Google’s Planetary Prediction Engine Automates Geospatial Modeling

PPE’s biggest advance is automating the messy data-discovery and geospatial-curation work that conventional AutoML largely assumes humans have already completed.

• –LLM orchestrators retrieve signals from Data Commons, Google Earth Engine, and open-web sources, then fuse them with PDFM and AlphaEarth embeddings.
• –Google reports strong first-party results, including 76.8% versus 60.0% mean R² on 21 CDC health indicators and 66.1% versus 31.5% R² for Nigerian food-security downscaling.
• –Its Ebola evaluation reached 83.3% Recall@10, but that result covers one outbreak and should not be treated as proof of operational readiness.
• –The system remains an early-stage research project with frozen foundation-model embeddings and unresolved challenges around causal validity, reproducibility, and human oversight, as [Google Research notes](https://research.google/blog/planetary-prediction-engine-automating-global-models-via-earth-ai/).