Science of the Total Environment 530531 (2015) 257262

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

Inuence of walking route choice on primary school children's exposure

to air pollution A proof of concept study using simulation
Anna Mlter a,, Sarah Lindley b
a
Department of Environmental and Radiological Health Sciences, Colorado State University, USA
b
School of Environment, Education & Development (Geography), The University of Manchester, UK

H I G H L I G H T S

Network analysis can be used to identify less polluted walking routes.

We carried out a simulation of 100,000 walking routes to primary schools.
For most routes relative reductions in exposure exceeded the increase in duration.
Identifying less polluted school routes could provide health benets for children.

a r t i c l e i n f o a b s t r a c t

Article history: This study developed a walking network for the Greater Manchester area (UK). The walking network allows
Received 6 February 2015 routes to be calculated either based on the shortest duration or based on the lowest cumulative nitrogen dioxide
Received in revised form 5 May 2015 (NO2) or particulate matter (PM10) exposure. The aim of this study was to analyse the costs and benets of faster
Accepted 26 May 2015
routes versus lower pollution exposure for walking routes to primary schools.
Available online 2 June 2015
Random samples of primary schools and residential addresses were used to generate 100,000 hypothetical
Editor: D. Barcelo school routes. For 60% (59,992) and 40% (40,460) an alternative low NO2 and PM10 route was found, respectively.
The median change in travel time (NO2: 4.5 s, PM10: 0.5 s) and average route exposure (NO2: 0.40 g/m3, PM10:
Keywords: 0.03 g/m3) was small. However, quantile regression analysis indicated that for 50% of routes a 1% increase in
Air pollution travel time was associated with a 1.5% decrease in NO2 and PM10 exposure.
Network analysis The results of this study suggest that the relative decrease in pollution exposure on low pollution routes tends to
GIS be greater than the relative increase in route length. This supports the idea that a route planning tool identifying
Walking routes less polluted routes to primary schools could help deliver potential health benets for children.
Journey-time exposure
2015 Elsevier B.V. All rights reserved.

1. Introduction
Walking to school is a simple and low-cost form of daily physical ac-
tivity for children (Foty et al., 2013) and is often promoted by policy
makers and public health groups (Mackett, 2013). Most children
spend a relatively small proportion of their day commuting to and
from school; however, they may be disproportionately exposed to air
pollution during this time period, due to travelling in close vicinity to
road trafc, especially in urban areas. Children are thought to be partic-
ularly sensitive to the adverse effects of air pollution exposure (Kim
Abbreviations: GIS, Geographic Information Systems; NO2, Nitrogen Dioxide; PM10,
Particulate Matter with an aerodynamic diameter less than 10 m; ITN, Integrated
Transport Network; RRI, Road Routing Information; OLS, Ordinary Least Squares
regression; IQR, Interquartile Range.
Corresponding author at: 1681 Campus Delivery, Colorado State University, Fort
Collins, CO 80523-1681, USA.
E-mail address: anna.molter@colostate.edu (A. Mlter).
et al., 2004), therefore home to school routes with lower pollution levels
could provide health benets for them.
Geographic Information System (GIS) packages will calculate the
shortest route between homes and schools through a road network
based on distance or on travel time. However, the underlying network
analysis algorithms could also be used to calculate routes with the low-
est air pollution exposure. To our knowledge only one online service
(www.walkit.com) currently provides low pollution routes and only
for a very limited spatial coverage (West and North London, UK; Cam-
bridge, UK). Few studies have compared shortest distance routes to
low pollution routes (Davies and Whyatt, 2014; Hatzopoulou et al.,
2013; Hertel et al., 2008; Sharker and Karimi, 2014) and most of these
studies focused on cycling rather than walking.
The aim of this study is to compare walking routes of primary school
children from home to school calculated on the basis of travel time and
lowest nitrogen dioxide (NO2) or particulate matter (PM10) exposure.
Random samples of primary schools and residential addresses were
used to generate 100,000 hypothetical school routes in the Greater

