Content uploaded by E. Stelcer
Author content
All content in this area was uploaded by E. Stelcer on Dec 16, 2015
Content may be subject to copyright.
Source apportionment of PM
2.5
at two receptor sites
in Brisbane, Australia
Adrian J. Friend,
A
Godwin A. Ayoko,
A
,
C
Eduard Stelcer
B
and David Cohen
B
A
International Laboratory for Air Quality and Health, Discipline of Chemistry, Queensland
University of Technology, QLD 4001, Australia.
B
Institute for Environmental Research, Australian Nuclear Science and Technology Organisation,
Locked Bag 2001, Kirrawee DC, NSW 2232, Australia.
C
Corresponding author. Email: g.ayoko@qut.edu.au
Environmental context. Fine particles affect air quality locally, regionally and globally. Determining the
sources of fine particles is therefore critical for developing strategies to reduce their adverse effects. Advanced
data analysis techniques were used to determine the sources of fine particles at two sites, providing information
for future pollution reduction strategies not only at the study sites but in other areas of the world as well.
Abstract. In this study, samples of particulate matter with aerodynamic diameter less than 2.5 mm (PM
2.5
) collected at
two sites in the south-east Queensland region, a suburban (Rocklea) and a roadside site (South Brisbane), were analysed for
H, Na, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Br, Pb and black carbon (BC). Samples were collected
during 2007–10 at the Rocklea site and 2009–10 at the South Brisbane site. The receptor model Positive Matrix
Factorisation was used to analyse the samples. The sources identified included secondary sulfate, motor vehicles, soil, sea
salt and biomass burning. Conditional probability function analysis was used to determine the most likely directions of the
sources. Future air quality control strategies may focus on the particular sources identified in the analysis.
Additional keywords: fine particles, Positive Matrix Factorisation, receptor modelling.
Received 11 July 2011, accepted 2 August 2011, published online 17 November 2011
Introduction
Particulate matter (PM) has been consistently linked to adverse
health effects upon a broad range of systems in the body.
[1]
Small particles with aerodynamic diameter less than 2.5 mm
(PM
2.5
) have been of particular interest as their size allows them
to penetrate the lungs.
[2]
Visibility, vegetation, materials and
climate change can also be adversely affected by air pollu-
tion.
[3,4]
Particles are emitted into the air from anthropogenic
and natural sources, either directly (as a primary source) or from
processes in the air that create new particles (secondary source).
Determining the identity, location and effect of a source can lead
to more targeted regulation as well as a better understanding of
the mechanisms of the effects of PM.
[5]
Current quality standards have been established by the
National Environmental Protection Council (NEPC) for various
pollution species.
[6,7]
For PM
2.5
the standards are: limits of
25-mgm
3
average for 24 h and 8-mgm
3
average for 1 year. To
monitor the level of these pollutants, the Queensland Depart-
ment of Environmental and Resource Management (QDERM)
has established more than 25 monitoring stations throughout
Queensland. In south-east Queensland, these include the
Rocklea and South Brisbane sites. The Queensland government
also releases regular reports on the State of the Environment.
[8]
These reports examine trends in the concentrations of the
detected pollutants and record the number of days that the
concentrations are higher than the national standards.
By combining these efforts with advanced data analysis, the
sources of the air pollutants could be determined.
Multivariate data analysis has been performed for source
determination and apportionment in locations around the
world.
[5,9–12]
Common receptor models include principal
component analysis/absolute principal component scores
(PCA/APCS),
[13]
UNMIX,
[14]
and Positive Matrix Factorisation
(PMF).
[15]
These models are based on a mass balance equation
and aim to reconstruct source emissions based on measurements
from monitoring sites.
[16]
PMF2 was the first PMF model and
was published by Paatero and Tapper in 1994.
[15]
The US
Environmental Protection Agency (EPA) has since developed
versions of both UNMIX and PMF (EPA PMF) that are freely
available (see http://www.epa.gov/scram001/receptorindex.htm,
accessed 3 October 2011). A non-negativity constraint and
individually weighted data points are used in PMF to determine
physically reasonable results. These models have been applied
to ambient PM
2.5
concentrations in numerous studies as outlined
by Watson.
[17]
In south-east Queensland there has been limited
application of modelling techniques for the determination of the
local sources of pollution.
[18–21]
The models applied in the
previous studies include target transformation factor analysis
and PMF.
For this study, PMF was applied to measurements recorded at
two monitoring sites, corresponding to a suburban and a road-
side site, over 3 and 1 years respectively. A range of factor
numbers (sources) was examined to determine the optimum
solution using evaluation tools.
[22]
The aims of this study were
to: (i) determine the identity and contributions of PM
2.5
to the
Brisbane airshed in order to facilitate the formation of pollution
CSIRO PUBLISHING
Environ. Chem. 2011,8, 569–580
http://dx.doi.org/10.1071/EN11056
Journal compilation ÓCSIRO 2011 www.publish.csiro.au/journals/env569
Research Paper
reduction strategies; (ii) compare the current results at Rocklea
with that obtained for a similar site in Rocklea from
1995–2003
[23]
; and (iii) compare the sources and source con-
tributions at two different sites in Brisbane in order to determine
whether it is important to have multiple air monitoring sites
around the city.
Methods and materials
Sampling program
Samples were collected at two sites in south-east Queensland,
including a suburban Rocklea site located at latitude
2783208.879400S and longitude 152859036.2400 E, and a roadside
South Brisbane site located at latitude 2782905.279400S and
longitude 15381055.6400E. The increasing population of Brisbane
should result in more anthropogenic sources and additional
pollution. South-east Queensland has high rainfall, winds and
humidity during the warm (summer) part of the year (November
to April), with dry, cold, low wind speeds during winter (May to
August). Stronger winds tend to prevail during summer and
reduce the pollution in the area, whereas during winter the
pollution does not disperse. This is accompanied by low altitude
temperature inversions that trap the pollution.
