Content uploaded by Ari B Friedman
Author content
All content in this area was uploaded by Ari B Friedman
Content may be subject to copyright.
Available via license: Public Domain Mark 1.0
Content may be subject to copyright.
Available via license: Public Domain Mark 1.0
Content may be subject to copyright.
Environmental Health Perspectives
•
v o l u m e 118 | n u m b e r 5 | May 2010
607
Research
Urban air pollution is responsible for an
estimated 800,000 annual deaths worldwide
(Cohen et al. 2004; Ezzati et al. 2002). e
burden of disease from air pollution exposure
is borne disproportionately by the growing
urban populations in the developing world,
where pollution is substantially higher than in
high-income countries (Cohen et al. 2004).
Research in cities in the United States and
Europe has demonstrated substantial spatial
variation in air pollution between and within
neighborhoods, in relation primarily to spe-
cific sources as well as to neighborhood socio-
economic status (SES) (Buzzelli and Jerrett
2004; Charron and Harrison 2005; Hoek
et al. 2001, 2002; Holmes et al. 2005; Kinney
and O’Neill 2006; Levy et al. 2000, 2001;
Loh et al. 2002; O’Neill et al. 2003; Su et al.
2007, 2008; Weijers et al. 2004).
Sources of air pollution in developing
country cities include those that are common
in high-income nations (e.g., transportation
and industrial pollution), as well as biomass
and coal fuel use for household and com-
mercial purposes (Barnes et al. 2005) and
resuspended dust from unpaved roads. A few
studies have examined spatial variability and
sources of air pollution in cities in developing
countries (Chowdhury 2004; Engelbrecht
et al. 2001; Etyemezian et al. 2005; Jackson
2005; Padhi and Padhy 2008; Saksena et al.
2003; van Vliet and Kinney 2007; Zheng
et al. 2005), but few have been in low-
income “slum” neighborhoods, have system-
atically examined variation in air pollution
within neighborhoods, or have analyzed
the effects of sources on local pollution lev-
els. e absence of data on air pollution in
urban communities in the developing world,
especially from slums, occurs despite the
evidence that other environmental factors,
such as sanitation infrastructure, are worse in
slum areas (Sclar et al. 2005; Songsore and
McGranahan 1998).
In the present study we systematically col-
lected and analyzed data on particulate matter
(PM) in four neighborhoods in Accra, Ghana,
with emphasis on within-neighborhood vari-
ability of PM pollution in relation to nearby
sources. Accra is a major city in sub-Saharan
Africa, the region with the highest urban
population growth rate in the world (United
Nations Department of Economic and Social
Affairs 2004).
Materials and Methods
Study location. Accra is the capital city of
Ghana and is located on the Gulf of Guinea,
with an elevation of 0–60 m above sea level.
The population of the Accra metropolitan
area (AMA) increased from 600,000 in 1970
to 1 million in 1984 and 1.7 million in 2000.
Our study took place in four Accra
neighborhoods: Jamestown/Ushertown (JT),
Asylum Down (AD), Nima (NM), and East
Legon (EL) (Figure 1). Study neighborhoods
were selected such that they lie on a nearly
straight line from the coast to the north-
ern boundaries of the AMA, and they had
varying SES based on data from the 2000
Population and Housing Census of Ghana
(Agyei-Mensah and Owusu 2009). JT is an
old inner-core area that lies between the coast
and the Accra business center; AD and NM
are located approximately 3 km inland, sepa-
rated from one another by the Ring Road
Central; EL is 10 km inland and lies just
north of Kotoka International Airport in
Accra. JT and NM are poor, densely popu-
lated communities where many residents live
in shared compounds along narrow alleys.
Biomass is the predominant fuel used for
household cooking and is also used for small-
scale commercial purposes, such as the smok-
ing of fish over wood fires (JT) and cooking
of food by street vendors (JT and NM). Both
JT and NM have markets and many small
vendors that sustain activities throughout the
day. A large, busy road with a central bus sta-
tion runs through NM. AD is a middle-class
neighborhood, with a combination of resi-
dential and commercial buildings. It is bor-
dered by the Ring Road Central, one of the
Address correspondence to M. Ezzati, Harvard
School of Public Health, 665 Huntington Ave.,
Boston, MA 02115 USA. Telephone: 1-617-432-
5722. Fax: 1-617-432-6733. E-mail: majid_ezzati@
harvard.edu
*ese authors contributed equally to the research
and manuscript.
This research was funded by National Science
Foundation grant 0527536 and by the Sustainability
Science Program at the Center for International
Development at Harvard University (through a
grant from the Italian Ministry for Environment,
Land and Sea).
e authors declare they have no actual or poten-
tial competing financial interests.
Received 22 August 2009; accepted 7 January 2010.
Within-Neighborhood Patterns and Sources of Particle Pollution:
Mobile Monitoring and Geographic Information System Analysis
in Four Communities in Accra, Ghana
Kathie L. Dionisio,1,2,* Michael S. Rooney,2,* Raphael E. Arku,3,4,* Ari B. Friedman,2 Allison F. Hughes,5
Jose Vallarino,1 Samuel Agyei-Mensah,4,6 John D. Spengler,1 and Majid Ezzati1,2
1Harvard School of Public Health, Boston, Massachusetts, USA; 2Harvard Initiative for Global Health, Cambridge, Massachusetts,
USA; 3Cyprus International Institute for the Environment and Public Health, Nicosia, Cyprus; 4Department of Geography and Resource
Development, 5Department of Physics, and 6Environmental Science Program, University of Ghana, Legon, Accra, Ghana
Background: Sources of air pollution in developing country cities include transportation and
industrial pollution, biomass and coal fuel use, and resuspended dust from unpaved roads.
oB j e c t i v e s : Our goal was to understand within-neighborhood spatial variability of particulate mat-
ter (PM) in communities of varying socioeconomic status (SES) in Accra, Ghana, and to quantify
the effects of nearby sources on local PM concentration.
