Uncertainty analysis of parameters in non‐point source pollution simulation: case study of the application of the Soil and Water Assessment Tool model to Yitong River watershed in northeast China

Abstract Uncertainty analysis of the model parameters in non‐point source pollution (NPSP) simulation is important because of its great effects on predictions and decision‐making. Understanding the main parameters that effect the uncertainty of NPSP is necessary to provide the basis for formulating control measures. In this study, two methods were applied to conduct parameter uncertainty analysis for Soil and Water Assessment Tool (SWAT). Sobol’ method was used to screen out the model parameters with great effects on the runoff, sediment, total nitrogen (TN) and total phosphorus (TP). The results obtained by sensitivity analysis were used subsequent model calibration and further uncertainty analysis. Monte Carlo (MC) method was employed to analyse the effects of parameter uncertainty on the model outputs. However, such problems are time‐consuming because the MC method required to invoke simulation model thousands of times. To address this challenge, a kriging surrogate model was developed to improve the overall calculation efficiency. The results obtained by sensitivity analysis showed that curve number value (CN2), soil evaporation compensation factor (ESCO), universal soil loss equation support practice factor (USLE_P) and initial organic nitrogen concentration in soil layer (SOL_ORGN) had significant effects on the SWAT outputs. The uncertainty analysis results showed that the uncertainty of runoff is the lowest, followed by TP and TN, and the uncertainty of sediment was the greatest. The kriging surrogate model has the ability to solve this time‐consuming problem rapidly with a high degree of accuracy, and thus it is very robust.