A 1.5 m resolution land cover map of South China created with Superpixel U-Net

We applied Superpixel_U-Net on 6.1 TB of very-high-resolution (VHR) remote sensing images from ESRI World Imagery covering the entirety of SC, to produce a 1.5m resolution land cover map for SC(named as the SCLC map). Towards this purpose, a total of 43,000 VHR images (512 × 512 pixels) were manually annotated into eight land cover classes. Vegetation and water indices derived from Sentinel-2 were used as auxiliary input datasets. Finally, the SCLC map was validated using pixel-level and object-level ground truth datasets, as well as the official land survey reports provided by the Chinese government. SCLC map demonstrates higher accuracy compared to existing high-resolution land cover products. Our dataset provides a solid foundation for further analyses that require accurate delineation of land cover classes in SC.