Google Earth AI Is Mapping Health Risks Near Canada

Matthew Leo · Published October 9, 2026 · AI

A public-health analyst studies a large paper map in a sunlit office.

Google is giving public-health researchers new ways to combine geography, mobility, environmental conditions and population signals. One of the first projects used patterns on both sides of the U.S.-Canada border to improve estimates of measles, mumps and rubella vaccination rates in U.S. counties.

That makes the October 6 announcement more than a distant global-health story. Infectious disease, wildfire smoke and access to care do not stop at a provincial or national border. The useful question for Canada is not whether a map looks impressive, but whether the underlying signals are representative, lawful to use and validated before they influence public decisions.

What Google Earth AI adds

The work combines several research systems rather than one consumer product. Google's Population Dynamics Foundation Model uses aggregated search trends, mobility patterns and environmental signals. Other components add satellite-derived information, population estimates and a prototype agent that can answer geographic questions in plain language.

Google says the World Health Organization's Africa regional office used a prototype during an Ebola response in the Democratic Republic of Congo. According to the company, the system identified 48 exposed settlements and more than 45,500 people at risk in minutes, a task that previously took weeks. Those are Google-reported operational results; the announcement does not provide an independent audit of the count or the resulting interventions.

The underlying research is available as preprints. That is useful for scrutiny, but it also means the work has not yet passed the complete peer-review process or been independently reproduced across health systems.

Why the border vaccination study matters

Researchers at Mount Sinai and Boston Children's Hospital used behavioural patterns on either side of the U.S.-Canada border to estimate county-level vaccination coverage. Cross-border movement can change the risk facing communities even when their health records are held in separate systems.

Canadian public-health agencies already work with surveillance data, census information and local reporting. A population model could help fill gaps between reporting cycles or identify an area that deserves attention. It should not replace the official vaccination record, a laboratory result or local epidemiology. A predicted low-coverage area is a reason to investigate, not proof that particular residents are unvaccinated.

Four questions Canadian health teams should ask

  1. What population is missing? Aggregated digital signals may underrepresent people with limited connectivity, different language habits, shared devices or less use of Google services.
  2. How local is the validation? A model that performs well in U.S. counties may behave differently in northern, rural, First Nations, Inuit or urban Canadian communities. Validation should use local outcome data and involve the communities affected.
  3. What decision will the model support? Forecasting where to send a mobile clinic is different from deciding who receives care. The more consequential the decision, the stronger the evidence and human review should be.
  4. Can the result be challenged? Public bodies need a record of inputs, model version, uncertainty and who approved the action. A conversational map should not become an untraceable policy decision.

Privacy is not answered by aggregation alone

Google describes the population and mobility signals as aggregated. Aggregation reduces some privacy risk, but Canadian organizations still need to know how small an area or group can become, whether data can be combined to infer sensitive traits and what contractual limits apply to reuse.

The concern is sharper in health work because a map can stigmatize a neighbourhood even when it contains no names. Publishing a predicted outbreak or low-vaccination area can affect trust, insurance discussions and local politics. Teams should decide in advance which results remain internal, how uncertainty will be communicated and when community partners should review a finding before release.

Mapletechie's look at PC Chat's health-data questions raised a similar principle: where data is stored is only one part of responsible use. Purpose, access, retention and the limits of the inference matter too.

Access is expanding, not universal

Google says commercial preview access is available through Population Dynamics Insights, while selected researchers can request no-cost access and eligible public organizations can apply for credits. Availability does not establish that a particular Canadian province, health authority or research project has adopted the tools.

The strongest near-term use is likely a second layer of evidence: flagging where field teams should look, helping analysts test scenarios and shortening the time needed to assemble geographic data. Public-health decisions still need current local records, domain experts and a clear account of uncertainty.

Sources

Tags: Google Earth AI, public health, geospatial AI, Canada, privacy

Read on Mapletechie