Marta Gonzalez, Visiting Associate Professor of Civil & Environmental Engineering
Jinhua Zhao, Associate Professor of Urban Studies & Planning
In the first year we have estimated the traffic surrounding 24 air quality (AQ) observation stations in Beijing, combining meteorological data, mobile phone data, road networks with a traffic assignment model we explore how to predict AQ in each station of the city and how to classify urban regions based on their traffic and air quality indexes. In addition, we deployed PAQ, a perceived air quality survey, which is a smartphone based survey relating to mobility and AQ of a set of users.
We showed that:
(i) A traffic model improves the prediction accuracy of the AQ index of the stations in the range of 5% – 15%. We show that in the summer, air pollution in the inner city is more severe than that in the suburbs, and urban traffic seems to be the main cause. Interestingly, in the winter, air pollution in the south of Beijing surpasses that in the inner city. We labeled stations where AQI is bad for reasons different than traffic, identifying those stations with poor AQ due to heating using coal.
(ii) The mobility patterns are not largely affected by the perceived air quality (PAQ) of the users. While there is agreement among users when the AQ is good, for poor AQ the replies of the survey tend to be very broad among the subjects.
In the second year we propose to integrate both research layers. While passive phone data have the potential to determine exposure of pollutants for the entire population, much less is known on how this affect the perception behavior of individuals.
We propose to use the passive phone data to calculate exposure per origin, and selected route of the travelers to identify most affected populations and to relate our estimates to the PAQ users to better explain their behaviors and AQ perceptions. The answers of the PAQ subjects could act as a validation of the exposure measured via ubiquitous mobile phone data.
Urban air pollution is one of the largest environmental health risks worldwide, and is expected to worsen over the coming decades as cities expand. Detailed, quantitative monitoring of urban air quality at high spatial and temporal resolution will prove critical to assessing risks and mitigating impacts. Although most existing monitoring networks lack the requisite spatial and temporal resolution, sensitive and inexpensive new technologies now enable the deployment of distributed air quality (AQ) networks to capture the full range of pollutant variability throughout an urban area. However, no proven techniques for analyzing large AQ datasets for metrics of interest—pollution sources, distributions and exposures—are yet well established.
To address this pressing data analysis need, we propose a pilot project to wed “big data” techniques with the methods of statistical physics. An 82-station, state-of-the-art AQ network in an area of Beijing will provide frequent measurements of pollutants, which will be overlain with local traffic information to create a unique, time-resolved dataset. A number of advanced data analysis techniques—e.g. frequent pattern mining and Markov random field modeling—will be applied to the data in order to gain quantitative insight into pollution sources, intensities and controlling factors.
During the first year of the project, we will focus our efforts on:
We expect this initial effort to generate positive public feedback and lead ultimately to the scaling up of the project to include the entire AQ monitoring network of greater Beijing, thereby setting stage for sustained air quality improvements.