A Monte Carlo approach to the inverse problem of diffuse pollution risk in agricultural catchments

Apportioning sources of chemicals of emerging concern along an urban river with inverse modelling

Concentrations of chemicals in river water provide crucial information for assessing environmental exposure and risks from fertilisers, pesticides, heavy metals, illicit drugs, pathogens, pharmaceuticals, plastics and perfluorinated substances, among others. However, using concentrations measured along waterways (e.g., from grab samples) to identify sources of contaminants and understand their fate is complicated by mixing of chemicals downstream from diverse diffuse and point sources (e.g., agricultural runoff, wastewater treatment plants). To address this challenge, a novel inverse modelling approach is presented. Using waterway network topology, it quantifies locations and concentrations of contaminant sources upstream by inverting concentrations measured in water samples. It is computationally efficient and quantifies uncertainty. The approach is demonstrated for 13 contaminants of emerging concern (CECs) in an urban stream, the R. Wandle (London, UK). Mixing (the forward problem) was assumed to be conservative, and the location of sources and their concentrations were treated as unknowns to be identified. Calculated CEC source concentrations, which ranged from below detection limit (a few ng/L) up to 1μg/L, were used to predict concentrations of chemicals downstream. Using this approach, >90% of data were predicted within observational uncertainty. Principal component analysis of calculated source concentrations revealed signatures of two distinct chemical sources. First, pharmaceuticals and insecticides were associated with a subcatchment containing a known point source of treated effluent from a wastewater treatment plant. Second, illicit drugs and salicylic acid were associated with multiple sources, interpreted as input from untreated sewage including Combined Sewer Overflows (CSOs), misconnections, runoff and direct disposal throughout the catchment. Finally, a simple algorithmic approach that incorporates network topology was developed to design sampling campaigns to improve resolution of source apportionment. Inverse modelling of contaminant measurements can provide objective means to apportion sources in waterways from spot samples in catchments on a large scale.

Soil erosion risk for farming futures: Novel model application and validation to an agricultural landscape in southern England

Increasingly, agricultural land managers are seeking new approaches for understanding the potential challenges posed by sediment connectivity across catchments from source to sink, and implications for delivery of ecosystem services determined by the condition of natural capital assets. Connectivity indices have been frequently applied in the calculation of risk in spatial and temporal assessment frameworks, and tools which facilitate rapid modelling and mapping of soil erosion risk using broad-scale environmental data are therefore of considerable interest. One such indicative tool is SCIMAP (Sensitive Catchment Integrated Mapping and Analysis Platform), which highlights where sediment runoff is likely to occur and be delivered to a watercourse by simulating the generation of saturation-excess overland flow. In this paper, we examine the utility of SCIMAP for exploring the changing nature of soil erosion risk as a function of land use change in the lower Rother catchment in West Sussex, southern England through the formulation of a suite of foresight scenarios informed by knowledge of historical land cover conditions and current management practice. The study area has previously been investigated at the field scale in terms of locating and quantifying sources of erosion and areas where in-stream sedimentation manifests. Output risk values from all simulations were quantified, mapped and compared to highlight areas of greatest/lowest risk. An area was identified immediately north of the main Rother channel that consistently exhibited greatest risk across each land cover scenario. We explore (i) the spatial and temporal variation in modelled risk and (ii) the utility value of SCIMAP for agricultural land-managers and policy-makers in generating robust risk estimates of soil erosion and in-stream sedimentation, and challenges with model verification in a foresight context.

Bayesian spatial analysis of crash severity data with the INLA approach: Assessment of different identification constraints

Multinomial logit models have been widely used in the analysis of categorical crash data. When the regional information of the data is available, the dependence structure needs to be incorporated into the model to accommodate for spatial heterogeneity. We consider a Bayesian multinomial structured additive regression model to analyze categorical spatial crash data and compare its performance with a fractional split multinomial model. We use the multinomial-Poisson transformation to apply the integrated nested Laplace approximation method for fitting the proposed model efficiently and fast. Moreover, we consider two different types of identifiability constraints to deal with the inherent identifiability problem of the multinomial models. The proposed models are studied through simulated examples and applied to a road traffic crash dataset from Mazandaran province, Iran.

Modeling of phosphorus loss from field to watershed: A review

Phosphorus (P) losses from nonpoint sources into surface water resources through surface runoff and tile drainage play a significant role in eutrophication. Accordingly, the number of studies involving the modeling of agricultural P losses, the uncertainties of such models, and the best management practices (BMPs) supported by the modeling of hypothetical P loss reduction scenarios has increased significantly around the world. Many improvements have been made to these models: separate manure P pools, variable source areas allowing the determination of critical source areas of P loss, analyses of modeling uncertainties, and understanding of legacy P. However, several elements are still missing or have yet to be sufficiently addressed: the incorporation of preferential flow into models, the modification of P sorption-desorption processes considering recent research data (e.g., pedotransfer functions for labile, active, or stable P, along with P sorption coefficients), BMP parameterization, and scale-up issues, as well as stakeholder-scientist and experimentalist-modeler interactions. The accuracy of P loss modeling can be improved by (a) incorporating dynamic P sorption-desorption processes and new P subroutines for direct P loss from manure, fertilizer, and dung, (b) modeling preferential flow, connectivity between field and adjacent water bodies, and P in-stream processes, (c) including an assessment of model uncertainty, (d) integrating field and watershed models for BMP calibration and scaling field results up to larger areas, and (e) building a holistic interaction between stakeholders, experimentalists, and modelers.

