Using Mutual Information for Global Sensitivity Analysis on Watershed Modeling

Abstract Global sensitivity analysis (GSA) often is applied to assess the of model outputs their inputs using ensemble simulations. However, increasing complexity and associated computational cost have limited use most GSA approaches for process‐based watershed models. We propose mutual information (MI) as a computationally efficient method modeling. Such MI computed from several hundred realizations usually can capture nonlinear relationships between interest. perform MI‐based analyses in studies Portage River Watershed Ohio American Washington. In these studies, used evaluate river discharges simulated by Soil Water Assessment Tool no less than 20 SWAT parameters each watershed. Our achieved convergence with about 300–500 realizations, small fraction size (i.e., thousands) required Sobol method. Nevertheless, two‐dimensional yields similar ranking compared Sobol's total‐order indices, especially sensitive parameters. study thus sheds new light on an affordable intensive models such hyperresolution, hydrobiogeochemical