Methods: We conducted 1 week of morning and afternoon mobile and stationary air pollution
measurements in four study neighborhoods. PM with aerodynamic diameters ≤ 2.5 µm (PM2.5) and
≤ 10 µm (PM10) was measured continuously, with matched global positioning system coordinates;
detailed data on local sources were collected at periodic stops. e effects of nearby sources on local
PM were estimated using linear mixed-effects models.
re s u l t s : In our measurement campaign, the geometric means of PM2.5 and PM10 along the mobile
monitoring path were 21 and 49 µg/m3, respectively, in the neighborhood with highest SES and
39 and 96 µg/m3, respectively, in the neighborhood with lowest SES and highest population den-
sity. PM2.5 and PM10 were as high as 200 and 400 µg/m3, respectively, in some segments of the
path. After adjusting for other factors, the factors that had the largest effects on local PM pollution
were nearby wood and charcoal stoves, congested and heavy traffic, loose dirt road surface, and trash
burning.
conclusions: Biomass fuels, transportation, and unpaved roads may be important determinants
of local PM variation in Accra neighborhoods. If confirmed by additional or supporting data, the
results demonstrate the need for effective and equitable interventions and policies that reduce the
impacts of traffic and biomass pollution.
ke y words: Africa, biomass, geographic information system, particulate matter, poverty, spatial,
urbanization. Environ Health Perspect 118:607–613 (2010). doi:10.1289/ehp.0901365 [Online
7 January 2010]
Dionisio et al.
608
v o l u m e 118 | n u m b e r 5 | May 2010
•
Environmental Health Perspectives
largest and busiest roads in Accra. Street food
vendors are less common in AD than in JT
and NM. EL is an upper-class, sparsely popu-
lated, residential neighborhood with most
families living in modern homes on large
plots of land. e streets are quiet during the
day. e main road in EL has heavier traffic
primarily during the morning and evening
commute periods. According to data from the
Ghana 2000 Population and Housing Census
(unpublished data), 81% of households in
JT, 75% in NM, 49% in AD, and 46% in
EL use charcoal and/or wood as their primary
cooking fuel.
Study design. A study of spatial and tem-
poral variability of air pollution would ideally
be based on continuous data using a dense
network of monitors placed at multiple loca-
tions, with additional information on sources
and meteorologic factors gathered at each
location. For PM, which is generally consid-
ered the best indicator of the health effects of
urban air pollution, such an approach would
be prohibitively costly and logistically dif-
ficult using current technologies, especially
when continuous (vs. temporally integrated/
averaged) data are needed. We used a com-
bination of mobile and stationary monitors
to examine the spatiotemporal patterns of air
pollution and the effects of sources in these
neighborhoods.
We conducted consecutive days of mobile
monitoring in each neighborhood (7 days in
most neighborhoods) (Table 1). ere were
two monitoring tours on each day, unless it
rained for more than approximately 30 min
or there was an equipment failure, in which
case the tour was postponed or cancelled.
In each monitoring tour, we walked slowly
along a predetermined path recording data at
1-min intervals with a continuous real-time
PM monitor and a global positioning sys-
tem (GPS) unit. e path for each neighbor-
hood was designed to traverse different areas,
including main highways/roads, local roads,
residential alleys and foot paths, and markets.
e paths ranged 7.7–9.4 km in length, and
the tours lasted 4.5–5.5 hr in duration in dif-
ferent neighborhoods (Figure 1). Each path
included approximately 20 predetermined
5-min stops, including one stop at ground
level in front of most fixed-monitoring sites
(Figure 1). At each stop, information on
PM sources within 10–15 m was recorded
through a visual sighting of the source or
presence of smoke using a standardized form
programmed into Palm Z22 handheld units
(Palm, Inc., Sunnyvale, CA, USA). The
information included nearby biomass stoves
(including fuel type and number of stoves),
traffic flow, other combustion sources (e.g.,
trash burning), and road surface.
We also measured 48-hr integrated con-
centrations of PM with aerodynamic diam-
eters ≤ 2.5 µm (PM2.5) and ≤ 10 µm (PM10)
at three roof-top fixed-monitoring sites in
each neighborhood [two in NM because an
earlier pilot study found nearly identical con-
centrations at two of the original three sites
(Arku et al. 2008)]. One fixed-monitoring
site in each neighborhood was located along
a main road, and the others were located in
residential areas (> 100 m from main roads,
although some were along smaller roads). e
monitors were 4–7 m above ground level so
that the air was relatively well mixed and less
likely to be strongly affected by a source in
the immediate vicinity. We also measured
PM2.5 and PM10 continuously at as many
fixed-monitoring sites as possible. Analysis of
between-neighborhood variation using data
from the fixed-monitoring sites are presented
elsewhere (Dionisio et al. 2010).
Measurements were done in April 2007
(before the main rainy season) in AD and in
July–August 2007 (after the main rainy sea-
son) in the other neighborhoods. ere were
no unusual meteorologic factors during the
measurement periods.
PM measurement methods. We used
DustTrak model 8520 monitors (TSI Inc.,
Shoreview, MN, USA) for continuous meas-
urement of PM2.5 and PM10. DustTrak
has an internal laser photometer that uses a
90° light-scattering laser diode to measure
PM concentration in air drawn by an inter-
nal pump. DustTrak monitors were oper-
ated at 1.7 L/min and used TSI-supplied inlet
Figure 1. Map of Accra metropolitan area, study areas, and mobile monitoring path, delimited by census
enumeration areas (EAs). EAs have nearly the same population; hence the area of an EA is inversely
related to population density. Local secondary roads are smaller roads whose traffic is primarily for the
purpose of reaching a local destination; connecting secondary roads are smaller roads used for passing
through the neighborhood.