http://dx.doi.org/10.1016/j.scitotenv.2015.05.118
0048-9697/ 2015 Elsevier B.V. All rights reserved.
258 A. Mlter, S. Lindley / Science of the Total Environment 530531 (2015) 257262
Manchester area (UK). These routes were analysed to answer the fol-
lowing research questions: what is the reduction in exposure on a
route optimized for low exposure? How much longer does it take to
travel along a route optimized for low exposure?
This is a proof of concept study to determine the costs and benets of
low pollution walking routes and to assess the potential value of a low
pollution route calculator.
2. Methods
2.1. Study area
The geographic setting for this research is the Greater Manchester
area in the North West of England. Greater Manchester is one of
the largest metropolitan areas in the UK covering approximately
1300 km2 with a population of approximately 2.7 million people
(ONS, 2011).
2.2. Development of walking network
The Ordnance Survey Mastermap Integrated Transport Network
(ITN) layer (scale 1:1250) is a vector dataset of the road network
and associated road routing information (RRI) for Great Britain
(Ordnance Survey, 2010). The ITN layer is geometrically aligned with
underlying topographic features available through the Ordnance Survey
Mastermap Topography Layer (scale 1:1250 in urban areas) (Ordnance
Survey, 2008). The Topography Layer is a polygon layer of the physical
extent of all surface features in Great Britain. Examples of features in
the Topography Layer include roads, sidewalks, buildings, and parks.
The line features of the ITN layer are typically located along the centre
line of the underlying road polygons (Topography layer), which limits
Fig. 1. Example illustrating the spatial difference between the ITN network and walking network.
its applicability in this study for the following reasons: (1) Pedestrians
do not walk in the centre of a road. Therefore, the spatial location of
the line features is unrepresentative of walking behaviour and intro-
duces positional inaccuracy. (2) Air pollution rapidly decreases perpen-
dicular to the road centre line (Gilbert et al., 2003), therefore
concentrations in the centre of a road would overestimate a pedestrian's
exposure. As a consequence we decided to create a new network
dataset specically for walking routes in Greater Manchester. For this
walking network we used ArcGIS 10.2 (Esri Inc.) to create 1 m buffers
around all road polygons (except motorways) and then converted
these into polylines. A large proportion of residential areas in Greater
Manchester were built during the Industrial Revolution in the
Victorian era. According to the UK Census of Population (ONS, 2011),
37.7% of all whole house dwellings in the study area are terraced and
45.8% are semi-detached. Many of the associated residential areas are
characterized by relatively narrow roads with narrow sidewalks. We
used a buffer distance of 1 m to ensure that the new walking routes
were located at a reasonable distance from the road edge, but at the
same time did not cross buildings or front gardens. The top and bottom
of the buffer polygons allowed connectivity at road junctions (Fig. 1).
A land use regression (LUR) model from a previous study (Mlter
et al., 2010) was used to predict NO2 and PM10 concentrations for the
new walking network. In brief, the LUR models were developed using
NO2 and PM10 concentrations at 278 sites in Greater Manchester provid-
ed by an air dispersion model. The potential predictors were variables
representing trafc intensity, background emission estimates, local
land use and physical geography. The nal LUR models had adjusted
R2 values of 0.7 and comprised eight predictor variables for the NO2
model and nine predictor variables for the PM10 model. The most inu-
ential predictors were trafc intensity on nearby motorways and A
roads.
 A. Mlter, S. Lindley / Science of the Total Environment 530531 (2015) 257262
The polylines were split at nodes to create straight line segments. For
each line segment shorter than 100 m the centre point was calculated.
Line segments longer than 100 m were split into 100 m segments and
the centre point was calculated. The LUR model was used to predict pol-
lutant concentrations at each centre point.
We assumed that children walk at a speed of 0.9 m/s (Gong and
Mackett, 2009). The length of each line segment described in the previ-
ous paragraph was divided by the walking speed to calculate the travel
time (in seconds) required to traverse it. The travel time was then mul-
tiplied with the pollutant concentration to create a pollution cost attri-
bute for each line segment. All line segments were used to build the
network. As this is a walking network, universal connectivity was
used so that movement was independent of direction and trafc control
measures and no turning restrictions were present. The cost attributes
were travel time (in seconds (s)) and cumulative NO2 and PM10 expo-
sure (in g/m3).
As it is unrealistic to assume that children move at a constant speed
and, in practice, cumulative exposures are affected by waiting times at
road crossings, we added cost barriers into the network analysis. Bar-
riers were added at the kerbside (i.e., the boundary between the side-
walk and the road), whenever the walking network crossed into the
road and vice versa. The LUR model was used to estimate NO2 and
PM10 concentrations for each barrier. It was assumed that crossing a
minor road incurred an added cost of 2 s (Pitcairn and Edlmann,
2000) and crossing a major road incurred an added cost of 15 s (de
Castella, 2013; York et al., 2011). These time costs were multiplied
with the NO2 and PM10 concentration to obtain cumulative exposure
cost attributes.
2.3. Routes
A random sample of primary schools (n = 500) within the Greater
Manchester area was selected and formed the endpoints of the routes.
The service area function in ArcGIS Network Analyst was used to create
catchment polygons for each primary school. Each catchment polygon
represented the area within 1500 m walking distance of the school
