Modification and validation of the InVEST erosion model for application at a national scale.

Soil erosion is an important ecological impact of human activities such as agriculture, and models can aid in identifying areas most at risk. However, to aid targeting of actions, predictions must simultaneously cover large spatial extents and account for site-specific variation in retention, while achieving adequate parameterisation and evaluation of models remains demanding. To address these issues, we adapted and assessed an erosion model using the InVEST platform, which applies the Revised Universal Soil Loss Equation (RUSLE), for Great Britain (GB). We parametrised the model using GB-specific input data, for multiple crop types, and integrating new factors including sub-annual periodicity, a GB-specific erosivity layer, and Normalized Difference Vegetation Index (NDVI) derived cover management metrics. Our modelled predictions validated well against sediment concentration measurements in rivers, with a predictive accuracy of 78% on normalised data and rank correlation of 0.46 between predictions and observations. However, absolute model values were overestimated by 28-times, especially at higher levels of sediment erosion. This seems related to a lack of measurements of sediment concentrations at peak flows in the validation data, combined with known RUSLE methodological issues. To allow for future model training, we call for improvements to national sediment load monitoring in rivers, by including measurements during peak flows especially, and inclusion of a within stream sedimentation function in the InVEST Sediment Delivery Ratio model. Our InVEST model parametrisation example provides a valuable tool for relative risk mapping, comparing regions and targeting where ecosystems are more prone to erosion. Cover management factors based on NDVI observations are a substantial methodological improvement for RUSLE modelling that can be readily reproduced in other locations. Whereas a sub-annual version did not lead to differences from an annual model in this all-year around rain area it could enhance modelling for more seasonal areas.