Nima (NM)
Path length: 7.7 km
Asylum Down (AD)
Path length: 8.9 km
Jamestown/Ushertown (JT)
Path length: 8.5 km
East Legon (EL)
Path length: 9.4 km
Fixed site
5-minute stop
Alley
Local secondary road
Connecting secondary road
Primary road
Divided multi-lane highway
PM spatial patterns and sources in Accra
Environmental Health Perspectives
•
v o l u m e 118 | n u m b e r 5 | May 2010
609
nozzles with a cutoff of 10 µm (aerodynamic
diameter) for PM10 measurement; for PM2.5
measurement, DustTrak monitors were fitted
with an external size-selective inlet containing
a level greased impaction surface and with a
cutoff of 2.5 µm (aerodynamic diameter) and
were operated at 0.8 L/min. The DustTrak
measures PM concentrations every second
and was set to average these measurements
and record at 1-min intervals. e DustTrak
monitors were set to zero against a zero fil-
ter on each measurement day. The factory-
specified resolution of the DustTrak monitor
is ± 0.1% of reading or ± 1 µg/m3, whichever
is greater.
PM measured using light-scattering tech-
nologies is subject to error because DustTrak
photometers are calibrated to aerosols whose
characteristics (e.g., shape, size, density, and
refractive index) may differ from those in our
study and because measured concentration
may be affected by factors such as humidity.
For the same reasons, measurement errors
can vary across days or neighborhoods. To
adjust for DustTrak measurement error, we
corrected continuous (DustTrak) PM con-
centrations by a correction factor (CF) calcu-
lated using the gravimetric data as described
below. PM concentration measured gravi-
metrically has substantially less measurement
error but, by definition, measures the average
concentration for the whole measurement
period. Gravimetric PM measurements are
described in detail elsewhere (Dionisio et al.
2010). In summary, gravimetric PM samples
were collected on a PTFE (polytetrafluoro-
ethylene) filter with ring (Teflo, 0.2 µm pore
size, 37 mm diameter; Pall Life Sciences, Port
Washington, NY, USA), back-supported
by a Whatman drain disc (Whatman Inc.,
Piscataway, NJ). PM10 measurements used a
Harvard impactor (Marple and Willeke 1976;
Marple et al. 1987) with a D50 (50% col-
lection efficiency) of 10 µm (aerodynamic
diameter) at 4 L/min (± 10%), with two con-
secutive pre-oiled impactor plates serving as
the impaction surface. PM2.5 measurements
used a modified Harvard impactor com-
bined with a polyurethane foam (PUF) PM2.5
size-selective inlet, with a D50 of 2.5 µm at
5 L/min (±10%), with a PUF pad serving as
the impaction surface. All PM concentrations
were blank corrected.
The CF was calculated separately for
PM2.5 and PM10 and for each 48-hr measure-
ment period. At each fixed-monitoring site we
used the gravimetric-to-DustTrak ratio as the
CF for the site itself. e CF for DustTrak
monitors used in mobile monitoring was cal-
culated as the geometric mean of the CFs
of the corresponding size fraction for all the
fixed-monitoring sites in that neighborhood
on the measurement day. e mean and 5th
and 95th percentiles of the individual fixed-
site CFs were 0.75, 0.35, and 1.19, respec-
tively; those of the mobile monitor CFs were
0.78, 0.48, and 1.25, respectively. Because
of interruptions of either the gravimetric or
the photometer measurements in JT, we had
only a single CF for this neighborhood. We
conducted the following sensitivity analyses to
ensure that the missing CFs did not affect our
overall conclusions: applying CFs calculated
from the sites in NM, where measurements
took place on the same days; and repeating
all analyses with uncorrected JT fixed-site and
mobile-monitor data, as has been done in at
least one previous study without gravimetric
data (Levy et al. 2001). e results from these
two analyses were within the 95% confidence
interval (CI) of each other, indicating that the
conclusions are robust.
Meteorologic/weather variables. PM con-
centration depends on meteorologic/weather
variables such as humidity, wind speed, and
recent rain. For this reason, we adjusted the
estimated effects of nearby sources on local
PM for these variables. Data on relative
humidity (RH), wind speed, and time since
last rain were from a station near the Kotoka
International Airport and maintained by the
U.S. Department of Commerce National
Oceanic and Atmospheric Administration
(http://www.ncdc.noaa.gov/oa/climate/
stationlocator.html). We predicted RH and
wind speed for hours with missing data using
simple linear models when data were missing
for < 3 hr. When > 3 hr of data were missing,
we used the average of RH for the same hour
during 5 days before and 5 days after the miss-
ing value. We then fit a cubic spline function
to hourly RH and wind speed to obtain RH
values for each minute. RH and wind speed
on days that we conducted mobile monitor-
ing were on average 79% (interquartile range,
72–86%) and 11 miles/hr (interquartile
range, 8–14 miles/hr).
Data management. Using the time
stamps of DustTrak monitors, we compiled
error-corrected continuous PM2.5 and PM10
data from fixed sites and mobile monitors
into a single data set, with each record rep-
resenting a unique date, minute, and loca-
tion. Geographic coordinates for each mobile
monitor data point were also included using
the GPS date/time stamp. Because GPS units
measure true location with error, GPS coordi-
nates were “snapped” to the nearest point on
the monitoring path; the location of the path
was ascertained using a Trimble GeoXT GPS
unit (Trimble Navigation Limited, Sunnyvale,
CA, USA) with a nominal error of < 1 m.
When a point was at or near the intersection
of two roads that were both on the path, the
snapping retained the temporal ordering of
data points. e average change in the posi-
tion of points snapped to the path was about
3 m. Weather variables were also added using
date and time stamps. We merged DustTrak
and Palm date and time stamps and GPS
geographic coordinates to identify the meas-
urements taken at each stop along the moni-
toring path. Measurements were averaged
over the approximately 5 min of recordings
for each stop visit and then merged with the
PM source data collected on the Palm units.
