Predicting impacts of commute mode and route
on exposures to traffic-related air pollution
Ryan Michael, Amy L. Stuart, and Getachew Dagne
University of South Florida, 13201 Bruce B. Downs Blvd, MDC-56, Tampa, FL 33612
[email protected], [email protected], [email protected]
Extended Abstract
INTRODUCTION
Motor vehicles emit a complex mixture of noxious chemicals that contribute to cardiovascular and
respiratory disease and excess mortality. Commuting may be important to pollution exposure because
concentrations of many traffic-related pollutants are higher near roads (HEI, 2010) and average time in
commute in the U.S. has been increasing (Jarosz and Cortes, 2014). Active travel, including bicycling and
walking, is one promising strategy for reducing emissions of traffic-related pollution, and improving
population health through daily exercise. However, studies suggest that active commuters could be at
higher risk of pollutant-related health effects due to higher breathing rates and longer transit times than
automobile commuters (Hatzopoulou et al., 2013; Johan de Hartog et al., 2010). Available routes of
travel for each mode could also play a role because they affect travel times. Hence, more data are needed
under real commuting conditions on the impacts of commute mode and route on exposure to traffic
pollution. Further, models that predict likely exposure are needed in order to design transportation
systems that are healthy for all modes of travel, and to inform peoples’ choices regarding commuting and
residential location.
The work described here is part of a larger project aimed at understanding differences in exposure and
short-term cardiovascular and respiratory responses to a few traffic-related air pollutants during car and
bicycle commuting in a mid-sized U.S. city (Good et al., 2015). For this work, we are developing and
applying a predictive commuter exposure modeling system that integrates mechanistic air pollution fate
and transport modeling, path-following exposure estimation using geographical information systems, and
Bayesian updating of predictions using available measurement data. Here, we describe the design of the
path-following exposure estimator and its application to characterize differences in exposure to select
pollutants for a sample of real commuters using alternative prescribed modes and routes of travel.
METHODS
Scope. The study area of Fort Collins, Colorado (shown in Figure 1) is a mid-sized suburban city with
approximately 160,000 residents. Air quality is generally good, but short-term excursions in pollutant
concentrations and personal exposures may adversely affect health. Automobiles emit a large portion of
ambient pollution in the city, but there is also a substantial network of both on- and off-road bike trails.
We focus on estimating exposures during automobile and bicycle commuting to carbon monoxide (CO)
and black carbon (BC) in the fine particle (PM2.5) size range. Both of these pollutants have known
cardiovascular and respiratory health effects and are considered markers of traffic pollution (HEI, 2010).
Model description. To estimate exposure to traffic-pollution during commuting, we combine temporally
and spatially varying pollutant concentrations estimated via dispersion modeling with sequential records
of individual human activity containing detailed temporal and spatial location coordinates. Using this
data, our exposure estimation module calculates exposure by numerically integrating pollutant
concentrations experienced by a commuter over the commuting time interval (T) as: 𝐸 = ∫𝑇 𝐶𝜎 (𝑡)𝑑𝑡,
1
Figure 1. Map of study area (Fort Collins, Colorado), showing
locations of major roadways, university area (CSU) and central
business district (CBD).
where E is the cumulative exposure (concentration × time), dt is the time step of the integration, and C
is the ambient pollutant concentration at the same location () as the commuter, which varies in time due
to temporal changes in ambient concentration and due to the commuter’s movement in space. The model
also calculates time-weighted-average exposure concentration (CE) by dividing E by T. Calculation of
personal intake is currently being implemented by incorporating exposure factors that vary between
commutes (including cabin infiltration fraction and breathing rate).
Figure 2 provides a flow chart of the exposure estimation process, which is implemented in the R
programming platform. There are three interacting process pathways: processing of activity data (dashed
lines), processing of concentration data (grey lines), and computation of exposure (bold lines). An input
file specifies information about the concentration and activity data and defines the scope of the population
of exposure to estimate. Possible scope selections include the commute mode (car/bicycle), route type
(high/low traffic), and time category including season, day type (weekday/weekend), and time of day
(am/pm, or hour). Processing of activity data involves appending temporal descriptors (hour of day, day
type, month, season), rediscretization of the sequential list of activity records into intervals of consistent
activity type, hour of day, and spatial movement less than 100 m, and calculation of activity durations for
each record. For concentration data, the estimator takes hourly-resolved, gridded ambient concentration
fields. Processing of this data involves splitting, repacking, and indexing the data by hour. For each
spatial field of hourly data, temporal identifiers are appended (hour of day, month, day, day type, season),
and an index table is created to facilitate fast searching and retrieval of hours satisfying a particular
exposure population scope. Pre-processed activity and concentration data saved from previous runs can
also be used.
2
Figure 2. Logic flow for estimating commute exposures
Once the two data sets are prepared, the estimator loops through the selected exposure scopes
(combinations of commute mode, route type, and time parameters). For each scope, the activity data are
filtered to extract those data that satisfy the temporal and activity parameters specified by the scope.
Similarly, subsamples of the concentration fields that match the temporal parameters of the scope are
extracted. Each discrete activity record interval is then matched to a concentration field for the same hour
of day; fields can be selected by random sampling from the available fields within the temporal scope, or
alternatively, a summary field (e.g. seasonal mean) for that scope can be used. The specific concentration
extracted for each activity interval is chosen by matching the spatial location within the field. After
matching, exposures are calculated for each discrete activity record and summed over each full commute.
Finally, summary statistics for each exposure population scope are generated.
