ZENITH International Journal of Multidisciplinary Research _______________ISSN 2231-5780 Vol.3 (7), July (2013) Online available at zenithresearch.org.in 317 COMPARATIVE ANALYSIS OF DIFFERENT SUPERVISED CLASSIFICATION TECHNIQUES USING LINEAR REGRESSION MODEL SUBHA CHAKRABORTY A , DEBALEENA MAJUMDAR B , SATIPRASAD SAHOO C A, B, C PROJECT ASSISTANT, DEPARTMENT OF CIVIL ENGINEERING, INDIAN INSTITUTE OF TECHNOLOGY, KHARAGPUR, WEST BENGAL, INDIA ABSTRACT The Sundarbans is a rich biodiversity with tidal mangrove forest in the world. It is a part of deltaic plain of fluvial marine deposits of Ganges–Brahmaputra basin. The main aim of this study is to identify the best Supervised Classification method using linear regression model. Thus main focus goes to three supervised classification methods; these are Minimum Distance, Maximum Likelihood and Parallelepiped. We use linear regression model with NDVI (Normalized Differenced Vegetation Index) value and different classification area. Here we found that Maximum Likelihood classification is more accurate comparison to others, depends upon regression coefficient and ground based observation. KEYWORDS: Linear Regression Model, Normalized Difference Vegetation Index (NDVI), Remote Sensing, Supervised Classification. ______________________________________________________________________________ 1. INTRODUCTION: In Remote Sensing and GIS research Vegetation indices considered as the property of being sensitive to a variety of biophysical vegetation canopy parameters and leaf chlorophyll concentration (Zhang L., et al 2005). Many researchers used these indices to quantify and monitor vegetation growth, health and associated environmental effects. Thereafter, Supervised Classification is a common technique to produce land use and land cover of a region using knowledge of features. This study has carried out in ERDAS Imagine, ENVI and Arc GIS platform on LANDSAT ETM+ satellite data. In our study we focused on the best Supervised Classification technique with the help of linear regression model. 2. MATERIALS AND METHODS: An area of approximately 38 km² that covers the reserve forest of Lothian Island, was chosen as a study area (Fig. 1), to analyze the comparison between best classification methods comparing to Vegetation Index. This present analysis requires knowledge of the physical characteristic of the land surface, remotely sensed satellite data and observed data recorded in ground verification. 2.1. OUTLINE OF THE STUDY AREA: The Lothian island, is a small island of The Sundarban, which extends from 88º18’10’’E to 88º21’30’’E longitude and 21º32’50’’N to 21º42’30’’N latitude. It is daily inundated by diurnal tide up to a certain distance from northern coast of the island. A north south curved road