Land Use Land Cover Classification Using a Novel Decision Tree Algorithm and Satellite Data Sets

Abstract

Since many changes are taking place on the surface of the earth for various reasons such as human activities, natural calamities, and environmental changes, the up-to-date regional land information of an area has got lot of importance for local authorities. Though this information can be collected using traditional land surveying methods it suffers from certain issues such as lot of man power involvement, more time consumption, etc. With the advent of technology the satellite data images available from various optical and polarimetric SAR (Synthetic Aperture Radar) sensors can be used for this land use land cover classification purpose. Since the polarimetric SAR (polSAR) data can provide more useful information than optical data it is more preferable for the land use land cover (LULC) classification. PolSAR data available from various airborne sensors and space-borne sensors which are launched into the space during the recent past can be best utilized for LULC classification of any area irrespective of seasonal effects. Though the available classification algorithms are efficient and produce good classification accuracy results they have their own drawbacks for some reasons. The decision tree algorithm developed for LULC classification in the present study is based on Gumbel distribution of nonlinear regression model [1]. Because of the flexibility in the design and other advantages this decision tree algorithm produced consistent classification accuracy results for all the data sets used in the present study [2].