1-s2 0-S1001074222001632-main

Prévia do material em texto

journal of environmental sciences 133 (2023) 8–22 
Available online at www.sciencedirect.com 
w w w . e l s e v i e r . c o m / l o c a t e / j e s 
Improving modeling of low-altitude particulate 
matter emission and dispersion: A cotton gin case 
study 
Zijiang Yang 1 , Michael N. Evans 2 , 3 , Michael D. Buser 4 , 
Cathleen J. Hapeman 5 , Alba Torrents 1 , Derek P. Whitelock 6 , ∗
1 Department of Civil and Environmental Engineering, University of Maryland College Park, 1173 Glenn L. Martin 
Hall, College Park, Maryland 20742, United States 
2 Department of Geology, University of Maryland College Park, 1212B Chemistry Building, College Park, Maryland 
20742, United States 
3 Earth System Science Interdisciplinary Center, 1120 Geology Building, University of Maryland, College Park, 
Maryland 20742, United States 
4 United States Department of Agriculture, Agricultural Research Service, Office of National Programs, 5601 
Sunnyside Ave (GWCC 4–2282), Beltsville, Maryland 20705, United States 
5 United States Department of Agriculture, Agricultural Research Service, Henry A. Wallace Beltsville Agricultural 
Research Center, 10300 Baltimore Avenue, Beltsville, Maryland 20705, United States 
6 United States Department of Agriculture, Agricultural Research Service, Southwestern Cotton Ginning Research 
Laboratory, PO Box 578, Mesilla Park, New Mexico 88047, United States 
a r t i c l e i n f o 
Article history: 
Received 7 December 2021 
Revised 30 March 2022 
Accepted 30 March 2022 
Available online 13 April 2022 
Keywords: 
Air quality 
Particulate matter 
Air dispersion modelling 
Cotton gin 
a b s t r a c t 
Monitoring and modeling of airborne particulate matter (PM) from low-altitude sources is 
becoming an important regulatory target as the adverse health consequences of PM become 
better understood. However, application of models not specifically designed for simulation 
of PM from low-altitude emissions may bias predictions. To address this problem, we de- 
scribe the modification and validation of an air dispersion model for the simulation of low- 
altitude PM dispersion from a typical cotton ginning facility. We found that the regulatory 
recommended model (AERMOD) overestimated pollutant concentrations by factors of 64.7, 
6.97 and 7.44 on average for PM 2.5 , PM 10 , and TSP, respectively. Pollutant concentrations were 
negatively correlated with height ( p < 0.05), distance from source ( p < 0.05) and standard de- 
viation of wind direction ( p < 0.001), and positively correlated with average wind speed ( p < 
0.001). Based on these results, we developed dispersion correction factors for AERMOD and 
cross-validated the revised model against independent observations, reducing overestima- 
tion factors to 3.75, 1.52 and 1.44 for PM 2.5 , PM 10 and TSP, respectively. Further reductions 
in model error may be obtained from use of additional observations and refinement of dis- 
persive correction factors. More generally, the correction permits the validated adjustment 
and application of pre-existing models for risk assessment and development of remedia- 
∗ Corresponding author. 
E-mail: derek.whitelock@usda.gov (D.P. Whitelock). 
https://doi.org/10.1016/j.jes.2022.03.048 
1001-0742/© 2022 The Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences. Published by Elsevier B.V. 
https://doi.org/10.1016/j.jes.2022.03.048
http://www.sciencedirect.com/science/journal/99999994
http://www.elsevier.com/locate/jes
mailto:derek.whitelock@usda.gov
https://doi.org/10.1016/j.jes.2022.03.048
journal of environmental sciences 133 (2023) 8–22 9 
tion techniques. The same approach may also be applied to improve simulations of other 
air pollutants and environmental conditions of concern. 
© 2022 The Research Center for Eco-Environmental Sciences, Chinese Academy of 
Sciences. Published by Elsevier B.V. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Introduction 
Particulate matter (PM) can lead to a variety of adverse health
effects, including asthma attacks, chronic bronchitis, cancer,
cardiovascular disease, diabetes, and other sources of mor-
bidity and mortality ( Löndahl et al., 2007 ; WHO, 2018 ), pos-
