An assessment of air pollution and its attributable mortality in Ulaanbaatar, Mongolia

Faculty of Health Sciences, Simon Fraser University, 8888 University Drive, Burnaby, BC V5A 1S6 Canada.
Air Quality Atmosphere & Health (Impact Factor: 1.8). 03/2013; 6(1):137-150. DOI: 10.1007/s11869-011-0154-3
Source: PubMed
ABSTRACT
Epidemiologic studies have consistently reported associations between outdoor fine particulate matter (PM2.5) air pollution and adverse health effects. Although Asia bears the majority of the public health burden from air pollution, few epidemiologic studies have been conducted outside of North America and Europe due in part to challenges in population exposure assessment. We assessed the feasibility of two current exposure assessment techniques, land use regression (LUR) modeling and mobile monitoring, and estimated the mortality attributable to air pollution in Ulaanbaatar, Mongolia. We developed LUR models for predicting wintertime spatial patterns of NO2 and SO2 based on 2-week passive Ogawa measurements at 37 locations and freely available geographic predictors. The models explained 74% and 78% of the variance in NO2 and SO2, respectively. Land cover characteristics derived from satellite images were useful predictors of both pollutants. Mobile PM2.5 monitoring with an integrating nephelometer also showed promise, capturing substantial spatial variation in PM2.5 concentrations. The spatial patterns in SO2 and PM, seasonal and diurnal patterns in PM2.5, and high wintertime PM2.5/PM10 ratios were consistent with a major impact from coal and wood combustion in the city's low-income traditional housing (ger) areas. The annual average concentration of PM2.5 measured at a centrally located government monitoring site was 75 μg/m3 or more than seven times the World Health Organization's PM2.5 air quality guideline, driven by a wintertime average concentration of 148 μg/m3. PM2.5 concentrations measured in a traditional housing area were higher, with a wintertime mean PM2.5 concentration of 250 μg/m3. We conservatively estimated that 29% (95% CI, 12-43%) of cardiopulmonary deaths and 40% (95% CI, 17-56%) of lung cancer deaths in the city are attributable to outdoor air pollution. These deaths correspond to nearly 10% of the city's total mortality, with estimates ranging to more than 13% of mortality under less conservative model assumptions. LUR models and mobile monitoring can be successfully implemented in developing country cities, thus cost-effectively improving exposure assessment for epidemiology and risk assessment. Air pollution represents a major threat to public health in Ulaanbaatar, Mongolia, and reducing home heating emissions in traditional housing areas should be the primary focus of air pollution control efforts.

Full-text

Available from: Tim K Takaro
An assessment of air pollution and its attributable mortality
in Ulaanbaatar, Mongolia
Ryan W. Allen & Enkhjargal Gombojav & Baldorj Barkhasragch aa &
Tsogtbaatar Byambaa & Oyuntogos Lkhasuren & Ofer Amram & Tim K. Takaro &
Craig R. Janes
Received: 21 January 2011 /Accepted: 20 July 2011
#
The Author(s) 2011. This article is published with open access at Springerlink.com
Abstract Epidemiologic studies have consistently reported
associations between outdoor fine particulate matter (PM
2.5
)
air pollution and adverse health effects. Although Asia bears
the majority of the public health burden from air pollution,
few epidemiologic studies have been conducted outside of
North America and Europe due in part to challenges in
population exposure assessment. We assessed the feasibility
of two current exposure assessment techniques, land use
regression (LUR) modeling and mobile monitoring, and
estima ted the mortality attributab le to air pollution in
Ulaanbaatar, Mongolia. We developed LUR models for
predicting wintertime spatial patterns of NO
2
and SO
2
based
on 2-week passive Ogawa measurements at 37 locations and
freely available geographic predictors. The models explained
74% and 78% of the variance in NO
2
and SO
2
,respectively.
Land cover characteristics derived from satellite images were
useful predictors of both pollutants. Mobile PM
2.5
monitor-
ing with an integrating nephelometer also showed promise,
capturing substantial spatial variation in PM
2.5
concentra-
tions. The spatial patterns in SO
2
and PM, seasonal and
diurnal patterns in PM
2.5
, and high wintertime PM
2.5
/PM
10
ratios were consistent with a major impact from coal and
wood combustion in the citys low-income traditional
housing (ger) areas. The annual average concentration of
PM
2.5
measured at a centrally located government monitor-
ing site was 75 μg/m
3
or more than seven times the World
Health OrganizationsPM
2.5
air quality guideline, driven by
a wintertime average concentration of 148 μg/m
3
.PM
2.5
concentrations measured in a traditional housing area were
higher, with a wintertime mean PM
2.5
concentration of
250 μg/m
3
. We conservatively estimated that 29% (95% CI,
1243%) of cardiopulmonary deaths and 40% (95% CI, 17
56%) of lung cancer deaths in the city are attributable to
outdoor air pollution. These deaths correspond to nearly 10%
of the citys total mortality, with estimates ranging to more
than 13% of mortality under less conservative model
assumptions. LUR models and mobile monitoring can be
successfully implemented in developing country cities, thus
cost-effectively improving exposure assessment for epidemi-
ology and risk assessment. Air pollution represents a major
threa t to public health in Ul aanbaatar, Mongolia, and
reducing home heating emissions in traditional housing areas
should be the primary focus of air pollution control efforts.
Keywords Satellite
.
Exposure
.
Nephelometer
.
Asia
.
Impact assessment
.
Coal
.
Combustion
Introduction
Epidemiologic studies have consistently reported associa-
tions between exposure to air pollution, including particu-
R. W. Allen (*)
:
T. Byambaa
:
T. K. Takaro
:
C. R. Janes
Faculty of Health Sciences, Simon Fraser University,
8888 University Drive,
Burnaby, BC V5A 1S6, Canada
e-mail: allenr@sfu.ca
E. Gombojav
School of Public Health, Health Sciences University of Mongolia,
Ulaanbaatar, Mongolia
B. Barkhasragchaa
National Agency for Meteorology and Environmental Monitoring,
Ulaanbaatar, Mongolia
O. Lkhasuren
World Health Organization,
Ulaanbaatar, Mongolia
O. Amram
Department of Geography, Simon Fraser University,
Burnaby, Canada
Air Qual Atmos Health
DOI 10.1007/s11869-011-0154-3
Page 1
late matter (PM), and adverse health effects (Pope and
Dockery 2006; HEI 2010; Brook et al. 2010). There is
evidence that fine particulate matter (PM
2.5
; PM with an
aerodynamic diameter 2.5 μm and smaller) generated by
combustion may be especially damaging to human health
(Pope and Doc kery 2006;Schlesingeretal.2 006).
Estimates of the annual global mortality attributable to
outdoor air pollution range from 0.8 million to over 4
million, with the majority of attributable deaths occurring in
Asia (Cohen et al. 2005; Anenberg et al. 2010). However,
due in part to challenges in population exposure assess-
ment, relatively few epidemiologic studies have been
conducted outside of North America and Europe (HEI
2004).
Mongolias population has undergone rapid urbanization
since the mid-1990s, and this shift has had a major impact on
the capital city, Ulaanbaatar, which is now home to 1.1 1 million
of the nations 2.74 million inhabitants (National Statistical
Of fice 2010). Population growth has led to major increases in
the citys air pollution emissions (Asian Development Bank
2006; Guttikunda 2007). Much of the population growth has
been in the citys low-income ger (traditional Mongolian
dwelling) areas where coal and wood are burned for heat
(World Bank 2004). Half of Ulaanbaatars population lives in a
ger (Asian Development Bank 2006), and the citys 160,000
gers each burn an average of 5 t of coal and 3 m
3
of wood per
year (Guttikunda 2007). Mobile sources also contribute to air
pollution in Ulaanbaatar. From 1995 to 2005, the number of
vehicles in Ulaanbaatar increased from 30,000 to 75,000 (Asian
Development Bank 2006), and Mongolia is 1 of only 17
countries where leaded gasoline is still legally available (HEI
2010). The citys other major air pollution sources include three
coal-fueled combined heat and power plants, approximately
400 heat-only boilers, and wind-blown dust (World Bank 2004;
Davy et al. 2011). A r ecent source apportionment study found
that the majority of PM
2.5
in Ulaanbaatar is produced by coal
combustion (Davy et al. 2011).
