Land Use Land Cover Mapping Using Random Forest and Multispectral Satellite Imagery in Google Earth Engine

Remote sensing has emerged as a critical tool for environmental monitoring, resource management, and urban planning, offering timely and cost-effective data to map and assess the Earth's surface. This study used Sentinel-2 surface reflectance imagery within the Google Earth Engine (GEE) platform to investigate land use land cover (LULC) classification in Ajmer, Rajasthan. GEE's cloud-based environment enabled efficient data handling, median composite generation, and near real-time analytical capabilities. The random forest (RF) algorithm, a robust ensemble machine learning (ML) method, was used to classify four primary land cover classes: water, barren, vegetation, and built-up areas. Training and testing datasets were created by merging geotagged training samples created on the GEE platform, with 80 % of samples dedicated to training and 20 % reserved for testing. The model achieved an overall accuracy (OA) of $\mathbf{9 5. 7 4 \%}$, proving the reliability of the RF approach. The Kappa Coefficient (KC) was calculated to be $\mathbf{9 4. 3 2 \%}$. These findings suggest that remote sensing-driven machine learning (ML) classifications can produce high-fidelity thematic maps, especially through platforms like Google Earth Engine (GEE). They also highlight the potential for replicating similar workflows in other regions, aiding resource allocation, policy formation, and sustainable development.