track was experiencing channelized flow and erosion in the pre‐mitigation period, it was not a major sediment source even at farm scale. Repeat source apportionment during the pre‐ and post‐mitigation periods showed that the targeted treatment train did not result in statistically significant decreases in predicted contributions from the farm track sources at either scale. Sediment sources must be determined at a range of spatial scales to support effective mitigation.

farm management, such as the timings of fertilizer spreading and overwinter housing of livestock, but can also involve improvements to farm infrastructure, such as roofing farm yards, clean and dirty water separation, resurfacing farm tracks, maintaining drainage ditches, and increasing the length and impermeability of hedgerows and riparian vegetation (e.g., Cuttle et al., 2007Cuttle et al., , 2016. However, farm-scale improvements to water quality through targeted mitigation of DWPA also need to deliver a positive impact at subcatchment and catchment scales before good ecological status can be achieved at the compliance reporting scales (e.g., WFD waterbodies) used for current policy delivery and assessment. It is important, therefore, that on-farm mitigation is effective enough to show an impact further downstream. Here, there are many common challenges for the signalto-noise effect, that is, isolating the impact of the targeted intervention from background variability in hydroclimatology, water quality, and sediment transport as landscape scale increases. Issues include targeting the most important on-farm pollutant sources and delivery pathways; the density of the on-farm measures across different landscape scales; the contribution of agricultural inputs to the water quality problem in the context of nonagricultural sources, including urban areas and domestic septic tanks; changing hydrological/biogeochemical process domains; and the maintenance of measures following implementation.
A challenge for managing DWPA concerns delivering robust empirical evidence on the efficacy of on-farm interventions at landscape scale (Lloyd, Freer, Collins, Johnes, & Jones, 2014). There is a lack of such evidence in the current literature (McGonigle et al., 2014), yet it is essential for keeping major stakeholders, including farmers, engaged in the direction of travel for environmental improvement. Here, lags in the response of conventional water quality data to targeted intervention (e.g., Boesch, Brinsfield, & Magnien, 2001;McDowell, Sharpley, & Folmar, 2003;Wang et al., 2016;Wang, Lyons, & Kanehl, 2002) pose a challenge for stakeholder engagement, because those lags can be up to decadal in duration, especially in the case of diffuse nutrient and sediment pollution.
In this context, a toolkit of monitoring methods is required to ensure that empirical data streams, with more sensitivity to targeted intervention, are collected. Against this background, sediment source fingerprinting is a useful tool for identifying the major sources of sediment and associated contaminants across scales (e.g., Collins et al., 2017;Collins, Walling, & Leeks, 1997;Collins, Walling, Webb, & King, 2010;Pulley, Foster, & Atunes, 2015;Walling, Collins, Jones, Leeks, & Old, 2006;Walling & Foster, 2016), as well as assessing the effectiveness of mitigation measures at farm and subcatchment scales by quantifying the source contribution before and after mitigation (e.g., Collins, Walling, McMellin, et al., 2010).
In England, the Demonstration Test Catchment (DTC) programme was established in December 2009 to test the efficacy of targeted onfarm interventions for water quality control at multiple (i.e., farm to landscape to catchment to national) scales (McGonigle et al., 2014). This programme is founded on testing on-farm interventions using a comparison of control and manipulated areas within a before-after-control-impact (BACI) experimental design and seeks to employ a toolkit of monitoring methods (e.g., Lloyd, Freer, Johnes, & Collins, 2016;Outram et al., 2014), rather than conventional water quality monitoring alone. More specifically, in the Hampshire Avon DTC, work as part of a PhD programme assessed the efficacy of targeted intervention at measure to landscape scales to provide valuable insight into the challenges of delivering improvements in water quality across these scales.

| Study area
The Hampshire Avon DTC drains an area of 1,700 km 2 , rising in Pewsey, Wiltshire, and flowing south into the English Channel in Christchurch, Dorset ( Figure 1). The River Avon and its tributaries are a Special Area of Conservation and a priority catchment as part of the catchment-sensitive farming programme for helping to deliver WFD environmental objectives.
The headwaters of the River Sem (~5 km 2 ), representing the Priors Farm subcatchment, were used for the study reported here because this area was identified as suffering from DWPA at the start of the DTC programme.
This subcatchment is underlain almost entirely by the Kimmeridge clay (Jurassic) formation, has slowly permeable soils (Wickham and Denchworth soil series) prone to seasonal waterlogging, and is characterized by very little topographical variation and flashy hydrology (Allen et al., 2014).
Annual average rainfall is~863 mm. Land use is dominated by dairy farming and low intensity mixed livestock grazing (91% of the subcatchment area).

