Characteristic Analysis of SWAT Model Parameter Values Based on Assessment of Model Research Quality

Abstract

The soil and water assessment tool (SWAT) model is currently one of the most widely used watershed models in China. Since the model has been developed with distributed parameters and is customized to satisfy the environmental characteristics of the U.S.A., determining appropriate parameter values that reflect local features for model application in China is crucial. Some studies have proposed parameter values for the SWAT model by summarizing reported values in the literature; however, these studies neither differentiate the literature with respect to its quality nor consider non-uniformity in parameter values and the impact of extreme values. To address this, an indicator system for assessing the quality of SWAT model research was established, taking into account the process of model development, parameter calibration, and model validation as well as model performance. This screening approach was applied to a total of 428 journal articles on SWAT model research published between 2015 and 2017 were retrieved from the China National Knowledge Infrastructure database. The reported values of 15 model parameters involved in hydrology and sediment and nutrient simulation were extracted from highly credible articles and analysed in terms of statistical distributions, differences among geographic regions, and discrepancies between calibrated and default values. Results showed that the 129 highly credible journal articles screened generally followed good modelling practice and consisted of case studies from different regions across China. The statistical distributions of the 15 model parameters derived from the SWAT model studies exhibited a range of features including positive and negative skewness, and those of 4 parameters showed significant differences among regions where the watersheds are located. Furthermore, the calibrated values of 12 out of 15 parameters were significantly different from their default values. Considering the statistical characteristics of these model parameters, recommended parameter values for SWAT model application in China are proposed in the form of confidence intervals, and specific suggestions are also provided based on data availability.