Multi-criteria validation of the SWAT hydrologic model in a small forested watershed
MetadataShow full item record
The goal of the study is to perform a multi-criteria automated calibration and validation of the Soil and Water Assessment Tool (SWAT) model using multiple observed datasets. A multi-criteria calibration uses multiple noncommensurable measures of information in order to improve the structural validity of the model. To achieve this goal two automated calibration methods, the Monte Carlo approach and the Multi-Objective Complex Evolution, are applied to a small watershed in western New York. Model calibration is performed in two stages. At the first stage a traditional manual calibration is employed. The purpose of the manual calibration is to ensure that the model provides an adequate representation of the catchment by modeling all relevant hydrologic processes, and to set the foundation and the basis of comparison with subsequent automated calibration. At the second stage an automated model calibration is performed using two strategies, a single-criteria and a multi-criteria. The single-criteria calibration for discharge at the outlet is performed with the Monte Carlo method. For the multi-criteria strategy the Multi-Objective Complex Evolution (MOCOM-UA) algorithm is employed to calibrate SWAT against several datasets of discharge and groundwater levels. The model is then validated using the split-sample and the proxy basin approaches. The study shows that multi-criteria calibration with the MOCOM-UA algorithm is able to utilize the information contained in the additional datasets to improve model performance. The effectiveness and efficiency of the MOCOM-UA calibration exceeds those of the single-objective calibration approach during both calibration and validation periods. It is demonstrated that the MOCOM-UA multi-objective calibration results in lower model uncertainty compared to the single-objective calibration. It is also shown that automated calibration with the MOCOM-UA and Monte Carlo methods is able to achieve better model performance than the traditional manual calibration.