Climate science in the Global South is severely hampered by label fragmentation. Geospatial labels critical for AI — crop types, land cover, building footprints, field polygons — exist across dozens of incompatible repositories, formats, and metadata standards.

uGAP is envisioned as an open-source platform that will unify scattered geospatial labels from the Global South into AI-ready pipelines. It comprises: (1) a label hub aggregating and standardizing multi-source labels using Croissant/GeoCroissant metadata standards; (2) cloud-optimized streaming in formats such as GeoParquet, COG, and Zarr; and (3) ready-to-use pipelines for training resource-efficient global models and fine-tuning geospatial foundation models on previously excluded Global South observations.

The project targets climate resilience & environmental science, food & agriculture, and climate & weather — with pilot deployments in Benin, Kenya, South Africa, Nigeria, and Senegal.