A STUDY ON METHOD OF EARLY STORM SURGE DAMAGE PREDICTION USING NEURAL NETWORK

Abstract

 It is concerned that storm surges amplified by climate change will increase damage in the future climate, and when a storm surge actually occurs, it is required to quickly understand the whole aspect of damage. The study examined two methods: 1) a loss function method, which is a conventional way; 2) a new method of early damage estimation using a neural network, which is presented in the present paper. First, a series of storm surge numerical simulations was carried out using the virtual 525 typhoons, and the amount of damage due to storm surge flooding along the Tokyo Bay coast was estimated. Next, we constructed the loss function and neural network based on above results. Finally, the damage amounts by two methods were compared with verification data and estimation accuracies were then assessed. As a result, the estimation accuracy for the damage amount is higher in the new method using the neural network than in the conventional loss function. The results reveal that a damage amount can be quickly and more accurately estimated for a devastating storm surge disaster.