Fuzzy C-means clustering for rainfall signature detection and towards understanding weather patterns

Abstract

This study investigates the use of fuzzy C-means (FCM) clustering algorithm to cluster Malaysia weather data for rainfall signature detection. Rainfall signature detection of several stations is vital to gain insight into the behaviour of the specific stations. Understanding rainfall behaviour has many advantages, such as preparing mitigation initiatives and developing early warning systems in specific areas to avoid abrupt changes that may affect the security and economy of the area. Three stations in the state of Selangor, Malaysia situated in Sepang, Subang, and Petaling Jaya collected rainfall data from 2009 to 2011. A comparison of the FCM with another fuzzy clustering algorithm, namely the Gustafson-Kessel (GK) proves that in terms of the number of iterations, the FCM consumes less processing time and gives the optimal number of clusters. Through verification tests using three different validity indices, the performance of the FCM could compete with that of the GK to produce a better validity index. Statistical analysis using the analysis of variance (ANOVA) test showed different parameters representing different stations, indicating the contributing factor in the formation of the cluster. This study gives new insight into analysing different signatures among the rainfall stations.