(Fig. 2). This corresponds to the upper threshold for walking distances
to school recommended for children in North America (Foty et al.,
2013; Oreskovic et al., 2009). Within each catchment polygon a random
sample of 200 residential addresses was selected resulting in 100,000
home (start point) to school (end point) pairs. For each home to school
pair ArcGIS Network Analyst was used to nd the fastest walking route
and the routes with the lowest cumulative NO2 or PM10 exposure.
2.4. Statistical analysis
Our walking network exposure assessment model automatically cal-
culates the total length, total time and cumulative exposure for each
route. To calculate the average pollutant concentration on each route
the cumulative exposure was divided by the total time. Subsequently,
the absolute and percentage difference in duration and average pollut-
ant concentration was calculated. Quantile regression was used to illus-
trate the change in pollutant concentration associated with the change
in travel time for different quantiles of the study sample (Cade and
Noon, 2003; Koenker and Hallock, 2001). All analyses were carried out
in SPSS v22 (IBM Corp.) and R v3.1.0 (The R Foundation).
3. Results
Network Analyst found routes for 99.94% of the home-school pairs
entered into the analysis. Even though only residences within the catch-
ment polygons were selected, 3312 routes were longer than 1500 m
(only routes b1500 m were included in subsequent analyses). For
around 60% (59,992) and 40% (40,460) of routes a less polluted route
was found, based on cumulative NO2 and PM10 exposure respectively.
Table 1 shows descriptive statistics of all routes. For individual pairs of
259
routes the increase in travel time and the decrease in exposure tended
to be small: the median increase in travel time on low NO2 routes was
4.5 s (IQR: 0.916.9 s) and on low PM10 routes was 0.5 s (IQR: 0.1
1.7 s). The median decrease in the average NO2 exposure on low NO2
routes was 0.40 g/m3 (IQR: 0.091.22 g/m3) and the median decrease
in the average PM10 exposure on low PM10 routes was 0.03 g/m3 (IQR:
0.010.08 g/m3). Fig. 3 shows the absolute and relative change in dura-
tion against the change in average pollutant exposure. The majority of
routes changed by less than 2 min and less than 5 g/m3 for NO2 and
less than 1 min and less than 2 g/m3 for PM10. The maximum change
in route duration was 540 s (9 min) for NO2 and 361 s (6 min) for
PM10. The maximum reduction in average exposure was 16.9 g/m3
for NO2 and 5.5 g/m3 for PM10. The regression lines and Table 2 show
that for 50% of routes a 1 s increase in route duration was associated
with 0.04 g/m3 decrease in NO2 and 0.03 g/m3 decrease in PM10 ex-
posure. For 95% of routes a 1 s increase in route duration was associated
with 0.02 g/m3 decrease in NO2 and 0.02 decrease in PM10 exposure.
The relative change uses the same unit (%) to compare the change in
length to the change in exposure. Fig. 4 and Table 2 show that for 50%
of routes a 1% increase in route duration was associated with a 1.50%
decrease in NO2 and a 1.53% decrease in PM10 exposure. For 95% of
routes a 1% increase in route duration is equivalent to or better than
the percentage decrease in exposure. Fig. 4 also indicates that for NO2
ordinary least squares regression (OLS) underestimates the average
change in exposure compared to quantile regression. For PM10 OLS
and quantile regression provide similar results. However, Fig. 4 also
shows a steep increase in exposure reduction for PM10 in the rst quar-
tile of routes, with 25% of routes achieving a 2% reduction in exposure
and 5% of routes achieving a 3.5% reduction in exposure.
4. Discussion
This simulation study of 100,000 walking routes shows that the per-
centage reduction in exposure exceeds the percentage increase in route
length for at least 50% of routes. Quantile regression of the absolute
change shows that walking along an alternative route which is 120 s
(2 min) longer results for 50% of routes in a 5 g/m3 decrease in NO2 ex-
posure and a 3.2 g/m3 decrease in PM10 exposure. These results suggest
that for a large proportion of alternative low exposure routes the benet
of lowered exposure is greater than the cost associated with walking for
a longer time. Furthermore, considering the health benets of physical
exercise, walking for an additional length of time in many cases may
not constitute a cost to a child.
A limited number of studies have investigated differences between
short routes and routes optimized for low exposure. A study in Copen-
hagen, Denmark, compared modelled exposures of 50 ctitious people
cycling from home to two destinations, either along the shortest route
or along a route with low annual daily trafc counts (Hertel et al.,
2008). The Copenhagen study found a mean decrease in the average
route concentration of 1014% in NO2 and 1113% in PM10. These per-
centage decreases are much larger than in our study, which is most like-
ly due to the fact that the routes used were up to 10 times longer
(15 km). A study in Montreal, Canada, used data from a travel survey
to identify start and end points for 2307 cycling trips (Hatzopoulou
et al., 2013). It used network analysis with a Google Maps front-end to
calculate the shortest route (based on length) and the cleanest route
based on NO2 exposure. Similar to our study, the Montreal study could
only nd alternative routes for 57% of the start-end point pairs. Limiting
the analyses to routes with a cleaner alternative found a mean de-
crease in NO2 concentration of 0.76 ppb (approximately 1.45 g/m3).
This reduction is much higher than in our study, which again is most
likely due to the greater length of cycle routes (468 m30.3 km) com-
pared to our walking routes. The only published study that compared
pollutant exposure on walking routes was a study in Lancaster, UK
(Davies and Whyatt, 2014). The Lancaster study investigated inhalation
journey time exposures based on PM2.5 concentrations, journey
260 A. Mlter, S. Lindley / Science of the Total Environment 530531 (2015) 257262

