Identifying critical source areas using multiple methods for e ﬀ ective di ﬀ use pollution mitigation

the ﬀ erent approaches are used for this assessment, comparisons of these are carried out. the CSAs identi ﬁ ed traditional surveys supported by three di erent approaches, high-lighting their bene ﬁ ts and disadvantages. a custom designed smartphone app; a desktop geo- graphic information system (GIS) and analysis-based SCIMAP (Sensitive Catchment Integrated Modelling and Analysis Platform) the use of a high spatial resolution drone dataset as an improved for SCIMAP modelling. the locations of the revealing similarities and di in the prioritisation of CSA features. The di ﬀ erences are due to the temporal and spatial resolution of the three methods such as the use of static land cover information, the ability to capture small scale features, such as gateways and the incomplete catchment coverage of the walkover survey. The relative costs and output resolutions of the three methods indicate that they are suitable for application at di ﬀ erent catchment scales in conjunction with other methods. Based on the results in this paper, it is recommended that a multi-evidence- based approach to di ﬀ use pollution management is taken across catchment spatial scales, incorporating local knowledge from the walkover with the di ﬀ erent data resolutions of the SCIMAP approach.


Introduction
Nutrient enrichment of water courses through 'cultural' eutrophication is a global environmental problem leading to a decline in water quality (Foley et al., 2005;Novotny, 2003). Frequently, for management purposes, the sources of nutrients and sediments responsible are divided into point sources (those that originate from single defined sites discharging directly into water courses), and diffuse or non-point sources (those that originate from across the land in a river or lake catchment) (Novotny, 2003). There have been significant improvements in water quality across Europe since the early 1990s, largely achieved by tackling point sources through policy mechanisms such as the European Union's Urban Wastewater Treatment Directive (Directive 91/271/EEC (Council of the European Communities, 1991)). However, to enable the additional restoration and remediation of water bodies required to achieve 'good status' under the more recent Water Framework Directive (WFD, Directive 2000/60/EC (Council of the European Communities, 2000)), the contribution from diffuse sources needs to be tackled. These sources of pollution are arguably more challenging to manage as there are difficulties in identifying and regulating their origin, despite the fact that they can represent a large proportion of the total pollution received by an individual water course (European Environment Agency, 2005 is temporally and spatially heterogeneous. The origin of nutrients or sediments often separated from the impacted water course by large distances or long periods of time (McDowell et al., 2004;Meals et al., 2010;Haygarth et al., 2012;Jarvie et al., 2013), making source identification especially challenging. Haygarth et al. (2005) proposed a simple conceptual breakdown of the movement of diffuse pollution within a catchment to address this, collectively referred to as the transfer continuum. It is made up of four sequential components: 1. Source, being the input of the pollutant (e.g. fertilizer) to the farm system; 2. Mobilisation, involving the initial dispersal of the polluting substance from the soil itself, either by chemical or biological solubilisation, physical detachment or incidental losses recently applied to the surface; 3. Delivery, sometimes referred to as 'transport' or 'pathway', relates to the hydrological transport of the substance through various surface or below-ground pathways, and finally; 4. Impact, where the economic or ecologic impact is manifest downstream in a river, lake or estuary, perhaps many years and kilometres from the 'source'. As well as being a conceptual model that aids understanding, this has also been adopted as a way of implementing and focusing the adoption of management measures (e.g. Haygarth et al., 2009;Cuttle et al., 2016). To implement such measures, it is crucial to correctly identify and map 'high risk' areas and characteristics in the landscape, to build a comprehensive picture of the potential terrestrial source areas and link them to the delivery pathways that lead to pollutant deposition in the water courses where impact occurs. This involves identification of sources as well as key landscape and land management factors which drive nutrient mobilisation and delivery (Collins and McGonigle, 2008). Equally important is the knowledge of where sinks and barriers along the diffuse pollution transfer continuum currently exist, as these features contribute to the disconnection of sediment and nutrient flow to watercourses. Currently, the main approaches used for identifying diffuse pollution risk areas are catchment walkover surveys and the spatial modelling of diffuse pollution risks using geomorphological and hydrological information derived from digital elevation models (DEMS; e.g. Reaney et al., 2011;Zhang et al., 2013). Walkover surveys have the potential to provide useful, highly detailed, site specific information on a range of potential sources, sinks and pathways of diffuse pollution, which can be used in mitigation planning (Newson, 2010). To achieve an adequate resolution of information at an appropriate scale for this purpose is, however, challenging, due to the broad range of areas to be surveyed, and the necessity to collect data across seasons and in all weather conditions. In particular, the spatial coverage that can be achieved by these surveys is likely to be limited relative to the 100's or 1000's of km 2 of catchment that may need to be assessed. Such data collection has traditionally been done noting observations using pen and paper, which has a number of associated problems. Data recorded on paper whilst in the field requires subsequent manual transcription into an electronic database to render data useful for analyses and mitigation planning purposes (Olson et al., 2014), presenting the opportunity for transcription and/or typographical errors. In addition, spatial location via a GPS (global position system) and photographic records are increasingly required for later analysis and data validation. This technique requires considerable time investment in the field and during data collation, resulting in additional costs.
