Improved ERRIS model for real-time correction of streamflow forecast based on deep learning
In order to improve the accuracy of streamflow forecast,the ERRIS model was improved based on LSTM,and the ERRIS-LSTM model was constructed for real-time correction of streamflow forecast.The Yarlung Zangbo River and Jiao River basins were taken as examples for comparative analysis.The results showed that,compared with the ERRIS model,the ERRIS-LSTM model increased the Nash-Sutcliffe efficiency coefficient by 4.1％and 1.1％,decreased the root mean squared error by 67.7％and 5.7％in streamflow forecast of the Yarlung Zangbo River and Jiao River basins,respectively.Especially for medium and low flows of the Yarlung Zangbo River Basin,the values of percent bias of streamflow forecast obtained by the ERRIS-LSTM model were reduced by 75.5％and 79.1％,respectively,and the statistical indexes of low flow in the Jiao River Basin obtained by the ERRIS-LSTM model were improved by more than 20％.The ERRIS-LSTM model could fully capture the continuity of the error series,and the ensemble forecasts generated by the ERRIS-LSTM model were more accurate,less uncertain,and more reliable than those of the ERRIS model,with the continuous ranked probability score reduced by more than 75％.In comparison with the deterministic corrected results of the LSTM model,the ERRIS-LSTM model can provide additional uncertainty information,which is promising in operational forecasting and decision-making in flood control.
streamflow forecastreal-time correctiondeep learningERRIS modelLSTM modelYarlung Zangbo River BasinJiao River Basin
万方数据
维普