Commute activity data. For the results presented here, we used data from personal monitoring of the
weekday commutes of 45 individuals over the period of September 2012 to February 2014. As described
in Good et al. (2015), activity data for each individual included up to 16 commutes on 8 days (one
morning commute and one evening commute each day), where 4 days involved driving commutes and 4
were bicycling commutes. For each mode, 2 of the commutes were on a direct route between the
participant’s home and work location along high-traffic roads (arterials and large collectors) and 2 were
on an alternative route along low-traffic paths (smaller collectors and some paved, off-road trails for
cycling). To estimate exposure for these commutes, we used sequential data recorded on the participant’s
GPS spatial location and time, at 10 s intervals, and categorical identifiers on the commute mode
(car/bike), route type (high/low traffic), and time of day (am/pm).
3
Concentration estimation. To generate concentration data for this study, we used the AERMOD
Gaussian plume model to simulate dispersion from all emissions sources in a 400 km2 domain centered on
Fort Collins. AERMOD is an established US Environmental Protection Agency (EPA) model that
estimates realistic hourly-varying concentrations from multiple source types at high spatial resolution.
(We chose this particular model for its simplicity – we are designing our system within a Bayesian data
assimilation framework in which probabilistic estimates of concentrations and exposures are generated;
this requires many repeated runs of the dispersion model to produce an adequate sample.) Emissions
input to the model were calculated from the 2008 National Emissions Inventory, with spatial allocation to
individual major roadway links and a regular grid of 0.25 km2 area sources using methods similar to those
described by Yu and Stuart (2016). Hourly varying meteorological data were obtained from National
Weather Station (NWS) hourly surface observations (Loveland Airport, WBAN 94062), twice-daily upper
air soundings (Denver/Stapleton Airport, WBAN 23062), and 10-minute onsite measurements (Colorado
State University, WBAN 53005). AERMOD was run to produce gridded concentrations with spatial
resolution of 500 m for a 100 km2 area centered on Fort Collins for each hour of the year.
Run specifications. Using this data, we ran the exposure estimator to calculate cumulative and timeweighted average exposure for each commute. The estimator was set to match concentrations to activity
records by sampling from all the dispersion model-generated fields (for the appropriate hour of the day).
Figure 3. Mean (with 95% confidence intervals) of raw individual differences in cumulative and time-weighted
average exposure to CO and BC for each mode/route combination compared with the direct (high traffic) car
commute, including both morning and evening commutes. BC = black carbon, CO = carbon monoxide, cum.exp =
cumulative exposure, twa.exp = time-weighted average exposure.
4
RESULTS AND DISCUSSION
Figure 3 shows the mean differences in raw individual cumulative and time-weighted average exposure
for each mode/route combination compared with the direct (high traffic) car commute, as calculated by
the exposure estimator. Mean cumulative exposures to BC and CO were estimated to be substantially
higher when cycling either route type compared with the direct route, and slightly higher for driving the
alternative route. This increased exposure is largely attributable to longer commute durations. Mean
change in time-weighted average exposures does not appear substantially different from zero across all
alternative mode/route combinations when compared to driving on a direct route. Ongoing work includes
incorporation of exposure and intake factors that differ by mode, microenvironment, and season (e.g.
vehicle infiltration and ventilation), analysis using linear mixed modeling to isolate the differences
attributable to mode and route in the context of repeated measures, and implementation of exposure
estimation within a Bayesian data assimilation framework.
SUMMARY
We are developing an exposure modeling system that estimates probabilistic commute exposures in Fort
Collins, Colorado. The exposure estimator module, which matches predicted concentrations with
personal activity records, was described here. It was also applied to explore differences in cumulative and
time-weighted average exposure to CO and BC between driving and bicycling along direct and alternative
routes for a sample of 45 commuters with measured activity data. Preliminary results suggest that
exposures to both pollutants may be higher when biking.
ACKNOWLEDGEMENTS
This work was funded by the National Institute of Environmental Health Sciences of the National
Institutes of Health under grant R01ES020017. The content is solely the responsibility of the authors and
does not necessarily represent the official views of the National Institutes of Health.
REFERENCES
Good N, Mölter A, Ackerson C, Bachand A, Carpenter T, Clark ML, Fedak KM, Kayne A, Koehler K,
Moore B, L'Orange C, Quinn C, Ugave V, Stuart AL, Peel JL, and Volckens J, 2015. The Fort Collins
Commuter Study: Impact of route type and transport mode on personal exposure to multiple air
pollutants. Journal of Exposure Science and Environmental Epidemiology. doi:10.1038/jes.2015.68.
Hatzopoulou M, Weichenthal S, Dugum H, Pickett G, Miranda-Moreno L, Kulka R, Andersen R, and
Goldberg M, 2013. The impact of traffic volume, composition, and road geometry on personal air
pollution exposures among cyclists in Montreal, Canada. Journal of Exposure Science and
Environmental Epidemiology, 23, 46-51; doi:10.1038/jes.2012.85.
HEI Panel on the Health Effects of Traffic-Related Air Pollution, 2010. Traffic-related Air Pollution: A
Critical Review of the Literature on Emissions, Exposure, and Health Effects. HEI Special Report
17Heath Effects Institute, Boston, MA.
Jarosz B and Cortes RT, 2014. In U.S., New Data Show Longer, More Sedentary Commutes. Population
Reference Bureau article, accessed at http://www.prb.org/Publications/Articles/2014/uscommuting.aspx, on February 11, 2016.
De Hartog JJ, Boogaard H, Nijland H and Hoek G, 2010. Do the health benefits of cycling outweigh the
risks? Environmental Health Perspectives, 118 (8), 1109-1116.
Yu H and Stuart AL, 2016. Exposure and inequality for select urban air pollutants in the Tampa Bay area.
Science of the Total Environment, 551–552, 474–483, doi:10.1016/j.scitotenv.2016.01.157.
5