Ulaanbaatar is located in a valley with mountains to the
northandsouth(AsianDevelopmentBank2006;Davyetal.
2011). The topography, extensive pollution emissions, and
frequent temperature inversions combine to cause very high
pollution concentrations in the winter months. Given the
limited applications of current exposure assessment techni-
ques in developing countries and the limited data on
Ulaanbaatars air quality in the literature (Davy et al. 2011),
our objectives were to (1) characterize air pollution concen-
trations and temporal patterns; (2) assess the feasibility of
using two current exposure assessment techniques, land use
regression (LUR) modeling and mobile monitoring, in a
developing country; (3) characterize spatial patterns in
pollutant indicators of specific sources to identify hot spots
and create exposure assessment tools; and (4) estimate the
mortality attributable to outdoor air pollution in Ulaanbaatar.
Methods
Fixed-site monitoring
The government air pollution monitoring network in
Ulaanbaatar has improved considerabl y in recent years.
PM
2.5
, the criteria pollutant most relevant to human health,
is now routinely monitored using tapered element oscillat-
ing microbalances (TEOMs) at four of the nine government
monitoring sites in the city (the locations of the government
monitoring sites are shown in Fig. 1). Air pollution data
from June 1, 2009 to May 31, 2010 were obtained from the
Ulaanbaatar City Environmental Monitoring Agencys four
PM monitoring sites (Fig. 1). Our main focus for analysis
was site #1 because (1) it had the most complete data record
for the period of interest (Table 1)and(2)itiscentrally
located and m ay, therefore, be the single site most
representative of overall population exposure in Ulaanbaatar
(Fig. 1). The PM
10
,PM
2.5
,andSO
2
data from this site
were used to characterize diurnal and seasonal patterns in
pollution concentrations and to estimate the annual
average PM
2.5
concentration for use i n the attributable
mortality calculation described below. SO
2
concentrations
are reported in units of micrograms per cubic meter, but
are also reported in parts per billion (assuming 0°C) for
comparison with Ogawa passive sampler measurements.
PM
2.5
data from the citys other three monitoring sites
were only 5077% complete (Table 1). To approximate the
long-term PM
2.5
concentrations at these sites, we replaced
missing observations with the site- and season-specific
median concentrations.
Land use regression modeling
LUR models have become a very common exposure
assessment tool in wealthy countries. These models are
developed based on relatively spatially dense monitoring of
one or more pollutants and the road configuration,
population density, land use, elevation, and other geograph-
ic characteristics surrounding the measurement sites (Hoek
et al. 2008). Based on the empirical relationship between
concentrations and predictors at the measurement locations,
it is possible to predict concentrations at unmeasured
locations (Hoek et al. 2008). To assess the wintertime
spatial patterns in traffic-related air pollution and pollution
produced by coal burning, we measured 14-day average
concentrations of NO
2
and SO
2
, respectively, using two-
sided passive Ogawa samplers attached to power poles,
telephone poles, etc., at approximately 3 m above ground
level. The samplers were deployed at 38 locations in
Ulaanbaatar on February 24 and 25, 2010 and retrieved in
the same order on March 10 and 11, 2010 (Fig. 1). Three of
the 38 sites were colocated with government monitoring
Air Qual Atmos Health
Page 2
sites, and the remaining locations were selected based on
local knowledge to cover the study area and capture a wide
range of NO
2
and SO
2
concentrations (Fig. 1). After
retrieval, the Ogawa samplers were shipped to Vancouver,
Canada and analyzed by ion chromatography at the
University of British Columbia School of Environmental
Health laboratory. Based on four field blanks, we deter-
mined the limits of detection (LOD, calculated as three
times the standard deviat ion of field blanks) for NO
2
and
SO
2
to be 0.8 and 2.5 ppb, respectively.
Given the relative lack of available geographic informa-
tion systems (GIS) data in Ulaanbaatar, we derived data
from several sources. A digital elevation model (DEM)
produced by Environmental Systems Research Institute
(Redlands, CA, USA) and provided with ArcGIS 9.3 was
used to calculate elevations. Land use data, which are
commonly used as predictors in LUR models (Hoek et al.
2008), were not avail able for Ulaanbaatar. Therefore, to
obtain information on land cover, we used Landsat
Enhanced Thematic Mapper Plus (ETM+) satellite images
( http://www.landcover.org), which have been used in
previous LUR modeling efforts (Su et al. 2008a, 2009).
The ETM+ images for Ulaanbaatar (path 131, row 27) were
captured on August 13, 2006 and orthorectified by the
United States Geological Survey. Using a tasseled cap
transformation (Crist and Cicone 1984), ETM+ bands 15
Months Site 1 (%) Site 2 (%) Site 3 (%) Site 4 (%)
Temperature JunAug 99.4 58.6 85.3 44.0
SepNov 89.5 46.6 48.1 47.3
DecFeb 98.0 97.2 89.1 90.8
MarMay 100.0 84.5 85.3 98.9
All 96.8 71.7 76.9 70.3
PM
2.5
and PM
10
JunAug 94.0 55.5 85.3 44.0
SepNov 90.5 46.6 48.1 47.3
DecFeb 95.8 89.2 89.0 88.6
MarMay 100.0 10.9 85.3 98.9
All 95.2 50.6 76.9 69.7
SO
2
JunAug 99.6 58.5 85.3 43.9
SepNov 89.6 46.6 20.2 47.3
DecFeb 98.0 53.7 89.0 0.0
MarMay 99.1 38.4 80.8 13.5
All 96.6 49.3 68.8 26.2
Table 1 Percent of possible
30-min measurements collected
from June 1, 2009 to May 31,
2010 at four monitoring sites
operated by the Ulaanbaatar
City Environmental Monitoring
Agency
Percentages are based on a
possible 4,380 30-min measure-
ments per 3-month period and
17,520 30-min measurements
per 1-year period. Site locations
are shown in Fig. 1
Fig. 1 Map of the study area
including Ogawa monitoring
locations and government-run
PM
2.5
monitoring sites
Air Qual Atmos Health
Page 3
and 7 were simplified into three dimensions: brightness (a
measure of soil reflectance), greenness (a measure of the
presence and density of green vegetation), and wetness (Su
et al. 2009; Crist and Cicone 1984). The locat ions of the
citys ger areas were determined based on the road network,
features in the DEM and ETM+ data, observations made
during Ogawa sampler deployment, and local knowledge
(Fig. 1). Data on roads were obtained from Open Street
Map (http://www.openstreetmap.org/), and minor modifica-
tions were made based on local knowledge and the location
of features in the ETM+ images. Roads were divided into
two categories: Peace Avenue (the citys busiest road and
main eastwest connector) and major roads. These GIS data
layers (Fig. 1) were used to derive 46 potential predictors of
NO
2
and SO
2
concentrations (Table 2).
LUR models for NO
2
and SO
2
were developed using
methods that have previously been applied to several North
American cities (Henderson et al. 2007 ; Poplawski et al.
2008; Allen et al. 2011). We first calculated correlations
between each potential predictor and the pollutant, then
ranked the predictors in each subcategory (Table 2) by the
absolute value correlation. We then removed any variables
in a subcategory that were correlated (r>0.6) with that
categorys highest ranking variable. All remaining variables
were entered into a stepwise multiple linear regression
model, and the models were rerun as necessary to include
only variables contributing at least 1% to the model R
2
and
coefficients consistent with a priori assumptions (e.g.,
positive coefficients for road variables in the NO
2
model).
Model performance was evaluated based on the model-
based R
2
and the R
2
and root mean square error from a
lea ve one out approach in which the model was
repeatedly calibrated based on all but one measurement
then used to predict the excluded measurement. Residuals
from both models were evaluated for normality and spatial
autocorrelation (Morans I statistic), and variance inflation
factors (VIF) were calculated for all models.
