City:

Oakland, California, US

Organization:

Environmental Science & Technology:

†Department of Civil, Architectural and Environmental Engineering, University of Texas at Austin, Austin, Texas 78712 United States

‡Environmental Defense Fund, New York, New York 10010 United States

§School of Population and Public Health, University of British Columbia, Vancouver V6T 1Z3 Canada

∥Energy Technologies Area, Lawrence Berkeley National Laboratory, Berkeley, California 94720 United States

⊥Aclima, Inc., 10 Lombard St., San Francisco, California 94111 United States

#Department of Civil and Environmental Engineering, University of Washington, Seattle, Washington 98195 United States

∇Institute for Risk Assessment Science, Utrecht University, Utrecht 3584 CM Netherlands

Project Start Date:

28 May 2015, 1 year pilot program 

Project End Date:

14 May 2016

Reference:

Apte, Joshua S.; Messier, Kyle P.; Gani, Shahzad; Brauer, Michael; Kirchstetter, Thomas W.; Lunden, Melissa M.; Marshall, Julian D.; Portier, Christopher J.; Vermeulen, Roel C.H.; Hamburg, Steven P.,"High-Resolution Air Pollution Mapping with Google Street View Cars: Exploiting Big Data", Source: Environmental Science and Technology, v 51, n 12, p 6999-7008, June 20, 2017

Problem:

Air pollution concentrations can vary sharply over short distances due to unevenly distribution sources and other urban factors.

Air Pollution global risk for ill-health and death factor.

Monitoring air pollution is crucial for epidemiology and air quality management yet, ground-based air pollution observations are limited. Additionally, air pollution concentrations can vary sharply over short distances due to unevenly distribution sources and other urban factors.

Technical Solution:

Data science methods:  Developed a series of data reduction algorithms to convert data set of ∼ 3 million instantaneous observations into estimates of median annual weekday concentrations for individual 30-m road segments.  Also, employed a “ snapping”  procedure to assign each 1- Hz measurement to the nearest 30 m road segment on the basis of measured GPS coordinates, allowing repeated measurements to be analyzed as a group. 

• Unsupervised Learning Data: The algorithms are set to discover and present a structure of the data collected.

• The Data Analysis Methods included: 
• Clustering Data
• Interclass Correlation: Indicate similarity in values
• Kolmogorov− Smirnov Test: Determine differences between data sets
• Affinity and Association:
• Data Mining spatial patterns: Finding frequent patterns and correlation 

Measurement Platform: Routine mobile monitoring with fleet vehicles (Two Google street view vehicles).

Datasets Used:

• Dataset 1: Mobile Monitoring data set, Google Street view Vehicles,  Aclima Inc., 2016. Mobile data set to characterize time averaged, sotropic (i.e., direction-independent) daytime distance-decay relationships.
• Dataset 3: Road line geometry shape-file, OpenStreetMaps, 2015

Outcome:

Previous Methods Performance (Different advantages and limitations)

• Satellite remote sensing: cannot categorize fine-scale gradients (1-10km resolution) such as traffic emissions including black carbon and other ultra-fine particles that a represent health risks
• Chemical Transport Models: Rely only in emissions inventories and are unable to discover new or unexpected sources
• Land-use Regression Models: Have limited temporal data 

Post Methods Performance

• High Resolution Mobile Mapping:  Revealed urban pollution gradients at very fine ∼ 30 m scales, or 10^4− 10^5 X ~  greater spatial resolution than with urban ambient monitors.

Issues that arose:

• None reported 

Status:

Terminated

Entered by:

Date: David Murayama, David.murayama@mail.utoronto.ca