Fig. 2. (A) Map of the study area showing the location of primary schools. (B) Example of a catchment polygon.

duration and respiratory minute volume for 16,440 hypothetical routes identical, which again is comparable to our study and the Montreal
on 60 days. PM2.5 concentrations included varying background and me- study. The Lancaster study found a mean reduction in the PM2.5 inhaled
teorological conditions, while journey duration allowed for differences dose of 1.8 g; however, it was also found that on a daily basis the least
in walking speed, and respiratory minute volume allowed for differ- polluted route varied depending on changes in local meteorology and
ences in physiology and activity level. The Lancaster study found that background concentrations. In addition, this study demonstrated differ-
for 51% of routes the shortest route and the least polluted route were ences in inhalation journey time exposure based on an individual's
physiology, physical activity and travel speed, which can also inuence
the choice of least polluted route.
Table 1
Median (IQR) travel time and exposure (n = 96,628). An obvious limitation of this study is that it is a simulation study and
therefore the start-end point pairs are hypothetical. However, the use of
Fastest route Low NO2 route Low PM10 route
simulations also allowed us to work with a large sample size. Studies
Time (s) 1144 (784, 1434) 1152 (789, 1446) 1144 (785, 1435) based on real-life start and end points are usually limited to much small-
NO2 exposure (g/m3) 32.9 (31.6, 34.6) 32.5 (31.1, 34.1) 32.9 (31.6, 34.6) er sample sizes, for example the Montreal study, even though it used
PM10 exposure (g/m3) 20.9 (20.6, 21.2) 20.8 (20.6, 21.1) 20.8 (20.6, 21.1)
data from a city wide survey, could only identify 2307 cycling trips
 A. Mlter, S. Lindley / Science of the Total Environment 530531 (2015) 257262 261

Fig. 3. Scatterplots of change in duration vs change in average exposure (NO2: n = 59,992; PM10: n = 40,460). Change is expressed as absolute change (difference in s or g/m3) and rel-
ative change (percent difference). Lines indicate the 50%ile and 95%ile coefcients obtained from quantile regression.

