Integrated Sample Dataset for Themeda: North Australian Savannas (1988–2020)

This dataset supports the study "Themeda: Predicting land cover change using deep learning" and integrates diverse data sources to predict future land cover based on historical observations and environmental predictors in the savannas of northern Australia. The dataset includes annual land cover data for 1988-2020 at 25 m resolution, reclassified into 23 classes, derived from Digital Earth Australia. It also incorporates environmental data, including rainfall, maximum temperature, elevation, land use, fire scars (early and late dry season), and soil properties (depth, clay content, and cation exchange capacity). Fire scar data is sourced from NOAA’s 1 km Fire History Mapping dataset, elevation from Australia's hydrologically conditioned Digital Elevation Model, climate data from ANUClimate 2.0, and soil properties from publicly available datasets. These variables were chosen for their relevance to savanna vegetation and land cover dynamics. The dataset enables replication of the Themeda study and further research into land cover prediction.