ABSTRACT
The combination of the machine learning and the numerical modeling is proposed in this study to predict the sediment budget in the watershed. HEC-HMS and SRH2D are the numerical models for hydrologic and hydraulic processes of the watershed system. The numerical models were calibrated and verified with field measurements which are difficult to acquire. Then, we simulated many scenarios with the models to create the artificial database. The ANN methods (RBFNN, BPNN, LSTM) conducted to build the relationship between hydrological measurements and the sediment concentration and then used as the simple equation to predict the sediment yield in the watershed. The Shimen watershed was used as the study case. Two storm events were applied to do the calibration and verification on numerical models. Five artificial scenarios were simulated to create the databased. One event is used to testify the prediction. The result shows that the proposed method can properly predict the event with RBFNN method.