To account for any bias in the Ogawa measurements, we
intended to adjust the Ogawa concentrations based on
colocated NO
2
and SO
2
measurements at three government
monitoring sites (Fig. 1). However, missing government
data during the 2 weeks of Ogawa monitoring did not allow
for such an adjustment. As a result, our LUR models
provide an assessment of relative concentrations across the
city, but the absolute concentrations have not been
independently verified.
Mobile monitoring
We used a mobile monitoring approach to assess spatial
patterns in PM
2.5
concentrations resulting primarily from
home heating (Larson et al. 2007;Suetal.2008b;
Lightowlers et al. 2008;Robinsonetal.2007). The details
of the method are presented elsewhere (Larson et al.
2007). Briefly, on three evenings (between approximately
20002300), we drove predetermined routes that w ere
selected based on local knowledge to capture different
land uses (including areas with high and low ger density)
and a wide range of PM
2.5
concentrations. We drove the
same route in opposite directions on February 24 and 25,
2010 and a different route on the evening of February 26,
2010. A nephelometer (Radiance Research M903, Seattle,
WA, USA), blower, and air preheater were placed in the
back seat of the vehicle and the nephelometersinletwas
extended out the window. The nephelometer recorded the
particle light scattering coefficient (b
sp
) at 15-s averages,
while a global positioning system receiver (Garmin
60CSx, Olathe, KS, USA) connected to a magnetic
antenna on the roof of the vehicle recorded the vehicles
location at 5-s intervals. Additional evenings of monitor-
ing would be needed to definitively characterize s patial
PM
2.5
concentration patterns; our goal was to assess the
feasibility of the mobile monitoring technique i n this
setting.
Table 2 Variables screened in developing LUR models for NO
2
and SO
2
Category Units Buffer radii (m) Subcategories Number of variables
Satellite-based
land cover
Average value in a circular
buffer
400, 500, 750, 1,000, 1,500, 2,000 Brightness, greenness 12
Ger area Hectares in a circular buffer 400, 500, 750, 1,000, 1,500, 2,000 N/A 6
Road length Kilometers in a circular buffer 50, 75, 100, 200, 300, 500, 750,
1,000, 1,500, 2,000
Peace Avenue,
major roads
20
Location Kilometers N/A Latitude, longitude 2
Proximity to
city center
a
Kilometers, 1/kilometers
2
,
log (kilometers)
N/A N/A 3
Elevation Meters N/A N/A 1
Proximity to
power plant
Kilometers to the nearest,
log (kilometers to the nearest)
N/A N/A 2
a
City center defined as Sukhbaatar Square (47.9188, 106.9176)
Air Qual Atmos Health
Page 4
We followed the approach of Larson et al. (2007) for
removing the influence of temporal variation on the mobile
measurements to allow for comparisons between measure-
ments made at different times and during different
evenings. Specifically, we used central-site TEOM data
(from the City Environmental Monitoring Agencys site #1)
to adjust for within- and between-evening temporal trends.
The temporally adjusted light scattering data were then
spatially smoothed by calculating the average value in a
100-m radius around each measurement location. The
temporally adjusted data from February 24 and 25 were
then averaged. Finally, we converted the light scatteri ng
data into approximate PM
2.5
concentrations using a
previously published b
sp
PM
2.5
mass relationship from a
study in Seattle, Washington (Liu et al. 2002) where wood
burning is a major source of PM
2.5
:
PM
2:5
mg=m
3

¼ b
sp
10
5

þ 0:39

=0:27 ð1Þ
Because this approach provides only a semiquantitative
estimate of PM
2.5
, resul ts were grouped into concentration
tertiles for mapping.
Estimation of mortality attributable to air pollution
Mortality data by age and cause, as indicated by the
International Classification o f Diseases, 10th Revision
(ICD-10), for the year 2009 were obtained from the
Statistical Department at the Mongolian Government
Implementing Agency/Department of Health. We used the
World Health Organizations (WHO) approach for environ-
mental burden of disease calculations (Cohen et al. 2005;
Ostro 2004) to estimate the deaths from lung cancer (ICD-
10 code C34) and cardiopulmonary causes (ICD-10 codes
I10I70 and J00J99) that are attributable to long-term
exposure to outdoor air pollution in Ulaanbaatar. First, we
estimated for both caus es of death, the air pollution-
attributable fraction, assuming mortality effects up to the
existing long-term PM
2.5
concentration (X=75 μg/m
3
, the
annual average concentration for June 1, 2009 to May 31,
2010 at Ulaanbaatar City Monitoring Agency site #1)
relative to a counterfactual concentration (X
o
), assuming a
log-linear concentrationmortality relationship (Ostro 2004;
Pope et al. 2009) derived from the American Cancer
Society (ACS) cohort study of air pollution and mortality
(Pope et al. 2002). X
o
was initially set at 7.5 μg/m
3
, the
lowest PM
2.5
concentration observed in the ACS study. The
number of deaths attributable to air pollution was then
determined based on the attributable fractions and the number
of deaths among those 30 years or older from lung cancer
(114) or cardiopulmonary causes (2,007) in Ulaanbaatar in
2009. As a sensitivity analysis, we also calculated attributable
mortality under alternative scenarios, including a linear
concentrationmortality relationship (Ostro 2004;Kunzliet
al. 2000), a counterfactual concentration (X
o
)of3μg/m
3
,
and maximum truncation concentrations (X)of96μg/m
3
(the annual average concentration in the more polluted
traditional housing areas approximated from measure-
ments at Ulaanbaatar City Monitoring Agency site #2)
and 50 μg/m
3
(i.e., we assumed no additional attributable
mortality above 50 μg/m
3
in consideration of the fact that
the ACS study was conducted in the U S where PM
2.5
concentrations are relatively low).
Results
Government fixed-site data
The annual average concentrations of PM
10
,PM
2.5
, and
SO
2
measured at the City Monitoring Agencys site #1 were
165.1, 75.1, and 50.5 μg/m
3
(17.7 ppb), respectively.
Concentrations were highest in winter (Fig. 2). For
example, the mean (±SD) 24-h PM
2.5
concentration in
summer (JuneAugu st) was 22.8±9.0 μg/m
3
, while in
winter (DecemberFebruary), the mean concentration was
147.8±61.2 μg/m
3
. The 24-h PM
2.5
/PM
10
ratios were also
highly variable between seasons (Fig. 2), with a mean ratio
of 0.26±0.11 in summer and 0.78±0.12 in winter.
In addition to seasonal variation, pollution levels also
varied diurnally with two concentration peaks per day. In
both summer and winter, the morning PM
2.5
concentration
peak occurred between approximately 0800 a nd 1000
(Fig. 3). In the summer, the maximum evening levels
occurred between approximately 2000 and 2300, while in
the winter, the highest evening concentrations were from
approximately 2200 to 0200.
After replac ing missing observations with site- and
season-specific median values, the annual average PM
2.5
concentrations at monitoring sites #2, #3, and #4 were 96,
67, and 57 μg/m
3
, respectively, with wintertime averages of
248, 172, and 153 μg/m
3
, respectively.
Land use regression models
Ogawa samplers were retrieved from 37 of the 38 sampling
locations. All NO
2
concentrations were above the LOD; two
SO
2
measurements below the LOD (2.5 ppb) were assumed
to have a concentration of 1.25 ppb (LOD/2). We colocated
our Ogawa monitors with government monitors at three
locations, but adjustment of the Ogawa measurements, which
are known to underestimate NO
2
concentrations at cold
temperatures (Hagenbjork-Gustafsson et al. 2010), was not
possible due to large gaps in the government data during the
2 weeks of Ogawa monitoring. NO
2
and SO
2
were normally
distributed with mean (±SD) concentrations of 10.7±5.8 and
Air Qual Atmos Health
Page 5
17.0±11.8 ppb, respectively, and the two polluta nts were
moderately correlated (r=0.50; p <0.01). There was
significant spatial autocorrelation in the measured concen-
trations of both NO
2
(Morans I=0.42, p<0.01) and SO
2
(Morans I=0.50, p<0.01).