While walkover surveys can be seen as a 'bottom up' approach to identifying diffuse pollution risks, computer based modelling usually involves a more 'top down' approach, generalising landscape risks through analysis of broader scale landscape features (Lane et al., 2006). A number of models have been developed to assess and quantify diffuse pollution at different scales using farm-based information (e.g. FARM-SCOPER, Zhang et al., 2012), process-based modelling such as PSYCHIC (Davison et al., 2008) and delivery coefficients (e.g. Phosphorus Export and Delivery in Agricultural Landscapes (PEDAL) (Zhang et al., 2013)). An established method in the UK that integrates a range of spatial scales to generate predictions of risk is the Sensitive Catchment Integrated Modelling And Prediction (SCIMAP) risk mapping framework (http:// www.scimap.org.uk/). It works by calculating the spatial patterns of relative potential erosion and hydrological connectivity at a particular location, defined by the resolution of the input data, and combines these data to identify possible critical source areas for diffuse pollution risk (Reaney et al., 2011). This risk is routed through the landscape and diluted according to the amount of water that would flow through the location based on the spatial pattern of observed rainfall depths. A map is then generated, showing where in the landscape the diffuse pollution risk would accumulate faster than the dilution potential, which may lead to observable in-stream water quality problems. Although the SCIMAP approach gives relatively detailed predictions of the spatial pattern of risk at the sub-field scale, these predictions are necessarily limited by the quality and type of input data, being based mainly on natural surface landscape features including slope and land use. Subtle and small scale features, such as those that might represent the enhancement or deceleration of diffuse pollution risk for mitigation planning, may not be represented in the spatial data. SCIMAP is normally applied using nationally available datasets, and within the UK the normal dataset is the NextMap 5 m DEM, Centre for Ecology & Hydrology (CEH) Land Cover map 2007 at 30 m (Morton et al., 2011) and the CEH GEAR 1 km Rainfall map (Tanguy et al., 2016). The spatial resolution of these data sets means that SCIMAP is applicable at the landscape extent with sub-field detail. However, subtle topographic features in the landscape, such as tracks, gateways and poached soil likely will not be represented in the dataset, and hence are missed from the analysis. Furthermore, SCIMAP predicts the spatial pattern of relative risk, i.e. areas labelled as high risk within a catchment are the highest within that catchment and may not be causing a problem, making it important to apply the approach to catchments with a known diffuse pollution issue. Predictions from such models are useful as tools for identifying potential areas of diffuse pollution, but they need testing to establish the importance of individual sites as potential areas of risk. Collecting and collating these data in a systematic way is currently a challenge for the validation of predictions from spatially based models.
The limitations of current walkover field surveys, alongside the need for higher resolution input data and readily useable validation data for spatial model predictions, pose challenges for mitigation planning and the management of diffuse pollution. However, the growing potential of mobile digital technologies and aerial drones with high resolution cameras in providing easy to use multi-purpose tools through the provision of geo-referencing applications (apps) and higher resolution input data could offer improved data quality for the assessment of diffuse pollution risk at the sub-field scale where management decisions are made. Mobile phone apps for the collection of scientific data have proliferated (e.g. Aanensen et al., 2009;Graham et al., 2011;Sunyoung et al., 2011;Olson et al., 2014), and can provide data for monitoring programmes (Olson et al., 2014) and engage citizen scientists (Graham et al., 2011) at relatively low costs. As data are recorded digitally, including location and image files, they are easily integrated into analysis systems such as Geographical Information Systems (GIS) for later assessment of quality control/assurance and visual assessment of small-scale catchment features. At the same time, the falling costs and increasing durability of UAVs or drones are providing new opportunities for detailed data collection using remote sensing (Schiffman, 2014). However, as yet, relatively few comparisons of these emerging techniques for data capture have been carried out, and their applicability for assessment of diffuse pollution risk remains untested.
Here we provide an evaluation of different techniques for data collection and analysis for diffuse pollution risk assessment along the transfer continuum. We focus on mobilisation and delivery, and provide a framework for integrating the use of these techniques in catchment management. Firstly, predictions of diffuse pollution risk were generated using three distinct methods: (1) running conventional input data sources through the SCIMAP model, giving a 5 m resolution output; (2) identifying locations using data collected with a bespoke mobile phone app from a walkover survey; and (3) identifying sites using drone data to build a high spatial resolution DEM and land cover map. Secondly, we assessed the performance of each approach from the perspective of potential improvements to the identification of diffuse pollution risk, the associated costs and their utility in mitigation planning and management. Finally, we show how, through the use of an integrated approach to diffuse pollution assessment, we can optimise these different techniques used across spatial scales to provide a framework for improved catchment management of diffuse pollution.

