Unlocking Earth AI’s planetary geospatial foundation models for global public health

Google Research introduced a public health paradigm using planetary geospatial foundation models, with Google Earth AI's Population Dynamics Foundation Model (PDFM) as a proof-of-concept. PDFM compresses privacy-preserving search trends, human mobility, built-environment density, and environmental determinants into monthly location embeddings. Without task-specific fine-tuning, these off-the-shelf embeddings matched or improved on conventional inputs across disease domains, geographic settings, and epidemiological tasks.

Five global health research partners conducted independent evaluations. At the US-Canada border (146 counties), Mount Sinai Health System and Boston Children's Hospital reported a 36% relative gain in explained variance (0.159 to 0.216, statistically significant) for MMR vaccination by capturing cross-border mobility and information spillovers. For cardiovascular disease mortality nowcasting across 3,091 US counties, NYU Grossman School of Medicine found accuracy comparable to census-based models (mean absolute error 18.7 vs 19.1 deaths per county; RMSE 46.00 vs 57.69; differences not statistically significant).

For dengue forecasting across roughly 2,450 Mexican municipalities, the University of Oxford and Tecnológico de Monterrey reported a statistically significant 1-month forecast gain (Delta Weighted Interval Score -0.0051) via TimesFM integration, with accuracy improved in up to 72% of active transmission municipalities. The University of Washington, using CDC PRAMS data on 332,970 respondents, reported a transferable gain in postpartum depression risk prediction (AUC +0.0020 on a 0.62 baseline in seen states; +0.0038 in unseen states). WHO AFRO, using Democratic Republic of the Congo surveillance data across 403 health zones over 89 weeks, reported statistically significant relative gains for cholera emergence: +9.7% in Area Under the Precision-Recall Curve at 4 weeks and +18.1% Precision@5 at 8 weeks.

Current limitations cited include static snapshots, which the researchers say are driving active research into temporally dynamic embeddings and geographic transfer learning for under-connected regions. PDFM embeddings are available in Preview as Population Dynamics Insights via Google Maps Platform, with no-cost access requestable for select non-operational research use cases.