[19]
According
to wind roses available at the website of the Bureau of
Meteorology (http://www.bom.gov.au/climate/averages/wind/
selection_map.shtml, accessed 16 June 2011), the summer is
dominated by south-south-easterly winds at 0900 hours and
east-north-easterly at 1500 hours, whereas the winter is domi-
nated by south-south-westerly at 0900 hours and west with east-
north-easterly at 1500 hours. Additional pollution enhancing
conditions in the area include the location of Brisbane city
within a valley with hills to the west, a temperature difference
between the land and sea, day–night stable wind cycles and the
location of industries in prevailing wind directions.
[24]
During
the period September–December, strong westerly winds from
the desert in the centre of Australia can cause dust storms. These
conditions affect the sources around the analysis site and
contribute to pollution conditions.
The sites were located in areas with numerous potential
sources that could influence the detected chemical levels
(Figs 1, 2). This includes major roads and highways, which lead
to an increase in the levels of the emissions from passing motor
vehicles. The Brisbane River and Oxley Creek are close to the
South Brisbane and Rocklea sites respectively and this may
indicate the proximity of potential sources of marine aerosols.
Biomass burning is a major concern in south-east Queensland
because of the practice of controlled burning, which is under-
taken in the dry stable winter period in an attempt to pre-
emptively reduce the risk of forest fires during summer. Railway
lines are also located close to the sites. The Brisbane Market is
,700 m from the Rocklea site. In addition to passenger vehicles
travelling along the adjacent road on weekends when the market
Rocklea site
Roads
Brisbane markets
Railways
Oxley Creek
Residential
Residential
South Brisbane
Rocklea
Fig. 1. Map of Rocklea site and surrounding areas.
Railways
Roads
Residential
Botanical
Gardens
Brisbane River
South-East
Freeway
South Brisbane
site
Residential
CBD
N
Fig. 2. Map of South Brisbane site and surrounding areas.
A. J. Friend et al.
570
operates, significant numbers of diesel trucks deliver produce to
the market.
The South Brisbane site was located between the Pacific
Motorway and the Stanley Street exit. Grassland and farming
areas surround the Rocklea site off Sherwood Road. PM
2.5
sampling was conducted with an IMPROVE cyclone sampler
adapted for the Aerosol Sampling Project (ASP) (Australian
Nuclear Science and Technology Organisation (ANSTO), Syd-
ney, NSW) using Teflon filters (PALL Life Science, Pall Corp.,
Ann Arbor, MI) at a height of 1.5 m.
[25]
Samples were collected
from midnight to midnight on Wednesdays and Sundays be-
tween June 2007 and June 2010 (Rocklea site) and May 2009
and June 2010 (South Brisbane site). This resulted in 316 and
109 samples collected from Rocklea and South Brisbane respec-
tively. Summary statistics are provided in Tables 1 and 2.
Elemental chemical analysis was then performed on the filters
at the Australian Nuclear Science and Technology Organisation
(ANSTO) in Sydney, Australia. Ion beam analysis using the
STAR accelerator (2.0-MV HVEE tandetron, High Voltage
Engineering Europa, Amersfoort, the Netherlands) was per-
formed to determine 20 chemical species
[25,26]
(H, Na, Al, Si,
P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Br and Pb) with
black carbon (BC) measured using a laser technique.
[27]
Before modelling techniques were applied, some of the data
points were removed because of the requirements of the PMF
model or other issues including mechanical faults and human
error during sampling. Samples were removed if the flow rate
was too low or if the filter was doubly exposed (sampled for 48h
instead of 24 h). By examining time series plots of each of the
elements, it became clear that some samples were outliers and
would affect the results of the modelling. Some of these samples
were removed if there was a known reason for their excessively
high concentrations. This data was then examined for trends at
each site and differences between the sites (suburban and
roadside sites).
Outliers and samples with excessively high and low flow
rates were also removed before the modelling was performed.
Consequently, 9 and 6 samples were removed from the Rocklea
and South Brisbane data, and this resulted in the retention of 307
samples for the PMF analysis at Rocklea and 103 for South
Brisbane.
PMF data analysis
To determine the sources of the chemical species examined in
the air samples, an assumption of mass balance (that the con-
centrations determined at the sites originated from sources) was
used so that the PMF could be applied. This is represented by the
following equation for an air sample analysed for each of the
chemical species:
wij ¼X
p
k¼1
gik fkj þeij ð1Þ
where i¼1, y,n;j¼1, y,m;k¼1, y,p;x
ij
is the jth
species concentration in the ith sample; g
ik
is the particulate
mass concentration from the kth source contributing to the ith
sample; f
kj
is the jth species mass fraction from the kth source;
and e
ij
is a residual associated with the jth species concen-
tration measured in the ith sample.
[28]
By applying the model to the samples collected and analysed,
the identified factors represent the sources of pollution prevalent
in the area at the time of sampling. An nnumber of samples by
mnumber of elements matrix is decomposed to determine
pindependent sources. These sources are defined by matrices
of g
ik
and f
kj
values representing the mass contributed from a
source to the collected samples and the profile of chemical
species present in a source respectively.
PMF is a receptor model developed in response to problems
associated with measurement uncertainties and rotational
Table 1. Summary statistics for the Rocklea site
Geometric mean and standard deviation (s.d.) are contained within parentheses
Chemical species Arithmetic mean (geometric mean) s.d. (geometric s.d.) Median Data zero or missing
(ng m
3
) (ng m
3
) (%)
H 164.7 (130.2) 130.5 (2.1) 122.1 0.0 %
Na 330.8 (262.5) 237.0 (2.1) 272.8 18.2 %
Al 41.6 (10.1) 236.5 (3.2) 11.2 2.8 %
Si 133.4 (50.0) 698.1 (2.6) 46.1 0.0 %
P 2.9 (1.7) 3.1 (3.4) 1.8 11.0%
S 282.4 (242.6) 159.4 (1.7) 245.7 0.0 %
Cl 231.5 (73.0) 300.9 (5.4) 105.0 1.9 %
K 54.7 (41.7) 83.9 (2.0) 39.9 0.0 %
Ca 25.0 (18.7) 46.1 (1.6) 19.0 0.0 %
Ti 7.2 (2.9) 39.2 (2.5) 2.9 0.0 %
V 0.8 (0.5) 1.0 (2.8) 0.5 14.7%
Cr 0.7 (0.4) 1.3 (2.3) 0.4 12.9%
Mn 5.3 (2.3) 9.4 (4.0) 2.1 0.0 %
Fe 85.4 (40.5) 329.5 (3.0) 38.3 0.0 %
Co 0.7 (0.4) 2.3 (3.0) 0.4 25.1%
Ni 0.5 (0.4) 0.6 (2.5) 0.4 17.9%
Cu 2.0 (1.5) 1.9 (2.2) 1.5 1.9 %
Zn 15.5 (7.9) 39.5 (2.5) 8.5 0.0 %
Br 3.0 (2.2) 2.4 (2.0) 2.4 2.2 %
Pb 5.0 (3.8) 4.2 (2.0) 4.1 4.4 %
Black carbon 878.6 (769.4) 482.4 (1.7) 743.8 0.9 %
Mass 5953.7 (4850.9) 10201.1 (1.6) 4675.0 0.0 %
Source apportionment of PM
2.5
at two receptor sites in Australia
571
ambiguity.
