Google is expanding the use of artificial intelligence and satellite imagery in public health through the Google Earth AI initiative, aiming to help researchers and health authorities detect areas of risk and predict disease trends before they become widespread crises. The initiative relies on combining local health data with environmental signals, satellite imagery, mobility data, and geographic foundation models.
Google says these tools address two recurring problems in health response: delayed reporting and areas not covered by traditional geographic data. Instead of waiting for information to become complete after an outbreak begins, teams can use predictive models to identify the communities most at risk and estimate the potential paths of the disease.
Testing the Tools During the Ebola Outbreak
During the Ebola outbreak in the Democratic Republic of the Congo, Google collaborated with the World Health Organization Regional Office for Africa and the country’s Epidemiological Modeling and Intelligence Unit at the National Institute for Biomedical Research. The partners used two initial research models: the geospatial reasoning agent, which enables spatial mapping operations through a natural-language conversation, and a planetary disease forecasting engine.
The WHO team used the agent to identify remote mining corridors that combined exposure risks with high human mobility. According to Google, the team identified 48 at-risk settlements and more than 45,500 people within minutes, a process that would have taken weeks using manual methods. The findings helped with planning the deployment of mobile laboratories and coordinating border surveillance.
Google also developed weekly models with the Congolese unit to estimate the likelihood of Ebola transmission to areas that had not recorded infections, relying on mobility flows, historical case trends, and Earth AI data. According to the company, these estimates provide additional time for planning before cases reach new areas.
From Acute Outbreaks to Broader Health Forecasting
The applications are not limited to infectious diseases. The Population Dynamics Foundation Model, or PDFM, combines aggregated search trends, mobility patterns, and environmental data to produce a high-resolution, continuously changing picture of communities. Researchers at NYU Langone Health used these signals to estimate cardiovascular disease mortality during the same year, with performance comparable to or better than traditional methods in some cases.
In other research, behavioral patterns on both sides of the border between the United States and Canada helped improve estimates of vaccination rates for measles, mumps, and rubella in U.S. counties. Combining PDFM with local climate models also improved dengue fever outbreak forecasts in Mexico, while integrating epidemiological records with PDFM improved the ability to identify areas at risk of cholera in the Democratic Republic of the Congo up to eight weeks in advance.
What Is Changing in Practice?
The practical value lies in turning scattered geographic and population data into actionable signals for prioritization: where to send mobile teams, which areas need border surveillance, and where early preventive interventions such as mosquito-larva control can be implemented. However, the tools do not eliminate the need for local data or public health expertise; they are research and predictive models that should complement existing surveillance systems, not replace field verification and medical decisions.
Google currently makes embedded PDFM data commercially available in preview under the name Population Dynamics Insights within Google Maps Platform, and researchers can request free access for specific uses. Eligible institutions can apply for Google Earth credits through Google Maps Platform’s public programs. The company also said that Google.org provided funding to the National Institute in the Democratic Republic of the Congo to help modernize local testing and disease surveillance.