Bayesian Central Clustering: Application to Classification of Landscapes in Western Ghats in India

People: 

Purdue University, Jayanta K. Ghosh
Indian Statistical Institute, Calcutta, Sourabh Bhattacharya
Indian Statistical Institute, Calcutta, Kajal Dihidar
Indian Statistical Institute, Calcutta, Tapas Samanta 

Description: For some years I have been analyzing satellite data for one of the biodiversity hotspots in India, namely Western Ghats in South India. India's other bio hot spot is the Eastern Himalays. Both are recognized by UNESCO. The data consists of about (55,000).(1000) pixels in four spectra. The object was to use cluster analysis to monitor and alert to degradation. 

Analysis led to questions of nonunique clusterings, comparing them, finding suitable metrics to measure similarity between two clusterings with uncertainty about labeling of clusters in two clusterings, assigning uncertainty to different clusterings via Bayesian methods, and presenting the results of analysis in a way that would be easy to comprehend. Each of these items is a major methodological problem. We have developed tentative solutions with a paper forthcoming. 

The data for this project was obtained from Madhab Gadgil, India's foremost ecologist and a foreign member of the US Academy of Science, and his postdoctoral student, N. Harini, who assisted in the data transfer from I.I. Sc, Bangalore to ISI , Calcutta.