Statistical analysis. We provide graphical
presentation as well as descriptive statistics for
mobile PM measurements. We also analyzed
the stop data, which included simultaneous
data on PM and sources at the same location,
for the effects of specific pollution sources and
meteorologic factors. All analyses used PM
data corrected for DustTrak measurement
error as described above.
We used regression analysis to quan-
tify the effects of sources on PM concentra-
tions at stops. Because the mobile monitors
reached stops at different times, the PM data
from mobile monitors alone cannot deter-
mine if observed variations across stops are
due to sources or due to changes in the overall
neighborhood PM level, which happened to be
lower or higher when the monitor reached a
specific stop. For example, if woodstoves were
disproportionately present at stops visited early
in the day in a specific neighborhood, and if
other factors (e.g., heavy traffic) led to high PM
concentration at those times in that neighbor-
hood, the effects of woodstoves on PM might
be overestimated. To deal with this issue, we
adjusted the local PM-source regressions for the
average neighborhood PM concentration at the
same time; average neighborhood concentra-
tion was estimated as the average of those at all
fixed-monitoring sites (which, by being fixed,
Table 1. Data used for the analysis, by PM size fraction and neighborhood.
Neighborhood
PM size fraction AD EL JT NM
PM2.5
No. of mobile monitoring toursa14 13 12 13
Total no. of stops contributing data to the regression analysis 279 247 288 286
No. of fixed sites contributing data to neighborhood average 3 3 1 2
PM10
No. of mobile monitoring toursa14 14 10 13
Total no. of stops contributing data to the regression analysis 279 265 240 286
No. of fixed sites contributing data to neighborhood average 2 1 1 1
aThere were two tours on most days unless one was postponed or canceled due to rain or equipment malfunction.
Dionisio et al.
610
v o l u m e 118 | n u m b e r 5 | May 2010
•
Environmental Health Perspectives
vary only in time). is approach may, how-
ever, lead to an underestimation of the effects
of sources if the neighborhood average itself
was influenced by that source and others like
it. For example, if the high neighborhood PM
concentrations in early mornings resulted from
the use of woodstoves, which happened to be
present at stops visited at those hours, adjusting
for neighborhood average could underestimate
the role of woodstoves. erefore, we present
our regression results both with and without
adjustment for the neighborhood average. We
estimated the following regression equation:
Ln(PMstop) = β0 + βX + δWeather
[+κ ln(PMfixedsite)] + ε, [1]
where X is a vector of source variables (data
collected using Palm units); Weather is a
vector of weather variables;
β
,
δ
, and
κ
are
regression coefficients; and ε is an error term.
We used a linear mixed effects model with a
random group effect for each neighborhood-
day (Davidian and Giltinan 1995; Laird and
Ware 1982). Neighborhood-day group effect
helps remove the influence of unobserved
factors that affect all measurements in each
neighborhood on the measurement day, for
example, unmeasured weather pattern or phe-
nomena that lead to more or less combustion.
PM concentrations were log-transformed to
ensure that model residuals were normally
distributed.
e five measurements at each stop were
averaged because source data were recorded
once for the 5-min duration of the stop, and
because averaging reduces random error due
to short-term fluctuations in PM. We also
smoothed the fixed-site continuous PM data
to retain salient temporal patterns and remove
minute-to-minute stochastic noise, which is
likely to be highly local. We used a nonpara-
metric regression [locally weighted scatterplot
smoothing (LOWESS) regression] for smooth-
ing, with a 60-min bounding radius, which
tends to eliminate perturbations sustained for
< 10 min but maintain patterns lasting more
than 30 min (Cleveland et al. 1992).
All analyses were done separately for PM2.5
and PM10 using the open-source statistical
Figure 2. Concentrations of PM2.5 and PM10 along the walking paths in the study neighborhoods. For each neighborhood and PM size fraction, data from all moni-
toring days/tours were combined in a moving average, with a 50-m averaging interval.
Distance (m)
0
200
400
600
800
1,000
1,200
PM2.5
PM10
Distance (m)
0
200
400
600
800
1,000
1,200
1,400
1,600
1,800
Distance (m)
0
200
400
600
800
1,000
Distance (m)
0
200
400
600
800
Asylum Down (AD) East Legon (EL) Jamestown (JT) Nima (NM)
PM2.5 (µg/m3)
< 10
10 − 20
20 − 30
30 − 40
40 − 50
50 − 60
60 − 70
70 − 80
80 − 90
90 − 100
100 − 150
150 − 200
> 200
PM10 (µg/m3)
< 20
20 − 40
40 − 60
60 − 80
80 − 100
100 − 120
120 − 140
140 − 160
160 − 180
180 − 200
200 − 300
300 − 400
> 400
Figure 3. Crude associations of residual PM, defined as the difference between PM measured at stops
and the neighborhood average, with sources present at the stop. In each plot, the middle line shows the
median, and the bottom and top of the box show the 25th and 75th percentiles of data. Numbers indicate
the number of stops contributing to each box plot.
−100
0
100
200
300
PM2.5
PM10
Stoves Traffic Road surface Burning trash
−100
0
100
200
300
827
117 65 27
64
412
13
254 178
160 44
76
357
73
594 1,042 12
816 111
60 23
51
383
13
256 174
154 42
72
343 60 586 1,004
11
No stove
Single charcoal
Multiple charcoal
Single wood
Multiple wood
No traffic
Stopped vehicle
< 2 cars/min
2−10 cars/min
Heavy moving
Heavy congested
Loose dirt
Packed dirt
Paved broken
Paved
Not present
Present
PM spatial patterns and sources in Accra
Environmental Health Perspectives
•
v o l u m e 118 | n u m b e r 5 | May 2010
611
analysis package R, version 2.6.1 (R Project
for Statistical Computing, Vienna, Austria).