(Hatzopoulou et al., 2013). A further limitation of this study is that route perceived as a risk, or not prioritized among the range of risks which
choice is affected by multiple factors not simply by travel time and pol- parents have to consider.
lution exposure. Previous studies on parents' choices regarding their A further consideration is that our results are based on exposure es-
child's travel mode to school have found that the main factors tend to timates generated from a LUR model rather than direct measurements
be convenience, road trafc dangers and personal safety (Faulkner of air pollution. The pollution models used in this study have been vali-
et al., 2010; Panter et al., 2010). A recent study on school travel patterns dated and potential limitations have been discussed in a previous pub-
in asthmatic children found that few parents were concerned about pol- lication (Mlter et al., 2010). The determination coefcients (R2) of
lution during their child's commute and a bivariate analysis showed the models were 0.71, which indicates good performance for a LUR
that it was not a signicant factor inuencing parents' choice of travel model and is comparable to other published LUR models (Beelen
mode (Oreskovic et al., 2009). These ndings might suggest that even et al., 2013; Eeftens et al., 2012). However, it should also be noted that
if a tool for low pollution routes was publicly available, it may not be direct measurements of 100,000 routes are not feasible and that a net-
widely used, unless coupled with awareness raising campaigns. At pres- work analysis of low pollution routes will always require some degree
ent, air pollution exposure during children's school commutes is not of pollution modelling. The air pollution models used in this study esti-
mated annual average concentrations; however, previous studies have
shown that diurnal and daily variability in pollution levels can change
Table 2
the location of a low pollution route (Davies and Whyatt, 2014; Hertel
Quantile regression coefcients and standard errors for 50%ile and 95%ile (per 1 s or 1%
increase in time). et al., 2008). This may suggest that a route planning tool based on a dy-
namic pollution model, i.e., a model that incorporates changes in trafc,
50%ile 95%ile
meteorology and background concentrations in real time, could provide
Coefcient Std error Coefcient Std error information about routes with even greater reductions in pollution
NO2 Absolute change (g/m3) 0.04 0.003 0.02 0.0001 exposure.
Relative change (%) 1.50 0.009 0.98 0.002 The walking network developed for this study was based on
PM10 Absolute change (g/m3) 0.03 0.0003 0.02 0.0001 walking along sidewalks next to the existing road network. The walking
Relative change (%) 1.53 0.019 1.03 0.003
network could be further improved by adding connectivity through
262 A. Mlter, S. Lindley / Science of the Total Environment 530531 (2015) 257262
Fig. 4. Quantile regression plot of relative change (NO2: n = 59,992; PM10: n = 40,460). The black line shows the regression coefcient for each quantile, the grey area indicates the 95%
condence interval. The red line indicates the regression coefcient derived from ordinary least squares regression.
pedestrianised areas, for example through parks. However, using the
datasets we had available this may require some manual calibration of
the network, which was impractical for the size of the study area. Final-
ly, it should be noted that during the network development a cost bar-
rier was added to either side of the road. Unfortunately, it was not
possible to set the barriers to only incur a cost when crossing from the
sidewalk into the road and not when crossing vice versa. Therefore,
crossing a road incurred a cost on either side. As this cost affected
both route duration and exposure it is unlikely that the second barrier
had an impact on the overall conclusions of this study. However, it is
possible that absolute route durations and cumulative exposures may
have been overestimated.
In conclusion, we found that for 4060% of our simulated sample of
school walking routes a less polluted alternative route was available.
Overall, differences in route duration and pollution exposure were rela-
tively small, but the results indicated that the percentage decrease in ex-
posure tends to be slightly larger than the percentage increase in route
length. Therefore, a publicly available tool to identify alternative low
pollution walking routes, could help deliver health benets for children
and adults. Ideally, this route planning tool should incorporate a dynam-
