In The Science of the total environment
The most widely implemented mitigation measure to reduce transfer of surface runoff pesticides and other pollutants to surface water bodies are vegetative filter strips (VFS). . The most commonly used dynamic model for quantifying the reduction by VFS of surface runoff, eroded sediment, pesticides and other pollutants is VFSMOD, which simulates reduction of total inflow (∆Q) and of incoming eroded sediment load (∆E) mechanistically during the rainfall-runoff event. These variables are subsequently used to calculate the reduction of pesticide load by the VFS (∆P). Since errors in ∆Q and ∆E propagate into ∆P, for strongly-sorbing compounds an accurate prediction of ∆E is crucial for a reliable prediction of ∆P. The most important incoming sediment characteristic for ∆E is the median particle diameter (d50). Current d50 estimation methods are simplistic, yielding fixed d50 based on soil properties and ignoring specific event characteristics and dynamics. We derive an improved dynamic d50 parameterization equation for use in regulatory VFS scenarios based on an extensive dataset of 93 d50 values and 17 candidate explanatory variables compiled from heterogeneous data sources and methods. The dataset was analysed first using machine learning techniques (Random Forest, Gradient Boosting) and Global Sensitivity Analysis (GSA) as a dimension reduction technique and to identify potential interactions between explanatory variables. Using the knowledge gained, a parsimonious multiple regression equation with 6 predictors was developed and thoroughly tested. Since three of the predictors are event-specific (eroded sediment yield, rainfall intensity and peak runoff rate), predicted d50 vary dynamically across event magnitudes and intensities. Incorporation of the improved d50 parameterization equation in higher-tier pesticide assessment tools with VFSMOD provides more realistic quantitative mitigation in regulatory US-EPA and EU FOCUS pesticide risk assessment frameworks. The equation is also readily applicable to other erosion management problems.
Reichenberger Stefan, Sur Robin, Sittig Stephan, Multsch Sebastian, Carmona-Cabrero Álvaro, López J Javier, Muñoz-Carpena Rafael
Global Sensitivity Analysis, Machine learning, Median particle size, Pesticides, Sediment trapping, Vegetated filter strips