降雨输入误差对新安江模型径流模拟影响研究OA

Rainfall is a critical input to hydrological models,and its data quality directly impacts the accuracy and reliability of runoff simulation.Investigating the influence of rainfall input errors on runoff simulation is essential for identifying the sources of hydrological simulation errors and improving simulation precision.This study systematically analyzes the comprehensive impact of multiple types of rainfall errors on runoff simulation using the Xin'anjiang(XAJ)model,integrated with explainable machine learning methods.A total of 12 typical flood events were selected,and a multi-dimensional rainfall error perturbation scheme—including systematic bias,intensity bias,and spatiotemporal distribution errors—was designed.Twelve rainfall error features were extracted,and XGBoost was applied to explore the relationship between rainfall errors and runoff simulation,thereby constructing a predictive model for the simulation accuracy(Nash-Sutcliffe efficiency,NSE,and Relative Error,RE)of the XAJ model.SHAP values were further used to quantitatively analyze the influence mechanisms of rainfall error features on runoff simulation accuracy,and to evaluate the global importance,directional effects,and interactions of each rainfall error feature.The results show that:The XGBoost model effectively predicts runoff simulation accuracy(NSE and RE),with an average coefficient of determination(R²)exceeding 0.96 on the test set.Relative Bias(RB)of rainfall is the dominant factor affecting runoff simulation accuracy,with heavy rainfall bias and Root Mean Square Error(RMSE)also being important influencing factors.RB exhibits a strong linear relationship with RE,but a nonlinear relationship with NSE,accompanied by a threshold effect:when RB exceeds 15％,NSE decreases significantly,and overestimated rainfall has a more pronounced impact on NSE than underestimated rainfall;When the rainfall RMSE exceeds 0.2 mm/h,it interacts with RB to jointly intensify runoff simulation uncertainty.The findings provide a new analytical framework and quantitative basis for controlling rainfall input errors in runoff simulation and improving the accuracy of flood forecasting.