Comparative Study of Supervised Classification for LULC Using Geospatial Technology

This paper presents the study on Land Use/Land Cover (LULC) classification of multispectral image in the Jaykwadi region, Aurangabad district, Maharashtra. A Landsat 8 level 2 image is selected for the study using the Minimum Distance (MD) and Spectral Angle Mapper (SAM) classifier and classified into five LULC classes, i.e., Vegetation, Fallow Land, Barren Land, Built-up and water. This study aims to explore and compare the performance of the MD and SAM classifiers for LULC classification by evaluating their accuracy, strengths, and limitations. The MD classifier has given an overall accuracy of 80.24% with a kappa coefficient of 0.70, while the SAM classifier has given the highest overall accuracy of 85.74% with a kappa coefficient of 0.81. It was found that MD and SAM provided similar results for water bodies and Vegetation, whereas significant differences were detected in Barren Land, Fallow Land and Built-up areas; due to their spectral signature similarity. The Landsat 8 image were captured on 6 May 2022, and this is a post harvesting period, so naturally there should be more Fallow land which is identified by SAM Classifier. Minimum Distance classifier identify 1.15% land as fallow land and 52.59% as barren land, whereas SAM classifier classified 39.83% land as fallow land and 17.48% as Barren Land, which is more accurate as compared to MD classifier. SAM provides satisfactory value for each type of LULC class as compared to MD.