Level I classification through computation of Normalized Difference Vegetation Indices (NDVI) is done to distinguish vegetative and non- vegetative areas in the district. Band-3 and Band-4 images were used to compute NDVI. NDVI is defined as

Where NIR is the spectral radiance in the near infrared band and RED is the spectral radiance in the red band. The index normalizes the difference between the bands so that the values range between -1 and +1. Vegetative and non-vegetative components from the NDVI image were delineated using Classification Trees approach. Classifi- cation Tree approach uses Decision Tree method. Trees provide a hierarchical and non-linear classification method and are suited to handling non-parametric training data as well as categorical or missing data. Classification Trees reduce the dimensionality of the data sets. Classification Trees use a set of independent variables to predict class membership. A tree is constructed by recursively partitioning a data set into purer homogeneous subsets. The method uses a deviance measure, the likelihood ratio statistic, to compare all possible splits of the data to find the one split that maximizes the dissimilarity among resulting subsets. Possible splits of each independent variable is examined and the particular split within a particular variable that produces the largest deviance measure is chosen to partition the dependent data (Madhu M.K., et.al. 2000).

Data coinciding with maximum vegetative cover were selected for identification of training sites. The training sites were selected such that they show distinct vegetative covers and are homogenous with large extent. The sites for which detailed ground data had been collected were located on the imagery and the digital numbers of blocks of pixels were extracted for the classification. Signature files were generated using the training data for each vegetative type. Maximum Likelihood classification was performed and the groups defined by this classification were then related to the ground data. Accuracy assessment was made for the classified image using signature files, classified image and PAN image.

Extensive mapping of fuelwood trees-Prosopis juliflora was carried out in Iragasandra and Huthur villages of Kolar taluk. The study was done in two phases. In the first phase all the areas with Prosopis juliflora were identified. A polygon was drawn surrounding this patch, taking the co-ordinates every 10 to 15 m (latitude and longitude) and the same is marked in village map (cadastral map). Details like age of a tree, its girth, type of soil and associated tree species, if any were noted down in a field map. Then the PAN data corresponding to these villages were analysed to get the spectral signatures corresponding to field data. There was a large variation in spectral response for each homogenous patch. Further field investigations were carried out to identify the factors responsible for variations. In the second phase, sub polygon was taken with in the polygon to indicate thick patches, patches with adult trees, patches with only juveniles and patches with trees sparsely spread. This approach enabled in identifying the factors for variation in spectral signatures. In order to extend this investigation to other taluks of Kolar district, spectral signatures corresponding to Prosopis juliflora were mapped in Gauribidanur taluk. Various parameters such as age, density, etc. were taken in to consideration for this purpose. In order to take advantage of spectral properties of LISS III and spatial resolution of PAN, LISS and PAN data were merged, which helped in identifying the patches of Prosopis juliflora.

In order to verify the mapping accuracy, sampling units were selected which includes all type of Prosopis juliflora patches along with parameters such as density, age, etc. In order to get village wise information, vector layer of Gauribidanur taluk with village boundaries was over laid. Road and stream layers were overlaid to identify the location of patches precisely.