Sobol′’s sensitivity analysis for a distributed hydrological model of Yichun River Basin, China

Abstract

Summary This paper aims to provide an enhanced understanding of the parameter sensitivities of the Soil and Water Assessment Tool (SWAT) using a variance-based global sensitivity analysis, i.e., Sobol′’s method. The Yichun River Basin, China, is used as a case study, and the sensitivity of the SWAT parameters is analyzed under typical dry, normal and wet years, respectively. To reduce the number of model parameters, some spatial model parameters are grouped in terms of data availability and multipliers are then applied to parameter groups, reflecting spatial variation in the distributed SWAT model. The SWAT model performance is represented using two statistical metrics – Root Mean Square Error (RMSE) and Nash–Sutcliffe Efficiency (NSE) and two hydrological metrics – RunOff Coefficient Error (ROCE) and Slope of the Flow Duration Curve Error (SFDCE). The analysis reveals the individual effects of each parameter and its interactions with other parameters. Parameter interactions contribute to a significant portion of the variation in all metrics considered under moderate and wet years. In particular, the variation in the two hydrological metrics is dominated by the interactions, illustrating the necessity of choosing a global sensitivity analysis method that is able to consider interactions in the SWAT model identification process. In the dry year, however, the individual effects control the variation in the other three metrics except SFDCE. Further, the two statistical metrics fail to identify the SWAT parameters that control the flashiness (i.e., variability of mid-flows) and overall water balance. Overall, the results obtained from the global sensitivity analysis provide an in-depth understanding of the underlying hydrological processes under different metrics and climatic conditions in the case study catchment.