Machine Learning-Driven Land Use Land Cover Classification using Different Algorithms on Sentinel-1 and Sentinel-2 Imagery

Land Use Land Cover (LULC) classification is crucial for understanding and managing the planet’s resources. The present study examines LULC classification using machine learning (ML) algorithms and geospatial data in the Karulai region of Kerala. Sentinel-1 and Sentinel-2A satellites provided Multispectral and radar imagery, offering high-resolution (10 m), frequent data. Three ML models Random Forest (RF), Classification and Regression Trees (CART), and K-Nearest Neighbors (KNN) were evaluated for classification accuracy. RF achieved the highest accuracy (93.87%) and Kappa coefficient (0.916), outperforming CART and KNN in complex land cover types, particularly forest and built-up areas. RF accurately detected 101.7 ha of water and 19,692.67 ha of forest, while CART and KNN showed variability in urban and plantation areas. Producer and user accuracy metrics further validated RF’s reliability, with 98% producer accuracy for teak plantations. Challenges emerged in classifying rubber plantations and built-up areas, but RF remained the most robust model. These findings highlight the importance of ML in LULC mapping, with applications in urban planning, forest monitoring, and agricultural management. This study improves LULC classification accuracy, aiding sustainable land management and planning.