Sentinel-2 Spatiotemporal Reconstruction Dataset

The Sentinel-2 Reflectance Reconstruction Dataset (S2RRD) provides prepared input data for deep-learning-based reconstruction of Sentinel-2 surface reflectance time series over the United Kingdom. The dataset supports the reconstruction of missing observations in the red (B4) and near-infrared (B8) bands using 15-day temporal composites at 100 m spatial resolution, together with meteorological, topographic, land-cover, and temporal predictors. It includes spatial–temporal patches prepared for 3D U-Net training, enabling research on cloud-gap reconstruction, temporal information recovery, and the contribution of environmental predictors to reflectance estimation. The related code can be found in: https://github.com/TolgaBkm/S2_Surface_Reconstruction