Effective Global Sensitivity Analysis for High-Dimensional Hydrologic and Water Quality Models

Abstract

Sensitivity analysis (SA) plays a vital role in hydrologic and water quality (H/WQ) model development and reliability assessment. This paper performed the first formal global SA (GSA) evaluation of the routing submodule of the Watershed Assessment Model (WAM), a spatially distributed high-dimensional H/WQ model, using a two-step GSA approach consisting of sequential application of the elementary effects (EE) parameter screening and the variance-based technique of Sobol’. The Taylor Creek Nubbin Slough basin (S-191 basin) in southcentral Florida is taken as a test case. Multioutput EE analysis successfully screened unimportant parameters, deeming the rest (27 of 297) likely influential. Sobol’ GSA with these 27 parameters showed that for flow and nutrient outputs, most of the model variability was attributed to only 12 parameters, fewer than 5% of the original parameter set. An approach to extract additional useful information about the output–parameter relationships, interactions, and monotonic/nonmonotonic behavior from the EE analysis with minimal additional calculations was explored. The proposed criteria for identifying parameter interactions from EE analysis showed good comparison with Sobol’ analysis results.