Sensitivity Analysis and Model Evaluation of Bifenthrin Surface Water Concentrations from California Urban Runoff

Aquatic species in the Sacramento-San Joaquin Delta are impacted by urban runoff. We analyse a commonly used pesticide regulatory model, the US EPA’s Pesticide Water Calculator (PWC version 1.59), to assess its performance in urban settings and to conditionally identify sensitive parameters. Bifenthrin is a major component of pyrethroid runoff to the Delta, particularly after storm events. It is used by pest control applicators for landscape application and as a perimeter treatment to create a chemical pest barrier for urban structures. This study employs deterministic and probabilistic approaches with the PWC model to simulate bifenthrin concentration in runoff from 4 urban storm drains located in Placer and Sacramento County. The deterministic approach used conservative (high) estimates for model inputs to calculate concentrations and represents what is typically done for a screening analysis. The probabilistic approach samples inputs from wide prior ranges to propagate variability through the model. A global sensitivity analysis was then administered to identify sensitive inputs with respect to model output variability. Partial correlation coefficients were used to measure input sensitivity and identify inputs whose variation substantially contributes to output variability and uncertainty. This provides the necessary information to prioritize efforts for uncertainty reduction by focusing on highly sensitive inputs, in addition to providing insight into model input-output relationships. We found that sensitivity results are driven by curve number assumptions, universal soil loss equation parameterization, benthic depth and application rates. Sensitivity dynamics are explored daily to visualize changes in sensitivity associated with major storm events. We found consistently strong agreement between model storm drain outfall water concentration predictions and measured values over a five-year period; observed concentrations were consistently within a 2 standard deviation bound of the median of the simulated concentrations.