S2GAIA: A Seasonally Aware Multi-Year Dataset for Deep Learning-Based Land Cover Mapping Using Sentinel-2 Imagery in Greece

S2GAIA is a multi-year, seasonally aware Sentinel-2 dataset developed for pixel-wise land use/land cover (LULC) mapping in Greece. It comprises 34,030 image patches of 256 × 256 pixels at 10 m spatial resolution, accompanied by dense pixel-level annotations across 22 land-cover classes and spanning the period 2017–2024. For each year, Sentinel-2 observations are organized into four seasonal periods representing winter, spring, summer, and autumn. Each seasonal composite contains seven Sentinel-2 spectral bands (B02, B03, B04, B08, B8A, B11, and B12), resulting in 28 spectral-temporal channels per image patch when the four seasons are combined. This multi-seasonal structure is designed to capture phenological and seasonal variability and facilitate the discrimination of temporally dynamic land-cover classes. The S2GAIA annotations were generated through a reproducible multi-source methodology that harmonizes four Copernicus Land Monitoring Service products with national datasets providing photovoltaic installation locations and annual wildfire perimeters. The annotation procedure combines class reclassification, spatial-priority rules, manual verification, and independent quality assessment. The resulting 22-class taxonomy preserves land-cover categories relevant to national-scale environmental monitoring and spatial planning, including photovoltaic installations and temporally updated burned areas. S2GAIA was designed to support the training, evaluation, and comparison of deep learning models for semantic segmentation, as well as research on multi-temporal land-cover classification and change detection. Its multi-year and seasonally explicit structure provides a resource for studying heterogeneous Mediterranean landscapes and supports applications in Greece and comparable Mediterranean and Southern European environments. Associated publication:Temenos, A., Verykokou, S., Papatheodorou, E., et al. (2026). S2GAIA: A seasonally aware multi-year dataset for deep learning-based land cover mapping using Sentinel-2 imagery in Greece. Remote Sensing Applications: Society and Environment, 102197.DOI: 10.1016/j.rsase.2026.102197