Development of Typhoon Damage Prediction Function Using a Logistic Distribution

Abstract

The purpose of this study is to develop a typhoon damage prediction function considering the correlation between damage amount due to typhoons and hydrometeorological factors estimated from disaster statistics in coastal districts. As a typhoon approaches, heavy rain, gale, and storms jointly and simultaneously cause severe damage. A total of 32 variables were selected to derive the function, such as factors that affect rainfall-runoff, maximum wind velocity, and central typhoon pressure. As observed, typhoon damages have a wide range-from hundreds of thousands of KRW to more than KRW 1 trillion-and accordingly have been classified into large and small groups depending on the amount of damage; thereafter, the typhoon damage prediction function was developed for each group. The logistic regression analysis was used for group classification of typhoon damages. Furthermore, the principal component regression analysis was employed for the typhoon damage prediction function. The principal component analysis effectively solved the problem of multicollinearity due to high correlation between 32-dimensional independent variables and rainfall-runoff related variables. According to the results of the typhoon damage prediction function, the Normalized Root Mean Squared Error (NRMSE) of the large group was 12.4%, while that of the small group was 18.5%. Keywords: Typhoon Damage Prediction Function, Logistic Regression, Principle Component Analysis, Coastal District