ing more danger to human health than ozone and carbon
monoxide ( Kim et al., 2015 ). For these reasons, PM with an
aerodynamic diameter less than or equal to 2.5 μm (PM 2.5 )
and PM with an aerodynamic diameter less than or equal
to 10 μm (PM 10 ) are regulated as ambient air pollutants.
As the adverse health consequences of PM become better
understood, regulatory standards for PM are becoming in-
creasingly strict ( US EPA, 2018 ). For example, in 2006 and
2012, the United States Environmental Protection Agency (US
EPA) twice revised the National Ambient Air Quality Stan-
dards (NAAQS) for PM ( Federal Register, 2006 ; Federal Reg-
ister, 2013 ), and most notably, the maximum allowable 24-
hour average PM 2.5 concentration was decreased from 65 to
35 μg/m 3 . 
Low-altitude emissions are of particular health and regu-
latory concern, especially if adjacent to population centers.
Although the term low-altitude emission has been frequently
used in the literature, it has been defined as no higher than
5 m ( Đor đevi ́c and Šolevi ́c, 2008 ), around 15 m ( Xin et al., 2011 ),
and not exceeding 40 m ( Piwowar and Dziku ́c, 2019 ). In this
paper, we define low-altitude emissions as occurring between
2 and 40 m, as is often the case in food and agricultural prod-
uct processing facilities (e.g., cotton gins, grain elevators, feed
mills) ( Whitelock et al., 2019 ; Venkataraman et al., 2018 ), ani-
mal operations ( Dai et al., 2020 ), distributed power generators
( Heath et al., 2005 ), and domestic combustion ( Piwowar and
Maciej, 2019 ). In the definition, 2 m is the upper limit for
ground-level emissions , such as parking lot ( Tomasi et al., 2015 ),
road and vehicles ( Misra et al., 2013 ; Brugge et al., 2007 ), an-
imal feeding houses with low-pitched roof ( Yao et al., 2018 ),
cattle feedlots ( Bonifacio et al., 2013 ), wastewater treatment
plant ( Behnami et al., 2019 ), and 40 m is the lower limit for
high-altitude emissions , such as industrial stacks ( Perry et al.,
2005 ). 
PM and other air pollutants from low-altitude emission
source have not been comprehensively investigated since air
pollutants from industrial stacks have been an overriding en-
vironmental concern. However, it is becoming increasingly im-
portant to understand the environmental fate of PM from low-
altitude emission sources and to assess their impact on hu-
man health and environmental quality. The optimal way to
investigate the dispersion of PM from low-altitude emission
sources is onsite monitoring. However, onsite measurement
is costly and time-consuming. Thus, air dispersion modelling
is used as an alternative method to assess the dispersion of
air pollutants and to determine if their concentrations down-
wind comply with local, regional, and national air quality
standards ( US EPA, 2021 ). This has created a pressing need for
models which can accurately simulate dispersion of air pol-
lutants in a variety of industrial, agricultural, and residential
settings. 
Among air dispersion models, the American Meteorolog-
ical Society (AMS)/Environmental Protection Agency (EPA)
Regulatory Model (AERMOD) is the regulatory recommended
model ( US EPA, 2019 ). AERMOD is a Gaussian-based steady-
state air dispersion model developed by the AMS/EPA Regu-
latory Model Improvement Committee ( US EPA, 2004 ). AER-
MOD is relatively simple and open source ( US EPA, 2019 ).
Thus, although AERMOD was originally designed for in-
dustrial stacks ( Cimorelli et al., 2005 ; Perry et al., 2005 ), it
is widely used for a variety of types of emissionsources
( Chen et al., 2009 ; Gibson et al., 2013 ; Jenkins et al., 2015 ;
Baawain et al., 2017 ; Damuchali and Guo, 2020 ; Liu and
Kim, 2019 ; Tong et al., 2020 ; Żeli ́nski et al., 2021 ; Tyovenda et al.,
2021 ), including low-altitude emissions from cotton gins
( Bairy et al., 2012 ), poultry facilities ( Hadlocon et al., 2015 ;
Pohl et al., 2017 ; Kelleghan et al., 2021 ), swine operations
( O’Shaughnessy and Altmaier, 2011 ), distributed electric gen-
eration facilities ( Yang et al., 2018 ), household solid fuel com-
bustion ( Mestl and Fang, 2003 ; Mestl et al., 2005 , 2006 ), and
