Abstract
Storm surge disasters have a tremendous impact on economic and social life, and the damage will be further exacerbated due to the intensification of typhoons by global warming in the future. From such a background, researches on real-time storm surge prediction are being conducted. However, because of uncertainty in meteorological forecasts during typhoon periods, it is difficult to accurately predict storm surges caused by the external meteorological forces. Ensemble experiments are required for predictions that take uncertainty into consideration, but it is not realistic to carry out a number of numerical simulations in real time from the viewpoint of computational costs. Therefore, we developed a probabilistic real-time storm surge prediction system using neural network that considers the uncertainty of typhoon prediction. The neural network model has a lower computational cost than that of the numerical simulation model. Additionally, the neural network model enables high-speed computation, but it requires a lot of training data for development. Thus, we extracted the typhoons from the database called “Database for Policy Decision-Making for Future Climate Change” created by the “Climate Change Risk Information Creation Program” implemented by the Ministry of Education. The general flow of the system is as follows. First, predict the future development of a typhoon using recurrent neural networks. Then create typhoon ensembles considering the error of the typhoon prediction model. Next, create storm surge ensembles by applying the storm surge prediction model to the typhoon ensembles in the above. Finally, evaluate storm surge ensembles probabilistically.
As a result of applying the system for typhoons to verify the model accuracy, we were able to fully demonstrate the effectiveness of this system. This made it possible for local governments to post useful information as “storm surge occurrence probability” when calling on residents to take appropriate actions and countermeasures when a typhoon strikes.
