Labs: Air Quality

Hamburg

In Hamburg, we partnered with the City Science Lab at HafenCity University Hamburg (HCU), Aclima, and a working group from the City of Hamburg. One electric Google Street View car was equipped with Aclima's mobile monitoring platform to measure air quality and greenhouse gases street-by-street. Driving took place Monday-Friday between 7:00 am and 7:00 pm from October 2021 through December 2022, so the dataset primarily represents typical daytime, weekday air quality. The car measured pollution on each street and highway at 1-second intervals, driving with the flow of traffic at normal speeds. The pollutants measured are: CO, CO₂, NO₂, NO, O₃, and PM2.5 (including counts of particle sizes from 0.3 - 2.5 μm). Engineering office Lohmeyer provided consulting services throughout the project.

Bengaluru

In Bengaluru, we partnered with Center for Study of Science, Technology and Policy (CSTEP), a Bengaluru-based think tank and ILK Labs to measure and analyze air quality in Bengaluru. CSTEP & ILK Labs modified a custom-fit compressed natural gas car to install portable air quality equipment and used it to measure on-road air pollution data in Bengaluru.

Scientists at CSTEP used research grade air quality monitors to measure on-road concentration levels, including TSI DustTrak DRX 8533 to measure fine particulate matter (PM 2.5), microAeth AE51 for black carbon (BC), and TSI CPC 3007 for ultrafine particles (UFPs). CSTEP drove and collected measurements over 900kms of city roads, which is 10% of the total road length in Bengaluru. Driving predominantly took place during daytime hours Monday-Friday between 9:00 am and 5:00 pm on non-rainy days from November 2021 through June 2022, so the dataset primarily represents typical daytime, weekday air quality.

The total study route was divided into 11 sections and each section was sampled at least four times. Three sections were sampled 8 times, and five were sampled 12 times. The total distance driven during the sampling campaign was approximately 10,600 km (through 665 drive hours), and more than 2 million measurements were collected. Instruments measured PM 2.5, BC, and UFP concentrations at 1-second intervals.

Dublin

In Dublin, we partnered with Dublin City and Aclima. One electric Google Street View car was equipped with Aclima's mobile monitoring platform to measure air quality and greenhouse gases street-by-street. Driving predominantly took place Monday-Friday between 9:00 am and 5:00 pm from May 2021 through August 2022, so the dataset primarily represents typical daytime, weekday air quality. The car measured pollution on each street and highway at 1-second intervals, driving with the flow of traffic at normal speeds. The pollutants measured are: CO, CO₂, NO₂, NO, O₃, and PM2.5 (including counts of particle sizes from 0.3 - 2.5 μm).

Copenhagen

In Copenhagen, we partnered with the City of Copenhagen and Utrecht University. Scientists at Utrecht University equipped one Google Street View car with a CAPS NO₂ monitor, an AE33 aethalometer measuring black carbon (BC) particles, and a Water-Based Condensation Particle Counter measuring ultrafine particles (UFP), and collected and analyzed the data in collaboration with the City of Copenhagen. The vehicles drove streets in Copenhagen, predominantly on weekdays between November 2018 and February 2020. Driving usually 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. The NO₂ data, in particular, has been temporally corrected, so it reflects the long-term averages representing the full day. The car measured pollution on each street and highway at 1-second intervals, and drove with the flow of traffic at normal speeds.

Amsterdam

In Amsterdam, we partnered with the City of Amsterdam and Utrecht University. Scientists at Utrecht University equipped two Google Street View cars with a CAPS NO₂ monitor, an AE33 aethalometer measuring black carbon (BC) particles, and a Water-Based Condensation Particle Counter measuring ultrafine particles (UFP), and collected and analyzed the data in collaboration with the City of Amsterdam. The vehicles drove streets in Amsterdam, predominantly on weekdays between May 2019 and February 2020. Driving usually took place during daytime hours — typically, between 9:00 am and 6:00 pm. All pollution measurements were temporally corrected so that they reflect long-term average concentrations representing the full day and weekdays and weekends. The cars measured pollution on each street and highway at 1-second intervals. Cars drove with the flow of traffic at normal speeds.

