What parameters really matter? Linking exposure and effect models to identify sensitive parameters for model-based aquatic risk predictions of pesticides

Environmental risk assessment (ERA) of pesticides is essential for protecting non-target organisms and supporting sustainable agricultural practices. This study links exposure and effect models for the prediction of pesticide risk and performs a global sensitivity analysis to identify the most influential parameters for maximum concentrations and risk indices. We coupled the FOCUS surface water Step 3 model (TOXSWA combined with PRZM) with the General Unified Threshold model of Survival (GUTS) to translate time-resolved exposure profiles into lethal profiles (LP 50 ) values, capturing time-dependent toxicity and recovery processes. This approach offers a dynamic and mechanistically-based quantification of risk compared to traditional static risk quotients, which are based on single time-point concentrations. After prescreening 20 model parameters using a Morris sensitivity analysis, we selected the sorption coefficient and water degradation time as substance-specific properties, water depth as a key water-body parameter, and catchment size to capture the influence of loadings from agricultural practices for further in-depth analyses. Sobol sensitivity analyses of these four parameters were performed to assess the global sensitivity to simultaneous changes in these parameters. We found that while the maximum predicted environmental concentrations (PEC max ) are primarily driven by water depth and catchment size, risk quantified by LP 50 values is most sensitive to catchment size and so to loadings. Furthermore, degradation time does not influence PEC max , but its sensitivity increased for LP 50 values associated with slower kinetics. These findings show that parameters to which peak exposure is most sensitive are not necessarily the same as those to which a time-resolved risk metric is most sensitive. Whether this added mechanistic detail improves the prediction of effects in the field remains to be tested by benchmarking against monitoring data and simpler metrics.