降雨时间分布对中长期旬极端径流模拟的影响OA

In long-term runoff forecasting,information within the forecast time step is often overlooked,such as temporal distribution of rainfall.To improve the simulation accuracy of extreme runoff in ten-day scale,an improved partial least squares regression model N-PLS and a neural network model LSTM are used to construct models for Wudongde Reservoir and Xiangsan catchment area,comparing the effects of different input features and data on these two models.The results show that:(a)R2 and NSE of the two models with different input schemes for Wudongde Reservoir are generally between 0.95 and 0.98,which is higher than that of Xiangsan catchment area;(b)the impact of the four sets of feature input schemes on the LSTM model is significantly weaker than that of the N-PLS model,and increasing characteristics within ten-day will slightly improve the performance of the LSTM model,which to some extent reduces the bias of overestimation in severe drought runoff simulation during flood season for Wudongde Reservoir and reduces the bias of underestimation in severe flood runoff simulation for Xiangsan catchment area;(c)adding detailed features can significantly improve the performance of the statistical model N-PLS,with a more pronounced performance in Xiangsan catchment area compared to Wudongde Reservoir,and can significantly reduce the bias of overestimation in severe drought runoff simulation and underestimation in severe flood runoff simulation;and(d)due to the stronger stability of the LSTM model,the overall accuracy of the model is higher than that of the N-PLS model,however,for the severe drought runoff during the dry season of Wudongde Reservoir and severe flood runoff during the flood season of Xiangsan catchment area,the non normalized N-PLS model considering the rainfall information and temperature characteristics within ten-day performs better.Overall,considering more hydrological characteristic inputs such as rainfall information and temperature within the ten-day period can improve the simulation accuracy of extreme runoff during the ten-day period.