National scale evaluation of the InVEST nutrient retention model in the United Kingdom

Abstract

A wide variety of tools aim to support decision making by modelling, mapping and quantifying ecosystem services. If decisions are to be properly informed, the accuracy and potential limitations of these tools must be well understood. However, dedicated studies evaluating ecosystem service models against empirical data are rare, especially over large areas. In this paper, we report on the national-scale assessment of a new ecosystem service model for nutrient delivery and retention, the InVEST Nutrient Delivery Ratio model. For 36 river catchments across the UK, we modelled total catchment export of phosphorus (P) and/or nitrogen (N) and compared model outputs to measurements derived from empirical water chemistry data.The model performed well in terms of relative magnitude of nutrient export among catchments (best Spearman's rank correlation for N and P, respectively: 0.81 and 0.88). However, there was wide variation among catchments in the accuracy of the model, and absolute values of nutrient exports frequently showed high percentage differences between modelled and empirically-derived exports (best median absolute percentage difference for N and P, respectively: ±64%, ±44%). The model also showed a high degree of sensitivity to nutrient loads and hydrologic routing input parameters and these sensitivities varied among catchments.These results suggest that the InVEST model can provide valuable information on nutrient fluxes to decision makers, especially in terms of relative differences among catchments. However, caution is needed if using the absolute modelled values for decision-making. Our study also suggests particular attention should be paid to researching input nutrient loadings and retentions, and the selection of appropriate input data resolutions and threshold flow accumulation values. Our results also highlight how availability of empirical data can improve model calibration and performance assessment and reinforce the need to include such data in ecosystem service modelling studies.