AI4Land: Scalable Deep Learning for Global High-Resolution Land Use Reconstruction

AI4Land is a data-driven framework that utilizes a U-Net architecture to generate high-resolution historical reconstructions and future projections of land use and land cover. It integrates coarse-resolution scenario data with static geophysical features, enabling the production of spatially explicit and physically consistent land surface patterns. Trained on Earth observation data using the MareNostrum5 supercomputer, AI4Land aims to enhance predictive capabilities in climate simulations by providing real-time, evolving land surface conditions for digital twin platforms, thereby addressing uncertainties in the terrestrial carbon cycle.