Development and improvement of a method for determining the worst-case typhoon path for storm surge deviation through Bayesian optimization

Abstract

A search for the worst-case typhoon path for storm surge is conducted to facilitate storm surge disaster prevention. A widely used method today is grid search, which is an exhaustive method in which the reference paths are paralleled and rotated in small steps, and storm surge simulations are performed for each path. However, the computational cost is enormous, and adequate typhoon paths may not be considered. Therefore, our goal is to propose a search method that can find the worst-case typhoon path at a lower cost. We have proposed a search method based on Bayesian optimization that is more efficient than grid search. However, issues such as the criteria for terminating the search and the complexity level of the function that can be applied remain unaddressed. Here, we examined the conditions for terminating the search using Bayesian optimization and the conditions for its application to the shape of the function. The results showed that Bayesian optimization can stably identify more dangerous typhoon paths than that found by grid search at a lower cost. We also confirmed that the search works for a wide range of function shapes. In addition, we searched for the worst-case typhoon path in real sea under the degrees of freedom that are too large to be searched by grid search in terms of computational cost. Consequently, a path approximately 5.5% larger than that found by grid search under low degrees of freedom was found. This means that the current search method underestimates the worst-case situation. The results confirm the usefulness of the search method using Bayesian optimization developed.