Evaluation of Sentinel-2 MSI and Landsat OLI data for land use and land cover classification across diverse geospatial regions using supervised machine learning classifiers

Abstract

Employing advanced remote sensing techniques, this paper undertakes a rigorous technical investigation into the comparative performance of Sentinel-2 and Landsat 9 datasets. Leveraging the analytical  power of four sophisticated machine learning algorithms – CART, Random Forest, Gradient Tree Boosting, and Support Vector Machine. The study scrutinizes land cover classification across diverse study areas  in Lucknow and Dehradun. Evaluating Sentinel-2’s multispectral sensor capabilities, particularly in the red-edge and near-infrared bands, the research dissects its prowess in vegetation-rich environments, exploiting the “red-edge effect” for precise vegetation assessment. Conversely, Landsat data, including Landsat 8  and Landsat 9, is assessed for its adaptability in areas characterized by diverse land cover types, including urban landscapes. Delving into algorithmic performance metrics, Random Forest and sentinel-2 emerge as  the preferred choice for most vegetated area like Dehradun about 90% accuracy across varied scenarios, Gradient Tree Boosting and Landsat data emerges as the preferred choice for most urbanized (populated)  areas like Lucknow about 92% accuracy across varied scenarios. This paper contributes novel insights into the comparative utility of Sentinel-2 and Landsat 9 datasets, bolstered by advanced machine learning methodologies, for refined land cover mapping and monitoring applications.