Mapping the Full Coastal Salinization Gradient from Upland to Salt Marsh by Satellite Hyperspectral Imaging: Evidence from EnMAP and Deep Learning

Dataset Description This dataset supports the study “Mapping the Full Coastal Salinization Gradient from Upland to Salt Marsh by Satellite Hyperspectral Imaging: Evidence from EnMAP and Deep Learning.” It contains a hyperspectral image and the corresponding ground-truth land-cover labels for coastal salinization gradient mapping. The dataset includes two MATLAB files: BW.mat: Hyperspectral remote sensing image converted from the original GeoTIFF image. gt2.mat: Pixel-level ground-truth label map corresponding spatially to the hyperspectral image. The ground-truth map contains six land-cover classes representing the environmental gradient from upland areas to tidal marshes: Class Land-cover type Number of samples Definition 1 Transition forest 17,375 Low-lying forests between marshes and upland forests where tree mortality caused by seawater intrusion has already begun. 2 Marsh 32,844 Tidal wetlands dominated by herbaceous hydrophytes, including cordgrass, rushes, and sedges. 3 Upland forest 23,480 Primary or long-established secondary forests characterized by a closed canopy and mature trees taller than 5 m. 4 Transition farmland 8,815 Low-lying agricultural land adjacent to marshes where soil salinity, periodic inundation, or vegetation stress indicates early-stage marsh encroachment. 5 Upland farmland 16,806 Agricultural land located at higher elevations beyond regular tidal influence, characterized by stable cropland conditions and relatively low soil salinity. 6 Water 8,283 Open water bodies, including tidal creeks, ponds, and channels, characterized by persistent inundation. The class labels in gt2.mat are encoded as integer values from 1 to 6. Pixels not assigned to any of the six classes may be represented by 0, depending on the preprocessing and labeling procedure. This dataset can be used for hyperspectral image classification, coastal land-cover mapping, coastal salinization gradient analysis, seawater-intrusion impact assessment, and the development and evaluation of machine-learning or deep-learning models.