Comparison of flood simulation capabilities of a hydrologic model and a machine learning model

Abstract

Abstract Machine learning models have been widely used for flood simulation. Few studies have compared the flood simulation capabilities of machine learning models and hydrologic models. This study compared the flood simulation capabilities of the SIMHYD hydrologic model and the LSTM machine learning model in 232 basins with different climate conditions. The results show that although the LSTM model significantly outperforms the SIMHYD model in the calibration period, it has significant performance degradation in the validation period. Basin characteristics had limited impacts on the performance difference between the LSTM model and the SIMHYD model. The extension of the calibration period improves the performance of the LSTM model, while it has a limited impact on the performance of the SIMHYD model. Thus, machine learning models are recommended for simulating floods if enough training data are available, otherwise, hydrologic models could be a better choice. This study is helpful to the choice of flood simulation models in different situations.