LONG-TERM EVALUATION OF STORM SURGES BY NEURAL NETWORK — CASE STUDY IN ISE BAY —

Abstract

 Although needs of long-term evaluation of storm surges have been increasing, number of observation of actual storms and storm surges are insufficient for modeling. Since simulation cost of storm surges by a numerical model is still high for long-term evaluation, statistical approaches can be effective methods to reduce the simulation cost. To overcome shortage of number of observed data, we used a stochastic tropical cyclone model (STM). We developed Neural Network using synthetic typhoons by STM as training data and evaluated long-term change in storm surge targeting Ise Bay in Japan. We also examined difference of training methods and data mainly considering typhoon intensity, time series and number of training data. We got result that training method which considered the bias of training data and temporal intensity change of typhoons could predict better relatively. As a result of the long-term evaluation, when the return period becomes longer, the storm surge in the future climate will increase until certainlevel and will saturate about 1.3 times higher than the present climate.