A global satellite-derived coral bleaching dataset (2016–2025)

This dataset provides monthly predicted coral bleaching for reef cells with a ca. 1km resolution across the global tropics, from 2016 to 2025. Bleaching is predicted using a recurrent (LSTM) neural network trained on survey- and aerial-survey-based bleaching observations, applied to monthly satellite features derived from Sentinel-2 imagery (brightness normalized against adjacent non-coral substrate) and MODIS ocean-colour. Reef locations are restricted to cells overlapping the Allen Coral Atlas benthic map. Each row corresponds to one reef grid cell in one month and reports its centroid coordinates, a mean predicted bleaching probability (predicted_prob), and a binary bleaching indicator (predicted_class). The dataset is derived from Sentinel-2 (Copernicus/ESA), MODIS (NASA), and the Allen Coral Atlas, with training labels from the van Woesik & Kratochwill global coral bleaching database and Great Barrier Reef aerial surveys. Please cite these sources and the accompanying manuscript when using this dataset. If you require access before the embargo date, please write to [e-mail].