Our Science
Most air quality maps smooth pollution across whole cities. Air Aware Labs combines trusted monitors, satellite data, traffic, weather and 3D city models where available with the Google Air Quality API to estimate pollution at street level, then converts it into a personal exposure score based on your route and journey.
Our Approach
Pollution changes road by road and hour by hour. We start with a global baseline, anchor it to trusted measurements, add the physical drivers that shape local pollution, and test the result against monitors the model has never seen.
Regulatory networks anchor the model to measured air quality.
Dense local sensors extend coverage and provide held-out validation.
Worldwide inputs give coverage even where ground sensors are sparse.
Road intensity, junctions, rush hours and seasonal rhythms explain local variation.
Wind, temperature and atmospheric mixing govern whether pollution builds or disperses.
3D scans capture street canyons and building form that trap or disperse pollution.
Monitors, satellite, traffic, weather, lidar and street data combined into one common picture.
Street-level and hourly corrections refine the global baseline.
Tested against independent monitoring networks held entirely out of training.
Route, timing and breathing rate turn street-level air into your inhaled dose.
We keep model architecture and training specifics proprietary. What we share publicly is the principle: trusted anchors, rich physical context and continuous validation against independent data.
Our Scientists
The model grew out of academic air quality research and is still led by the people who built it.
Chief Scientific Officer
Air quality PhD and Imperial College London postdoctoral researcher. Leads model design, validation and scientific interpretation.
Chief Technology Officer
Leads the engineering and machine-learning systems that turn the science into scalable products.
FAQ
Most apps show a single city-wide reading from the nearest monitor. AirTrack estimates the air on the specific streets you use, adjusts for whether you are outdoors, in a vehicle, or indoors, and accounts for how long you were there and how hard you were breathing. The result is an estimate of your personal exposure and dose which we translate into an AirTrack Score, not just a map colour.
The outdoor layer is built on the Google Air Quality API, which fuses reference-grade monitors, low-cost sensors, satellite retrievals and traffic-informed atmospheric modelling. We add route geometry and city data to sharpen that estimate at street level. Indoor exposure uses an infiltration model based on established indoor air quality literature. In-vehicle exposure is estimated from your route, travel mode and vehicle cabin assumptions.
The AirTrack Score is a time-weighted exposure score, not a single pollution snapshot. It looks at where you were, how polluted the air was, how long you spent there, and - where activity data is available - how hard you were breathing. More time in polluted air, or more intense activity in that air, lowers the score. Cleaner routes, better timing and lower-infiltration indoor environments can improve it.
Yes. Your breathing rate changes with activity. Running or cycling usually means breathing in more air per minute than walking, and far more than sitting still. That means two people in the same air can receive different doses depending on what they are doing. AirTrack accounts for this when activity data is available, which is why the same route can produce different scores for different activities.
For outdoor air, we validate our models against independent monitoring networks that are kept out of model training. In our latest Breathe London holdout test, Air Aware Labs’ model improved accuracy by up to 35% compared with the raw global baseline across 145 sensors. We report validation metrics such as R² and RMSE for transparency, and accuracy improves where richer local data is available, such as traffic, land-use and dense sensor networks.
For outdoor air, we validate our models against independent monitoring networks that are kept out of model training. In our latest Breathe London holdout test, Air Aware Labs’ model improved accuracy by up to 35% compared with the raw global baseline across 145 sensors. We report validation metrics such as R² and RMSE for transparency, and accuracy improves where richer local data is available, such as traffic, land-use and dense sensor networks.
Research Partners & Funders
Air Aware Labs
We turn rigorous environmental science into tools people, employers and cities can act on every day.