Identifying critical source areas using multiple methods for effective diffuse pollution mitigation

Diffuse pollution from agriculture constitutes a key pressure on the water quality of freshwaters and is frequently the cause of ecological degradation. The problem of diffuse pollution can be conceptualised with a source-mobilisation-pathway (or delivery)-impact model, whereby the combination of high source risk and strong connected pathways leads to 'critical source areas' (CSAs). These areas are where most diffuse pollution will originate, and hence are the optimal places to implement mitigation measures. However, identifying the locations of these areas is a key problem across different spatial scales within catchments. A number of approaches are frequently used for this assessment, although comparisons of these assessments are rarely carried out. We evaluate the CSAs identified via traditional walkover surveys supported by three different approaches, highlighting their benefits and disadvantages. These include a custom designed smartphone app; a desktop geographic information system (GIS) and terrain analysis-based SCIMAP (Sensitive Catchment Integrated Modelling and Analysis Platform) approach; and the use of a high spatial resolution drone dataset as an improved input data for SCIMAP modelling. Each of these methods captures the locations of the CSAs, revealing similarities and differences in the prioritisation of CSA features. The differences 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 different catchment scales in conjunction with other methods. Based on the results in this paper, it is recommended that a multi-evidence-based approach to diffuse pollution management is taken across catchment spatial scales, incorporating local knowledge from the walkover with the different data resolutions of the SCIMAP approach.

Predicting diffuse microbial pollution risk across catchments: The performance of SCIMAP and recommendations for future development

Microbial pollution of surface waters in agricultural catchments can be a consequence of poor farm management practices, such as excessive stocking of livestock on vulnerable land or inappropriate handling of manures and slurries. Catchment interventions such as fencing of watercourses, streamside buffer strips and constructed wetlands have the potential to reduce faecal pollution of watercourses. However these interventions are expensive and occupy valuable productive land. There is, therefore, a requirement for tools to assist in the spatial targeting of such interventions to areas where they will have the biggest impact on water quality improvements whist occupying the minimal amount of productive land. SCIMAP is a risk-based model that has been developed for this purpose but with a focus on diffuse sediment and nutrient pollution. In this study we investigated the performance of SCIMAP in predicting microbial pollution of watercourses and assessed modelled outputs of E. coli, a common faecal indicator organism (FIO), against observed water quality information. SCIMAP was applied to two river catchments in the UK. SCIMAP uses land cover risk weightings, which are routed through the landscape based on hydrological connectivity to generate catchment scale maps of relative in-stream pollution risk. Assessment of the model's performance and derivation of optimum land cover risk weightings was achieved using a Monte-Carlo sampling approach. Performance of the SCIMAP framework for informing on FIO risk was variable with better performance in the Yealm catchment (r_s=0.88; p<0.01) than the Wyre (r_s=-0.36; p>0.05). Across both catchments much uncertainty was associated with the application of optimum risk weightings attributed to different land use classes. Overall, SCIMAP showed potential as a useful tool in the spatial targeting of FIO diffuse pollution management strategies; however, improvements are required to transition the existing SCIMAP framework to a robust FIO risk-mapping tool.

Evaluating a Bayesian modelling approach (INLA-SPDE) for environmental mapping

Understanding the uncertainty in spatial modelling of environmental variables is important because it provides the end-users with the reliability of the maps. Over the past decades, Bayesian statistics has been successfully used. However, the conventional simulation-based Markov Chain Monte Carlo (MCMC) approaches are often computationally intensive. In this study, the performance of a novel Bayesian inference approach called Integrated Nested Laplace Approximation with Stochastic Partial Differential Equation (INLA-SPDE) was evaluated using independent calibration and validation datasets of various skewed and non-skewed soil properties and was compared with a linear mixed model estimated by residual maximum likelihood (REML-LMM). It was found that INLA-SPDE was equivalent to REML-LMM in terms of the model performance and was similarly robust with sparse datasets (i.e. 40-60 samples). In comparison, INLA-SPDE was able to estimate the posterior marginal distributions of the model parameters without extensive simulations. It was concluded that INLA-SPDE had the potential to map the spatial distribution of environmental variables along with their posterior marginal distributions for environmental management. Some drawbacks were identified with INLA-SPDE, including artefacts of model response due to the use of triangle meshes and a longer computational time when dealing with non-Gaussian likelihood families.