[29,30]
Thus uncertainties are defined individually for
each of the data points. In this study, uncertainties were
calculated using the following equation
[31]
:
Uncertainty ¼concentration error þMDL ð2Þ
where MDL is the method detection limit and the error is the
determined error for each data value.
The PMF2 process initially developed by Paatero was used in
this study.
[15]
For PMF2, the results are resolved by alternating
regression fits where each row of source contribution is deter-
mined while the source profile is kept constant and each column
of source profile is determined while keeping the source
contribution constant. A non-negativity constraint is applied
for the input data, as well as the G and F results because in
environmental analysis it is more reasonable for the concentra-
tion to be positive. This also reduces the rotational ambiguity in
the results.
An object function (Q) was used to first identify the
potential number of factors by minimising and comparing with
an ideal Q:
Q¼X
n
i¼1
X
m
j¼1
ðeij=sij Þ2ð3Þ
where s
ij
is an uncertainty estimate in a data point with the jth
element measured in the ith sample and e
ij
is the amount of
measured mass that was not explained by the model.
[32]
Source profiles (F matrix) and source contributions
(G matrix) are calculated by performing multilinear regression:
wij ¼X
p
k¼1
ðskgik Þðfkj=skÞð4Þ
where s
k
is determined by regressing the total PM
2.5
mass con-
centration in the ith sample against estimated source contribu-
tion values.
[28]
The determined F, G and regression coefficient
values are important in determining the number of factors.
The ideal number of factors (sources) were determined by
attempting to fit the data with a defined number of factors,
examining the results and then comparing with alternative
results obtained using other numbers of factors. Some research-
ers have outlined how to determine the number of factors.
[33]
The determined Qvalue was compared with an ideal Qvalue,
with the latter defined as the degrees of freedom of the data
matrix.
[34]
As part of PMF2, the parameter FPeak is available to
reduce the rotational ambiguity of the results.
[35]
The minimum
Qvalue is then determined and compared with the global
minimum Qvalue.
[36]
If the coefficients determined during
the multilinear regression in Eqn 4 are negative, it then indicates
that too many factors have been selected.
[36,37]
The factors are
examined to see if they are ‘physically reasonable’. This is
performed by comparing the identified sources with previous
PMF analysis at other locations, characteristic elements defined
by measurements taken at the same type of source and the
potential sources around the site. Also, if too many or too few
factors are used, sources may be split or combined into non-
existent sources.
[36]
Conditional probability function (CPF) analysis
Meteorological measurements of wind speed and direction were
combined with results from the PMF analysis to determine the
direction of the sources around the site using the following
equation
[38]
:
CPF ¼mDy
nDy
ð5Þ
Table 2. Summary statistics for the South Brisbane site
Geometric mean and standard deviation (s.d.) are contained within parentheses
Chemical species Arithmetic mean (geometric mean) s.d. (geometric s.d.) Median Data zero or missing
(ng m
3
) (ng m
3
) (%)
H 235.1 (183.7) 255.6 (2.1) 159.3 0.0%
Na 332.1 (284.1) 181.8 (2.2) 301.3 27.5%
Al 104.2 (21.8) 474.6 (3.0) 21.5 0.0 %
Si 324.7 (89.1) 1428.4 (2.6) 81.7 0.0 %
P 1.9 (1.1) 2.1 (3.0) 1.2 15.6 %
S 291.9 (257.4) 156.2 (1.6) 255.5 0.0%
Cl 246.9 (73.5) 294.0 (3.8) 147.1 0.9 %
K 81.8 (54.1) 177.1 (2.1) 49.7 0.0 %
Ca 50.2 (33.7) 104.0 (1.6) 32.4 0.0 %
Ti 20.7 (8.2) 87.3 (2.1) 7.7 0.0 %
V 1.3 (0.8) 2.3 (2.5) 0.9 4.6 %
Cr 0.8 (0.6) 0.9 (2.1) 0.6 5.5 %
Mn 4.9 (2.5) 14.3 (2.3) 2.1 0.0 %
Fe 241.9 (132.3) 761.2 (1.9) 124.5 0.0 %
Co 1.3 (0.6) 5.2 (1.7) 0.6 9.2 %
Ni 0.4 (0.3) 0.3 (2.3) 0.4 11.9 %
Cu 9.9 (8.0) 7.7 (1.9) 7.8 0.0 %
Zn 14.2 (10.1) 13.1 (2.1) 10.7 0.0 %
Br 3.8 (3.2) 2.2 (1.5) 3.5 0.0 %
Pb 5.6 (4.5) 3.8 (1.9) 4.6 0.9 %
Black carbon 1529.4 (1399.1) 671.0 (1.6) 1403.5 1.8 %
Mass 10225.2 (7245.2) 24867.3 (1.5) 6741.0 0.0 %
A. J. Friend et al.
572
where m
Dy
is the number of events from wind sector Dythat are
greater than the 75th percentile of the fractional contribution
from each source; and n
Dy
is the total number of events from the
same wind sector.
[39]
The wind direction bins (Dy) were set at an
angle of 248in this study so that there were 15 wind sectors.
