Tree model based eco-climatic vegetation classification and fuzzy mapping in diverse tropical deciduous ecosystems using multi-season NDVI

Abstract

Many vegetation classification strategies in tropical ecosystems involving conventional image processing of original satellite imagery bands require considerable prior site knowledge, statistical assumptions, and are difficult, expensive and inconsistent. In this paper we show that the intra-annual variation and rates of change in NDVI for different parts of a large forest area in combination with rules derived from a tree model can be used for detailed vegetation mapping. We used three-date NDVI data for the Biligiri Rangaswamy Temple Wildlife Sanctuary in Karnataka, southern India comprising mean NDVI, coefficient of variation (CV) and two NDVI change vectors corresponding to intraseasonal NDVI differences. A rule-based classification using a tree model was developed at two levels. The overall kappa statistic is 0.61 at level 1 classification, indicating a strong correspondence with the raster version of the available vector reference map based on ground data, even though the two maps are not strictly comparable. Several limitations of the available reference map have been highlighted by the new technique, especially the absence of ecotones. At level two the tree model map has provided detailed classification of dry deciduous forests and new classes not available in the reference map. The method in combination with reference data also provides a framework for fuzzy classification. This technique offers a relatively simple cost-effective alternative to existing classification strategies, especially for areas with diverse evergreen and deciduous vegetation elements.