Google Research introduced S2Vec on March 24, 2026 as a self-supervised way to turn built-environment data into general-purpose embeddings. The framework aims to predict socioeconomic and environmental patterns from how cities are physically organized.
Google on Mar 12, 2026 introduced Groundsource, a Gemini-powered method for turning public reports into historical disaster data. The company says the system identified more than 2.6 million flood events across over 150 countries and now supports urban flash-flood forecasts up to 24 hours in advance.
AI at Meta says it is open-sourcing CHMv2, a high-resolution global forest canopy mapping model built with the World Resources Institute. Meta says the release uses DINOv3 Sat-L for satellite imagery and improves accuracy, detail, and global consistency.