Unlocking Earth AI’s planetary geospatial foundation models for global public health
Google says PDFM embeddings improved or matched conventional public-health models across five partner-run studies without task-specific fine-tuning.
The tests covered MMR vaccination near the U.S.-Canada border, cardiovascular mortality in U.S. counties, dengue forecasting in Mexico, postpartum depression risk in U.S. survey data, and cholera emergence in the Democratic Republic of the Congo. The blog says the embeddings package search trends, mobility, built-environment density, weather, air quality, and other signals into monthly location “fingerprints.” Gains were strongest in cases where official data lagged, crossed borders poorly, or missed local context. Google says the dataset is commercially available in preview, with no-cost access available for some academic and public-health research uses. Google Research's note
The tests covered MMR vaccination near the U.S.-Canada border, cardiovascular mortality in U.S. counties, dengue forecasting in Mexico, postpartum depression risk in U.S. survey data, and cholera emergence in the Democratic Republic of the Congo. The blog says the embeddings package search trends, mobility, built-environment density, weather, air quality, and other signals into monthly location “fingerprints.” Gains were strongest in cases where official data lagged, crossed borders poorly, or missed local context. Google says the dataset is commercially available in preview, with no-cost access available for some academic and public-health research uses. Google Research's note
score 5