Data Science and GeoAI

Development Seed's Data Science and GeoAI service, branded 'AI for Earth', designs, trains and deploys custom AI pipelines that bring together satellite imagery, drone imagery and ground data, from cleaning and annotation through model training to delivery. The team builds and fine-tunes geospatial foundation models, including the open-source Clay model it built with the non-profit Clay and NASA's Prithvi models, and uses their embeddings for similarity search, segmentation, object detection and change analysis. It also builds agent-powered assistants that let non-technical users query Earth data in plain language. Its work includes Global Nature Watch with Land & Carbon Lab (convened by WRI and the Bezos Earth Fund), SkyTruth's Cerulean oil-slick detection from Sentinel-1 radar, Sentinel-2 water-quality indicators for EDP Labelec, and the PEARL land-cover mapping tool for Microsoft's Planetary Computer. Engagements are bespoke and open source, and include building the client team's own capacity to apply the models.