Results
Figure 2 shows the gravimetric-corrected
concentrations of PM2.5 and PM10 along the
walking path, averaged over all monitoring
days/tours. In our measurement campaign,
the EL walking path had the lowest levels of
PM and the JT walking path the highest, with
geometric means of PM2.5 and PM10 of 21
and 49 µg/m3, respectively, along the EL path
and 39 and 96 µg/m3, respectively, along the
JT path. In fact, the less polluted segments
of the JT walking path had PM2.5 and PM10
values that were similar to the average for all
of EL. AD and NM walking paths had PM
pollution levels that fell between the other
two neighborhoods, with geometric means of
PM2.5 and PM10 of 35 and 86 µg/m3 for AD
and 41 and 58 µg/m3 for NM. In AD and
NM, pollution was highest along the largest
roads/highways. Our observations during data
collection indicate that the primary pollu-
tion source along the main highway in AD
was traffic (cars, minibuses, and trucks) and
in NM a combination of traffic and roadside
biomass use.
Figure 3 shows the crude associations of
nearby sources with residual PM, defined as
the difference between PM measured during
5-min stops and the neighborhood average
in the same 5 min. PM2.5 and PM10 meas-
urements at stops with multiple woodstoves
were, respectively, 30 µg/m3 and 85 µg/m3
higher than the neighborhood average at the
same time (median residual); the residual PM2.5
and PM10 were smaller for stops that had one
woodstove (8 µg/m3 and 32 µg/m3) or one
or more charcoal stoves. When a stop had no
stoves, residual PM2.5 and PM10 were only
0 µg/m3 and 14 µg/m3. Similarly, we gener-
ally found a gradient of residual PM pollution
with increasing local traffic density. Residual
PM2.5 and PM10 at stops near congested traf-
fic were, respectively, 12 and 46 µg/m3 greater
than the same metric for stops near light traffic
(< 2 cars/min). However, residual PM2.5 (but
not PM10) at stops with no traffic was higher
than at stops with light and medium traffic
(see below).
The results in Figure 3 show only crude
associations; that is, they do not consider the
possibility that some sources may be more/less
likely to co-occur at the same place. For exam-
ple, it may be the case that stops with no traf-
fic had higher PM2.5 because charcoal or wood
stoves were present near them. Table 2 shows
the adjusted association of PM with sources
and demonstrates a number of features of local
PM pollution in these neighborhoods. First,
adjusting for average neighborhood pollution
had some, but limited, influence on either the
magnitude of the effects of individual sources
or their statistical significance. Second, the
presence of multiple woodstoves (Figure 4)
had the unequivocal largest effect on nearby
PM2.5 and PM10 concentrations. In the log-
transformed model, multiple nearby wood-
stoves would be associated with nearly three
times (297%; 95% CI, 247–357%) higher
PM2.5 levels and more than two times (227%;
95% CI, 189–272%) higher PM10 levels after
adjustment for all other source and meteo-
rologic variables (262% and 197% higher
if neighborhood averages were also adjusted
for). e next most important determinants
of local pollution were the presence of a single
woodstove or multiple charcoal stoves, heavy/
congested traffic, having loose dirt road sur-
face, and trash burning (Figure 4). e coef-
ficients of trash burning were significant only
Table 2. Regression coefficients (95% CIs) for multivariate analysis of the association of PM with sources,
road surface, and meteorologic covariates.
Model 1 Model 2
Variable Coefficient p-Value Coefficient p-Value
Dependent variable: ln(PM2.5)
Constant –0.169 (–0.846 to 0.508) 0.63 0.897 (0.190 to 1.603) 0.013
Ln(neighborhood average) 0.522 (0.422 to 0.623) < 0.001
Distance to nearest main road (km) –0.061 (–0.342 to 0.221) 0.67 0.044 (–0.244 to 0.332) 0.764
Trash burning 0.414 (0.082 to 0.746) 0.02 0.465 (0.118 to 0.812) 0.009
Traffic flow
No traffic 0.0 NA 0.0 NA
Idling vehicle –0.171 (–0.510 to 0.168) 0.32 –0.214 (–0.568 to 0.141) 0.24
Light (< 2 cars/min) 0.095 (–0.012 to 0.202) 0.08 0.097 (–0.014 to 0.209) 0.09
Medium (< 10 cars/min) 0.181 (0.059 to 0.302) 0.004 0.174 (0.047 to 0.300) 0.007
Heavy moving 0.326 (0.185 to 0.468) < 0.001 0.339 (0.192 to 0.487) < 0.001
Congested/stopped heavy traffic 0.466 (0.260 to 0.673) < 0.001 0.496 (0.280 to 0.711) < 0.001
Biomass stoves
No stove 0.0 NA 0.0 NA
Single charcoal stove 0.153 (0.038 to 0.269) 0.009 0.155 (0.035 to 0.275) 0.01
Multiple charcoal stoves 0.294 (0.144 to 0.445) < 0.001 0.313 (0.156 to 0.469) < 0.001
Single wood stove 0.373 (0.134 to 0.613) 0.002 0.365 (0.116 to 0.614) 0.004
Multiple wood stoves 0.962 (0.783 to 1.142) < 0.001 1.089 (0.905 to 1.273) < 0.001
Road surface
Paved 0.0 NA 0.0 NA
Paved broken 0.119 (–0.054 to 0.292) 0.18 0.161 (–0.019 to 0.342) 0.08
Packed dirt 0.009 (–0.095 to 0.114) 0.86 0.029 (–0.080 to 0.138) 0.60
Loose dirt 0.337 (0.175 to 0.498) < 0.001 0.384 (0.216 to 0.552) < 0.001
Meteorologic factors
Wind speed (miles/hr) –0.013 (–0.027 to 0.002) 0.09 –0.043 (–0.057 to –0.029) < 0.001
RH 2.216 (1.600 to 2.832) < 0.001 3.415 (2.806 to 4.024) < 0.001
Ln(hours since rain + 1) 0.049 (–0.046 to 0.145) 0.31 0.089 (–0.023 to 0.201) 0.12
Dependent variable: ln(PM10)
Constant 0.456 (–0.101 to 1.012) 0.11 1.582 (0.917 to 2.248) < 0.001
ln(neighborhood average) 0.531 (0.455 to 0.606) < 0.001