transportation tunnels ( Onay et al., 2019 ). 
However, the evaluation and validation of models such as
AERMOD for the simulation of low-altitude emissions still
present important challenges, such as lack of onsite measured
emission factors and the lack of systematically designed sam-
pling campaigns. A better model evaluation and validation
study would include onsite measured emission factors and
systematic and targeted sampling campaigns. Otherwise, un-
certainties may be introduced into the modelling process, bias
the validation results and result in ineffective regulatory pro-
cedures and practical applications. 
Therefore, we conducted a sampling campaign to collect
PM emission, dispersion and associated meteorological data
from a typical cotton gin facility. As a case study to illustrate
the potential for improving model accuracy, we empirically
developed and validated correction factors for the simulation
of low-altitude emissions of PM from the cotton gin by AER-
MOD. We first report patterns in observations of the local en-
vironment that suggest systematic model correction factors.
We then apply atmospheric dispersion modelling with on-
site measured emission factors ( Boykin et al., 2013a , 2013b ;
Buser et al., 2013a , 2013b , 2013c ; Whitelock et al., 2013a , 2013b ,
2013c ) to evaluate the performance of the modified regula-
tory recommended model and compare the results to the un-
modified model simulations. We then discuss the potential to
use this simple, easy-to-use, and cost-effective method for the
improved simulation of particulate matter and other atmo-
spheric pollutants in a variety of industrial, agricultural, and
residential settings. 
10 journal of environmental sciences 133 (2023) 8–22 
1
1
A
U
s
p
w
a
b
p
i
d 
a
l
a
s 
a
i
t
s 
a
a
r
a
3
o
s
a
t
o
m
1
A
N
l
c
T
(
m
W
1
3
t
s
d
t
I
d 
2
1
P
( 
2 
p
l
l
m
t
t 
U
t
c
v
r
c
a
c
i
l
C
w
c
t
p
a
f
f
w
c
a
t
r
1
I 
L
M
m
m
s
M 
E
t
t
m
T 
2 
t
a
p
c
a
C
. Material and methods 
.1. Sampling campaign 
 sampling campaign was conducted in the Mid-South of the 
S from September 20 to 29, 2010 at a typical cotton gin with 
tack heights that ranged from about 9 to 12 m. Ten sampling 
eriods (experiments) of ca. 10 hr were conducted each day 
hile the gin was operating from 07:00 to 17:00 and are noted 
s EXP01 – EXP10. A test run was also carried out on Septem- 
er 13 from 11:00 to 24:00 (EXP00). The summary of the ex- 
eriment time periods is shown in Appendix A Table S1. Dur- 
ng preliminary calculations, the average meteorological con- 
itions (pressure, relative humidity, temperature, wind speed,
nd wind direction) of EXP00 were not identified as an out- 
ier based on probability density, so EXP00 was included in the 
nalysis. 
The cotton gin was surrounded by a sampling array con- 
isting of samplers located on three concentric circles 60, 120,
nd 180 m from the main cyclone bank at 30 ° intervals shown 
n Fig. 1 . Some samplers were moved or not installed due 
o site restriction (e.g., buildings and roads). The middle ring 
amplers were installed on towers at heights of 1, 2, 3, 4.5, 7.25,
nd 10 m, while only 2-m samplers were deployed on the inner 
nd outer rings. Samplers were designated with a letter for the 
ing (“I” for inner, “O” for outer, and “T” for middle towers) and 
 number for the bearing from North (1 for 0 ° or North, 2 for 
0 °, and so forth). An additional indicator for sampler height 
n the tower was also added (e.g., T01-1 for tower 01 at 0 ° and 
ampler at the first level or 1-m height and T07-5 for tower 07 
t 180 ° and sampler at the fifth level or 7.25-m high, etc.). Due 
o potential multidirectional wind and spatial heterogeneity 
f PM concentration, background PM concentration was not 
easured. For more discussion see Section 2.10 . 
.2. Meteorological data 
n onsite weather station was located approximately 200 m 
orth of the cotton gin. Onsite meteorological data were col- 
ected using a Hobo H21-001 weather station datalogger with 
ompatible sensors (Onset Computer Corporation, USA): (1) S- 
HA temperature and relative humidity sensor at 2-m height; 
2) S-TMB-M002 temperature sensors at 1, 3, 4.5, 7.25 and 10- 
 height; (3) S-BPA-CM10 barometric pressure sensor; (4) S- 