The LUR model predictors for NO
2
were satellite-based
greenness, ger areas, major roads, and distance to city center
(Table 3). The model-based R
2
was 0.74 and the cross-
validation R
2
was 0.66 (Table 3). Of the 46 potential
predictor variables, average greenness in a 1,000-m buffer
had the strongest bivariate relationship (R
2
=0.47) with NO
2
.
The VIF for predictors in the NO
2
LUR model were 1.41
and there was no significant spatial autocorrelation in the
model residuals (Morans I=0.03).
The final SO
2
LUR model included satell ite-based
greenness and ger areas as predictors (Table 3). The model
explained 78% of the variability in SO
2
concentrations,
with a cross-validation R
2
of 0.75 (Table 3). Ger area in a
2,000-m buffer had the strongest bivariate relationship with
SO
2
(R
2
=0.67). The average satellite-based brightness in a
2,000-m buffer was also highly correlated with SO
2
(R
2
=
0.55), although this variable was also correlated with ger
areas and, therefore, did not appear in the final LUR model.
The SO
2
model predictors had VIF=1.1 and there was no
significant spatial autocorrelation in the model residuals
(Morans I=0.14).
The two LUR models captured the different spatial
patterns for these pollutants, with higher NO
2
concen-
trations around the city center and near major ro ads and
higher SO
2
concentrations in the ger areas north of the
city (Fig. 4). There was a strong correlati on (r=0.96)
between the modeled SO
2
concentrations and wintertime
(DecemberFebruary) PM
2.5
concentrations mea sured at
four government-r un sites, although this correlation was
J
u
n
A
ug
O
ct
De
c
F
e
b
Apr
Temperature (
0
C)
-30
-20
-10
0
10
20
30
a
Temperature
Jun
A
u
g
Oct
De
c
Fe
b
Apr
PM
10
Concentration (
µ
g/m
3
)
0
100
200
300
400
500
b
PM
10
J
un
Aug
Oct
D
e
c
Feb
Apr
PM
2.5
Concentration (
µ
g/m
3
)
0
50
100
150
200
250
300
c
PM
2.5
Jun
Aug
O
ct
Dec
F
eb
Apr
PM
2.5
: PM
10
Ratio
0.0
0.2
0.4
0.6
0.8
1.0
d
PM
2.5
:PM
10
Ratio
Jun
A
u
g
Oct
De
c
Fe
b
Apr
SO
2
Concentration (
µ
g/m
3
)
0
20
40
60
80
100
120
140
e
SO
2
Fig. 2 Monthly distributions
of 24-h average a temperature, b
PM
10
, c PM
2.5
, d PM
2.5
/PM
10
ratio, and e SO
2
measured at the
Ulaanbaatar City Environmental
Monitoring Agencys site #1
from June 1, 2009 to May 31,
2010
Air Qual Atmos Health
Page 6
driven primarily by one influential observation from
government site #2, which had relatively high mo deled
SO
2
and a wintertime average PM
2.5
concentration of
248 μg/m
3
.
Table 3 Wintertime LUR models for NO
2
and SO
2
in Ulaanbaatar
Pollutant Model
a
β SE p value VIF Partial R
2
Model R
2
CV R
2
CV RMSE (ppb)
NO
2
b
(ppb) Intercept 8.79 2.77 <0.01 –– 0.74 0.66 3.4
Average greenness in a 1,000-m buffer 0.43 0.08 <0.01 1.19 0.47
1/squared distance to city center
c
0.78 0.24 <0.01 1.16 0.11
Length of Peace Avenue in a 75-m buffer 33.77 11.31 <0.01 1.41 0.06
Length of major roads in a 100-m buffer 17.99 5.39 <0.01 1.15 0.05
Ger area in a 750-m buffer 0.02 0.01 0.02 1.10 0.05
SO
2
b
(ppb) Intercept 13.02 4.56 <0.01 –– 0.78 0.75 5.9
Ger area in a 2,000-m buffer 0.03 0.00 <0.01 1.10 0.66
Average greenness in a 1,000-m buffer 0.56 0.13 <0.01 1.10 0.12
Variables are listed by decreasing partial R
2
VIF variance inflation factor, CV leave one out cross-validation, RMSE root mean square error
a
See Table 2 for variable units
b
See Fig. 4 for modeled concentrations
c
For mapping (Fig. 4a), this variable was capped at a distance of 0.265 km from the city center to correspond with the distance of the closest
Ogawa monitoring site
Hour of the Day
Normalized PM
2.5
Concentration
0.0
0.5
1.0
1.5
2.0
2.5
June - August
a
Hour of the Da
y
7 8 9 1011121314151617181920212223 0 1 2 3 4 5 6
7 8 9 1011121314151617181920212223 0 1 23 4 5 6
Normalized PM
2.5
Concentration
0.0
0.5
1.0
1.5
2.0
2.5
December - February
b
Fig. 3 Diurnal patterns in PM
2.5
concentrations from a June to
August and b December to
February at the Ulaanbaatar City
Environmental Monitoring
Agencys site #1. PM
2.5
con-
centrations are expressed as the
ratio of hourly concentration to
average concentration over the
3-month period (June August
average, 23 μg/m
3
; December
February average, 148 μg/m
3
)
Air Qual Atmos Health
Page 7
Mobile monitoring
Mobile monitoring was conducted on cold, calm evenings that
were typical for the time of year (Fig. 2a). The mean
temperature and wind speed on February 24, 25, and 26
were 22.5°C and 1.4 m/s, 19.6°C and 0.5 m/s, and 14.3°C
and 0.8 m/s, respectively. We observed a wide range of light
scattering values across Ulaanbaatar; the interquartile ranges
of temporally adjusted approximate PM
2.5
concentrations
(converted from b
sp
) measured during mobile monitoring on
February 24/25, 2010 and February 26, 2010 were both
85 μg/m
3
(25th75th percentiles, 11019 5 μg/m
3
on February
24/25 and 85170 μg/m
3
on February 26) (Fig. 5). The spatial
patterns captured by mobile monitoring were generally similar
to SO
2
patterns predicted by the LUR model (Fig. 5). The
mobile monitoring routes passed within 250 m of 25 Ogawa
monitoring sites, and at these sites, the correlation between the
2-week average SO
2
concentration and the nearest temporally
adjusted light scattering coefficient was 0.55 (p<0.01). F or
NO
2
, the correlation was 0.15 (p=0.47).
Mortality attributable to long-term air pollution exposure
in Ulaanbaatar
There were 6,426 total deaths in Ulaan baatar in 2009,
including 1,885 (29.3%) from cardiovascular disease (ICD-
Fig. 4 LUR model predictions
of wintertime a NO
2
and b SO
2
in Ulaanbaatar
Air Qual Atmos Health
Page 8
10 codes I10I70), 269 (4.2%) from respiratory causes
other than lung cancer (ICD-10 codes J00J99), and 117
(1.8%) from lung cancer (ICD-10 code C34). Among those
30 years or older, we estimated that 40% (95% CI, 17
56%) of lung cancer deaths and 29% (1243%) of
cardiopulmonary deaths in Ulaanbaatar are attributable to
outdoor air pollution. These attributable fractions corre-
spond to 578 (232857) cardiopulmonary deaths and 45
(1964) lung cancer deaths annually, or 9.7% of the total
mortality in Ulaanbaatar, attributable to air pollution
(Table 4). Cal culations using alternative assumptions
resulted in estimates that generally deviated from the base
scenario by <50%. For example, a counterfactual concen-
tration of 3.0 μg/m
3
(instead of 7.5 μg/m
3
) increased the
estimated number of attributable deaths by 27% (to 792, or
12.3% of total mortality).