| On-farm mitigation implemented by the DTC programme
The headwaters of the River Sem flow through a dairy farm (Hays Farm), before continuing downstream to a neighbouring lowland grazing farm ( Figure 2). Catchment walkover surveys at the start of the DTC programme identified a degraded farmyard (clean and dirty water separation and lack of roofing issues) and a track linking that farmyard to the stream on Hays Farm. The degraded farm track was producing and delivering sediment and associated contaminants down slope towards a drainage ditch connected to the river, as well as off a bridge crossing into the river directly (Figures 3 and 4). Targeted intervention was implemented between June and July 2013 whereby a pollution control cascade comprising the farmyard and track linking the yard to the stream was funded by the DTC programme. Work involved resurfacing the steepest (upper) part of the farm track (FTU; Figure 4) and digging a swale to one side, which was connected to a retention pond at the foot of the slope (Figure 3). The drainage ditch running beside the lower part of the degraded farm track (FTL; Figure 3) was also dredged ( Figure 5), to improve storage capacity and help reduce delivery of sediment and associated contaminants to the stream. DTC funding was not sufficient to resurface and improve FTL substantially, although the surface was rolled to remove any major erosion channels.
The banks of the drainage ditch were allowed to revegetate naturally to trap run-off and sediment from the track, encourage uptake of contaminants, and increase flow retention ( Figure 6). V-notch weirs were also installed in the drainage ditch to further increase flow retention ( Figure 5). It should also be noted that the channel banks of the River Sem through this site are steep and prone to fluvial scour during flashy run-off that characterizes this subcatchment. In 2012, before the study began, the channel banks were re-profiled and fencing was installed along either side to prevent poaching from cattle and to allow the development of a vegetated buffer. As this intervention was implemented before research began, it was not possible to analyse the differences in sediment contribution between pre-and post-mitigation; however, the change in overall contribution over time could still be examined.

| Field work
The impact of the targeted on-farm interventions at Hays Farm in the headwaters of the River Sem was monitored following the BACI approach (e.g., Roley et al., 2012;Stewart-Oaten, Murdoch, & Parker, 1986). To assess the impact of the on-farm interventions, finegrained sediment (<63 μm) stored on the riverbed was collected at sampling locations upstream (A) and downstream of the bridge crossing (B) and ditch (C) confluence, as well as further downstream at the subcatchment outlet (D) used by this study (Figure 2). Bed sediment disturbance is commonly used to provide sediment samples for the analysis of sediment properties and provenance (Duerdoth et al., 2015;Lambert & Walling, 1988;Naden et al., 2016) and was one of the methods employed in this study. A hard plastic stilling well, 70 cm in height and 50 cm in diameter, was pushed firmly into the riverbed until a seal was created within the well. The depth of the water was measured, then the water and top~5 cm of the riverbed   substrate was manually agitated for around 1 min with a wooden pole until the stored sediment was suspended in the water (e.g., Walling et al., 2003). Five 500 ml polyethylene bottles, secured together in a line, were then immediately plunged into the agitated water and filled.
The disturbance measurements were repeated in three areas at each monitoring location, to achieve a spatial representation of sediment stored within the reach (e.g., Walling, Owens, & Leeks, 1998). The three repeat areas were selected to represent the erosional and depositional areas at the sampling location; measurements were not repeated in the exact same positions each month, due to constraints with creating a seal and the need for an adequate flow depth for water sampling, but recent tests of this method have underscored its reliability even in the context of such factors (Duerdoth et al., 2015). Bed sediment distur- periods. The intervening period of July to October 2013 encompassed the on-farm works to deliver the treatment train.
Sediment source sampling was conducted to determine the provenance of the in-stream sediment. Source samples were collected from eroding channel banks, damaged road verges, topsoil sources (e.g., poached pasture soils), and Hays Farm track sources (upper pre-mitigation, upper post-mitigation, and lower track). These potential sources were identified using topographic maps and walkover surveys of the subcatchment to identify areas of potential connectivity with the river. Samples were obtained by collecting surface scrapes to approximately 2 cm depth (e.g., Collins et al., 2012), to collect material likely to be mobilized by water (Collins, Walling, Webb, et al., 2010;Gruszowski, Foster, Lees, & Charlesworth, 2003;Walling, Collins, & Stroud, 2008). Channel bank samples were collected from the entire bank profile (e.g., Collins, Walling, Webb, et al., 2010)