CA-M003 wind speed and direction smart sensor (for u < 
7 m/sec: ± 0.5 m/sec for 17 m/sec < u < 30 m/sec: ± 3%; for 
0 m/sec < u < 44 m/sec: ± 4%; ± 5 degrees for wind direc- 
ion); (5) S-LIB-M003 silicon pyranometer solar radiation sen- 
ors, one facing upward and one facing downward, for net ra- 
iation. Data were collected every 5 min. Surface hourly me- 
eorological data was from a station at Blytheville, AR (Station 
D 53869, 47 km from the site) and upper air meteorological 
ata was from the North Little Rock station (Station ID 3952,
62 km from the site). 
.3. Particulate matter measurements and analysis 
articulate matter concentration and particle size distribution 
PSD) were determined based on previous work ( Buser et al.,
007a , 2007b , 2007c ; Buser et al., 2009 ). Briefly, total suspended
article (TSP) samples were collected on Teflon filters using 
ow-volume (2.78 × 10 −4 m 3 /sec) TSP sampler inlets. After col- 
ection, filters were preserved and delivered for analysis. The 
ass percent of PM 2.5 and PM 10 of each TSP sample was de- 
ermined by PSD analysis (Beckman Coulter L230 laser diffrac- 
ion system with software version 3.29, Beckman Coulter Inc.,
SA). PM 2.5 and PM 10 concentrations were calculated by mul- 
iplying the determined percentage by the corresponding TSP 
oncentration. In cases where the PM concentrations were 
ery low, and insufficient PM was collected on the filter to de- 
ive a PSD, the corresponding PM 2.5 and PM 10 concentrations 
ould not be obtained. These data were noted as N/A, treated 
s non-detects, and removed from further analyses. 
The concentration measurements were cumulative con- 
entrations over the whole sampling period. To eliminate the 
nfluence of time, an hourly average concentration was calcu- 
ated by: 
 o = 
∑ 
C o ∑ 
T 
· 60 (1) 
here, C o (μg/m 3 ) refers to the hourly average observed con- 
entration; 
∑ 
C o (μg/m 3 ) is the cumulative particle concentra- 
ion during the sampling period; 
∑ 
T (min) is the total sam- 
ling period. 
Outliers can bias the summary statistics and may have 
 significant impact on the results, so to eliminate their ef- 
ects, potential outliers were identified for potential removal 
rom further analyses using the following procedure. The data 
ere log-transformed ( Ott, 1990 ), and then Rosner’s statisti- 
al outlier test was applied to identify the statistical outliers 
t α = 0.05 significance level ( US EPA, 2000a ). Finally, scien- 
ific judgement was applied to determine whether or not to 
emove a statistical outlier. 
.4. AERMOD and air dispersion modelling 
n this study, AERMOD View modelling package (version 9.7.0,
akes Environmental, Waterloo, Ontario) was used to run AER- 
OD (v.18081). Surface hourly meteorological data, upper air 
eteorological data, and hourly averaged onsite measured 
eteorological data were pre-treated by AERMET View (ver- 
ion 9.7.0, Lake Environmental, Waterloo, Ontario). The AER- 
ET version used was that included in AERMOD (v.18081).
mission rate (ER) and PSD were based on stack sampling of 
he thirteen point sources adopted from previous onsite cot- 
on ginemission monitoring study publications and using PSD 
ethodology described above and summarized in Appendix A 
able S2 ( Boykin et al., 2013a , 2013b ; Buser et al., 2013a , 2013b ,
013c ; Whitelock et al., 2013a , 2013b , 2013c ). PSD was assumed
o be log-normal to derive mass fractions, and density was set 
s 2.65 g/cm 3 ( Buser et al., 2013b ). 
AERMOD View was used to predict the hourly-averaged 
ollutant concentration of each receptor (details of model 
onfiguration see Appendix A Table S3). Then, weighted hourly 
verage of the whole sampling period was calculated by: 
 p = 
∑ n 
i =1 f t i · C p i ∑ n 
i =1 f t i 
(2) 
journal of environmental sciences 133 (2023) 8–22 11 
Fig. 1 – Layout of sampler sites. O refers to outer samplers; T to middle sampling towers; and I to inner samplers. Inner 
samplers, middle towers, and outer samplers were about 60 m, 120 m, and 180 m from the source, respectively. Middle 
towers were equipped with 6 samplers at 1, 2, 3, 4.5, 7.25 and 10-m heights. Outer and inner samplers were deployed at 2 m. 
 
