Exploring Machine Learning Techniques for Classifying Land Use and Land Cover: A Comparative Analysis Using Google Earth Engine

Abstract

Sentinel 2, which provides a Level-1C dataset with a spatial resolution of 10 meters, is used in this work to assess a variety of machine-learning approaches for identifying land use and land cover. Gradient Boosting (GTB), Random Forest (RF), Support Vector Machines (SVM), Classification, and Regression Trees (CART) are a few of the techniques. The Google Earth Engine (GEE) platform was utilized for categorization. The results show that different algorithms classify land cover differently. For example, RF and CART identify agriculture as the major land cover, SVM indicates forest cover, and GTB emphasizes the significance of Agriculture. The algorithms’ performance was assessed by accuracy evaluation, which took into account indicators including Kappa coefficient, producer, consumer, and total correctness. The best overall accuracy and agreement with reference data are shown by SVM. GEE is useful for classifying LULCs, and the study offers land management and planning insights.