Distance to nearest main road –0.248 (–0.485 to –0.010) 0.04 –0.206 (–0.466 to 0.054) 0.12
Trash burning 0.123 (–0.184 to 0.429) 0.43 0.189 (–0.133 to 0.511) 0.25
Traffic flow
No traffic or stopped vehicle 0.0 NA 0.0 NA
Idling vehicle –0.084 (–0.383 to 0.216) 0.58 –0.126 (–0.442 to 0.190) 0.44
Light (< 2 cars/min) 0.110 (0.014 to 0.206) 0.03 0.112 (0.011 to 0.213) 0.03
Medium (< 10 cars/min) 0.247 (0.138 to 0.357) < 0.001 0.247 (0.131 to 0.362) < 0.001
Heavy moving 0.370 (0.242 to 0.498) < 0.001 0.383 (0.248 to 0.518) < 0.001
Congested/stopped heavy traffic 0.537 (0.350 to 0.724) < 0.001 0.528 (0.331 to 0.725) < 0.001
Biomass stoves
No stove 0.0 NA 0.0 NA
Single charcoal stove 0.116 (0.012 to 0.220) 0.03 0.104 (–0.006 to 0.214) 0.06
Multiple charcoal stoves 0.225 (0.088 to 0.363) 0.001 0.243 (0.099 to 0.387) 0.001
Single wood stove 0.277 (0.051 to 0.504) 0.02 0.287 (0.049 to 0.524) 0.02
Multiple wood stoves 0.677 (0.505 to 0.849) < 0.001 0.818 (0.638 to 0.999) < 0.001
Road surface
Paved 0.0 NA 0.0 NA
Paved broken 0.213 (0.049 to 0.377) 0.011 0.243 (0.072 to 0.414) 0.005
Packed dirt 0.062 (–0.031 to 0.156) 0.19 0.036 (–0.063 to 0.135) 0.47
Loose dirt 0.223 (0.078 to 0.367) 0.003 0.264 (0.111 to 0.416) 0.001
Meteorologic factors
Wind speed (miles/hr) –0.011 (–0.024 to 0.003) 0.12 –0.036 (–0.050 to –0.023) < 0.001
RH 1.590 (1.074 to 2.105) < 0.001 3.117 (2.560 to 3.674) < 0.001
Ln(hours since rain + 1) 0.132 (0.075 to 0.190) < 0.001 0.150 (0.043 to 0.256) 0.006
NA, not applicable. Model 1 is adjusted for neighborhood average (estimated as average of smoothed concentrations at
all fixed sites) and Model 2 is without this variable. See “Materials and Methods” for details.
Dionisio et al.
612
v o l u m e 118 | n u m b e r 5 | May 2010
•
Environmental Health Perspectives
for PM2.5, possibly because this source was
present at fewer stops than were other com-
bustion sources. The coefficients of nearby
stoves were ordered with larger effects from
woodstoves than from charcoal stoves and
larger effects from multiple stoves than from
single ones. Further, for each stove category,
the coefficients of the log-transformed regres-
sion were larger for PM2.5 than for PM10,
that is, larger proportional effects on PM2.5.
The coefficients of nearby traffic flow also
rose monotonically with apparently compara-
ble proportional effects on PM2.5 and PM10.
Stops where road surface was loose dirt had
significantly higher PM2.5 and PM10 concen-
trations, and those with broken paved surface
had higher PM10 concentrations after adjust-
ment for combustion sources; PM did not
seem to vary with other road surface materi-
als. e slightly larger proportional effects of
loose dirt road surface on PM2.5 compared
with PM10 is unexplained and may be attrib-
utable to the presence of unrecorded sources
(e.g., stoves inside homes that were not visible
to us). After controlling for traffic and other
local sources at stops, the coefficient of dis-
tance from the main road was nonsignificant
in most models.
Discussion
is study provides one of the first systematic
measurements showing how PM pollution
varies within neighborhoods of varying SES
in a developing country city and the role of
specific combustion sources in local pollution
patterns. Our results showed significant spa-
tial variability in PM concentrations within a
small geographic area in these neighborhoods
(each ~ 1–2 km diameter). In our measure-
ment campaign, the walking path in the lower
SES neighborhood of JT had the highest pol-
lution, followed by segments of the path along
the primary road in NM and the Ring Road
Central in AD. PM2.5 and PM10 were as high
as 200 and 400 µg/m3, respectively, in some
segments of the path.
Combinations of stationary and mobile
measurements have been used to investigate
variations in air pollution levels in relation to
important sources in high-income countries
(Larson et al. 2007; Levy et al. 2001; Su et al.
2007; Weijers et al. 2004). Our study is an
innovative application of such a design by
conducting measurements in a developing
country city, choosing multiple neighbor-
hoods with varied SES, and assessing the role
of sources. Prior studies of local PM pollution
have used different metrics (e.g., particle
count vs. particle mass; fine vs. ultrafine par-
ticle mass). erefore, our results can be com-
pared only with selected other studies that
measured PM2.5 and PM10 in urban micro-
environments. This comparison shows that
during this campaign PM pollution along pri-
mary roads was comparable to or higher than
the most polluted urban microenvironments,
for example, in buses and trolleys and near
bus stations (Levy et al. 2000, 2001), and
substantially higher than those in wood burn-
ing areas of the Pacific Northwest or roadside
sites in European cities (Harrison et al. 2004;
Hoek et al. 2002; Su et al. 2007). We could
not locate other studies of small-area spatial
variability and sources in developing country
cities for direct comparison. More broadly,
studies in Nairobi, Kenya, and Bolpur, India,
found higher PM along major traffic routes
than in nontraffic areas (Padhi and Padhy
2008; van Vliet and Kinney 2007), but these
studies did not examine the presence of non-
transportation combustion sources; a study
at multiple sites in Addis Ababa, Ethiopia,
also found spatial variation in short-term PM
measurements but did not collect data on
nearby sources (Etyemezian et al. 2005).
