The Camp Fire–the deadliest, most destructive wildlife in California history–started around 6:30 a.m. on the morning of November 8. By 8 a.m., it reached Paradise. By the afternoon, as the fire raged, the smoke reached cities like San Leandro, nearly 200 miles away. As the air filled with smoke, two Google Street View cars drove around the East Bay, using air quality sensors to map the pollution.
The level of smoke sometimes varied dramatically even in the same neighborhood or on the same road. On the first day, one car recorded a jump in air pollution as it crossed the Bay Bridge. On the second day, as a car drove back and forth in the city of Richmond, it captured data showing concentrations of PM 2.5–tiny particles of pollution–that were twice as high in one area as a location just down the street. “In a space of two to three blocks, you’re in significantly dirtier air,” says Melissa Lunden, chief scientist at Aclima, the startup that designed the air quality platform now used in certain Street View cars.
By the next week, the smoke had covered the entire Bay Area. One car drove to the top of Mt. Diablo in the East Bay; once it reached a certain elevation, the pollution reading suddenly dropped. “You can see a line at which the air above is relatively clean, and below it, you’re in the soup,” says Lunden. In 2017, when another catastrophic fire burned in Santa Rosa, the cars gathered similar data about the smoke drifting down to the East Bay.
The cars typically measure everyday smog; the high-quality sensors gather data as cars cover cities, helping build a far more detailed picture of pollution than the current handful of official, stationary sensors. Even on a typical day, pollution is hyperlocal. In one past study, the cars found that one end of a block was five to eight times more polluted than the other end.
In an event like a wildfire, the data from the cars could potentially be used to build models of how plumes of smoke travel through the area to help better prepare for future disasters. “If you could get more sophisticated models of how smoke starts to move around in areas, you could really put out location-specific alerts with that modeling compared with other types of measurements,” Lunden says. “It’s clear that we’re just going to continue to have fires now, unfortunately. So the driving helps support neighborhood-focused data.”