Marginal utility of conditional sensitivity analyses for dynamic models
Dynamic ecological processes may be influenced by many factors. Simulation models that mimic these processes often have complex implementations with many parameters. Sensitivity analyses are subsequently used to identify critical parameters whose uncertainties can be further reduced or better described and prediction variability minimized. In this study, we compare the ability of partial derivative based local sensitivity analysis methods to a variance decomposition based global sensitivity analysis technique (Sobol’ method). We use an ecological exposure model that is used to estimate pesticide concentrations in runoff and vertical soil compartments. Daily simulations are performed that fully explore the input parameter space. Estimated concentrations are compared to data collected over the course of a growing season from an experimental site in Georgia and representative local and global sensitivity analyses are conducted and compared.
Our results demonstrate that parameter sensitivity is conditional for dynamic models and should be examined at appropriate spatial and temporal resolution with global methods to avoid omitting important parameters. Sensitivities can and do change when they are evaluated temporally, with depth, as well as taking other related conditions into account. This is a different approach than traditional sensitive analysis, whose conclusions are drawn by essentially “averaging” across all conditions thus could result in omitting important parameters. For instance, to better predict pesticide's residual concentration at shallow depths (0-15cm), one should focus on pesticide application rates, pesticide decay rate on foliage and rain intensity that have more direct impact on predicted pesticide concentrations. While in deeper soil compartments (>15cm), more attention should be paid to the partitioning coefficient-organic component, the decay rate in the soil and water, and the accuracy of the rainfall time series. Global approaches can yield a better understanding about the interplay between sensitivity/uncertainty and model dynamics in non-monotonic, non-linear systems.