Assessment of land use and land cover mapping using object-based classification techniques for the eastern districts of Tamil Nadu

LULC (Land use and land cover) mapping is crucial for understanding environmental monitoring, supporting sustainable development, and managing natural resources. This study evaluated the accuracy of object-based LULC classification using Sentinel-2 data and machine learning classifiers in the Ariyalur, Perambalur, and Mayiladuthurai districts of Tamil Nadu during the kharif season of 2023. OBIA (Object-based image analysis) clusters pixels based on their spectral and spatial characteristics, utilizing segmentation to generate masks that effectively represent the image content. The OBIA methodology involves multiresolution segmentation using eCognition software to delineate homogeneous image objects based on spectral, spatial, and contextual characteristics. Several widely used machine learning algorithms, including Random forest (RF), Support vector machine (SVM), Decision Tree (DT), Naive bayes (NB) and k-nearest Neighbor (k-NN), were evaluated to improve classification accuracy. The classification results varied across the districts, with the RF algorithm consistently demonstrating high performance. The Perambalur and Mayiladuthurai RF achieved an overall accuracy of 88 %, with a kappa coefficient of 0.76 and 83 % and a kappa coefficient of 0.66. In Ariyalur, the DT model was used, with an accuracy of 85 % and a kappa coefficient of 0.70. The NB and k-NN classifiers achieved lower accuracies in all districts. In contrast, the RF algorithm was the most reliable for LULC classification in these areas, highlighting its strength and efficiency in accurately identifying complex land cover patterns.