Discussion
The annual average concentrations of PM
10
and PM
2.5
in
central Ulaanbaatar (165.1 and 75.1 μg/m
3
, respectively)
are approximately seven to eight times the WHO air quality
guidelines of 20 and 10 μg/m
3
, respectively (Krzyzanowski
and Cohen 2008). Although there is no WHO guideline for
annual SO
2
concentrations, the annual average concentra-
tion of 50.5 μg/m
3
(17.7 ppb) in Ulaanbaatar far exceeds
even the 24-h guideline of 20 μg/m
3
(7 ppb). Importantly,
the concentrations measured in the city center are consid-
erably lower than those measured in one of the citys ger
areas.
These PM concentrations place Ulaanbaatar among
the most polluted c ities in the world ( HEI 2004). For
example, Ulaanbaatars annual average PM
10
concentra-
tion of 165 μg/m
3
is comparable to late 1990s levels
(approximated as half the concentration of total suspended
particles; Cohen et a l. 2005) in megacities such as
Kolkata, Delhi, and Beijing and exceeds levels in cities
such as Mexico City and Buenos Aires (Gurjar et al.
2008). Despite its extraordinarily high air pollution
concentrations, Ulaanbaatar has received very little re-
search attention (Davy et al. 20 11).
The high pollution concentrations are driven by con-
ditions during winter, when 24-h PM
2.5
concentrations
frequently exceed 150 μg/m
3
(and approach 250 μg/m
3
in
traditional housing areas) and SO
2
levels are frequently
above 80 μg/m
3
. The high PM
2.5
/PM
10
ratios (0.65) in
winter are comparable to previous wintertime measure-
ments in polluted urban areas such as 0.71 in Beijing
(Zhang et al. 2010) and 0.69 in Shanghai (Zhang et al.
2006) and suggest a major contribution from combustion-
derived particles (Davy et al. 2011). The lower summertime
PM
2.5
/PM
10
ratios (0.35) are similar to observations in
arid locations impacted by wind-blown dust (Eliasson et al.
2009) and suggest a relatively large influence of crustal
particles. The diurnal PM
2.5
patterns varied by season, with
an evening peak in the winter that occurs later in the day
and lasts longer than the evening peak in summer. The
wintertime diurnal pattern in Ulaanbaatar is similar to
developed country communities impacted by emissions
from residential wood combustion (Robinson et al. 2007;
Krecl et al. 2008).
Rapidly developing cities often have different urban
designs and air pollution sources than cities in high-income
regions, and few studies have atte mpted to characterize
spatial patterns of air pollution in developing cities (Padhi
and Padhy 2008; Dionisio et al. 2010; Etyemezian et al.
2005). We developed LUR models for NO
2
, a marker of
traffic emissions, and SO
2
, a marker of coal combustion, in
Ulaanbaatar. The model-based R
2
of our NO
2
model was
0.74, which is within the wide range (0.510.97) reported
in previous studies (Hoek et al. 2008 ). Our SO
2
model-
based R
2
of 0.78 is higher than the values (0.66 and 0.69)
reported in the two previously published LUR SO
2
models
(Wheeler et al. 2008; Atari et al. 2008).
The vast majority of existing L UR models were
developed in high-income countries (Hoek et al. 2008),
and very few LUR models have been developed for Asian
cities (Kashima et al. 2009;Chenetal.2010). One
challenge to LUR modeling in developing settings is the
lack of data on spatial predictors, but satellite-based ETM+
data, which have global coverage and are freely available,
show promise for overcoming this limitation. In our
analysis, average greenness in a 1,000-m buffer explained
47% of the variance in NO
2
, while brightness in a 2,000-m
buffer explained 55% of the SO
2
variance. ETM+ may even
be useful for LUR models in developed countries, as
demonstrated by the inclusion of satellite-based soil
brightness in recent LUR models for Los Angeles (Su et
al. 2009).
A few limitations of our LUR models should be
considered. First, the models are based on 37 observations,
which is fewer than the 4080 recommended in a recent
LUR review (Hoek et al. 2008). Additional observations
may have resulted in a model with more variables that
captured additional complexity in the spatial patterns,
particularly for NO
2
due to its high spatial variability
(Fig. 4b). Second, although ETM + data from 2006 were the
most recent available, they may have missed important
recent land cover changes in this rapidly growing city.
Nevertheless, the high correlations between the ETM+
ground cover classifications and both NO
2
(greenness
predicted 47% of the variance) and SO
2
(brightness
predicted 55% of the variance) indicate the value of these
data for predicting spatial pollution gradients. Third, we
were unable to calibrate our Ogawa measurements with
Air Qual Atmos Health
Page 9
government monitors, so our LUR surfaces can be used for
assessing spatial patterns but the absolute concentrations
may be inaccurate. Finally, LUR models are generally
developed based on multiple sampling campaigns to assess
long-term conditions (Hoek et al. 2008), but our models are
based on a single monitoring session. While additional
monitoring would more definitively characterize wintertime
spatial patterns, we have demonstrated the feasibility of
developing LUR models in a rapidly developing Asian city.
Spatially dense fixed-location monitoring is an expen-
sive and logistically challenging way to capture within-city
spatial variations in PM
2.5
, and mobile monitoring repre-
sents a promising alternative, particularly in developing
countries with limited resources for environmental moni-
toring. For example, Dionisio et al. (2010) measured spatial
PM
2.5
patterns in Accra, Ghana by walking 7.79.4 km
paths while recording PM
2.5
and latitude/longitude at 1-min
resolution. They identified nearby wood and charcoal stoves,
congested and heavy traffic, loose dirt road surface, and trash
burning as important PM
2.5
sources. In Ulaanbaatar, we
piloted a vehicle-based mobile nephelometer monitoring
technique that was originally developed for capturing
Fig. 5 Temporally adjusted
light scattering tertiles obtained
during mobile monitoring in
Ulaanbaatar on a February 24
and 25, 2010 and b February 26,
2010. Approximate PM
2.5
concentrations for each tertile
are given in parentheses. For
comparison, the mobile data are
superimposed on the modeled
SO
2
surface shown in Fig. 4b
Air Qual Atmos Health
Page 10
Table 4 Estimates of mortality attributable to long-term PM
2.5
air pollution exposure in Ulaanbaatar under different model scenarios
Shape of
concentrationmortality
relationship
Counterfactual PM
2.5
concentration (μg/m
3
)
Maximum
truncation
concentration
a
(μg/m
3
)
Attributable fraction
of lung cancer deaths
b
(95% CI)
Attributable lung
cancer deaths
c
(95% CI)
Attributable fraction of
cardiopulmonary
b
deaths (95% CI)
Attributable
cardiopulmonary
deaths
c
(95% CI)
Percentage of Ulaanbaatar
deaths attributable to
outdoor air pollution
d
Deviation
from base
scenario
Log-linear 3.0 96
e
52.3% (23.970.1%) 60 (2780) 39.0% (16.455.5%) 783 (3291,114) 13.1% 35.3%
75
f
49.5% (22.367.2%) 56 (2577) 36.7% (15.352.7%) 736 (3061,057) 12.3% 27.1%
50 44.6% (19.661.9%) 51 (2271) 32.6% (13.347.6%) 655 (268956) 11.0% 13.3%
7.5 96
e
43.2% (18.860.2%) 49 (2169) 31.5% (12.846.1%) 632 (257926) 10.6% 9.3%
75
f
39.9% (17.156.4%) 45 (1964) 28.8% (11.642.7%) 578 (232857) 9.7%
50 34.0% (14.249.3%) 39 (1656) 24.3% (9.636.6%) 487 (192734) 8.2% 15.7%
Linear 3.0 96
e
69.2% (33.185.8%) 79 (3898) 56.4% (25.974.4%) 1,132 (5191,493) 18.8% 94.4%
75
f
59.8% (26.778.0%) 68 (3089) 47.4% (20.765.1%) 952 (4151,308) 15.9% 63.7%
50 44.9% (18.462.8%) 51 (2172) 34.3% (14.049.7%) 688 (282998) 11.5% 18.6%
7.5 96
e
67.4% (31.884.4%) 77 (3696) 54.6% (24.872.6%) 1,096 (4981,458) 18.3% 88.3%
75
f
57.5% (25.375.8%) 66 (2986) 45.3% (19.562.8%) 908 (3921,260) 15.2% 56.3%
50 41.6% (16.859.1%) 47 (1967) 31.6% (12.846.3%) 634 (257930) 10.6% 9.3%
The conservative base scenario is shown in italics
a
The PM
2.5
concentration above which no additional attributable mortality is assumed to result
b
Among those >30 years old based on methods described in Ostro (2004) and PM
2.5
concentrationmortality relationships estimated in Pope et al. (2002)
c
Based on the calculated attributable fraction and 2009 mortality statistics for those >30 years in Ulaanbaatar: 1,859 cardiovascular disease deaths, 148 respiratory/pulmonary deaths, and 114 lung
cancer deaths
d
Calculated as the sum of attributable deaths from lung cancer and cardiopulmonary causes divided by the 6,426 total deaths in 2009
e
Assumes no truncation of PM
2.5
effects on mortality up to the measured 96 μg/m
3
20092010 annual average PM
2.5
concentration at Ulaanbaatar City Environmental Monitoring Agencys site
#2
f
Assumes no truncation of PM
2.5
effects on mortality up to the measured 75 μg/m
3
20092010 annual average PM
2.5
concentration at Ulaanbaatar City Environmental Monitoring Agencys site
#1
Air Qual Atmos Health
Page 11
spatial patterns of wood smoke PM
2.5
in North American
cities (Larson et al. 2007;Suetal.2008b). Although 10
20 evenings of monitoring may be needed to definitively
identify spatial patterns (Larson et al. 2007;Suetal.