Calm wind speed samples (,1ms
1
) were excluded from the
analysis.
The obtained outcomes that could be examined from the
PMF and CPF analysis include: the F matrix, which shows
the chemical species responsible for the sources (known as the
source profiles); and the G matrix of mass contributed from a
source to an individual sample, which can then be taken over the
entire sampling period and shown as a time series. From the
G matrix results, an average mass over a time period can be
determined and compared (e.g. by contrasting the average
summer concentration with the average winter concentration).
So, seasonal trends and differences between weekdays (Monday
to Friday) and weekends (Saturday and Sunday) can be ascer-
tained. The likely locations of the sources are indicated by the
high CPF values and displayed as a polar plot. Finally, the
percentage contribution (i.e. the mass contributed from an
individual source compared to that from all of the sources
combined) establishes how important the source is at a particular
site. From these results, the difference between two sites can be
evaluated.
Results and discussion
Basic data evaluation was performed to examine trends in the
data. Fig. 3 shows the PM
2.5
concentration over the sampling
period for both the Rocklea and South Brisbane sites. It was
observed that the same 2 days for both sites exceeded the NEPC
standard for 24-h PM
2.5
concentrations of 25 mgm
3
. These
samples were taken during severe dust storms in the area and this
was corroborated by the high concentrations for the character-
istic soil elements of Al, Si, K, Ca and Ti. Because of the outlier
nature of these points, they were excluded from the PMF
analysis.
These samples also affected the yearly average. In 2009–10
(June 2009–July 2010) for the South Brisbane site, the annual
average PM
2.5
level exceeded the NEPC standard of 8 mgm
3
.
Statistical analysis showed that the mean difference between
the years and sites had no statistical significance at the 0.05
level. However, after these samples were removed from the
data, the average concentration decreased to become lower
than the standard (from 10.2 4.6 mgm
3
(95 % confidence
interval, CI) to 7.6 0.7 mgm
3
). For the Rocklea site, the
average PM
2.5
concentration was lowest for the 2008–09 period
(4.8 0.3 mgm
3
); however, overall there was no significant
difference between these years. South Brisbane had a higher
average annual concentration compared to Rocklea (Rocklea
2009–10 ¼5.6 0.6 mgm
3
, South Brisbane 2009–10 ¼7.6
0.7 mgm
3
) and the difference between the sites was statistically
significant.
The average monthly PM
2.5
concentration (for June 2009–
June 2010) illustrated in Fig. 4 showed that between July and
September there was a significant increase in concentrations.
During this period, the conditions in the area were more stable
with lower wind speeds and less rainfall, which causes
the pollution to be dispersed more slowly. The sources of the
particles can be affected by this condition in different ways as
demonstrated in the modelling results.
The number of factors identified in the analysis was six
factors for the Rocklea site and five factors for the South
Brisbane sites. The Qvalues determined were 5026 for Rocklea
and 2743 for South Brisbane compared to the ideal Qvalues of
4479 and 1543 respectively. This is a ratio of 1.12 and 1.78 for
the two sites. FPeak was determined to be 0 for the sites by
varying the value and examining the Qvalues and results
produced. The scaled residuals were also examined to find out
whether the values were between 3 and 3. Both sites had five
sources in common with an additional motor vehicle source
identified at Rocklea. Figs 5 and 6 are the source profile results
for the sites. Table 3 contains the seasonal and weekly variation
with percentage contribution results. Table 4 contains the
percentage contribution values for each season and Table 5
has the weekday and weekend percentage contributions for the
sources. These trends are very similar to those observed in the
contribution comparisons. The trends were established by
averaging the source contribution values over a period of time
0
5
10
15
20
25
30
35
40
PM2.5 concentration (µg m⫺3)
Date
01 Jun 2007 01 Dec 2007 01 Jun 2008 01 Dec 2008 01 Jun 2009 01 Dec 2009 01 Jun 2010
Rocklea Standard South Brisbane
Fig. 3. Time series plot for the PM
2.5
concentration (mgm
3
) for the Rocklea and South Brisbane sites.
Source apportionment of PM
2.5
at two receptor sites in Australia
573
H
Na
Al
Si
P
S
Cl
K
Ca
Ti
V
Cr
Mn
Fe
Co
Ni
Cu
Zn
Br
Pb
BC
0.0001
0.001
0.01
0.1
1
0.0001
0.001
0.01
0.1
1
Source fraction
0.0001
0.001
0.01
0.1
1
0.0001
0.001
0.01
0.1
1
0.0001
0.001
0.01
0.1
1
H
Na
Al
Si
P
S
Cl
K
Ca
Ti
V
Cr
Mn
Fe
Co
Ni
Cu
Zn
Br
Pb
BC
0.0001
0.001
0.01
0.1
1
Sea salt
Soil
Biomass burning
Secondary sulfate
Motor vehicle 1
Motor vehicle 2
Fig. 5. Source profile for the Positive Matrix Factorisation (PMF) analysis for the Rocklea
site. Error bars represent 95 % confidence interval.
0
2
4
6
8
10
12
14
16
18
Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec
PM2.5 concentrattion (µg m⫺3)
Month
Rocklea South Brisbane
Fig. 4. Average monthly PM
2.5
concentration for the Rocklea and South Brisbane sites from June 2009 to June
2010. Error bars represent 95 % confidence interval.
A. J. Friend et al.
574
H
Na
Al
Si
P
S
Cl
K
Ca
Ti
V
Cr
Mn
Fe
Co
Ni
Cu
Zn
Br
Pb
BC
0.0001
0.001
0.01
0.1
1
0.0001
0.001
0.01
0.1
1
Source fraction
0.0001
0.001
0.01
0.1
1
0.0001
0.001
0.01
0.1
1
H
Na
Al
Si
P
S
Cl
K
Ca
Ti
V
Cr
Mn
Fe
Co
Ni
Cu
Zn
Br
Pb
BC
0.0001
0.001
0.01
0.1
1
Sea salt
Soil
Biomass burning
Secondary sulfate
Motor vehicle
Fig. 6. Source profile for the Positive Matrix Factorisation (PMF) analysis for the South
Brisbane site. Error bars represent 95 % confidence interval.
