AI-Based Land Use and Land Cover Classification Using Sentinel-2 Imagery in California, USA

Abstract

Land use and land cover (LULC) data are very important for achieving sustainable urban planning, environmental monitoring, agriculture management, and natural resource conservation. In this study, a framework based on artificial intelligence (AI) technology is presented for LULC classification through free-to-use Sentinel-2 multispectral imagery in California, USA, which is considered to be one of the most environmentally diverse places in the world. Cloud-free Sentinel-2 Level-2A images of 2025-2026 will be used, and spectral bands will be combined with vegetation and water indices such as NDVI, NDWI, and NDBI for better discrimination of land cover classes. Reference samples will be taken from the ESA World Cover dataset, and then training (80%) and test (20%) datasets will be created. The Random Forest classifier will be applied in the Google Earth Engine (GEE) environment because of its efficient processing speed and high classification performance. Evaluation metrics will include overall accuracy, precision, recall, F1-score, and Kappa coefficient. The result of this study is expected to be a high- resolution map showing major land cover classes like urban areas, croplands, forests, grasslands, water, wetlands, and bare lands with more than 90% overall accuracy. The suggested approach provides an affordable, flexible, and replicable process based solely on freely available satellite images, cloud computing and AI methods. It is anticipated that the results will provide evidence-based data for land use management, environmental surveillance, and sustainable development, thus showing the capabilities of using AI in remote sensing for fast and accurate LULC mapping.