2008b), we captured a wide range of PM
2.5
concentrations
during three evenings of monitoring. In spite of the
differences in technique, pollutant, and averaging time,
the spatial patterns in PM
2.5
identified by mobile moni-
toring and SO
2
patterns identified by Ogawa measure-
ments and an LUR model were generally similar (Fig. 5).
The identification of PM
2.5
and SO
2
hot spots in the
citys ger areas is consistent with major emissions of these
pollutants in these areas and with source apportionment
results, suggesting that coal is the dominant s ource o f
PM
2.5
in Ulaanbaatar (Davy et al. 2011). Moreover, the
similarities between SO
2
and PM
2.5
spatial patterns
(Fig. 5), agreement between modeled SO
2
and measured
PM
2.5
at four government monitoring sites, and similari-
ties between Ogawa SO
2
measurements and mobile light
scattering coefficient measurements suggest that our SO
2
LUR model may provide a tool for PM
2.5
exposure
assessment in Ulaanbaatar, although this needs to be
verified with additional PM
2.5
monitoring.
Based on 2009 mortality statistics, we conser vatively
estimated that 623 deaths in Ulaanbaatar wer e attributable
to air pollution. This represents 9.7% of the 6,426 total
deaths in the city and, notably, 4.0% of the 15,522 annual
deaths for the entire country. Calculations using alternative
assumptions produced estimates that generally dev iated
from the base scenario by <50%. The exceptions were
estimates based on a linear concentrationresponse rela-
tionship and no truncation, which were up to 94% higher
than our base scenario estimate. These scenarios may be
unrealistic (Ostro 2004), given evidence suggesting that
PM
2.5
mortality effects are nonlinear across the wide range
of concentrations considered here (Pope et al. 2009).
Our estimate of attributable mortality probably under-
estimates the true public health burden of air pollution in
Ulaanbaatar for several reasons (Kunzli et al. 2000, 2008).
First, we did not consider the effects of indoor air pollution
or outdoor pollutants other than PM
2.5
(Anenberg et al.
2010; HEI 2004; Rylance et al. 2010). In addition, due
primarily to data limitations, we did not consider non-
mortality endpoints that have been linked to air pollution
such as cardiovascular disease, impaired lung development,
incident asthma, asthma exacerbations, bronchitis, hospital-
izations, and school absences (Pope and Dockery 2006;
Brook et al. 2010; Allen et al. 2009; Gauderman et al.
2004; Clark et al. 2010; Perez et al. 2009). We also
considered mortality impacts only among those 30 years or
older, thus excluding, for example, attributable infant
mortality (Kaiser et al. 2004). Finally, we made the a priori
decision to use PM
2.5
data from a centrally located
government monitoring site to estimate outdoor concen-
trations for the attributable mortality calculation. Our SO
2
LUR model and mobile PM
2.5
monitoring (Figs. 4b and 5)
suggest that this site is located in a relatively unpolluted
area of Ulaanbaatar. As a result, both outdoor concen-
trations and attributable mortality may be underestimated,
especially given that half the citys population lives in a ger
(Asian Development Bank 2006), and the higher attribut-
able mortality estimates (10.613.1% of total mortality)
based on concentrations at site #2 may be more appropriate.
The strengths and weaknesses of q uantitative impact
assessment methods have bee n discussed extensiv ely
(Brunekreef et al. 2007; Perez and Kunzli 2009; Sahsuvaroglu
and Jerrett 2007; O'Connell and Hurley 2009). Estimates
of attributable mortality are often misinterpreted as
avoidable deaths, but it is more appropriate to interpret
these as estimates of postponable deaths (Brunekreef et
al. 2007). Some have suggested that changes in life
expectancy (calculated from life tables) are a better and
more interpretable indicator of the mortality impacts of
long-term air pollution exposure (Brunekreef et al. 2007;
Perez and Kunzli 2009; Boldo et al. 2006). If age-specific
population and death statistics can be obtained, changes in
life expectancy can be calculated (e.g., WHO AirQ 2.2.3
or http://www.iom-world.org/research/iomlifet.php). Un-
fortunately, the mortalit y data provided by the Statistical
Department at the Mongolian Government Implementing
Agency/Department of Health were aggregated for those
older than 65 years. Therefore, we restricted our impact
assessment to the attributable mortality calculations.
Despite its limitati ons, attributable mortality i s a com-
monly used metric for impact assessment. For example, it
also allows for a comparison of the 623 deaths attributable
to air pollution annually in Ulaanbaatar with the mortality
attributable to other risk factors such as suicide (199
deaths in 2009), transpo rtation accidents (185), and
homicide (179).
Conclusions
Due in part to challenges and limitations in population
exposure assessment, few epidemiologic studies of air
pollution and health have been conducted in developing
countries. We successfully applied current, cost-effective
exposure assessment techniques in Ulaanbaatar, Mongolia,
one of the most polluted cities in the world, which suggests that
these techniques are feasible in other rapidly developing cities.
Based on satellite-based land cover and other predictors, we
developed LUR models that identified strong spatial concen-
tration gradients consistent with a major contribution from
home heating in Ulaanbaatars low-income traditional housing
(ger) areas. Temporal patterns, mobile PM
2.5
monitoring, and
Air Qual Atmos Health
Page 12
PM
2.5
/PM
10
ratios supported this finding. Air pollution
represents a major threat to public health in Ulaanbaatar,
and reductions in home heating emissions should be the
primary focus of future air pollution control efforts.
Acknowledgments We are grateful to the students and staff at the
School of Public Health, Health Sciences University of Mongolia for
their assistance with data collection. Air pollution data were provided
by both the Mongo lian Nati onal Mon itoring Age ncy and the
Ulaanbaatar City Environmental Monitoring Agency, and mortality
data were provided by Statistical Department at the Mongolian
Government Implementing Agency/Department of Health. We thank
Dr. Winnie Chu and the staff at the University of British Columbias
School of Environmental Health laboratory for analyzing Ogawa
samplers. Funding for this work was provided by the BC Environmental
and Occupational Health Research Network and Health Canada.
Open Access This article is distributed under the terms of the
Creative Commons Attribution Noncommercial License which per-
mits any noncommercial use, distribution, and reproduction in any
medium, provided the original author(s) and source are credited.
References
Allen RW, Criqui MH, Roux AVD et al (2009) Fine particulate matter
air pollution, proximity to traffic, and aortic atherosclerosis.
