Assessment of the MHYDAS-Pesticide-1.0 model in simulating pesticide concentrations in surface waters at plot-scale continuously over decades

The use of pesticides in agriculture leads to widespread contamination of various environmental compartments. Surface water contamination represents a major environmental and health risk to manage. Models are a valuable tool for quantifying levels and temporal dynamics of contamination and for identifying and locating the physico-chemical processes involved in pesticide fate. The use of a model requires to know its performance for a given objective. Models simulating pesticide transfers to surface waters have generally been validated either in the laboratory or on a small scale (Gao et al., 2004, Joyce et al., 2010), or at field scale on mass export of discrete floods (Young and Fry, 2019), or at catchment scale on average concentrations or mass exports with minimum daily resolution (Wang et al., 2019). To our knowledge none have been evaluated on multi-year and field-scale databases. MHYDAS-Pesticide-1.0 is a continuous distributed mechanistic model used to simulate water and pesticide transfers from agricultural plots to the river system. The aim of this study is to evaluate at field scale the performance of this model to reproduce multi-year chronicles of runoff water contamination by a post-emergence herbicide intensively used in viticulture, glyphosate. The model was applied to a vineyard plot belonging to the Long-Term Observatory OMERE (Molénat et al., 2018), which outlet runoff discharge and glyphosate concentration have been monitored since 2001. The evaluation was conducted sequentially: the model was first evaluated on its performance in reproducing the runoff hydrographs observed at the plot outlet, then on its performance in simulating the temporal dynamics of glyphosate concentrations measured in runoff water. In both cases, the evaluation included a calibration and a validation step. The simulations were compared to data acquired between 2001 and 2017. The first three years were used for calibration and the following years for validation. To compare simulation results to observations, we considered variables of interest at flood event scale : the flood volume, the peak outflow and the mean glyphosate concentration. Goodness of fit was evaluated considering classical statistical performance indicators such as percent bias and Nash-Sutcliffe efficiency. These indicators were adapted to decrease the weight of the highest values relative to the lowest values. First calibration-validation results of saturated conductivity and height of surface detention enabled to identify a change of soil surface state during the assessment period that was corroborated by field observations. Taking this change into account by specific calibrations, the model reproduced very well runoff heights and peaks during the validation periods. The reproduction of concentration dynamics was also very satisfactory after calibration of glyphosate degradation kinetic and sorption coefficients and depth of the mixing layer between topsoil and surface runoff. This study demonstrates the ability of MHYDAS-Pesticide-1.0 to simulate multi-year time series of pesticide concentrations in surface water. It also provides a better understanding of the hierarchy of transfer processes occurring at the plot scale.