Spatial and attribute filtering as a complementary measure in the statistical prediction of tropical cyclone rainfall

Abstract

The increasing rate of tropical cyclone (TC) rainfall has put populations in the Western North Pacific Region at greater risk of TC rainfall-induced disasters. Statistical methodologies have shown potential in complementing existing prediction approaches. With TC track prediction accuracy significantly improving, statistical predictions have turned to TC tracks as a measure of similarity between TCs. Several studies have utilized Fuzzy C Means (FCM) to this end. However, FCM alone does not provide guidance on how many similar TCs should be used for predicting rainfall through ensemble averaging. While various number of ensemble members have been used to check the average error, such an approach yields only one number, which may not always be the most appropriate. In this study, we proposed a spatial and attribute filter to complement FCM identification of similar TCs. This filter excludes similar TCs with central pressure differences greater than 5% at strategic TC locations near land. The use of the filter yielded better rainfall prediction values than using FCM alone, as demonstrated in this study and validated against previous research findings. Our proposed model offers a reliable means of predicting TC rainfall when used in conjunction with accurately predicted TC tracks, representing a valuable complementary approach to existing prediction methods.