Site description
The study site, a 2.2 km 2 headwater agricultural catchment in the River Eden valley, Cumbria, UK ( Fig. 1), was within the Eden Demonstration Test Catchment (DTC) project area, and has diffuse pollution issues associated with livestock production combined with high risk rainfall-runoff patterns (Owen et al., 2012;Perks et al., 2015). The catchment predominantly consisted of improved grassland for dairy and sheep, acid grassland and some arable. The elevation range was 95 m and the mean slope gradient was 6.5°. Soils were mainly clay loam and sandy clay loam; locally deep and well-drained in the headwaters, seasonally wet in the central elevations, moving through to slowly permeable and seasonally waterlogged in lower parts of the catchment. Post-glacial till dominated the superficial geology, forming a relatively continuous cover on the limestone bedrock, which intersected with the surface at a number of points. The mean annual rainfall was 1200-1272 mm based on the 1980-2010 average from the CEH GEAR dataset (Tanguy et al., 2016), with an annual average potential evapotranspiration of 636 mm, based on the Thornthwaite equation, showing the temperate nature of the catchment and the resultant importance of lateral hydrological flow pathways.

Mobile phone app and walkover survey
The diffuse pollution mobile phone app enabled multiple data records to be captured and stored offline, enabling its use in rural areas with little or no mobile network coverage. Each record is assigned a unique ID with time and date stamp. The latitude, longitude and altitude of each survey point is returned from the GPS unit of the phone and logged as part of the record. A series of pages with explanatory text at each step enables data entry relative to the feature being recorded, logging whether the feature is enhancing (a source/pathway) or  Table 1). 'Wheelings' are the result of farm machinery operated on fields and are narrow unvegetated areas of the field that are often compacted (Silgram et al., 2015). This compaction can result in increased runoff generation and transmission and associated erosion. The final data capture step gave the option of adding photographs of the feature. Once all steps have been completed, the information was summarised so the user can check that input is as comprehensive and accurate as possible, before the data are saved. When the mobile phone was next connected to the internet, stored data can be uploaded to a central web-hosted database, which allowed users to map, visualise and analyse submitted information. This feature enables data to be quality assured, before being exported as a.csv (comma separated values) file. The.csv format facilitated the dataset use in a variety of analysis packages, including GIS software, for visual analysis to enable, for example, mitigation planning.
A walkover survey of the study site was conducted to assess sources, barriers and connectivity, i.e. factors which may accelerate or decelerate pollutant delivery to the watercourses. The diffuse pollution phone app was used to record features regarded as either enhancing or decelerating pollutant mobility, and resulting data were mapped using ESRI ArcGIS software. To determine the visual coverage of the catchment, visibility analysis was undertaken with the Viewshed tool within ArcGIS and the 0.4 m digital surface model (DSM) from the drone mapping. This DSM included landscape features that may block views, such as walls, woodlands and buildings. The analysis shows how much of the catchment could have been observed from the walkover survey observation points.

