NDVILand: A Dataset of NDVI-Enhanced EuroSAT Multispectral Satellite Images for Land Use Mapping

NDVILand is a structured, multi-class geospatial dataset designed for satellite imagery-based surface characterization and multi-terrain analysis. The dataset comprises 27,000+ image patches of uniform 64 × 64 pixel resolution, collected from high-resolution Sentinel-2 multispectral satellite imagery. Each geographical location is provided in its raw, uncompressed 16-bit GeoTIFF format to preserve complete radiometric sensor depth for scientific computing. The dataset captures a balanced representation of 10 distinct terrain types, including dense forests, seasonal annual crops, residential suburban areas, commercial industrial layouts, asphalt highway infrastructure, and open sea/lake water bodies. Features are pre-selected across targeted electromagnetic bands, specifically visible Red (Band 4) and Near-Infrared (Band 8) to enable localized evaluation of vegetation vitality and surface absorption properties. The NDVILand dataset is mapped across the RdYlGn (Red-Yellow-Green) color spectrum. It is primarily meant for distributed geospatial benchmarking, downstream machine learning semantic segmentation tasks, and precision land cover classification purposes.