Rainfall patterns over India: Classification with Fuzzy c-means method

Seasonal (June through September) percentage departure from normal rainfall patterns over India for the period 1871-1994 have been classified using Fuzzy c-means method (FCM) to identify the dominant modes of spatio-temporal variability in the Indian monsoon rainfall. Unlike the hard clustering methods, for example the Map-to-Map (MM) correlation method and the k-means (KM) clustering method, this method does not force a pattern to get classified into only one cluster but assigns varying membership to every cluster. Thus marginal patterns get classified into all clusters with different memberships. Patterns for the 124-year period are represented by the four dominant clusters. The spatial patterns associated with the extreme (deficient/excess) Indian monsoon rainfall (IMR) get high membership in one of the clusters only, while the patterns associated with the normal IMR get almost equal membership to all clusters. Even the spatial patterns during the El Niño/La Nina episodes show high preference to a particular cluster. Time variation of each cluster shows that there are epochs where a particular cluster dominates. Possible dynamic causes leading to the clusters are examined. Merits and demerits of the FCM method are also discussed