Challenges of spatially extrapolating aquatic pesticide pollution for policy evaluation

Aquatic pesticide pollution is an important issue worldwide. Countries rely on monitoring programs to observe water bodies quality and on models to evaluate pesticide risks for entire stream networks. Measurements are typically sparse and discontinuous which lead to issues in quantifying pesticide transport at the catchment scale. Therefore, it is essential to assess the performance of extrapolation approaches and provide guidance on how to extend monitoring programs to improve predictions. Here we present a feasibility study to predict pesticide levels in a spatially explicit manner in the Swiss stream network based on the national monitoring program quantifying organic micropollutants at 33 sites and spatially distributed explanatory variables. Firstly, we focused on a limited set of herbicides used on corn crops. We observed a significant relationship between herbicide concentrations and the areal fraction of hydrologically connected cornfields. Neglecting connectivity revealed no influence of areal corn coverage on the herbicide levels. Considering chemical properties of the compounds slightly improved the correlation. Secondly, we analysed a set of 18 pesticides widely used on different crops and monitored across the country. In this case, the areal fractions of arable or crop lands showed significant correlations with average pesticide concentrations. Similar results were found with average annual discharge or precipitation if two outlier sites were neglected. The correlations found in this paper explained only about 30 % of the observed variance leaving most of the variability unexplained. Accordingly, extrapolating the results from the existing monitoring sites to the Swiss river network comes with substantial uncertainty. Our study highlights possible reasons for weak matches, such as missing pesticide application data, limited set of compounds in the monitoring program, or a limited understanding of factors differentiating the loss rates from different catchments. Improving the data on pesticide applications will be essential to progress in this regard.