Epidemiology 20(2):254264
Allen RW, Amram O, Wheeler AJ, Brauer M (2011) The transferabil-
ity of NO and NO
2
land use regression models between cities and
pollutants. Atmos Environ 45:369378
Anenberg SC, Horowitz LW, Tong DQ, West JJ (2010) An estimate of
the global burden of anthropogenic ozone and fine particulate
matter on premature human mortality using atmospheric model-
ing. Environ Heal Perspect 118(9):11891195
Asian Development Bank (2006) Country synthesis report on urban
air quality management: Mongolia
Atari DO, Luginaah I, Xu XH, Fung K (2008) Spatial variability of
ambient nitrogen dioxide and sulfur dioxide in Sarnia, Chemical
Valley, Ontario, Canada. J Toxicol Environ Health-Part A-Curr
Issues 71(24):15721581
Boldo E, Medina S, LeTertre A et al (2006) Apheis: health impact
assessment of long-term exposure to PM2.5 in 23 European
cities. European J Epidemiol 21(6):449458
Brook RD, Rajagopalan S, Pope CA et al (2010) Particulate matter air
pollution and cardiovascular disease. An update to the scientific
statement from the American Heart Association. Circulation 121
(21):23312378
Brunekreef B, Miller BG, Hurley JF (2007) The brave new world of
lives sacrificed and saved, deaths attributed and avoided.
Epidemiology 18(6):785788
Chen L, Bai ZP, Kong SF et al (2010) A land use regression for
predicting NO
2
and PM10 concentrations in different seasons in
Tianjin region, China. J Environ Sci-China 22(9):13641373
Clark NA, Demers PA, Karr CJ et al (2010) Effect of early life
exposure to air pollution on development of childhood asthma.
Environ Health Perspect 118(2):284290
Cohen AJ, Anderson HR, Ostro B et al (2005) The global burden of disease
due to outdoor air pollution. J Toxicol Environ Health-Part A-Curr
Issues 68(1314):13011307
Crist EP, Cicone RC (1984) A physically-based transformation of
thematic mapper datathe TM tasseled cap. IEEE Trans Geosci
Remote Sens 22(3):256263
Davy PK, Gunchin G, Markwitz A et al (2011) Air particulate matter
pollution in Ulaanbaatar, Mongolia: determination of composi-
tion, source contributions and source locations. Atmos Pollut Res
2:126137
Dionisio KL, Rooney MS, Arku RE et al (2010) Within-
neighborhood patterns and sources of particl e pollution:
mobile monitoring and geographic info rmation system anal-
ysis in four communities in Accra, Ghana. Environ Heal
Perspect 118(5):607613
Eliasson I, Jonsson P, Holmer B (2009) Diurnal and intra-urban
particle concentrations in relation to windspeed and stability
during the dry season in three African cities. Environ Monit
Assess 154(14):309324
Etyemezian V, Tesfaye M, Yimer A et al (2005) Results from a pilot-
scale air quality study in Addis Ababa, Ethiopia. Atmos Environ
39(40):78497860
Gauderman WJ, Avol E, Gilliland F et al (2004) The effect of air
pollution on lung development from 10 to 18 years of age. New
Engl J Med 351(11):10571067
Gurjar BR, Butler TM, Lawrence MG, Lelieveld J (2008) Evaluation
of emissions and air quality in megacities. Atmos Environ 42
(7):15931606
Guttikunda S (2007) Urban air pollution analysis for Ulaanbaatar, the
World Bank Consultant Report
Hagenbjork-Gustafsson A, Tornevi A, Forsberg B, Eriksson K
(2010) Field validation of the Ogawa diffusive sampler for
NO
2
and NO
x
in a cold climate. J Environ Monit 12(6):1315
1324
HEI (2004) Health Effects Institute special report 15, health effects of
outdoor air pollution in developing countries of Asia: a literature
review
HEI (2010) Health Effects Institute special report 17, traffic related air
pollution: a critical review of the literature
Henderson SB, Beckerman B, Jerrett M, Brauer M (2007) Application
of land use regression to estimate long-term concentrations of
traffic-related nitrogen oxides and fine particulate matter. Environ
Sci Technol 41(7):24222428
Hoek G, Beelen R, de Hoogh K et al (2008) A review of land-use
regression models to assess spatial variation of outdoor air
pollution. Atmos Environ 42(33):75617578
Kaiser R, Romieu I, Medina S, Schwartz J, Krzyzanowski M, Kunzli
N (2004) Air pollution attributable postneonatal infant mortality
in U.S. metropolitan areas: a risk assessment study. Environ
Health 3(1):4
Kashima S, Yorifuji T, Tsuda T, Doi H (2009) Application of land use
regression to regulatory air quality data in Japan. Sci Total
Environ 407(8):30553062
Krecl P, Larsson EH, Strom J, Johansson C (2008) Contribution of
residential wood combustion and other sources to hourly winter
aerosol in Northern Sweden determine d by positive matrix
factorization. Atmos Chem Phys 8(13):36393653
Krzyzanowski M, Cohen A (2008) Update of WHO air quality
guidelines. Air Qual Atmos Health 1:713
Kunzli N, Kaiser R, Medina S et al (2000) Public-health impact of
outdoor and traffic-related air pollution: a European assessment.
Lancet 356(9232):795801
Kunzli N, Perez L, Lurmann F, Hricko A, Penfold B, McConnell R
(2008) An attributable risk model for exposures assumed to cause
both chronic disease and its exacerbations. Epidemiology 19
(2):179185
Larson T, Su J, Baribeau AM, Buzzelli M, Setton E, Brauer M (2007)
A spatial model of urban winter woodsmoke concentrations.
Environ Sci Technol 41(7):24292436
Lightowlers C, Nelson T, Setton E, Keller CP (2008) Determining the
spatial scale for analysing mobile measurements of air pollution.
Atmos Environ 42(23):5933 5937
Air Qual Atmos Health
Page 13
Liu LJS, Slaughter JC, Larson TV (2002) Comparison of light
scattering devices and impactors for particulate measurements in
indoor, outdoor, and personal environments. Environ Sci Technol
36(13):29772986
National Statistical Office of Mongolia (2010) Mongolian statistical
yearbook, 2009. Ulaanbaatar
O'Connell E, Hurley F (2009) A review of the strengths and
weaknesses of quantitative methods used in health impact
assessment. Public Health 123(4):306310
Ostro B (2004) Outdoor air pollution. Assessing the environmental
burden of disease at national and local levels. World Health
Organization, Geneva
Padhi BK, Padhy PK (2008) Assessment of intra-urban variability in
outdoor air quality and its health risks. Inhal Toxicol 20(11):973
979
Perez L, Kunzli N (2009) From measures of effects to measures of
potential impact. Int J Public Health 54(1):4548
Perez L, Kunzli N, Avol E et al (2009) Global goods movement and
the local burden of childhood asthma in southern California. Am
J Public Health 99(Suppl 3):S622S628
Pope CA, Dockery DW (2006) Health effects of fine particulate air
pollution: lines that connect. J Air Waste Manag Assoc 56
(6):709742
Pope CA, Burnett RT, Thun MJ et al (2002) Lung cancer, cardiopulmo-
nary mortality, and long-term exposure to fine particulate air
pollution. Jama-J Am Med Assoc 287(9):11321141
Pope CA, Burnett RT, Krewski D et al (2009) Cardiovascular mortality
and exposure to airborne fine particulate matter and cigarette
smoke shape of the exposureresponse relationship. Circulation
120(11):941948
Poplawski K, Gould T, Setton E et al (2008) Intercity transfer-
ability of land use regress ion models for estimating ambient
concentrations of nitrogen dioxide. J Expo Anal Environ
Epidemiol 19:107117
Robinson DL, Monro JM, Campbell EA (2007) Spatial variability and
population exposure to PM2.5 pollution from woodsmoke in a
New South Wales country town. Atmos Environ 41(26):5464
5478
Rylance J, Fullerton DG, Semple S, Ayres JG (2010) The global
burden of air pollution on mortality: the need to include exposure
to household biomass fuel-derived particulates. Environ Heal
Perspect 118(10):A424
Sahsuvaroglu T, Jerrett M (2007) Sources of uncertainty in calculating
mortality and morbidity attributable to air pollution. J Toxicol
Environ Health-Part A-Curr Issues 70(34):243260
Schlesinger RB, Kunzli N, Hidy GM, Gotschi T, Jerrett M (2006) The
health relevance of ambient particulate matter characteristics:
coherence of toxicological and epidemiological inferences. Inhal
Toxicol 18(2):95125
Su JG, Brauer M, Buzzelli M (2008a) Estimating urban morphometry
at the neighborhood scale for improvement in modeling long-
term average air pollution concentrations. Atmos Environ 42
(34):78847893
Su JG, Buzzelli M, Brauer M, Gould T, Larson TV (2008b) Modeling
spatial variability of airborne levoglucosan in Seattle, Washington.
