EMULATION OF URBAN RUNOFF MODEL BY DEEP LEARNING FOR BENCHMARK VIRTUAL HYETO AND HYDROGRAPH

Abstract

 In recent years, flood disaster in urban area have frequently occurred, and improving the accuracy of urban runoff prediction is a pressing issue. The urban runoff mechanism is complicated, and it is difficult to constract an accurate prediction model. So, in this paper, we aim to confirm whether the urban runoff model can be emulated by using the deep learning model, first of all, runoff volume (virtual hydrograph) using the urban runoff model and virtual rainfall (virtual hyetograph) was constructed. Then, using the created virtual hyetograph and virtual hydrograph, we constructed a deep neural network model and verified the reproducibility in the training data and the validation data. In addition, since the observation data of floods used as input data is limited, the reproduction characteristics when the number of training data was reduced were examined.