Evaluation of Real-Time Water Level Prediction Technology Using Statistical Models for Reducing Urban Flood Risk

The frequency of localized short-term torrential rains that exceed the planned rainfall is increasing along with inundation damage due to inland flooding. Step wise inundation measures utilizing existing stock and disaster prevention/mitigation for excessive rainfall are required. In this study, we describe the results of empirical research using a statistical model constructed based on rainfall and water level observation data as a highly accurate water level prediction method suitable for real-time prediction. This is aimed at application in flood control activities and operation support of pump facilities. By comparing and verifying the prediction accuracy between the water level prediction model and the statistical model by Convolutional Neural Network (CNN), which is generally used as an image recognition technology, the usefulness of the statistical model was confirmed. Further improvement in accuracy and widespread use of these water level prediction models are expected.