Sensitivity Analysis And Interdependence Of The SWAT Model Parameters

Abstract

In contrast to lumped-parameter models, the distributed and processed-based hydrologic models take into account the spatial distribution of the hydrologic processes but became highly parameterized. In the Soil and Water Assessment Tool (SWAT) for example, the watershed is subdivided into spatial units (subbasins and hydrologic response units, HRU’s) and each spatial unit has its own unique parameters that are utilized in SWAT simulation. Sensitivity analyses had been used as screening tools for reducing the number of parameters in model calibration. The objective of this study was to analyze the sensitivity of the objective functions to changes in parameters used in the multiobjective automatic calibration of the SWAT model. We used a Bayesian network to estimate the interdependencies of the SWAT parameters. The direct and indirect effect of the parameters on the model output was also explored. Where there are multiple objectives, the parameters and their interaction in searching for the Pareto optimum change with position along the Pareto front. The information derived from the Bayesian network requires redefining sensitivity to include a description of the interaction of parameters in the calibration search process.