Diffuse pollution risk mapping methods
The diffuse pollution risk mapping tool SCIMAP uses a 'minimum information requirement' approach to identifying where in the catchment diffuse pollution is most likely to be originating (Reaney et al., 2011). SCIMAP has been successfully applied to determining the spatial pattern of diffuse pollution risk on salmon and trout (Reaney et al., 2011), nitrogen and phosphorus (Milledge et al., 2012), FIOs (Porter et al., 2017) and sediment (Perks et al., 2017). SCIMAP is based on the critical source area concept, and hence calculates maps of source and mobilisation risk and the hydrological connectivity to capture the pathways across the catchment. This information was then used to route and dilute the diffuse pollution risk across the landscape to find areas where there is a greater diffuse pollution risk than water to dilute it.
The location of the high source and mobilisation risk areas (sourceRisk) were defined by the balance between the erosive energy of overland flow, determined by stream power and the erodibility of the soil surface, in this application regulated by land cover: where K is the erodibility of the surface on a value range 0-1, with 1 being the highest source and mobilisation risk; uca is the upslope contributing area; and B is the local slope gradient. The ease with which material can be transmitted along a pathway was calculated with the Network Index algorithm (Lane et al., 2009). This algorithm works by tracing the flow pathway from each point in the DEM to the river channel to identify drier, disconnecting locations and thereby determines how wet the catchment needs to be for each location to be runoff generating and to have a connected pathway to the river channel.
The source and connectivity maps are scaled between 0 and 1 and were then multiplied together to give the location of the CSAs in the landscape. The diffuse pollution risk values were then accumulated through the catchment based on a weighted flow accumulation algorithm to give a risk loading the channels. The loads were converted to a concentration based on the rainfall weighted upslope contributing area. The areas where there was more diffuse pollution risk than water to dilute it were highlighted as probable key source areas. SCIMAP produced maps of the source and mobilisation risk, showing the hydrological connectivity representing the ease with which material can move along a pathway and the integrated relative in-channel concentrations of risk, which enables the selection of high-risk sub-catchments for mitigation works.

Desktop SCIMAP application
For the desktop application of SCIMAP to the Newby Beck mitigation catchment, the NextMap 5 m DEM was used in combination with the CEH Land Cover map 2007 dataset. The land cover classes were grouped into eight main functional covers and the source and mobilisation risk values were assigned using the SCIMAP standard risk weights (Table 2). Due to the small size of the catchment (2.2 km 2 ), rainfall was assumed not to have a spatial pattern. These datasets are the standard inputs to SCIMAP since they have national coverage within the UK.

Drone data collection methods
The aim of the drone data collection was to build a high spatial resolution DEM and land cover map captured within the same season as the app-based walkover survey. The dataset was built using a 'Structure from Motion' photogrammetry approach by using a drone was used to capture multiple overlapping images of the catchment which were then processed to build a 0.4 m DEM and visible colour orthophoto. The drone was a DJI Phantom 4, with a 12-megapixel camera, which was flown at an altitude of 80 m above the land surface. This altitude was maintained over the topography to ensure that there was consistent detail and resolution in the source images. The images were captured with 90% forward overlap and 70% sideways overlap. A set of ground control points were placed across the catchment and their location recorded with Leica dGPS with an accuracy of 0.01 m. The image capture was performed over two days of consistent weather conditions and resulted in 5500 images. These images were processed within Agisoft Photoscan Pro to produce the 3D point cloud representing the catchment. This point cloud dataset was classified into ground and vegetation points and the ground points were used to build the DEM and the visible colour orthophoto.
The SCIMAP approach needs information of the spatial pattern of source areas and this map can be based on vegetation cover providing protection to the soil surface. Since near-infrared data were not available, the Visible Atmospherically Resistant Index (VARI), which enables the mapping of vegetation patterns (Gitelson et al., 2002), was used: The drone-based datasets were then processed within SCIMAP using the same options and parameters as for the desktop-based application.

