Google Earth AI: Foundation models and geospatial reasoning at planetary scale
Google unveiled significant advances in Google Earth AI with new imagery and population foundation models that elevate geospatial analysis to a new level. The platform now combines vision and language models with a geospatial reasoning agent powered by Gemini models.
The key innovation lies in the ability to answer complex questions like: Where is a hurricane likely to make landfall? Which communities are most vulnerable and how should they prepare? — requiring combined reasoning about imagery, population, and environment.
Major advances announced:
Imagery Models — New Remote Sensing Foundations models simplify satellite image analysis with three capabilities: vision-language models, open-vocabulary object detection, and adaptable vision backbones. They achieved over 16% improvement on image search tasks and more than doubled accuracy in novel object detection.
Population Models — Population Dynamics Foundations introduces globally consistent embeddings across 17 countries with monthly updates capturing changes in human activity.
The geospatial reasoning agent acts as an intelligent orchestrator: it decomposes complex questions into multi-step plans, executes each step by calling foundation models, queries massive datastores, and fuses results into holistic answers.
Source: Google Research Blog