Can information on past and near-future weather and field conditions predict the safest pesticide application day?

• Weather conditions near application day affect pesticide leaching to drains. • Machine-learning models trained on simulated data can predict the safest app. day. • Safest days are predicted with an 80% accuracy with accurate weather and field data. Pesticides applied on drained fields can be transported through drain-connected macropores to surface waters, as affected by weather and field conditions around spraying time. We investigated whether field conditions and past and near-future weather could be predictors of the ecotoxicological risk of pesticide leaching to surface waters and help identify the safest application day. We developed a methodological approach for one climate-soil setting and two herbicides. The agro-hydrological model Daisy and a 3,000-year series of generated weather were used to simulate a total of 369,326 pesticide application days and their resulting hourly drain concentrations, used in the risk calculation. Recurrent neural networks were trained on the simulated data. The trained meta -models selected the safest application day within a 5-day period with an average accuracy of 60–80%. The effective risk reductions were only of 3–11% for the investigated climate-soil-pesticide settings. Nevertheless, they represented 46–86% of the achievable reductions according to Daisy. [ABSTRACT FROM AUTHOR]