Methodology: How we collected data

Data collection

EDF partnered with Google Earth Outreach, the University of Texas at Austin (UT Austin) and Aclima, a San Francisco-based company that specializes in environmental sensor networks, to outfit the cars with mobile sensing equipment and to collect and analyze the data.

The study used two Google Street View cars equipped with Aclima's sensing platform, providing real-time air quality data for the maps. The vehicles drove streets in Oakland, California, on most weekdays between June 2015 and May 2016 (about 150 days). Driving routes, data collection, and processing were managed by software tools designed by Aclima to support large-scale measurement with vehicles. This new method and its results have been published in a peer-reviewed paper for the Environmental Science & Technology journal.

The vehicles drove streets in Oakland, California, on most weekdays between June 2015 and May 2016 (about 150 days). Driving took place during daytime hours  — typically, between 9:00 am and 6:00 pm — so the dataset is primarily representative of daytime, weekday air quality.

Scientists at UT Austin designed the daily driving plan for cars to ensure that each neighborhood was systematically sampled at different times of the day, week and year. The cars repeatedly measured pollution on every street and highway within a 30 km2 area of Oakland. Cars drove in the flow of traffic at normal speeds. Over the course of a year, the team sampled each road in this area (~700 km of roads) an average of 30 times. In total, the cars collected about 3 million unique measurements while driving about 23,000 kilometers. This dataset is one of the largest sets of mobile air pollution measurements ever assembled.

While scientists have measured pollution using mobile technology for about two decades, these studies have generally lasted for a few days or weeks, or only covered select routes within a city.

Vehicles and instrumentation

The Google Street View cars are custom-modified, gas-powered vehicles. The Google Street View cars included both 3D camera equipment and the Aclima's sensing platform that provided sensing hardware, data management, quality control and visualization functions during the study. This system used fast-response laboratory-grade instruments to precisely measure pollution concentrations every second. A photoacoustic extinctiometer made measurements for black carbon (BC) particles, a chemiluminescence analyzer measured nitric oxide (NO) and a cavity-attenuation phase shift spectroscopy measured nitrogen dioxide (NO2).

In addition to the air quality instruments, the mobile platform included two GPS units and an on-board data management system. The cars continuously uploaded their data to a central server via a 4G internet link, so that scientists at UT Austin and Aclima could monitor conditions in real-time.

Data analysis

To generate the pollution maps, scientists at UT Austin and EDF developed an algorithm for aggregating and summarizing the repeated measurements made on each street.

The analysts started with the full data set of about 3 million measurements. Each individual data point, representing one second of pollution observations at a specific location, was assigned to a corresponding 30-meter length of road, using a unique algorithm developed by scientists at the University of Texas. These algorithms incorporated the mobile monitoring data from the cars as well as information on the hourly variation to account for time of day variability in regional air quality as measured by the Bay Area Air Quality Management District network of fixed urban air quality monitoring sites.

On average, the team drove each street in the study area 30 times (or “drive passes”), over the course of a year, collecting about 200 unique observations for each 30-meter road section. The pollution maps presented here, updated in January 2018, were generated by first calculating the average concentration within each 30-meter road section during each drive pass, and then calculating the median of averages across multiple drive passes for that 30-meter segment. This algorithm is an update to the originally-published method and reduces the influence of individual extreme samples if, for example, a Street View car happened to be driving behind a truck during one pass of a street. This method allows for a precise, robust estimate of median concentration even at very fine spatial scales.

Over time, some locations are more consistently impacted by major pollution sources (e.g., traffic, industries) than others. The effect of these pollution sources leads to the different median concentration measurements shown in the maps. Based on statistical analyses of uncertainty, UT Austin scientists estimate that measurements for each 30-meter road segment are precise to within ± 10-20%. Given that concentrations on some city blocks vary by as much as a factor of six or more, this precision is adequate to identify the key spatial patterns of air pollution within a city.

Interactive map details

The scales shown on the maps presented here were chosen to emphasize the spatial variability in air pollution levels observed on surface streets. The maximum values on the scales for black carbon (BC) and NO2 (1.0 microgram per cubic meter and 24 parts per billion, respectively) are the 90th percentiles of all readings in the Oakland study area, meaning that the top 10% of readings for each of these pollutants are represented in the darkest red. The map for NO is presented on the same scale as NO2 (the 90th percentile reading for NO was 26 ppb).