Clustering of rainfall stations and distinguishing influential factors using PCA and HCA techniques over the western dry region of India

Abstract

This study used hierarchical cluster analysis (HCA) to delineate the spatial patterns of monthly, seasonal and annual rainfall by clustering 62 stations in the western arid region of India based on a 55 year (1957–2011) data set. The statistical properties of clusters were computed and box–whisker plots plotted. Furthermore, the relative influence of three geographical factors (longitude, latitude and altitude) and five statistical parameters (the mean, standard deviation (SD), co-efficient of variation (CV), and maximum and minimum rainfall) on mean rainfall was investigated using principal component analysis (PCA). The use of HCA resulted in four rainfall clusters geographically located at a distinct position. Cluster I, characterized by the lowest mean rainfall and highest CV, was located in the western portion, whereas mean rainfall was the highest for cluster IV situated in the eastern portion. Box–whisker plots revealed a slight skewness, although the monsoon and annual rainfall followed a normal distribution. The PCA results indicted two to three significant principal components (PCs) with eigenvalues > 1. In four clusters, two PCs explained the major variance, ranging from 69.41% (June) to 91.83% (August) in monthly rainfall, from 63.62% (monsoon) to 93.30% (post-monsoon) in seasonal rainfall, and from 71.48% to 90.73% in annual rainfall. In monthly and seasonal rainfall, first PC 1 is termed the “mean rainfall component”, which has strong to moderate associations with longitude, and is equally opposed by the CV. These findings are vital for planners and decision-makers to formulate strategies to manage unusual rainwater quantities.