Author(s): Kejia Wen; Chengshuai Liu; Caihong Hu
Linked Author(s):
Keywords: Flash flood forecasting PE-FFM Model Storage-Infiltration Compatible Model HEC-HMS LSTM Multi-watershed verification
Abstract: Flash flood modeling is of great significance in flash flood early warning and prevention. This article constructs the Parameter Estimation-based Flash Flood Forecasting Model (PE-FFM), compares and explores the application effects with the storage-infiltration compatible model, HEC-HMS model, and LSTM model. It uses 236 flood events in 12 small watersheds for simulation, with 70% used for calibration and the remaining 30% for validation. The relative error of peak flow (REQ), absolute error of peak occurrence time, and Nash efficiency coefficient (NSE) are used as evaluation indicators to assess the model's simulation accuracy. The results show that In the research area, all four different models meet the requirements of flood forecasting. However, compared to the other three models, the PE-FFM model has fewer parameters, stronger applicability, and a more refined ability to control peak floods. The storage-infiltration compatible model has strong applicability in wet and semi-arid regions, but the disadvantages of having more parameters and complex computational processes affect its widespread application. The HEC-HMS model has a strong ability to handle different situations but has higher requirements for input data. The LSTM model performs well for long sequence rainfall data, but actual measured data for flash floods in small watersheds are often limited, which to some extent restricts the application of the LSTM model in flash flood forecasting. This study can provide guidance for future applications of flash flood forecasting models.
Year: 2025