tool· draft
Google Earth AI
Google's family of geospatial AI models and agents, covering satellite imagery, weather, floods and population, reached through Google Earth, Earth Engine, BigQuery, Maps Platform and Google Cloud.
Cities & placesPeople & societyWeather & climateWaterEnergy
Google Earth AI is the name Google gives to a set of geospatial foundation models plus an agent that orchestrates them, presented by Google Research in October 2025. The named parts are Remote Sensing Foundations (vision-language and detection models for satellite imagery), the Population Dynamics Foundation Model (location embeddings from anonymised Search and Maps trends, covering 17 countries and updated monthly as of 2025), AlphaEarth Foundations (a unified representation of Earth observation data), the WeatherNext weather models, the flood forecasting models behind Flood Hub, the Groundsource flash-flood dataset, and the experimental Geospatial Reasoning agent and Planetary Prediction Engine. Google's product page also lists conservation models such as SpeciesNet and Perch, the NeuralGCM atmospheric model, air quality and pollen APIs and a traffic simulation API under the same banner.
Uses Google names include WHO AFRO's outbreak response in the Democratic Republic of the Congo, GiveDirectly's anticipatory cash assistance ahead of floods, Bellwether's prediction of property damage before storms, the Milken Institute's solar feasibility screening, and University of Oxford researchers who reported R² for twelve-month dengue predictions in Brazil rising from 0.456 to 0.656 when the Population Dynamics embeddings were added.
How it runs depends on the piece. Imagery models are in the Vertex AI Model Garden; AlphaEarth Foundations is in Earth Engine; WeatherNext is in BigQuery and Earth Engine; Flood Hub is a free public site; the Geospatial Reasoning agent and Planetary Prediction Engine are by early-access application. Some components are open source, including the hydrology framework behind Flood Hub (Apache 2.0), SpeciesNet and NeuralGCM. Terms and prices vary by component and are not collected in one place.
Status: a mix of operational products used inside Google Search, Maps and Earth, and experimental research pieces; the agent and prediction engine were still in limited preview in October 2026.
- homepage
- https://ai.google/earth-ai
- form
- model family
parts
-
Experimental Gemini-based agent from Google Research that plans a multi-step spatial analysis, calls Earth AI models and geospatial data sources, and returns the result with its plan and sources shown.
-
Part of Google Earth AI; given a user's local outcome data such as disease rates or crop yields, it picks relevant population and satellite signals and trains models for spatial prediction, downscaling and forecasting.
Tags: geospatial, satellite, weather, flood, population, foundation model