Atmos Environ 42(22):55195525
Su JG, Jerrett M, Beckerman B, Wilhelm M, Ghosh JK, Ritz B (2009)
Predicting traffic-related air pollution in Los Angeles using a
distance decay regression selection strategy. Environ Res 109
(6):657670
Wheeler AJ, Smith-Doiron M, Xu X, Gilbert NL, Brook JR (2008)
Intra-urban variability of air pollution in Windsor, Ontario
measurement and modeling for human exposure assessment.
Environ Res 106(1):716
World Bank (2004) environment monitor: environmental challenges of
urban development
Zhang YX, Zhang YM, Wang YS et al (2006) PIXE characterization of
PM10 and PM2.5 particulate matter collected during the winter
season in Shanghai city. J Radioanal Nuclear Chem 267(2):497499
Zhang WJ, Zhuang GS, Guo JH et al (2010) Sources of aerosol as
determined from elemental composition and size distributions in
Beijing. Atmos Res 95(23):197209
Air Qual Atmos Health
Page 14
    • "Rasters of ger density were created using the Point Statistics tool to count the number of gers in each of the 11 circular buffers, and then dividing by the area of the circular neighbourhood area (in hectares). Although the focus of this paper is the methods for deriving LUR model predictors, we used measured concentrations of NO 2 and SO 2 made at 37 monitoring sites across Ulaanbaatar, collected as part of a pilot study conducted in the winter of 2010 (Figure 1) (Allen et al. 2013), to evaluate the potential for these variables to serve as useful predictors in LUR models. First, we estimated correlation coefficients between each of the 55 predictor variables and measured concentrations. "
    [Show abstract] [Hide abstract] ABSTRACT: Air pollution is a major risk factor for death and disease, particularly in low- and middle-income countries (LMIC) where concentrations are relatively high and large populations are exposed. High-quality exposure assessment is integral to both air pollution epidemiologic studies and impact assessments. Land use regression (LUR) modelling is a powerful exposure assessment technique that uses the relationships between air pollution concentrations at discrete monitoring locations and the surrounding characteristics of those locations to model small-scale spatial concentration gradients within cities. Regardless of whether they are calibrated based on local measurements or transferred from another location, LUR models require spatially resolved data on predictor variables that may be unavailable or of insufficient quality in many settings. We describe methods for deriving LUR model predictors, including land cover, road locations, and ger (Mongolian yurt) locations, from satellite data and high-resolution imagery in Ulaanbaatar, Mongolia. These methods may allow LUR models to be developed for more locations in LMIC, potentially improving the quality of air pollution exposure assessments in those settings.
    No preview · Article · May 2016 · Canadian Geographer / Le Géographe canadien
  • Source
    • "However, while previous studies have estimated the human health impacts from ambient air pollution due to fossil fuel combustion (Anenberg et al., 2010), open biomass burning (Johnston et al., 2012;Marlier et al., 2013), and wind-blown dust (Giannadaki et al., 2014), fewer studies have quantified the impact of residential combustion on ambient quality and human health.Lim et al. (2012)estimated that 16 % of the global burden of ambient PM 2.5 was due to RSF sources but did not estimate premature mortality. Another study concluded that ambient PM 2.5 from cooking was responsible for 370 000 deaths in 2010 (Chafe et al., 2014), but it did not include residential heating emissions, which will cause additional adverse impacts on human health (Johnston et al., 2013;Allen et al., 2013;). Here we use a global aerosol microphysics model to make an integrated assessment of the impact of residential emissions on atmospheric aerosol, radiative effect, and human health. "
    [Show abstract] [Hide abstract] ABSTRACT: Combustion of fuels in the residential sector for cooking and heating results in the emission of aerosol and aerosol precursors impacting air quality, human health, and climate. Residential emissions are dominated by the combustion of solid fuels. We use a global aerosol microphysics model to simulate the impact of residential fuel combustion on atmospheric aerosol for the year 2000. The model underestimates black carbon (BC) and organic carbon (OC) mass concentrations observed over Asia, Eastern Europe, and Africa, with better prediction when carbonaceous emissions from the residential sector are doubled. Observed seasonal variability of BC and OC concentrations are better simulated when residential emissions include a seasonal cycle. The largest contributions of residential emissions to annual surface mean particulate matter (PM2.5) concentrations are simulated for East Asia, South Asia, and Eastern Europe. We use a concentration response function to estimate the human health impact due to long-term exposure to ambient PM2.5 from residential emissions. We estimate global annual excess adult (> 30 years of age) premature mortality (due to both cardiopulmonary disease and lung cancer) to be 308 000 (113 300–497 000, 5th to 95th percentile uncertainty range) for monthly varying residential emissions and 517 000 (192 000–827 000) when residential carbonaceous emissions are doubled. Mortality due to residential emissions is greatest in Asia, with China and India accounting for 50 % of simulated global excess mortality. Using an offline radiative transfer model we estimate that residential emissions exert a global annual mean direct radiative effect between −66 and +21 mW m−2, with sensitivity to the residential emission flux and the assumed ratio of BC, OC, and SO2 emissions. Residential emissions exert a global annual mean first aerosol indirect effect of between −52 and −16 mW m−2, which is sensitive to the assumed size distribution of carbonaceous emissions. Overall, our results demonstrate that reducing residential combustion emissions would have substantial benefits for human health through reductions in ambient PM2.5 concentrations.
    Full-text · Article · Jan 2016 · Atmospheric Chemistry and Physics
  • Source
    • "In general, the availability of highresolution satellite images offer a good potential to derive appropriate land use predictors when readily available GIS data are lacking. The ease of use of GIS and remote sensing tools, as has been shown by other researchers (Maxwell, 2010; Mao et al., 2012; Allen et al., 2013), further facilitates the exploitation of satellite imagery in air pollution modelling and exposure assessments in low income countries. Many applications, including air pollution modelling and exposure assessment, often require population data at finer spatial resolutions; however, population data are usually collected only at coarse administrative area levels. "
    [Show abstract] [Hide abstract] ABSTRACT: Land use regression (LUR) modelling is a common approach used in European and Northern American epidemiological studies to assess urban and traffic related air pollution exposures. Studies applying LUR in Africa are lacking. A need exists to understand if this approach holds for an African setting, where urban features, pollutant exposures and data availability differ considerably from other continents. We developed a parsimonious regression model based on 48-hour nitrogen dioxide (NO2) concentrations measured at 40 sites in Kaédi, a medium sized West-African town, and variables generated in a geographic information system (GIS). Road variables and settlement land use characteristics were found to be important predictors of 48-hour NO2 concentration in the model. About 68% of concentration variability in the town was explained by the model. The model was internally validated by leave-one-out cross-validation and it was found to perform moderately well. Furthermore, its parameters were robust to sampling variation. We applied the model at 100 m pixels to create a map describing the broad spatial pattern of NO2 across Kaédi. In this research, we demonstrated the potential for LUR as a valid, cost-effective approach for air pollution modelling and mapping in an African town. If the methodology were to be adopted by environmental and public health authorities in these regions, it could provide a quick assessment of the local air pollution burden and potentially support air pollution policies and guidelines.
    Full-text · Article · Nov 2015 · Geospatial health
Show more