 
 
 
 
C 
 
 
 
 
 
 
 
 
 
 
 
where, C p (μg/m 3 ) refers to the hourly-average predicted con-
centration; n refers to the total hours of the sampling period;
C p i refers to the predicted concentration of the i -th hour; and
f t i is the fraction of effective time of the i -th hour. The values of
f t i for each run are summarized in Appendix A Table S4. 
1.5. Formulation and validation of dispersion correction 
factor 
A dispersion correction factor is modeled as a multiplier to
correct the original AERMOD predicted concentrations: 
 cp = f c · C p (3)
where, C cp (μg/m 3 ) is the corrected model-predicted concen-
tration of the air pollutant, and C p (μg/m 3 ) is the original
model-predicted concentration. f c is the dispersion correction
factor. The value of f c can be calculated by the ratio of pre-
diction to observation (noted as R p ), which is defined as the
ratio of model-predicted concentration over observed concen-
tration: 
R p = 
C p 
C o 
(4)
where, R p is a measure of the extent of over/under prediction.
R p > 1 indicates overestimation and R p < 1 indicates under-
estimation. Based on Eq. (4) , if R p can be estimated, then the
estimated R p can be used to calculate f c : 
f c = 1 ˆ R p 
(5)
where, ˆ R p is the estimate of R p . 
12 journal of environmental sciences 133 (2023) 8–22 
w
t
p
a
u
i
m
R
w
b
c
a
c
s
θ
s
a
d
w
d
p
(
1
o
o
a
1
M
v
b
(
(
(
 