| Laboratory methods
In the laboratory, all samples were dried at 40°C, disaggregated with a pestle and mortar, and sieved to <63 μm, the size fraction primarily associated with higher concentrations of pollutants (Horowitz, 1991).
The samples were weighed for mass before and after sieving, and then the <63-μm fraction was analysed for several fingerprint properties.
First, geochemistry, using inductively coupled plasma mass spectrometry (ICP-MS) after acid (aqua regia) digestion following the methods from Pulley et al. (2015);~0.8 g of sample sediment was digested in 10 ml of aqua regia at 180°C for 45 min in a CEM Mars 6 microwave digestion unit, before being measured using a Thermo Scientific iCAP

| Data analysis
Composite fingerprints using geochemistry, mineral magnetism, and environmental radionuclides were determined using a two-stage statistical procedure (Collins et al., 1997), comprising a Kruskal-Wallis H test and discriminant function analysis, to test the ability of the fingerprints to discriminate between the individual potential sediment sources identified in the subcatchment. This method has been used extensively in previous fingerprinting studies (e.g., Collins et al., 1997;Collins, Walling, McMellin, et al., 2010;Collins, Walling, Webb, et al., 2010;Pulley et al., 2015;Walling et al., 2006  3,000 iterations for each sediment sample using the median ± one median absolute deviation of each fingerprint property for each potential source group. Goodness-of-fit between the source-weighted predicted and measured sediment sample fingerprint property concentrations was used to assess the reliability of the unmixing model predictions. Any model iteration with a goodness-of-fit below 80% was deemed potentially unreliable and was therefore not used for further analysis (e.g., Pulley et al., 2015). Further detailed discussion of the sediment fingerprinting methodology and modelling used here can be found in Collins et al. (2017). For this specific study, Kruskal-Wallis H tests were used to test for statistically significant differences in the overall contribution of sediment sources between the farm scale (Site C) and subcatchment scale (Site D), to highlight any contrasts in mitigation effectiveness as scale increases. As the constraints of this study did not allow for equal timescales for pre-and post-mitigation, additional statistical tests were conducted to compare January to March of both the pre-and post-mitigation periods to account for potential seasonal differences in sediment mobilization and delivery from the sources under scrutiny.
3 | RESULTS Figure 7 shows the range in the averaged median predicted contributions from the individual sediment sources in the River Sem subcatchment for the pre-and post-mitigation monitoring periods.
These ranges reflect the unmixing model predictions for the individual sampling dates comprising each time period (i.e., pre-or post-mitigation). Table 2 presents the corresponding overall averaged median source contributions at each bed sediment sampling site, again for the pre-and post-mitigation periods. The data show that pre-mitigation, the major predicted source contribution, was from eroding channel banks, with an overall averaged median at A of 91%, at B of 91%, at C of 88%, and further downstream at the subcatchment scale at D of 75% (see Figures 2 and 3 for locations of these bed sediment sampling sites). Post-mitigation, the predicted contribution from eroding channel banks, remained high at 80% A, 81% B, 84% C, and a statistically significant decrease at D to 65% (p = .05; Table 2). Predicted contributions from eroding topsoil sources were far lower at the farm scale. In the pre-mitigation period, there was an overall averaged median predicted contribution from topsoils of 7% to A, 6% to B, and 8% to C, but a statistically significant increase to 20% at D at the subcatchment scale (p = .00; Table 2). In the post-mitigation period, the corresponding overall averaged median predicted contribution to A was 17% but only 5% at B and 4% at C, with a statistically significant increase to 30% at D at the subcatchment scale (p = .04; Table 2). Corresponding predicted contributions from damaged road verges were far lower, not exceeding 3% in either the pre-or post-mitigation periods at any site (Table 2). Table 2 shows that there was a relatively low contribution from the farm track sources (FTUO, FTL, and FTUN) at both the farm and subcatchment scales. In the pre-mitigation period, the overall averaged median predicted contribution to A from the upper farm track (FTUO) was 1%, at B 3%, at C 2%, and at D at the subcatchment scale 0%. The corresponding contributions from the FTL were predicted at 1% for A, 0% for B, 1% at C, and 4% at D (Table 2). In the post-mitigation period, the overall averaged median predicted contribution from the upper farm track (FTUO) to A was 0%, at B 14%, at C 12%, and at D a statistically significant decrease to 2% at the subcatchment scale (p = .00; Table 2). From the FTL, there was an overall averaged median predicted contribution of 0% to all sites during the post-mitigation period. There was no predicted contribution from the new, resurfaced FTU to any site during the pre-or post-mitigation periods (Table 2). To account for differences in timescale between the pre-(6 months) and post-mitigation (17 months) periods, a subset of months was compared. This subset comprised January to March 2013 in the pre-mitigation period and January to March 2014 in the post-mitigation period (Table 2). In the pre-mitigation period, the overall averaged median contribution from eroding channel banks decreased from 89% to 75% between Sites C and D with a corresponding decrease from 3% to 0% for FTUO. In contrast, the predicted contribution from topsoils increased from 8% to 24% (  Eroding topsoils were shown not to be an important source of fine-grained sediment by the fingerprinting work at the farm scale. However, scaling up from farm to subcatchment scale, the source tracing data for both the pre-and post-mitigation periods exhibited a statistically significant increase in the relative contribution from eroding topsoils. This is consistent with the area of topsoils at risk of erosion and delivery to the river channel increasing with scale across this agricultural landscape. The study subcatchment is heavily underdrained, which has been shown in previous studies to deliver significant quantities of mobilized topsoil to rivers (e.g., Bilotta et al., 2008;Chapman, Foster, Lees, Hodgkinson, & Jackson, 2001;Foster et al., 2003;McDowell & Wilcock, 2004;Zhang, Collins, & Hodgkinson, 2016). Several areas of heavily poached soils were also noted during walkover surveys, and some of these were directly connected to the river channel either due to proximity or as a result of surface run-off pathways, thereby also increasing the signal from eroding pasture topsoils as scale increases from farm to subcatchment level. In the context of the results for eroding channel banks discussed above, the source The results reported here are highly relevant to the use of treatment trains for mitigating DWPA. Such approaches are increasingly encouraged by policy initiatives and on-farm advice programmes in that they technically help deliver multiple lines of defence against water pollution. However, the evidence at different scales presented herein underscores the need for a dual approach using treatment trains. One approach needs to target obvious pollutant delivery pathways such as the example targeted in this study linking a polluting farmyard to the stream system, whereas the other approach needs to take due account of pollutant source and process domains across a range of scales, designing cascades or trains of measures on that basis.
In the case study used in this paper, there is clear evidence of increasing sediment inputs from eroding pasture topsoils with increasing  spatial scale, meaning that an appropriate treatment-train approach targeting the most common configurations of risk in the landscape needs to be rolled out on multiple farms throughout the subcatchment.
On the basis of field observations from walkover surveys, the latter will need to combine grassland compaction management and grazing management during wet weather/winter, with feeder ring management and maintenance of buffer strips. The latter intervention will also assist in managing bank erosion associated with cattle poaching that was observed below the headwater study farm that implemented bank fencing works.