Table 3. Seasonal and weekly trends with percentage contributions for the identified sources
Statistical significance at the 0.05 confidence level is shown in parentheses with a value less than 0.05 considered to be significant
Source identified Seasonal variation Weekly variation Percentage contribution (%)
Rocklea South Brisbane Rocklea South Brisbane Rocklea South Brisbane
Sea salt Summer Summer Constant Constant 13.42 10.78
(2.6 10
7
) (0.01) (0.41) (0.67)
Soil Spring Spring Weekday Weekday 6.36 6.51
(0.008) (0.05) (0.25) (0.4)
Biomass burning Winter Winter–spring Weekend Weekend 41.8 22.05
(1.5 10
12
) (1.3 10
5
) (0.03) (0.21)
Secondary sulfate Spring–summer Spring–summer Constant Weekend 24.57 30.98
(9.0 10
9
) (0.15) (0.48) (0.08)
Motor vehicle 1 Winter Winter Weekday Weekday 6.93 29.68
(4.8 10
6
) (6.6 10
6
) (1.8 10
19
) (2.810
7
)
Motor vehicle 2 Winter Weekday 6.91
(9.6 10
4
) (1.2 10
11
)
Source apportionment of PM
2.5
at two receptor sites in Australia
575
(e.g. average for summer) and compared to the other seasons.
For the purpose of these comparisons, the seasons were sepa-
rated as summer (December–February), autumn (March–May),
winter (June–August) and spring (September–November)
whereas weekdays (represented by Wednesday) were compared
with weekends (represented by Sunday). The statistical signifi-
cance was determined for each of the trends by applying one-
way ANOVA with a significance level of 0.05 for the seasonal
and two tailed t-test for the weekly variation. A value less than
0.05 indicated that the difference between the means was
statistically significant. The CPF analysis results are shown in
Figs 7 and 8, and the six identified sources, their trends, possible
locations, differences between the methods and sites and finally,
differences between this period of sampling and that described
by Friend et al.
[23]
are outlined below.
Sea salt or marine aerosol was identified as the first source, as
sodium and chlorine had the highest determined concentra-
tions.
[40]
Summer was found to have a significantly higher
average contribution because of the increased formation of sea
salt, as higher temperatures and higher wind speeds lead to more
sea spray. In addition, the warmer conditions of the land over the
sea contributed to this trend. Consequently, the wind direction
changes so that sea spray is blown towards the land. The wind
changes from south-west in winter to east from the ocean in
summer. No significant difference was determined between
weekdays and weekends, which is reasonable as marine aerosol
is a natural source and anthropogenic activities would have little
effect. This source had the third highest contribution at Rocklea
and the fourth highest at South Brisbane. CPF analysis showed
that both sites indicated the Pacific Ocean to the east of the sites
as the likely location. These CPF results are also consistent with
those observed in an earlier study conducted at Rocklea by
Friend et al.
[23]
The second source identified consistently across both analy-
sis and sites contained Al, Si, K, Ca, Ti and Fe, which were
characteristic of a soil source.
[40]
Some differences were
observed in the other elements measured but the characteristic
elements strongly indicate this identity. Seasonal trends found
that spring was the period with the highest average contribution.
This is consistent with the monthly average basic trends analysis
and the time series, which showed large peaks in September.
Dust storms are frequent during this period as indicated by the
outlier event on 23 September 2009. A higher average weekday
trend was observed but it was not statistically significant. Soil
was the lowest contributor at both sites. This relatively low value
may be attributed to soil events such as occasional dust storms
rather than a consistent source. The Rocklea site was located in a
grassed area with limited activity that would increase soil
suspension. Friend et al.
[23]
identified a similar source, however,
higher hydrogen and BC concentrations were observed possibly
indicating the influence of road dust. Also, winter was the
highest contributing season and weekdays were significantly
higher than weekends. This may be because of the influence of
road dust that would come from motor vehicle traffic travelling
along nearby roads. A similar percentage contribution was
observed for both sites. The CPF analysis indicated west and
north as the direction of the source, but as the contributions for
both sites were dominated by a few high soil days and all of these
days indicate the south-west to west, long distance dust travel-
ling from the desert may be a major contributor.
For the third source, H, K, S and BC were the elements with
the highest concentrations. It is well known that potassium is
characteristic of a biomass burning source.
[32]
However, other
elements such as Br also had relatively high fractions in the
source profile, possibly indicating an influence from motor
vehicles. This is especially true at the South Brisbane site,
which is close to a major freeway. A significantly higher
seasonal average concentration was observed in winter and is
because of the stable weather conditions in the area during this
period caused by low windspeeds and low rainfalls. Brisbane’s
climate is too warm even in winter for indoor burning to be a
significant influence but forest fires can be a problem in the area.
To reduce the likelihood of forest fires during summer, the
government routinely practices controlled burning of the poten-
tially vulnerable areas. This is practiced during the winter. PMF
identified this as the second highest contributor behind second-
ary sulfate at South Brisbane and the highest contributor at
Rocklea consistent with the State of the Environment report
released by the Queensland Government every four years.
[8,41]
Also, at the Rocklea site, this method identified a higher average
contribution on the weekend. Gildemeister et al.
[42]
also identi-
fied a weekend trend but it was not statistically significant,
whereas Kim et al. identified a slightly significant weekend
trend but attributed this to residential heating.