London

In London, we partnered with the Breathe London project to outfit two Google Street View cars with a cavity-attenuation phase shift spectroscopy instrument measuring nitrogen dioxide (NO₂), and to collect and analyze the data in collaboration with the Greater London Authority, Environmental Defense Fund Europe, C40, and other partners. The vehicles drove streets in London, on most weekdays between August 2018 and August 2019. Driving took place during daytime hours — typically, between 4:00 am and 11:00 pm — so the dataset is primarily representative of daytime, weekday air quality. The cars measured pollution on each street and highway at one-second intervals. Cars drove with the flow of traffic at normal speeds.

Pollutants

The measured pollutants are black carbon (BC), and ultrafine particulate matter (UFP), and nitrogen dioxide (NO₂).

• Nitrogen dioxide (NO₂) is formed primarily by the burning of fuel, often from cars, trucks, and power plants. It's associated with respiratory problems, including increased asthma attacks and reduced lung function.

• Black carbon particles come from burning fuel, especially diesel, wood and coal. High exposure is associated with heart attacks, stroke and some forms of cancer.

• Ultrafine particulate matter (UFP, sometimes called PM 0.1) are particles of nanoscale size (less than 0.1 microns in diameter). Ultrafine particles are emitted from traffic and other forms of transportation, including aviation and shipping. Ultrafine particles penetrate deep into the lungs and cause heart attacks, strokes, asthma, and bronchitis, as well as premature death from heart ailments, lung disease, and cancer.

Learn more about these pollutants and their effects

Hamburg

These maps show spatially-resolved ambient concentration estimates for each pollutant over the study period from October 2021 to December 2022. The values represent the typical pollutant levels for each location over weekday business hours that were measured during the study period. Aclima's methodology for generating these estimates is a statistical aggregation of measurements, as described in Apte et al. (2017) and Miller et al. (2020). The vehicles measured a geolocated concentration every second (1 Hz) during daily mapping driving. The time-resolved 1 Hz data points are aggregated in approximately 50m road segments, and all the visits to an individual segment are combined to produce the long-term ambient concentration estimate. The median number of visits across the city is 10. All segments included in the map have at least 6 unique visits over the course of the study period, where a unique visit is defined as passing through the segment at least once in a 4-hour time period. The map offers users the ability to filter to roads with at least 6, 10, 15 and 20 unique visits, over the study period. Confidence in the single segment concentration estimate generally improves with the number of visits and care should be taken when interpreting estimates for individual segments with less than 10 visits. While confidence is somewhat lower for segments with fewer visits, the larger spatial trends observed in the map across the larger collection of segments are generally robust (Apte et al., 2017).

Bengaluru

The on-road 1 Hz pollution data was gridded to 50-meter uniform road segments. On a daily basis, all 1 Hz measurements falling in the grid were averaged. Using daily gridded data, a grid-wise median of daily means was computed to represent a particular grid pollution level. In the measured data, the day time variability was observed during the measurement campaign which was adjusted using factors derived from the ambient measurements made at CSTEP (for detailed methodology, see Apte et al., 2017). Spatial representativeness of the various pollutants increased with the increase in the number of rides, which increased the confidence of the pollution maps. The on-road air pollution levels for PM2.5, BC, and UFPs were found to be highest on major roads, followed by arterial and residential roads. The spatial gradient in PM2.5 (across various road types) was shallower compared to that of BC and UFPs.

Dublin

These maps show spatially-resolved ambient concentration estimates for each pollutant over the study period from May 2021 to August 2022. The values represent the typical pollutant levels for each location over weekday business hours that were measured during the study period. Aclima's methodology for generating these estimates is a statistical aggregation of measurements, as described in Apte et al. (2017) and Miller et al. (2020). The vehicles measured a geolocated concentration every second (1 Hz) during daily mapping driving. The time-resolved 1 Hz data points are aggregated in approximately 50m road segments, and all the visits to an individual segment are combined to produce the long-term ambient concentration estimate. The median number of visits across the city is 14. All segments included in the map have at least 6 unique visits over the course of the study period, where a unique visit is defined as passing through the segment at least once in a 4-hour time period. The map offers users the ability to filter to roads with at least 6 unique visits, and at least 10 unique visits, over the study period. Confidence in the single segment concentration estimate generally improves with the number of visits and care should be taken when interpreting estimates for individual segments with less than 10 visits. While confidence is somewhat lower for segments with fewer visits, the larger spatial trends observed in the map across the larger collection of segments are generally robust (Apte et al., 2017).