| CONCLUSIONS
The findings of this study underscore that it is vital that the major sources of sediment are identified at a variety of spatial scales within any given landscape prioritized for mitigation of DWPA, so that interventions can be targeted correctly. Failure to consider sediment sources and process domains across a range of spatial scales, from individual farms to landscape scale, is likely to reduce the efficacy of the on-farm interventions, especially at those scales currently used for water quality compliance reporting. This highlights the potential benefits of collaboration between farmers, coordinating multiple farm-scale interventions within a subcatchment to ensure overall improvement at increasing landscape scales. It also underscores the need for on-farm pollution management advice delivered to any individual holding within a landscape to be placed carefully in the context of the scaling issues highlighted herein. Farm advisors therefore need to be equipped with tools and information for such considerations and to be trained accordingly, to help deliver maximum impact for environmental sustainability. The pre-and post-mitigation source tracing data for farm track sources highlight the risk of contributions at both farm and landscape scale being elevated as a consequence of on-farm remedial works, at least in the short term (1 to 2 years) during and immediately after implementation. Longer term studies are clearly required to convince farmers that such deviations in the outcomes arising from targeted interventions are indeed short term and must therefore be placed in a longer term management perspective. Longer term studies would also enable short-term variability in weather and climate to be evaluated in relation to changing sediment sources independent of the applied mitigation. This is important because hydro-climatic variability has the potential to govern mitigation impacts meaning that monitoring programmes must span the range of hydro-climatic variation to deliver robust assessments.

ACKNOWLEDGEMENTS
The co-funding provided for a PhD studentship (MB) by the Department of Environment, Food and Rural Affairs (Defra project WQ0225; awarded to ALC) and the University of Northampton