Selection of the top critical source areas
All the accelerating features identified as part of the walkover survey were selected and assigned to one of two categories indicating their dominant role as a CSA: mobilisation or connectivity. For each of the SCIMAP based maps, the top ten critical source areas were selected to represent the key locations within the catchment where mitigation measures would have the maximum effect. The mobilisation and connectivity grids were multiplied together and areas with a combined score of > 0.9 were selected. These areas were converted into single units based on four-way neighbourhood connections, and then the largest ten areas were selected.

Walkover survey
The area covered by the walkover survey accounted for just over 60% of the catchment based on the visibility analysis (Fig. 2). In total, 75 enhancing features were identified in 12 of the 14 categories available in the mobile phone app, with only septic tank outflow and animals in the watercourse not being observed. Of these 75 enhancing features, six were identified as 'enhancing feature other', four of which were related to animal feeding or silage, while the remaining two were an overflowing yard drain and an arable field with no buffer strip. These points were further classified into 39 mobilisation features and 36 connectivity features ( Fig. 2; Table 1). The app also collected information on 52 decelerating features, including the presence of woodland, fenced buffer strips and settling ponds. A number of these measures had been installed in response to a previous desk-based SCIMAP assessment which identified areas of high diffuse pollution risk. These included online and offline sediment settling ponds, in-ditch hard-core sediment traps and fenced off riparian buffer zones with newly planted trees. This information can also be used to validate and cross reference with the modelling approaches.

Desktop modelling with SCIMAP
The 5 m DEM identified the main topographic highs at the southern edge of the catchment, and the lows, including the main stream channels and steeper slopes along the sides of the stream at the headwaters (Fig. 3). The Land Cover map 2007 indicated that the dominant land use within the catchment is improved (63%) or acid grassland (19%) followed by arable/horticulture (11%) and woodland (5%; Fig. 3). Some misclassification of suburban area is apparent for one area of this map. The steeper slopes linked to the stream channels drives the areas with high values on the mobilisation and connectivity maps (Fig. 3 c-d), particularly when associated with arable/horticultural land uses.

Drone data inputs and SCIMAP modelling outputs
The results from the 'Structure from Motion' processing are shown in Fig. 4. Comparison of the model to the ground control points showed a calculated error of 75 mm in the x -y plane, and 53 mm in the z plane, giving an overall error statistic of 92 mm. These errors show that the model is suitable for application in the SCIMAP processing. The maximum resolution of the DEM ground pixel size and orthophoto are 140 mm and 34 mm, respectively. However, the final version of the dataset was produced with a 400 mm ground resolution to enable compatibility with the standard terrain analysis algorithms, whilst still capturing subtle details in the catchment.
The improved resolution of the drone DEM enables the delineation of woodland areas, farm buildings and field boundaries (Fig. 4a). The bare ground identified (Fig. 4c) also revealed the dynamic pattern of land cover occurring in the catchment when compared to location of the arable/horticultural land use from the desktop study. The much greater resolution of results is clear in the mobilisation and connectivity maps (Fig. 4d and e), where the between and within field connectivity and mobilisation can be clearly identified. However, the SCIMAP model has difficulty with correctly classifying the connectivity risk in areas of woodland with these data.

Comparison of the three methods
The effort required for each of the three methods used to classify diffuse pollution risk in the catchment is provided in Table 3. The desktop study was the least resource intensive and includes full catchment coverage, but has the lowest resolution and, therefore, a greater level of uncertainty in the results. In contrast, the walkover survey provides very detailed information on local scale source and connectivity effects, with visual validation of the data. However, it required a larger surveying commitment and does not attain the same coverage as either of the remote sensing methods. The drone mapping combined a higher spatial resolution than the desktop mapping and 100% catchment coverage, although validation of the data was still required through aerial image analysis or ground based checks, and it represented the most expensive option of the three.

