Comparison and Correction of Runoff Simulation Based on XGBoost and SWAT

Abstract

Runoff simulation has always been a research hotspot in hydrological science, which has important guiding significance for water resources research. Based on the extreme gradient boosting (XGBoost) algorithm and soil and water assessment tool (SWAT) model, a runoff simulation model was constructed by using the observed hydrological data of Waizhou hydrological station in Ganjiang River Basin, and the simulation effects of the two models were compared. Meanwhile, based on the XGBoost model, the residual correction method was used to correct the runoff simulation results of the SWAT model. The results show that: ① The simulation effect of the XGBoost model is good, and the Nash-Sutcliffe efficiency (NSE) of daily runoff simulation of the XGBoost model is 15.66% higher than that of the SWAT model. ② The SWAT model is underestimated at high flow rates, while the XGBoost model results in basically consistent simulation values and observed values, and the simulated hydrograph has a high degree of fitting, showing a good correlation. ③ After XGBoost correction, the simulation accuracy of the SWAT model is obviously improved, which can effectively improve the runoff simulation effect. When the residual error is corrected, the NSE value of runoff simulation can reach 0.96, an increase of 15.66%.

Keywords

• XGBoost
• SWAT
• runoff simulation
• error correction