Powering new research with hyperlocal air pollution data

By Karin Tuxen-Bettman, Program Manager, Google Earth Outreach, and Lead for Project Air View

Today, for the first time, Google and Aclima are making the entire hyperlocal air pollution dataset of California — measured and analyzed over 4 years — available to scientific researchers. Scientists and researchers are now invited to request access to the complete dataset we’re releasing starting today for air quality studies here.

Four years ago, we announced the beginning of a groundbreaking initiative: we were partnering with start-up Aclima to equip a few dedicated Street View cars with laboratory-grade air quality measurement equipment, to map hyperlocal air pollution in a few regions across California.

Our goals were based in science:

1. To test if mapping air quality with a fleet of vehicles was possible.
2. To share the validated data with scientific researchers to help move the field forward.
3. To compare and fine-tune Aclima’s mobile air sensing technology, in order to enable broader scaling to additional Street View cars and additional regions around the world.

Last summer, we completed this initial experiment. We drove a few cars over four years. We measured Carbon Dioxide (CO2), Ozone (O3), Nitrogen Dioxide (NO2), Nitric Oxide (NO), Methane (CH4), Particle Number below 2.5 microns in size (PN2.5), Black Carbon (BC), and Ultrafine Particles (UFP) at different times over the course of the experiment.

With our partners at Aclima, we are excited to share the complete dataset with researchers to inspire more scientific discovery. Here is more about the dataset, by the numbers:

Already, the data has been used for some groundbreaking research. In 2017, Josh Apte from the University of Texas-Austin led his collaborators to use data from two of the cars which drove for a year in Oakland, CA, to publish a first-of-its-kind mobile mapping methodology paper in Environmental Science & Technology. This work demonstrated that street-level mapping with vehicles was possible. And the results showed that air quality can vary up to 5–8x from one end of a city block to another. Environmental Defense Fund (EDF) was the lead researcher on this project, and released hyperlocal air quality maps of black carbon, nitric oxide (NO), and nitrogen dioxide (NO2) for West Oakland and East Oakland.

Researchers demonstrated that street-level mapping with vehicles was possible, and that the results showed that air quality can vary up to 5–8x within one city block. Source: EDF

Then, in 2018, the collaborators took the science one step further: they published a study in Environmental Science & Technology that compared two approaches for efficiently mapping spatial air quality patterns in a city: a “data-only” approach which applied summary statistics to the entire dataset, compared with a modeling approach (specifically, a commonly used land-use regression-kriging model) to predict at unobserved locations. They found that “data-only” worked better with more passes, and they found that the modeling captured general spatial air pollution trends but not the full variability of concentrations.

Researchers compared different methods for efficient mapping. Source: Messier et al 2018

That same year in 2018, EDF and Kaiser Permanente published a study in Environmental Health showing a link between the street-level air pollution in Oakland and heart disease among the elderly. EDF released maps visualizing relative risk for residents living in specific parcels due to the hyperlocal pollution.

Researchers combined the Google-Aclima hyperlocal air pollution data with electronic health records and showed how location impacts health. Source: EDF and Kaiser.

Now, for the first time today, Google and Aclima are sharing the entire dataset — from over 4 years — with researchers: 42 million measurement locations with multiple pollutants and other parameters measured at each location, across 357,235 kilometers.

To date, we’ve shared air pollution data with over 150 scientists. We’ve already seen some great use of the data. For instance, CalTrans awarded Professor Antonio Bento at the USC school of public policy a grant to combine Google and Aclima research data with Performance Measurement System (PeMS) data to estimate the environmental benefits of congestion pricing in the Los Angeles metropolitan area through study of the link between speeds and pollution. The grant is part of an initiative to design new performance metrics for the Caltrans Strategic Management Plan to improve resilience and protect the environment, identifying the opportunities and barriers of various congestion pricing strategies.

Another example: Vahid Moosavi, a senior researcher at ETH Zurich, used the data to train a neural network machine learning model to map the nonlinear relationship between different urban factors (e.g. types of activities, land use, building geometry, and density, etc.) and urban air quality, similar to previous work they did on urban flood mapping.

Researchers trained a neural network ML model to map San Francisco air quality. Source: ETH Zurich

After building this foundation of research and validation, we’re now scaling street-level air quality measurement with Aclima around the world.

I invite scientists and researchers to request access to the complete dataset we’re releasing starting today for air quality studies here. Once given access to the data, you’ll be invited to opt into our Google Air View Data Community.