Smoke from wildfires burning across western Ontario, Canada, pushed New York City’s air quality into unhealthy territory again this month, prompting a joint advisory from New York City Emergency Management and the city’s Department of Health and Mental Hygiene, while weather.com’s live wildfire smoke tracker showed poor air quality persisting across parts of the Northwest and Western United States as of its August 10, 2026 update. Behind many of the alerts reaching phones during episodes like this one sits a machine-learning system Google has spent several years building: a neural network that reads satellite imagery in near real time to map wildfire boundaries and estimate burnt area, then feeds that information directly into Google Search’s SOS Alerts and Google Maps.
How the AI Actually Tracks a Fire
Google’s wildfire boundary tracker pulls imagery from a rotating set of satellites depending on geography — GOES-16 and GOES-18 over North and South America, Himawari-9 and GK2A around Australia, and Suomi NPP and NOAA-20 with their VIIRS imagers elsewhere. Because any single satellite pass can be blocked by cloud cover or smoke itself, the system processes the three most recent images in sequence rather than relying on one snapshot, letting the neural network work around temporary obstructions to keep its boundary estimates current. The tool now operates across a notably wide footprint, covering fire-prone regions in Greece, Cyprus, Spain, Portugal and France in Europe; Kenya and Rwanda in Africa; Argentina, Brazil and Chile in South America; and large parts of the US, Canada, Mexico and Australia.
Why This Matters for Ordinary Health Decisions
The output of that satellite analysis is not just a map for firefighters — it is increasingly the invisible layer behind the AQI numbers and smoke alerts that determine whether millions of people decide to keep windows shut, cancel outdoor plans, or bring an inhaler to work. Health officials note that people with respiratory or cardiovascular disease, children, older adults, pregnant women, outdoor workers and lower-income residents face disproportionately higher risk from wildfire smoke exposure, which is why timely, geographically precise alerts matter more than a single national air-quality number. Fine particulate matter, PM2.5, is the pollutant driving most of these advisories, since it is small enough to penetrate deep into the lungs and bloodstream.
Filling a Data Gap With Synthetic Fires
One of the harder problems in training reliable wildfire-prediction models is that real fires, thankfully, are relatively rare events compared to the enormous range of conditions — wind, humidity, terrain, fuel type — that can shape how one spreads. To address that data scarcity, Google built what it describes as a large-scale, high-fidelity wildfire simulator capable of generating synthetic training data across a much wider range of fire scenarios than historical records alone could supply. Google’s research team has paired that simulated data with a collaboration with the US Forest Service’s Fire Lab to develop next-generation fire-spread prediction models that combine live weather data with satellite imagery, aiming to forecast not just where a fire is now but where it is likely to go next.
The View From Emergency Managers
Local agencies have leaned on these AI-assisted tools as fires have become a more routine cross-border health disruption rather than an occasional regional event — smoke drifting from Ontario into New York City is now a recurring late-summer pattern rather than a once-a-decade anomaly. NYC Emergency Management’s advisory this month specifically flagged the source as ‘significant, still-spreading wildfires in western Ontario,’ language that reflects how far downstream a fire’s health impact can travel from its point of origin, and why continent-scale satellite tracking rather than local monitoring alone has become necessary to give affected cities useful lead time.
Where the Technology Still Falls Short
AI boundary detection and smoke forecasting are not without limits: satellite-based systems can lag behind fast-moving fire fronts, cloud and smoke cover itself can degrade image quality even with multi-pass processing, and near-surface smoke forecasts remain probabilistic rather than precise down to the neighborhood level. Weather.com’s own tracker, updated multiple times daily through August, still frames its smoke and air-quality maps as forecasts alongside official fire-weather warnings like Red Flag Warnings, issued when critical fire conditions are expected within 12 to 24 hours, underscoring that AI-generated estimates are meant to supplement, not replace, on-the-ground fire weather monitoring by services like NOAA.
What Comes Next as Fire Seasons Lengthen
With wildfire smoke increasingly crossing international borders and affecting population centers far from the fires themselves, the pressure is growing on tech companies and public health agencies to tighten the loop between satellite detection, AI-driven spread prediction, and the advisories that reach people’s phones before smoke arrives rather than after air quality has already degraded. Google’s continued expansion of its tracker’s geographic coverage and its Forest Service Fire Lab partnership suggest the next phase of this work will focus less on simply mapping where fires are burning now and more on giving cities like New York enough advance warning to issue health guidance before, rather than during, a smoke event.
