PREDICTING FLOOD INUNDATION AREA BY RAINFALL-RUNOFF-INUNDATION MODEL EMULATOR

An emulator was developed for a runoff inundation model predicting the range of inundation using a machine-learning method based on the results of a rainfall-runoff inundation model experiment. The 1500-year equivalent of the current reproduction results of large-scale ensemble climate experiment data was used as input. The model was validated using the Omono River basin, which has frequently flooded in the past, as an example. It was found that the machine-learning method could reproduce the experimental results of the model with approximately 80%–90% accuracy for the major inundation areas near the river channel. Comparing the results of inputting the actual observed rainfall during the July 2017 floods into the emulator obtained by machine learning shows that the prediction results were comparable to those of the experiments using the rainfall-runoff inundation model. However, the accuracy of the reproduction varied depending on the machine-learning method.