Figure 4. Woodstoves (A), charcoal stoves (B), trash burning (C), and congested traffic (D) in study neighborhoods.
PM spatial patterns and sources in Accra
Environmental Health Perspectives
•
v o l u m e 118 | n u m b e r 5 | May 2010
613
Our study has a number of innovations
and strengths. We used a combination of geo-
referenced pollution and source data from
mobile monitoring to investigate both the
within-neighborhood spatial patterns of PM2.5
and PM10 pollution and the effects of nearby
sources on local pollution. e data were from
four neighborhoods that covered the full range
of community SES in Accra. We used a com-
bination of fixed-site and mobile-monitor con-
tinuous data to account for the background
temporal pattern of air pollution that may con-
found the data from mobile monitors. Finally,
we used integrated PM measurement to correct
for the measurement error of continuous data
measured with DustTrak monitors.
The data used in this study also have a
number of limitations. First, data were collected
during about 1 week in each neighborhood.
e measurements in three neighborhoods (JT,
NM, and EL) were conducted within a few
weeks, and in the fourth (AD) a few months
prior. Although there were no unusual meteoro-
logic factors during data collection, it would be
ideal to have multiple measurement campaigns
in each neighborhood, in different seasons.
Because of lack of data from different seasons,
our results should not be used to estimate the
usual or average pollution in these neighbor-
hoods. However, our analysis of the effects of
sources on local PM are unlikely to be affected
by macro-level PM changes because we adjusted
for average neighborhood pollution from fixed
sites and used a mixed-effects model with
neighborhood-day group effect. Second, PM
measured with DustTrak monitors is subject to
error. Although we systematically applied a CF
to PM data, the steps involved in calculating
CFs introduce additional uncertainty. Third,
using mobile monitoring alone did not allow
us to separate temporal and spatial changes in
pollution. We relied on continuous PM data
at fixed sites to adjust for temporal changes in
neighborhood PM. If low-cost and low-power
PM monitors were available, it would be ideal
to have a large number of stationary monitors
in the neighborhood instead of mobile ones.
Conclusions
We found that, after adjusting for other fac-
tors, the factors wood and charcoal stoves,
congested and heavy traffic, and trash burn-
ing had large and significant effects on local
PM pollution in these Accra neighborhoods.
Biomass fuels are a source of energy for house-
holds and small commercial purposes in urban
sub-Saharan Africa, especially in low-income
and marginalized neighborhoods (Bailis et al.
2005; Barnes et al. 2005); older vehicles are
also common in sub-Saharan African cities.
If other studies in Accra and other developing
country cities show that the effects of these
common sources on local pollution observed
in our measurement campaign are typical
of their usual contributions, there is need to
identify and implement effective and equitable
transportation regulations and policies that
reduce the impacts of traffic pollution, and
technological and policy innovations that can
reduce air pollution from biomass fuels with-
out restricting what may be the only energy
source available to poor households.
Re f e R e n c e s
Agyei-Mensah S, Owusu G. 2009. Segregated by neighbour-
hoods? A portrait of ethnic diversity in the neighbourhoods
of the Accra metropolitan area, Ghana. Popul Sp ace
Place; doi: 10.1002/psp.551 [Online 6 May 2009].
Arku RE, Vallarino J, Dionisio KL, Willis R, Choi H, Wilson JG,
et al. 2008. Characterizing air pollution in two low-income
neighborhoods in Accra, Ghana. Sci Total Environ 402(2–
3):217–231.
Bailis R, Ezzati M, Kammen DM. 2005. Mortality and green-
house gas impacts of biomass and petroleum energy
futures in Africa. Science 308(5718):98–103.
Barnes DF, Krutilla K, Hyde WF. 2005. The Urban Household
Energy Transition: Social and Environmental Impacts in
the Developing World. Washington, DC:RFF Press.
Buzzelli M, Jerrett M. 2004. Racial gradients of ambient air
pollution exposure in Hamilton, Canada. Environ Plann A
36(10):1855–1876.
Charron A, Harrison RM. 2005. Fine (PM2.5) and coarse
(PM2.5–10) particulate matter on a heavily trafficked London
highway: sources and processes. Environ Sci Technol
39(20):7768–7776.
Chowdhury MZ. 2004. Characterization of Fine Particle Air
Pollution in the Indian Subcontinent. Atlanta, GA:Georgia
Institute of Technology.
Cleveland WS, Grosse E, Shyu WM. 1992. Local regression
models. In: Statistical Models in S (Chambers JM, Hastie
TJ, eds). Pacific Grove, CA:Wadsworth, 309–376.
Cohen A, Anderson R, Ostro B, Pandey KD, Krzyzanowski M,
Künzli N, et al. 2004. Urban ambient air pollution. In:
Comparative Quantification of Health Risks: Global and
Regional Burden of Disease Attributable to Selected Major
Risk Factors (Ezzati M, Lopez AD, Rodgers A, Murray CJL,
eds). Geneva:World Health Organization, 1353–1433.
Davidian M, Giltinan D. 1995. Nonlinear Mixed Effects Models for
Repeated Measurement Data. London:Chapman and Hall.
Dionisio KL, Arku RE, Hughes AF, Vallarino J, Carmichael H,
Spengler JD, et al. 2010. Air pollution in Accra neighbor-
hoods: spatial, socioeconomic, and temporal patterns.