[43]
CPF analysis
showed the prominent direction at Rocklea to be to the east with
a peak to the north, whereas at South Brisbane, the peaks
occurred in most directions, from south-west to north-east and
Table 4. Seasonal trends in percentage contributions for the identified sources
Source identified Rocklea South Brisbane
Summer Autumn Winter Spring Summer Autumn Winter Spring
Sea salt 20.80 % 11.79 % 7.81 % 13.56 % 17.82 % 11.27 % 3.73 % 11.67 %
Soil 4.29 % 3.70 % 4.24 % 10.54 % 2.82 % 2.97 % 6.00 % 10.77 %
Biomass burning 23.60 % 34.92 % 60.72 % 39.30 % 11.69 % 16.96 % 27.75 % 29.62 %
Secondary sulfate 39.48 % 29.31 % 11.00 % 24.55 % 38.32 % 30.43 % 28.16 % 20.73 %
Motor vehicle 1 5.40% 8.37 % 9.37 % 6.42 % 29.36 % 38.38 % 34.36 % 27.23 %
Motor vehicle 2 6.44% 11.91 % 6.86 % 5.63 %
Table 5. Weekly trends in percentage contributions for the identified
sources
Source identified Rocklea South Brisbane
Weekday Weekend Weekday Weekend
Sea salt 12.79 % 14.83 % 10.50% 10.99 %
Soil 7.90 % 4.58 % 8.39 % 4.79 %
Biomass burning 37.21 % 46.13 % 19.92 % 24.00 %
Secondary sulfate 21.02 % 28.52 % 26.51 % 35.31 %
Motor vehicle 1 10.64 % 2.18 % 34.68 % 24.90 %
Motor vehicle 2 10.45 % 3.77 %
A. J. Friend et al.
576
a peak to the south-east. This may indicate that this source is
located in multiple directions. Biomass burning was found to be
the highest contributor at a site close to the Rocklea site.
[23]
Secondary sulfate, sulfate that has reacted in the air, was
identified as the fourth source. H, S, Na and BC were consis-
tently observed at both of the sites and are characteristic for
secondary sulfate.
[44]
A significant difference between the sites
was observed for the seasonal average concentrations. Spring
and summer were found to be significantly higher than the other
seasons at the Rocklea site whereas the same trend was found at
the South Brisbane site, but it was not statistically significant.
A summer variation would be consistent with known trends in
secondary sulfate because sunlight is important for the reaction
of chloride to form sulfate in the atmosphere. Secondary sulfate
Secondary sulfate
0.00
0.25
0.50
0.000.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Vehicle 1
0.00
0.25
0.50
0.00
0.250.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Biomass burning
0.00
0.25
0.50
0.000.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Vehicle 2
0.00
0.25
0.50
0.00
0.250.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Soil
0.00 0.25 0.50
0.00
0.25
0.50
0.00
0.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Sea salt
0.00 0.25 0.50
0.00 0.25 0.500.00 0.25 0.50
0.00 0.25 0.500.00 0.25 0.50
0.00
0.25
0.50
0.00
0.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Fig. 7. Conditional probability function (CPF) results for the Rocklea site.
Source apportionment of PM
2.5
at two receptor sites in Australia
577
is a regional source,
[45]
and owing to the location of the Rocklea
site in an open area, it may be more easily influenced by this
source. There was no significant difference between the week-
days and weekend observed, which is consistent with a second-
ary source. Again PMF determined secondary sulfate to be the
highest contributing source at the South Brisbane site and the
second highest at Rocklea. Peaks to the north and north-east at
the Rocklea site and to the north-east at the South Brisbane site
were found in the CPF analysis. Power generation and
petroleum refining is the main source of sulfur in Brisbane,
[8]
although there is an oil refinery located to the north-east of the
South Brisbane site. In the previous analysis of the data from
the Rocklea site, a similar source was identified as a mixture of
aged sea salt and secondary sulfate.
[23]
The high concentration
of sodium in the source profile would be consistent with the
chloride in sea salt reacting in the atmosphere.
The fifth and sixth sources were identified as motor vehicles
because of the high concentration of BC. Fe was the other major
Secondary sulfate
0.00 0.25 0.50
0.00
0.25
0.50
0.000.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Motor vehicle
0.00 0.25 0.50
0.00
0.25
0.50
0.00
0.250.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Biomass burning
0.00 0.25 0.50
0.00
0.25
0.50
0.000.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Soil
0.00 0.25 0.50
0.00
0.25
0.50
0.00
0.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Sea salt
0.00 0.25 0.50
0.00
0.25
0.50
0.00
0.25
0.50
0.00
0.25
0.50
0
30
60
90
120
150
180
210
240
270
300
330
Fig. 8. Conditional probability function (CPF) results for the South Brisbane site.
A. J. Friend et al.
578
element identified in the fifth source and was observed at both
sites and so this factor may be from petrol vehicles. The sixth
source, however, had zinc, bromine and sulfur, which indicated
that the motor vehicle source was from diesel vehicles. The high
iron concentration in this source may also indicate a railway
influence as this source was only observed at the Rocklea site.
As expected, Motor vehicle 1 was significantly higher at the
South Brisbane site than at Rocklea. Passenger vehicles may
be responsible for this source and the close proximity of the
motorway to the South Brisbane site would support this inter-
pretation. Motor vehicle 2 was not observed in the data at the
second site. Both sources showed the same trends with high
average concentrations in the winter. The stable weather con-
ditions during the winter affect motor vehicle emissions and trap
them for a longer period of time. Concentrations were signifi-
cantly higher during the weekdays in keeping with the higher
volumes of vehicle traffic as people travel to and from work
during the week. CPF analysis showed that motor vehicle 1 was
from the north of the Rocklea site, which is the direction of the
road, whereas for the South Brisbane site the peak was to the
west consistent with the location of a major highway to the west
of this site. For motor vehicle 2 the peak at Rocklea was from the
south and to the east. A railway exchange is located to the east
with the Ipswich Motorway far to the south. Very similar
sources were observed by Friend et al.,
[23]
with similar trends
and percentage contributions for the Rocklea site although it was
identified as railway.
Conclusions
The concentrations of PM
2.5
elements were determined for two
sites in the south-east Queensland region and analysed using the
receptor model PMF. These analyses were performed to identify
the sources in the area and determine how influential they were
on the ambient pollution levels. Based on the PMF results, the five
common sources of PM
2.5
at both sites were motor vehicle
emissions, biomass burning, secondary sulfate, sea salt and soil,
with secondary sulfate as the most significant contributor of PM
2.5
aerosols at the South Brisbane site and biomass burning the most
significant at the Rocklea site. In addition, significant dust storms
that caused the PM
2.5
concentration to exceed the NEPC standard
were observed at both sites during the sampling period. However,
Rocklea and South Brisbane sites showed the most significant
differences in the motor vehicle sources. Owing to the close
proximity of the highway to the South Brisbane site, the common
vehicle emission source had a significantly higher contribution to
this site.A second vehicle emission wasidentified at Rocklea and
this is most likely because of the prevalence of diesel trucks at the
Brisbane market, which is close to this site. CPF analysis com-
bined meteorological data with the PMF results to determine the
locations of the sources. It is important to have multiple sampling
sites around a city to compare the different sampling environ-
ments (urban, rural or roadside). The results highlight the need to
take both regional and local sources of pollution into consider-
ation when formulating pollution control strategies.
