Evolutionary algorithms for multi-objective, automatic calibration of a semi-distributed hydrologic model

Calibration of hydrologic models has traditionally been performed with respect to a single performance metric. This type of calibration, however, is often inadequate to properly evaluate the simulation of important characteristics of a hydrologic system. This paper presents a multi-objective, automatic calibration model, which is developed using an evolutionary optimization technique known as Strength Pareto Evolutionary Algorithm 2 (SPEA2). SPEA2 is a multi-objective search algorithm that employs the concept of Pareto dominance for selection of better solutions. The calibration model is integrated with the U.S. Department of Agriculture's Soil and Water Assessment Tool (SWAT). SWAT is a physically-based, semi-distributed hydrologic model that was developed to predict the long term impacts of land management practices on water, sediment and agricultural chemical yields in large complex watersheds with varying soils, land use, and management conditions. Being a semi-distributed model, SWAT consists of a large number of parameters to capture the various physical characteristics of a hydrologic-defined unit. In order to reduce the number of calibrable parameters to a manageable size, parameterization was performed. In this study, SWAT was calibrated for daily streamflow and sediment. The calibration process was formulated as a multi-objective optimization problem, and it involves parameter specification, whereby sensitive model parameters are identified, and parameter estimation. The methodology is demonstrated using Big Creek watershed in southern Illinois. The model finds Pareto optimal solutions for the calibration problem, which are sets of model parameters. Application results show that model predictions are significantly improved and that the use of multiple calibration objectives results in better model performance.