| dc.description.abstract | Pesticides play an important role in modern-day agriculture by protecting produce from pests and diseases. Nonetheless, exposure to these biological agents poses a significant public health concern, highlighting the necessity for exposure assessments. In recent years, spatial simulation models have emerged as an effective approach to estimate the extent and distribution of pesticide drift due to their ability to consider a range of factors, such as wind directions. Here, the typical approach is to derive this type of information from meteorological stations closest to the pesticide application area, introducing considerable inaccuracies. In order to provide a more robust and versatile alternative, this study aimed to examine spatial interpolation methods that may improve pesticide exposure estimates using wind field records from the Netherlands in 2017. In doing so, five spatial interpolation models were adopted to estimate wind directions at unobserved sites, namely naïve interpolation, nearest neighbour, inverse distance weighting, universal kriging and random forest. Performance of these models was evaluated using an out- of-sample circular root-mean-squared error (CRMSE) that was obtained through spatial 𝑘-fold cross validation. A sensitivity analysis examined the influence of varying observations on the performance of each model. Results showed distinct visual patterns that aligned with previous studies. Nonetheless, limited variability in hourly wind field measures resulted in a relatively similar performance across the employed models. All in all, the inverse distance weighting demonstrated the lowest out-of-sample error for interpolating wind directions. This finding suggested that the adoption of this model in pesticide drift simulations provides a more valid representation of wind fields at the application areas compared to the approach often employed. In turn, this may improve the accuracy of pesticide exposure estimates obtained from these simulations. | |
