Classification of Daily Torrential Rainfall Patterns Based on a Robust Correlation Measure

Abstract

The objective of this study is to identify the main spatial distribution patterns associated with torrential rainfall days that linked to the topography of Peninsular Malaysia.This is done by applying cluster analysis on the most relevant principal directions extracted from a principal components analysis of the between day correlation. However, the characteristic of rainfall data in Peninsular Malaysia involve skewed observations which only take positive values and are skewed towards higher values. Thus, applying PCA based Pearson correlation on rainfall data set could affect cluster partitions and generate extremely unbalanced clusters. Tukey's biweight correlation is introduced to overcome the problem where the weight function down weights data values that is far from the center of the data. The findings indicate that ten rainfall patterns obtained are quite definite and clearly display the dominant role extended by the complex topography and exchange monsoons of the peninsular.