Environ Sci Technol 44(7):2270–2276.
Engelbrecht JP, Swanepoel L, Chow JC, Watson JG, Egami RT.
2001. PM2.5 and PM10 concentrations from the Qalabotjha
low-smoke fuels macro-scale experiment in South Africa.
Environ Monit Assess 69(1):1–15.
Etyemezian V, Tesfaye M, Yimer A, Chow JC, Mesfin D, Nega T,
et al. 2005. Results from a pilot-scale air quality study in
Addis Ababa, Ethiopia. Atmos Environ 39(40):7849–7860.
Ezzati M, Lopez AD, Rodgers A, Vander Hoorn S, Murray CJ.
2002. Selected major risk factors and global and regional
burden of disease. Lancet 360(9343):1347–1360.
Harrison RM, Jones AM, Lawrence RG. 2004. Major component
composition of PM10 and PM2.5 from roadside and urban
background sites. Atmos Environ 38:4531–4538.
Hoek G, Fischer P, Van Den Brandt P, Goldbohm S,
Brunekreef B. 2001. Estimation of long-term average
exposure to outdoor air pollution for a cohort study on
mortality. J Expo Anal Environ Epidemiol 11(6):459–469.
Hoek G, Meliefste K, Cyrys J, Lewne C, Bellander T, Brauer M,
et al. 2002. Spatial variability of fine particle concentrations
in three European areas. Atmos Environ 36:4077–4088.
Holmes NS, Morawska L, Mengersen K, Jayaratne ER. 2005.
Spatial distribution of submicrometre particles and CO
in an urban microscale environment. Atmos Environ
39(22):3977–3988.
Jackson MM. 2005. Roadside concentration of gaseous and
particulate matter pollutants and risk assessment in Dar-es-
Salaam, Tanzania. Environ Monit Assess 104(1–3):385–407.
Kinney PL, O’Neill MS. 2006. Environmental Equity. In: Air
Quality Guidelines: Global Update 2005 (World Health
Organization, ed). Rheinbach, Germany:Druckpartner
Moser, 135–152.
Laird NM, Ware JH. 1982. Random-effects models for longitudi-
nal data. Biometrics 38(4):963–974.
Larson T, Su J, Baribeau AM, Buzzelli M, Setton E, Brauer M.
2007. A spatial model of urban winter woodsmoke concen-
trations. Environ Sci Technol 41(7):2429–2436.
Levy JI, Houseman EA, Ryan L, Richardson D, Spengler JD.
2000. Particle concentrations in urban microenvironments.
Environ Health Perspect 108:1051–1057.
Levy JI, Houseman EA, Spengler JD, Loh P, Ryan L. 2001. Fine
particulate matter and polycyclic aromatic hydrocarbon con-
centration patterns in Roxbury, Massachusetts: a community-
based GIS analysis. Environ Health Perspect 109:341–347.
Loh P, Sugerman-Brozan J, Wiggins S, Noiles D, Archibald C. 2002.
From asthma to AirBeat: community-driven monitoring of
fine particles and black carbon in Roxbury, Massachusetts.
Environ Health Perspect 110(suppl 2):297–301.
Marple VA, Rubow KL, Turner W, Spengler JD. 1987. Low flow
rate sharp cut impactors for indoor air sampling: design
and calibration. JAPCA 37(11):1303–1307.
Marple VA, Willeke K. 1976. Impactors design. Atmos Environ
10:891–896.
O’Neill MS, Jerrett M, Kawachi I, Levy JI, Cohen AJ, Gouveia N,
et al. 2003. Health, wealth and air pollution: advancing the-
ory and methods. Environ Health Perspect 111:1861–1870.
Padhi BK, Padhy PK. 2008. Assessment of intra-urban variabil-
ity in outdoor air quality and its health risks. Inhal Toxicol
20(11):973–979.
Saksena S, Singh PB, Prasad RK, Prasad R, Malhotra P,
Joshi V, et al. 2003. Exposure of infants to outdoor and
indoor air pollution in low-income urban areas—a case
study of Delhi. J Expo Anal Environ Epidemiol 13(3):219.
Sclar ED, Garau P, Carolini G. 2005. The 21st century health
challenge of slums and cities. Lancet 365(9462):901–903.
Songsore J, McGranahan G. 1998. The political economy of
household environmental management: gender, environ-
ment and epidemiology. World Dev 26(3):395–412.
Su JG, Brauer M, Ainslie B, Steyn D, Larson T, Buzzelli M.
2008. An innovative land use regression model incorporat-
ing meteorology for exposure analysis. Sci Total Environ
390(2–3):520–529.
Su JG, Larson T, Baribeau AM, Brauer M, Rensing M,
Buzzelli M. 2007. Spatial modeling for air pollution monitor-
ing network design: example of residential woodsmoke.
J Air Waste Manag Assoc 57(8):893–900.
United Nations Department of Economic and Social Affairs
(Population Division). 2004. World Urbanization Prospects:
The 2003 Revision. New York:Population Division, Department
of Economic and Social Affairs, United Nations Secretariat.
van Vliet E, Kinney P. 2007. Impacts of roadway emissions on
urban particulate matter concentrations in sub- Saharan
Africa: new evidence from Nairobi, Kenya. Environ Res Lett
2(4); doi:10.1088/1748-9326/2/4/045028 [Online 21 December
2007].
Weijers EP, Khlystov AY, Kos GPA, Erisman JW. 2004. Variability
of particulate matter concentrations along roads and
motorways determined by a moving measurement unit.
Atmos Environ 38(19):2993–3002.
Zheng M, Salmon LG, Schauer JJ, Zeng L, Kiang CS, Zhang Y,
et al. 2005. Seasonal trends in PM2.5 source contributions
in Beijing, China. Atmos Environ 39(22):3967–3976.
