Global, spatial and temporal sensitivity analysis for a complex pesticide fate and transport model

Background/Question/Methods

As one of the most heavily used exposure models by U.S. EPA, Pesticide Root Zone Model (PRZM) is a one-dimensional, dynamic, compartment model that predicts the fate and transport of a pesticide in the unsaturated soil system around a plant’s root zone. Like other complex environmental models, PRZM may be over-parameterized and significant uncertainties exist for both its inputs and outputs. Therefore, a number of studies have been conducted to identify the most influential parameters of PRZM model. Unfortunately, they were based on local sensitivity analysis methods which failed to capture input parameters' joint efforts in determining modeling outputs. The objective of this study is to identify the most sensitive PRZM inputs using a global sensitivity technique (Sobol’s method) to fully explore physically possible parameter spaces and to consider variables' interactions at the same time. 

Results/Conclusions

The results suggest that for a complex environmental fate and transport model, the sensitivity of its inputs should be examined spatially and temporally. Spatially (i.e., one-dimension vertically), at shallow depths (0-15cm), pesticide application rates, pesticide decay rates on foliage and rain intensity have greater impact on the predicted pesticide concentrations. At greater depths (>15cm), dominant variables become rain intensity, partition coefficient, pesticide soil and water decay rate, and their interactions. When coupled with the timing of pesticide applications, pesticide application rates, and soil and water decay rates are more powerful in determining sensitivity. During precipitation events, the most sensitive variables become rainfall intensity, and pesticide decay rate on foliage. One of this study’s contributions is to provide a framework for ecological and environmental modelers of how to quantify a complex model's uncertainties globally, spatially, and temporally. In addition, the proper sensitivity analysis for the PRZM model better informs the pesticide label registration process.