Copenhagen

To generate the Copenhagen pollution maps, scientists at Utrecht University developed a “mixed model” that mixes, or combines, both (1) a statistical aggregation of repeated observed measurements made on each street and (2) a predictive regression model, as described in Kerckhoffs et al., 2022 research methodology. This algorithm brings together previously published statistical aggregation used to create measurement-only maps (Apte et al., 2017, Miller et al., 2020) and state-of-the-art land-use regression modeling, which includes predictive independent land-use and other built environment characteristics that contribute to air pollution. The model is a “mixed model” because it mathematically combines the two approaches. Based on the amount of uncertainty in the observed measurements, the model weighs either the land-use regression model prediction or the statistical aggregation of measurements on a particular road segment more heavily. This method allows for a precise, robust estimate of average concentration even at very fine spatial scales. Given that concentrations on some city streets vary by an order of magnitude, this precision is useful to identify the key spatial patterns of air pollution within a city. Scientists at Aarhus University also collaborated on the data analysis and methodology.

Amsterdam

To generate the Amsterdam pollution maps, scientists at Utrecht University developed a “mixed model” that mixes, or combines, both (1) a statistical aggregation of repeated observed measurements made on each street and (2) a predictive regression model, as described in Kerckhoffs et al., 2022 research methodology. This algorithm brings together previously published statistical aggregation used to create measurement-only maps (Apte et al 2017, Miller et al 2020) and state-of-the-art land-use regression modeling, which includes predictive independent land-use and other built environment characteristics that contribute to air pollution. The model is a “mixed model” because it mathematically combines the two approaches. Based on the amount of uncertainty in the observed measurements, the model weighs either the land-use regression model prediction or the statistical aggregation of measurements on a particular road segment more heavily. This method allows for a precise, robust estimate of average concentration even at very fine spatial scales. Given that concentrations on some city streets vary by an order of magnitude, this precision is useful to identify the key spatial patterns of air pollution within a city.

London

To generate the London pollution maps, scientists at Environmental Defense Fund used an updated algorithm for aggregating and summarizing the repeated measurements made on each street, based on Apte et al. (2017) research methodology. Aggregation of repeated measurements reduces the influence of individual extreme samples on the estimate of expected pollution (median) if, for example, a Street View car happened to be driving behind a truck during one pass of a street. This method allows for an estimate of median concentration at 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. We required a minimum of 10 visits to a road location in order to calculate a median. All roads were not sampled the same number of times, at the same times, or on the same day. Different weather conditions, which drive regional background concentrations at the time of sampling, introduce uncertainty in our estimates of median concentration as much as +/- 50% of the median NO₂ value. Roads with a higher number of visits have uncertainty that is substantially lower than this range. For segments with 30 drives, uncertainty decreases to less than +/-30%. The uncertainty is expected to reduce with on-going effort to refine analytical methods.

Air quality results for Hamburg can be found on Hamburg Open Science.

Air quality results for Bengaluru can be found on CSTEP's website.

Air quality results for Dublin can be found on Dublinked, Dublin's Open Data platform.

Air quality model results for Copenhagen can be found at the Open Data DK website.

Air quality model results for Amsterdam can be found on the Healthy Urban Living Data and Knowledge Hub (DKH GSL).

Air quality data for London can be found on the OpenAQ platform.

The use of this data is subject to Google's Terms of Service. Feel free to include the data from the Environmental Insights Explorer in other analysis, materials, reports, and communications with the following attribution:

Source for Bengaluru data: CSTEP & ILK Labs 2022 via Google Environmental Insights Explorer (<current month / year>)

Source for Hamburg and Dublin data: Aclima & Google 2022 via Google Environmental Insights Explorer (<current month / year>)

Source for Copenhagen and Amsterdam data: Utrecht University & Google, 2021, via Google Environmental Insights Explorer (<current month / year>)

Source for London data: BreatheLondon 2020, via Google Environmental Insights Explorer (<current month / year>)