Direct Treatment of Uncertainty: II—Applications in Pesticide Runoff,Leaching and Spray Drift Exposure Modeling

Abstract

Mechanistic numerical models are used throughout the scientific community to assist in understanding physical and chemical phenomena. However, results generated by a wide variety of deterministic models represent only a single value without any consideration as to the uncertainty associated with the prediction. A promising technique, known as the Deterministic Equivalent Modeling Method (DEMM), has been gaining popularity as an alternative to random sampling techniques such as Monte Carlo (MC) or Latin Hypercube Sampling. DEMM allows for the calculation of uncertainty in output parameters based upon the direct effect of every uncertain input parameter. Any model having an input/output file structure can be coupled with DEMM/collocation techniques without resorting to source code modifications. This paper explores the capability and functionality of DEMM uncertainty propagation techniques for a pesticide environmental fate model (PRZM3) and a pesticide aerial drift model (AGDRIFT). Results for chlorpyrifos and atrazine pesticides are presented. Comparisons between DEMM and Monte Carlo (MC) for pesticide runoff, leaching, and aerial drift indicate that the range and magnitude for estimated cumulative probability functions predicted by PRZM3 and AGDRIFT are achievable with DEMM while requiring an order of magnitude less iterations than with the MC counterpart. For the problems investigated, DEMM does break down for at the higher percentiles (>90%) when compared to the more numerically intense MC. However, if a quick order of magnitude analysis for uncertainty and parameter sensitivity is sought, or if a model is of such complexity that CPU time is an issue, then DEMM, with limitations, can become an alternative to MC for characterizing approximate model predictions when parametric uncertainty is addressed.