Comparison of Three Machine Learning Algorithms Using Google Earth Engine for Land Use Land Cover Classification

Abstract

Google Earth Engine (GEE) is presently the most innovative international open-source platform for the advanced-level analysis of geospatial big data. In this study, we used three machine learning algorithms to apply this cloud platform for Land Use Land Cover (LULC) research in the Mardan, Pakistan. The machine learning algorithm in GEE is the most advanced technique to generate reliable and informative LULC maps from various satellite data to present reliable results. The primary goal of the present study is to compare the performance of various machine learning models (i.e., classification and regression trees [CART], support vector machine [SVM], and random forest [RF]) in GEE for the reliable four classes LULC maps using the Sentinel-2 imageries of 2022. In the current study, three satellite indices like the Normalized Difference Vegetation Index, Modified Normalized Difference Water Index, and Normalized Difference Built Index were applied to detect the features (i.e., vegetation, built, barren land, and water bodies in the study area). The performance of all three models was evaluated by validation and accuracy assessments. The Kappa coefficients of CART, SVM, and RF for Sentinel-2 images were 94%, 95%, and 97%, while the average overall accuracy is 96.25%, 97%, and 98.68%, respectively. The present study illustates that in this classification and comparison, RF performed better than SVM and CART. The current research study revealed that GEE has speedily processed the satellite imageries to develop the four classes of reliable LULC maps of the study area with the best accuracy results and deliver excellent support for further analysis.