The development of flood forecasting technology is crucial to flood control. Therefore, it is very essential to use the method of real-time error correction to improve the accuracy and reliability of the flood forecasting model. For flood forecasting, this study evaluated the performance of a single Excess Infiltration and Excess Storage (EIES) flood forecast model and the forecast model after error correction using the linear Auto Regressive, Auto Regressive Moving Average with exogenous inputs (ARMAX), and Long Short-term Memory Network (LSTM), and then compared the performance of each model forced with historical flood data in the upper reaches of Jingle station of the Fen River in China. These EIES-standalone, EIES-AR, EIES-ARMAX, and EIES-LSTM frameworks are field-tested for 1- to 6-hours lead-time flood forecasting with historical flood data. The capability of the four models are compared using the mean absolute error (MAE), Nash–Sutcliffe efficiency (NSE), Pearson’s correlation coefficients (r), and Percent error in volume (Evol). The evaluation measures analysis reveal that EIES-AR and EIES-ARMAX perform acceptable when the lead time is 1 hour(NSE>0.7), but poorly when the lead time is 2-6 hours; EIES-LSTM model performs well and is the best approach of these models for short to medium range flood forecasting with up to 6 hours lead-time(NSE≥0.75).