Identified critical source areas
The identified enhancing features and top ranked critical source areas for each method are shown in Fig. 5 a-c and in Fig. 6. The walkover survey (a) showed a clear linear distribution of features, partially reflecting the route of the surveyor and the stream network. Sites also clustered around the farm buildings located within the catchment. The top 10 CSAs identified by the desktop (b) and dronebased (c) SCIMAP clearly differ, reflecting differences in the land cover and bare ground used in the analysis, as well as the resolution of the input data. The desktop mapping also identifies grassland areas with steeper slopes adjacent to the stream network as being potentially high risk.

Discussion
The three approaches to identifying critical source areas for diffuse pollution provide different levels of temporal and spatial coverage and resolution. The desktop and drone-based SCIMAP model results both provide full coverage of the catchment. The more rapid and lower cost desktop-based analysis was, however, limited by the static nature of the land use data from 2007 and the 5 m DEM. This implies that the identification of diffuse pollution risk features will only be at a generic level, limited by the resolution of the input data and the assumption of a fixed land use pattern within the catchment. Comparing the 2007 land cover map data with the drone orthophoto clearly shows that in the landscape context of this catchment the changing pattern of bare fields is missed, which could act as a key CSA at particular times . To some extent, the identification of land use or risk areas by the other two methods also face this limitation, but they also offer the opportunity for rapid reassessment and ground-truthing that is not possible with a national scale map. The high level of spatial and temporal heterogeneity associated with diffuse pollution risks illustrates one of the reasons it is considered a 'wicked' environmental problem (Patterson et al., 2013;Thornton et al., 2013).
In contrast to the comprehensive coverage of the modelling, the walkover survey provided very detailed information at individual points within the catchment. This included information on the agricultural activity at the time of the survey and details of CSAs around farm buildings, which are features that cannot be picked up by SCIMAP modelling. As is clear from the routes taken by the observers, the identification of features occurs along linear pathways, which may result in CSAs not in the immediate vicinity of a particular route being missed. However, obtaining the same level of coverage as the spatial modelling would likely require a large amount of additional effort, which is unlikely to be a cost-effective option due to time requirements, or be particularly feasible where accessibility of land may be challenging due to crop cover, topography or presence of livestock.
Each method helped identify priority CSAs (Fig. 6). This is a direct result of the different resolutions of input data or scale of observation and the ephemeral nature of CSAs in the landscape. The dominant drivers of the CSAs identified in the Newby catchment are the mobilisation risks caused by the presence of bare ground (drone-based SCIMAP), land use and slopes adjacent to the stream network (deskbased SCIMAP) and features associated with pugging/poaching (mobilisation) and roads/tracks (connectivity; walkover survey). Pugging is  the creation of hoof imprints in the soil and poaching is the slurry like soil conditions created when saturated soil is trampled by livestock (Bilotta et al., 2007). These conditions represent very typical CSA features within largely grassland agriculture dominated catchments White et al., 2009). Understanding the context of these individual CSAs will be essential when identifying the priority locations for mitigation efforts that will maximise impact downstream (Doody et al., 2012). The desktop assessment SCIMAP based on the 5 m DEM dataset provided the most rapid assessment of diffuse pollution risk in the Newby catchment. However, the resolution of the DEM meant that it did not capture the subtle landscape features that can affect the sources, mobilisation pattern and pathways taken by diffuse pollution and hence may miss certain CSAs. This is particularly clear within the scale of the sub-catchment (~2 km 2 ) when comparing the map outputs in Fig. 3 with those in Fig. 4. The limitation of only using this approach to characterise CSAs is that relatively large areas are likely to be assigned similar risk categories, which may make the prioritisation of sites for mitigation measures more difficult at the field scale.
The drone-based SCIMAP assessment captured small scale features across the whole catchment relatively well, however, results in Fig. 4 identify problems with the representation of land under woodland. The photogrammetry-based Structure from Motion approach does not capture the ground surface and hence calculates the surface as being the top of the canopy. For individual trees, the approach is able to calculate the ground surface due to being able to see the land under the tree from multiple angles. An alternative approach to Structure from Motion would be the use of data capture using LiDAR, which has the advantage that it is able to detect gaps in the canopy and record the ground surface. The trade-off in this case is that a LiDAR approach would represent a significantly higher data collection cost.