Acknowledgements
The authors acknowledge the work of the Australian Nuclear Science and
Technology Organisation (ANSTO) in the analysis. The Queensland
Department of Environmental Management (QDERM) is acknowledged for
allowing access to the sites and additional assistance. Funding for the
sampling and analysis was provided by the Australian Institute of Nuclear
Science and Engineering (AINSE).
References
[1] Environmental and Workplace Health – Health Effects of Air Pollu-
tion 2006 (Health Canada). Available at http://www.hc-sc.gc.ca/
ewh-semt/air/out-ext/effe/health_effects-effets_sante_e.html [Verified
23 December 2010].
[2] H. Shaka’, N. A. Saliba, Concentration measurements and chemical
composition of PM
10–2.5
and PM
2.5
at a coastal site in Beirut, Lebanon.
Atmos. Environ. 2004,38, 523. doi:10.1016/J.ATMOSENV.
2003.10.009
[3] P. Brimblecombe, The Effects of Air Pollution on the Built Environ-
ment 2003 (Imperial College Press: London).
[4] S.-T. Kim, Y. Maeda, Y. Tsujino, Assessment of the effect of air
pollution on material damages in northeast Asia. Atmos. Environ.
2004,38, 37. doi:10.1016/J.ATMOSENV.2003.09.045
[5] S. J. Dutton, S. Vedal, R. Piedrahita, J. B. Milford, S. L. Miller, M. P.
Hannigan, Source apportionment using positive matrix factorization
on daily measurements of inorganic and organic speciated PM
2.5
.
Atmos. Environ. 2010,44, 2731. doi:10.1016/J.ATMOSENV.2010.
04.038
[6] National Environmental Protection (Ambient Air Quality) Measure
1998 (National Environmental Protection Council: Adelaide).
[7] Variation to the National Environment Protection (Ambient Air
Quality) Measure for Particles as PM 2003 (National Environmental
Protection Council: Adelaide).
[8] J. Freeman, W Webber (Eds), State of the Environment Queensland
2007 (National Environmental Protection Council: Brisbane).
[9] P. Khare, B. P. Baruah, Elemental characterization and source
identification of PM
2.5
using multivariate analysis at the suburban
site of north-east India. Atmos. Res. 2010,98, 148. doi:10.1016/
J.ATMOSRES.2010.07.001
[10] S. Raja, K. F. Biswas, L. Husain, P. K. Hopke, Source apportionment
of the atmospheric aerosol in Lahore, Pakistan. Water Air Soil Pollut.
2010,208, 43. doi:10.1007/S11270-009-0148-Z
[11] J.-M. Lim, J.-H. Lee, J.-H. Moon, Y.-S. Chung, K.-H. Kim, Source
apportionment of PM
10
at a small industrial area using positive matrix
factorization. Atmos. Res. 2010,95, 88. doi:10.1016/J.ATMOSRES.
2009.08.009
[12] P. Lestari, Y. D. Mauliadi, Source apportionment of particulate matter
at urban mixed site in Indonesia using PMF. Atmos. Environ. 2009,43,
1760. doi:10.1016/J.ATMOSENV.2008.12.044
[13] G. D. Thurston, J. D. Spengler, A quantitative assessment of source
contributions to inhalable particulate matter pollution in metropolitan
Boston. Atmos. Environ. 1985,19, 9. doi:10.1016/0004-6981(85)
90132-5
[14] R. C. Henry, Multivariate receptor modeling by N-dimensional edge
detection. Chemom. Intell. Lab. Syst. 2003,65, 179. doi:10.1016/
S0169-7439(02)00108-9
[15] P. Paatero, U. Tapper, Positive matrix factorization – a nonnegative
factor model with optimal utilization of error-estimates of data values.
Environmetrics 1994,5, 111. doi:10.1002/ENV.3170050203
[16] R. Tauler, M. Viana, X. Querol, A. Alastuey, R. M. Flight, P. D.
Wentzell, P. K. Hopke, Comparison of the results obtained by four
receptor modelling methods in aerosol source apportionment studies.
Atmos. Environ. 2009,43, 3989. doi:10.1016/J.ATMOSENV.2009.
05.018
[17] J. G. Watson, L. W. A. Chen, J. C. Chow, P. Doraiswamy, D. H.
Lowenthal, Source apportionment: findings from the US Supersites
Program. J. Air Waste Manage. Assoc. 2008,58, 265. doi:10.3155/
1047-3289.58.2.265
[18] Y. C. Chan, R. W. Simpson, G. H. Mctainsh, P. D. Vowles, D. D.
Cohen, G. M. Bailey, Source apportionment of PM
2.5
and PM
10
aerosols in Brisbane (Australia) by receptor modelling. Atmos.
Environ. 1999,33, 3251. doi:10.1016/S1352-2310(99)00090-4
[19] Y.-C. Chan, D. D. Cohen, O. Hawas, E. Stelcer, R. Simpson,
L. Denison, N. Wong, M. Hodge, E. Comino, S. Carswell, Apportion-
ment of sources of fine and coarse particles in four major Australian
cities by positive matrix factorisation. Atmos. Environ. 2008,42, 374.
doi:10.1016/J.ATMOSENV.2007.09.030
Source apportionment of PM
2.5
at two receptor sites in Australia
579
[20] A. J. Friend, G. A. Ayoko, Multi-criteria ranking and source appor-
tionment of fine particulate matter in Brisbane, Australia. Environ.