Impact of legacy soil phosphorus on losses in drainage and overland flow from grazed grassland soils

Rates and quantities of legacy soil phosphorus (P) lost from agricultural soils, and the timescales for positive change to water quality, remain unclear. From 2000 to 2004 five 0.2ha grazed grassland plots located on a drumlin hillslope in Northern Ireland, received chemical fertiliser applications of 0, 10, 20, 40, 80kgPha^-1yr^-1 resulting in soil Olsen P concentrations of 19, 24, 28, 38 and 67mgPL^-1, respectively, after which applications ceased. Soil Olsen P and losses to overland flow and drainage were monitored from 2005 to 2011 on an event and weekly flow proportional basis, respectively. Soluble reactive P and total P time series were synchronised with daily rainfall and modelled soil moisture deficits. From 2005 to 2011 soil Olsen P decline was proportional to soil P status with a 43% reduction in the plot at 67mgPL^-1 in 2004 and a corresponding 12% reduction in the plot with lowest soil P. However, there was no significant difference in the flow-weighted mean concentration for overland flow among plots, all of which exceeded 0.035mgL^-1 in >98% of events. Strong interannual and event variations in losses were observed with up to 65% of P being lost during a single rainfall event. P concentrations in drainage flow were independent of Olsen P and drain efficiency was potentially the primary control on concentrations, with the highest concentrations recorded in the plot at 38mgL^-1 Olsen P in 2004 (up to 2.72mgL^-1). Hydrological drivers, particularly antecedent soil moisture, had a strong influence on P loss in both overland and drainage flow, with higher concentrations recorded above a soil moisture deficit threshold of 7mm. This study demonstrates that on some soil types, legacy P poses a significant long term threat to water quality, even at agronomically optimum soil P levels.

A Bayesian approach of high impaired river reaches identification and total nitrogen load estimation in a sparsely monitored basin

In this study, a modeling framework based on the theory of SPAtially Referenced Regression On Watershed attributes (SPARROW) model was developed to identify impaired river reaches with respect to total nitrogen (TN) and estimate the TN sources in the Xin'anjiang River basin, which had limited monitoring sites. A Bayesian approach was applied to estimate the mean values and uncertainties of parameters, including land use export coefficients and in-stream attention rates. Based on the parameters, the midranges (25-75 %) of annual TN concentrations were assessed by the model and 4.5 % of river reaches in the basin were found to be with higher impaired probabilities (namely [TN] > 1.5 mg/l) than other reaches. The amount and yields of TN discharged from diffuse sources were estimated for each county in the basin. The results suggested that Tunxi City had the highest TN yields from farm land and population, while the highest TN yields in Huangshan City were from tea plantations. The outcomes of this study will guide the implementation of practical management measures to reduce TN loads.

Recent trends in water quality in an agricultural catchment in Eastern Scotland: elucidating the roles of hydrology and land use

Across the EU, programmes of measures have been introduced as part of river basin management planning as a means of tackling problems of diffuse pollution from agriculture. Evidence is required to demonstrate the effectiveness of these measures and with this overarching objective, monitoring of an agricultural catchment in Eastern Scotland was initiated in 2007. As a precursor to evaluating the effect of new management measures it is essential to understand how other factors, including hydrology and land use changes, could have influenced water quality. This study undertook an analysis of the trends in concentrations and loads of nitrate, soluble reactive phosphorus (SRP), suspended solids (SS) and turbidity measured at six points in the catchment over a six year period. The results identified both differing trends between determinands and differing trends occurring over varying spatial scales. The only direct relationships between land use and water quality that could be identified based on annual data was a positive link between arable cropping and nitrate concentrations. At the sub-catchment scale some temporal changes in land use and management explained short-term trends in nitrate but not in SRP. Lags in the system were identified due to soil adsorption, in-stream/loch processing and groundwater transport making the identification of cause and effect problematic. The results have implications for the demonstration of effectiveness of measures over the shorter term and the timescales of recovery from diffuse pollution. Longer term monitoring at small scales will be important in this regard.

Understanding nutrient biogeochemistry in agricultural catchments: the challenge of appropriate monitoring frequencies

We evaluate different frequencies of riverine nutrient concentration measurement to interpret diffuse pollution in agricultural catchments. We focus on three nutrient fractions, nitrate-nitrogen (NO3-N), total reactive phosphorus (TRP) and total phosphorus (TP) observed using conventional remote laboratory-based, low-frequency sampling and automated, in situ high-frequency monitoring. We demonstrate the value of low-frequency routine nutrient monitoring in providing long-term data on changes in surface water and groundwater nutrient concentrations. By contrast, automated high-frequency nutrient observations provide insight into the fine temporal structure of nutrient dynamics in response to a full spectrum of flow dynamics. We found good agreement between concurrent in situ and laboratory-based determinations for nitrate-nitrogen (Pearson's R = 0.93, p < 0.01). For phosphorus fractions: TP (R = 0.84, p < 0.01) and TRP (R = 0.79, p < 0.01) the relationships were poorer due to the underestimation of P fractions observed in situ and storage-related changes of grab samples. A detailed comparison between concurrent nutrient data obtained by the hourly in situ automated monitoring and weekly-to-fortnightly grab sampling reveals a significant information loss at the extreme range of nutrient concentration for low-frequency sampling.