ic pollution model; though a validated empirically-based model like the
one used in this study could still provide useful information about aver-
age exposure estimates on candidate routes. Finally, as parents do not
seem to perceive air pollution as a risk factor during school commutes,
further work may be necessary to achieve behavioural changes related
to the choice of school route.
Acknowledgements
We acknowledge the use of Ordnance Survey data Crown copy-
right and database right [2013]. This study was funded through a Uni-
versity of Manchester strategic studentship. We would like to thank
the two anonymous reviewers for their helpful insights and suggestions
during the review of this manuscript.
References
Beelen, R., Hoek, G., Vienneau, D., Eeftens, M., Dimakopoulou, K., Pedeli, X., et al., 2013. De-
velopment of NO2 and NOx land use regression models for estimating air pollution
exposure in 36 study areas in Europe the ESCAPE project. Atmos. Environ. 72,
1023.
Cade, B.S., Noon, B.R., 2003. A gentle introduction to quantile regression for ecologists.
Front. Ecol. Environ. 1, 412420.
Davies, G., Whyatt, J.D., 2014. A network-based approach for estimating pedestrian
journey-time exposure to air pollution. Sci. Total Environ. 485486, 6270.
de Castella, T., 2013. Does pressing the pedestrian crossing button actually do anything?
BBC News Magazine. BBC
Eeftens, M., Beelen, R., de Hoogh, K., Bellander, T., Cesaroni, G., Cirach, M., et al., 2012. De-
velopment of land use regression models for PM2.5, PM2.5 absorbance, PM10 and
PMcoarse in 20 European study areas; results of the ESCAPE project. Environ. Sci.
Technol. 46, 1119511205.
Faulkner, G.E., Richichi, V., Buliung, R.N., Fusco, C., Moola, F., 2010. What's quickest and
easiest?: parental decision making about school trip mode. Int. J. behav. Nutr.
Phys. Act. 7, 62.
Foty, R.G., Nelligan, K., To, T., Stieb, D.M., Dell, S.D., 2013. Children with asthma are less
likely to walk to school. Pediatr. Allergy Immunol. Pulmonol. 26, 8187.
Gilbert, N.L., Woodhouse, S., Stieb, D.M., Brook, J.R., 2003. Ambient nitrogen dioxide and
distance from a major highway. Sci. Total Environ. 312, 4346.
Gong, Y., Mackett, R., 2009. Visualizing children's walking behavior using portable global
positioning (GPS) units and activity monitors. In: Lin, H., Batty, M. (Eds.), Virtual Geo-
graphic Environments. Science Press, pp. 295311.
Hatzopoulou, M., Weichenthal, S., Barreau, G., Goldberg, M., Farrell, W., Crouse, D., et al.,
2013. A web-based route planning tool to reduce cyclists' exposures to trafc pollu-
tion: a case study in Montreal, Canada. Environ. Res. 123, 5861.
Hertel, O., Hvidberg, M., Ketzel, M., Storm, L., Stausgaard, L., 2008. A proper choice of route
signicantly reduces air pollution exposure a study on bicycle and bus trips in
urban streets. Sci. Total Environ. 389, 5870.
Kim, J.J., Shannon, M.W., Best, D., Binns, H.J., Johnson, C.L., Mazur, L.L., et al., 2004. Ambient
air pollution: health hazards to children. (Policy Statement). Pediatrics 114, 1699.
Koenker, R., Hallock, K.F., 2001. Quantile regression. J. Econ. Perspect. 15, 143156.
Mackett, R.L., 2013. Children's travel behaviour and its health implications. Transp. Policy
26, 6672.
Mlter, A., Lindley, S., de Vocht, F., Simpson, A., Agius, R., 2010. Modelling air pollution for
epidemiologic research part I: a novel approach combining land use regression and
air dispersion. Sci. Total Environ. 408, 58625869.
ONS, 2011. Census: aggregate data (England and Wales) [computer le]. In: Statistics OfN
(Ed.), UK Data Service Census Support.
Ordnance Survey, 2008. OS Mastermap Topography Layer User Guide and Technical
Specication.
Ordnance Survey, 2010. OS Mastermap Integrated Transport Network Layer User Guide
and Technical Specication.
Oreskovic, N.M., Sawicki, G.S., Kinane, T.B., Winickoff, J.P., Perrin, J.M., 2009. Travel pat-
terns to school among children with asthma. Clin. Pediatr. 48 (6), 632640.
Panter, J.R., Jones, A.P., Van Sluijs, E.M., Grifn, S.J., 2010. Neighborhood, route, and school
environments and children's active commuting. Am. J. Prev. Med. 38, 268278.
Pitcairn, T.K., Edlmann, T., 2000. Individual differences in road crossing ability in young
children and adults. Br. J. Psychol. 91 (Pt 3), 391410.
Sharker, M.H., Karimi, H.A., 2014. Computing least air pollution exposure routes. Int.
J. Geogr. Inf. Sci. 28, 343362.
York, I., B, S., Beesley, R., Webster, D., Knight, P., Hopkin, J., 2011. Pedestrian Countdown at
Trafc Signal Junctions (PCaTS) Road Trial. Transport Research Laboratory.