Chem. 2009,6, 398. doi:10.1071/EN09035
[21] Y.-C. Chan, O. Hawas, D. Hawker, P. Vowles, D. D. Cohen,
E. Stelcer, R. Simpson, G. Golding, E. Christensen, Using multiple
type composition data and wind data in PMF analysis to apportion and
locate sources of air pollutants. Atmos. Environ. 2011,45, 439.
doi:10.1016/J.ATMOSENV.2010.09.060
[22] E. Lee, C. K. Chan, P. Paatero, Application of positive matrix
factorization in source apportionment of particulate pollutants in
Hong Kong. Atmos. Environ. 1999,33, 3201. doi:10.1016/S1352-
2310(99)00113-2
[23] A. J. Friend, G. A. Ayoko, S. G. Elbagir, Source apportionment of fine
particles at a suburban site in Queensland, Australia. Environ. Chem.
2011,8, 163. doi:10.1071/EN10112
[24] S. Khan, R. Simpson, Mesoscale trajectory modeling for the Brisbane
airshed. Environ. Model. Assess. 1997,2, 201. doi:10.1023/
A:1019013428121
[25] D. D. Cohen, J. Crawford, E. Stelcer, V. T. Bac, Characterisation and
source apportionment of fine particulate sources at Hanoi from 2001
to 2008. Atmos. Environ. 2010,44, 320. doi:10.1016/J.ATMOSENV.
2009.10.037
[26] D. D. Cohen, E. Stelcer, O. Hawas, D. Garton, IBA methods for
characterisation of fine particulate atmospheric pollution: a local,
regional and global research problem. Nucl. Instrum. Methods Phys.
Res. B 2004,219–220, 145. doi:10.1016/J.NIMB.2004.01.043
[27] G. Taha, G. P. Box, D. D. Cohen, E. Stelcer, Black carbon measure-
ment using laser integrating plate method. Aerosol Sci. Technol. 2007,
41, 266. doi:10.1080/02786820601156224
[28] E. Kim, P. K. Hopke, E. S. Edgerton, Source identification of Atlanta
aerosol by positive matrix factorization. J. Air Waste Manage. Assoc.
2003,53, 731.
[29] R. C. Henry, Current factor analysis receptor models are ill-posed.
Atmos. Environ. 1987,21, 1815. doi:10.1016/0004-6981(87)90122-3
[30] P. Paatero, Least squares formulation of robust non-negative factor
analysis. Chemom. Intell. Lab. Syst. 1997,37, 23. doi:10.1016/S0169-
7439(96)00044-5
[31] S. Eberly, EPA PMF Version 1.1. 2005 (US Environmental Protection
Agency: Durham, NC).
[32] Y. Song, S. D. Xie, Y. H. Zhang, L. M. Zeng, L. G. Salmon, M. Zheng,
Source apportionment of PM
2.5
in Beijing using principal component
analysis/absolute principal component scores and unmix. Sci. Total
Environ. 2006,372, 278. doi:10.1016/J.SCITOTENV.2006.08.041
[33] R. Lorenzo, R. Kaegi, R. Gehrig, B. Grobety, Particle emissions of a
railway line determined by detailed single particle analysis. Atmos.
Environ. 2006,40, 7831. doi:10.1016/J.ATMOSENV.2006.07.026
[34] A. V. Polissar, P. K. Hopke, R. L. Poirot, Atmospheric aerosol over
Vermont: chemical composition and sources. Environ. Sci. Technol.
2001,35, 4604. doi:10.1021/ES0105865
[35] P. Paatero, P. K. Hopke, X.-H. Song, Z. Ramadan, Understanding and
controlling rotations in factor analytic models. Chemom. Intell. Lab.
Syst. 2002,60, 253. doi:10.1016/S0169-7439(01)00200-3
[36] D. Ogulei, P. K. Hopke, L. A. Wallace, Analysis of indoor particle size
distributions in an occupied townhouse using positive matrix factori-
zation. Indoor Air 2006,16, 204. doi:10.1111/J.1600-0668.2006.
00418.X
[37] D. Ogulei, P. K. Hopke, L. Zhou, J. Patrick Pancras, N. Nair, J. M.
Ondov, Source apportionment of Baltimore aerosol from combined
size distribution and chemical composition data. Atmos. Environ.
2006,40(Suppl. 2), 396. doi:10.1016/J.ATMOSENV.2005.11.075
[38] J. H. Lee, P. K. Hopke, Apportioning sources of PM
2.5
in St Louis,
MO, using speciation trends network data. Atmos. Environ. 2006,
40(Suppl. 2), 360. doi:10.1016/J.ATMOSENV.2005.11.074
[39] E. Kim, P. K. Hopke, Comparison between conditional probability
function and nonparametric regression for fine particle source direc-
tions. Atmos. Environ. 2004,38, 4667. doi:10.1016/J.ATMOSENV.
2004.05.035
[40] X. H. Song, A. V. Polissar, P. K. Hopke, Sources of fine particle
composition in the northeastern US. Atmos. Environ. 2001,35, 5277.
doi:10.1016/S1352-2310(01)00338-7
[41] State of the Environment Queensland 2003 (Queensland Environmen-
tal Protection Agency: Brisbane).
[42] A. E. Gildemeister, P. K. Hopke, E. Kim, Sources of fine urban
particulate matter in Detroit, MI. Chemosphere 2007,69, 1064.
doi:10.1016/J.CHEMOSPHERE.2007.04.027
[43] E. Kim, K. Turkiewicz, S. A. Zulawnick, K. L. Magliano, Sources of
fine particles in the South Coast area, California. Atmos. Environ.
2010,44, 3095. doi:10.1016/J.ATMOSENV.2010.05.037
[44] H. Guo, A. J. Ding, K. L. So, G. A. Ayoko, Y. S. Li, W. T. Hung,
Receptor modeling of source apportionment of Hong Kong aerosols
and the implication of urban and regional contribution. Atmos.
Environ. 2009,43, 1159. doi:10.1016/J.ATMOSENV.2008.04.046
[45] K. L. So, H. Guo, Y. S. Li, Long-term variation of PM
2.5
levels and
composition at rural, urban, and roadside sites in Hong Kong:
increasing impact of regional air pollution. Atmos. Environ. 2007,
41, 9427. doi:10.1016/J.ATMOSENV.2007.08.053
A. J. Friend et al.
580