Robust Quantitative Techniques for Validating Pesticide Transport Models

Abstract

Quantitative techniques used in validating pesticide and other solute transport models fall into three main categories: comparison of summary statistics, hypothesis testing, and measures of goodness-of-fit based on the analysis of residual errors. Most studies have used quantitative techniques under the assumption that the observed data conform to a Gaussian (normal) distribution. Presented are robust quantitative techniques, based on nonparametric statistical methods, that can be used when the distribution of the observed data is non-Gaussian or when the sample size is not large enough to determine the underlying data distribution. A nonparametric method to test whether the model predictions fall within a prescribed factor of true values using confidence intervals based on the sign statistic, and robust measures of goodness-of-fit, are proposed. The robust quantitative techniques are illustrated using a model testing example with the nonpoint source pollution model, Opus, based on observed and predicted atrazine concentration profiles from an agricultural field in the Virginia Coastal Plain.