Land use changes over time potentially represent an important contributory factor to the changing location of CSAs within a catchment. The land cover data used in the desktop SCIMAP assessment dates from 2007, and is almost a decade older than the drone and walkover survey carried out in 2016. This discrepancy in age resulted in very different land use categories being assigned by the two SCIMAP methods, with the importance of bare ground as a CSA from the dronebased assessment, revealing that what may be considered relatively subtle changes in land cover at a catchment scale may be strong drivers of diffuse pollution risk at the catchment and field scale. At shorter timescales, the ephemeral nature of CSAs can be revealed by repeated walkover visits using the app (Fig. 7), which illustrates the importance of considering the weather conditions and the time of year when surveys are conducted.
The three methods used to identify CSAs in the landscape represent a trade-off between a contrasting level of detail versus coverage and effort. The desk-based modelling represents the coarsest detail with high coverage and lowest effort, followed by the drone-based modelling with high detail and coverage with highest effort. The highest detail but lowest coverage was delivered by the walkover survey with intermediate effort. The cost of the drone survey renders the method suitable for sub-catchments that have been identified as high risk by the other methods. This indicates that each method has a potential application for different scales and purposes for identifying and mitigating CSAs. To improve confidence in the results and address uncertainties in the modelling outputs and observations, we recommend using a multiple evidence assessment for CSAs to maximise catchment coverage whilst allowing for ground assessment and verification in targeted locations.
Ultimately the purpose of identifying CSAs for diffuse pollution is to assist in the prioritisation of sites for targeting diffuse pollution mitigation management. The outputs of the key CSAs from the three tools ( Fig. 6) can be combined by identifying the spatial clustering of features identified by one or more methods to investigate options for mitigation at that location. Features identified by the app can be ranked according to priority or likely severity of impact and assessed alongside the relative risk scores of the SCIMAP outputs. Review of the images captured during the walkover survey offers the potential to increase confidence in deciding where mitigation measures should be located. They can also be utilised in post mitigation assessments of effectiveness in a before/ after comparison. Previous management measures in the catchment were targeted at locations along the stream channel and these align closely with the features identified by the app.
The challenge of mitigating diffuse pollution problems in catchments as part of the River Basin Planning approach in the Water Framework Directive, requires that areas of greatest diffuse pollution risk be identified and prioritised for mitigation from the catchment scale down to individual landscape features affecting river reaches. Considering these results, which demonstrate the wide range of features in a relatively small landscape area that contribute to diffuse pollution risk, it is recommended that a multi-evidence-based approach to identifying CSAs is taken, which incorporates local knowledge from the walkover app with the survey based SCIMAP approach to adequately capture these source areas across scales. This method can be conceptualised as a tiered approach to diffuse pollution mitigation management, which starts with an assessment of the entire river basin using national scale data and progressively scaling downwards to high-risk sub-catchment assessment using a drone survey, and ultimately to field scale verification and the identification of sites for mitigation using the app walkover survey can be performed (Fig. 8). Each of these methods provides a different view of the CSA identification problem and hence provide complimentary between the approaches. This tiered approach to assessment would enable river basin managers to effectively map diffuse pollution risks in their catchments and prioritise areas for management interventions in a more objective and cost-effective way.

Conclusion
The management of diffuse pollution, through the mitigation of sources, and features that drive mobilisation and delivery, is challenging due to the temporal and spatial heterogeneity of the problem. However, the increasing importance of managing this type of pollution to address persistent poor water quality requires effective approaches to identifying and prioritising the location of management interventions. This study assesses three techniques for the identification of critical source areas for diffuse pollution risk at a catchment scale to highlight how the different spatial scales of observation and coverage can alter which areas or features are identified as high risk. In addition, temporal variation in the age of input data, changing land use and the ephemeral nature of some diffuse pollution critical source areas all increase the challenge of identifying and prioritising mitigation of the most risky areas. In order to optimise the use of different datasets, techniques and to minimise the cost of these assessments, we advocate the use of a multi-tiered, multi-evidence approach to critical source area and mitigation measure identification from the river catchment to the sub-field scale.