Stability and reproducibility evaluation of different windspeed spatial interpolation models to assist pesticide dispersion estimates.

Abstract

Prolonged pesticide exposure is known to raise health risks to respiratory, reproductive, neurological, endocrine, and circulatory systems. To give some indication about the degree of exposure near households in the Netherlands, a mixed model (OBOmod) is being developed by Utrecht University's Institute of Risk Assessment Sciences (IRAS) which considers many variables to determine indoor concentration of different pesticides.

The effect of including meteorological estimates (specifically windspeed) alongside the Gaussian plume-based pesticide dispersion model part of OBOmod has not yet been studied. This paper compared seven spatial interpolation models using a total of ten metrics and recommended the use of a hyperbolic trend surface model to minimize bias caused by random error and trends in estimates, which Gaussian plume models are known to be most sensitive to. The metrics were evaluated using ideal annual model hyperparameters determined using Bayesian optimization with a logarithmic loss function.