Study on Ensemble Calibration of Flood Forecasting Based on Response Curve of Rainfall Dynamic System and LSTMStudy on Ensemble Calibration of Flood Forecasting Based on Response Curve of rainfall dynamic system and LSTML. Tiana et al.

To improve flood forecasting accuracy, the dynamic system response curve correction method was employed to invert and establish an error time series of areal rainfall in the Shouxi River Basin in Sichuan Province and the Qingyangcha Basin in Shaanxi Province. The areal rainfall in the watershed was corrected using the obtained error time series. The corrected areal rainfall was then used as input for flood forecasting using the excess storage and excess infiltration simultaneously model. Additionally, a hierarchical optimization method and LSTM error output correction method were applied to calibrate the three sources of errors. The results showed that the accuracy of flood peak discharge improved after the correction of areal rainfall. Specifically, in the validation set of the Shouxi River Basin, the absolute error of flood peak discharge decreased by 0.56% to 6.3% for 12 out of 15 flood events. The Nash–Sutcliffe Efficiency (NSE) of flood discharge increased by 0.002 to 0.015 for 13 flood events, and the time lag of two flood peaks shortened by 1 h. In the validation set of the Qingyangcha Basin, the absolute error of flood peak discharge decreased by 0.23% to 5.49% for 5 out of 6 flood events. The NSE of flood discharge increased by 0.01 to 0.071 for 5 flood events, and the time lag of two flood peaks shortened by 1 h. Overall, the results demonstrate that this method can reduce the forecast error and improve the accuracy of flood forecasting in the watershed.