 
(
(
 
(
t
F
w 
F
s
l
w 
(
a
f
i
a
s
v
h 
s
a
i
h
m 
N 
( 
r 
i
s
t
t
1
P
t
m
( 
m
s
o
i
v
p
s
o 
i
w
v
2
2
A
p
h
m
A
m
f
t
U
e
Potential predictors of R p were investigated, and a model 
as built to estimate R p . Based on correlation analyses, poten- 
ial variables that may be used for R p estimation were used as 
redictors, and stepwise ordinary least square regression was 
pplied to determine the variables. A model was constructed 
sing log-transformed R p , and results were back transformed 
nto the linear scale to calculate R p . The generalized statistical 
odel for R p is as follows: 
 p = exp ( b 0 + b 1 h + b 2 d + b 3 u + b 4 σ + b 5 θ) (6) 
here, b i ( i = 0,1,2,3,4,5) refer to the regression coefficients. If 
 i was not significantly different from zero ( α = 0.05), then the 
orresponding variable was not included in the model. Vari- 
ble h (m) is height of receptor; d (m) is the distance of re- 
eptor from emission source; u (m/sec) is the ambient wind 
peed; σ (deg) is the standard deviation of wind direction; and 
(deg) is the deviation from wind direction, defined as the ab- 
olute difference between the direction of the wind velocity 
nd the direction from the source to the receptor, 0 ≤ θ ≤ 180 
eg ( Venkatram et al., 2004 ). 
The developed dispersion correction factors were coupled 
ith AERMOD to create a modified model, and then vali- 
ated by k -fold cross-validation to evaluate out-of-sample 
redictive accuracy, and compared with the original model 
 Gelman et al., 2014 ; Hooten and Hobbs, 2015 ). Since there were 
1 independent observations, k was set to 11, i.e., each subset 
f data contains data from one experiment, and the set of 11 
ut-of-sample predictions of f c may be evaluated for precision 
nd accuracy, as described below. 
.6. Model performance evaluation 
odel evaluation was based on the statistics proposed by pre- 
ious air quality modelling literature ( Hanna and Chang, 2012 ) 
elow: 
 1 ) Fractional bias ( FB ) : FB = C o − C p 
0 . 5 ·
(
C o + C p 
) (7) 
 2 ) Geometric mean ( MG ) : MG = exp 
(
ln C o − ln C p 
)
(8) 
 3 ) Normalized mean square error ( NMSE ) : NMSE = ( C o − C p ) 
2
C o · C p 
(9)
 4 ) Geometric v ariance ( VG ) : VG = exp 
[
( ln C o − ln C p ) 2 
]
(10) 
 5 ) Normalized absolute difference ( NAD ) : NAD = 
∣∣C o − C p ∣∣
C o + C p 
(11)
6) Fraction of predictions within a factor of two of observa- 
ions ( FAC 2): 
AC 2 = fraction of data that satisfy : 0 . 5 ≤ C p 
C o 
≤ 2 . 0 (12) 
here the “overbar” refers to the average over the dataset.
ractional bias (FB) and geometric mean (MG) are measures of 
ystematic bias, and the difference is that FB is based on the 
inear scale while MB is on the log-scale. A perfect model 
ould have FB = 0 and MG = 1. Normalized mean square error
NMSE) and geometric variance (VG) are measures of scatter 
nd can reflect systematic and random errors. Again, the dif- 
erence is that NMSE is based on the linear scale while VG 
s on the log-scale. A perfect model would have NMSE = 0 
nd VG = 1. Normalized absolute differenc e (NAD) is a mea- 
ure of normalized difference between prediction and obser- 
ation. NAD is between 0 and 1, and a perfect model would 
ave NAD = 0. Fraction of predictions within a factor of two of ob-
ervations (FAC2) is a measure of the total effect of systematic 
nd random error, and it is the most robust measure because 
t is not influenced by extreme values. A perfect model would 
ave FAC2 = 1. 
In terms of acceptance criteria for an air quality dispersion 
odel, previous literature suggested that | F B | < ∼ 0 . 30 | FB | < ∼ 0 . 30 ,
MSE < ∼ 3 NMSE < ∼ 3 , FAC 2 > ∼ 0 . 5 FAC 2 
> 
∼ 0 . 5 , NAD < ∼ 0 . 30 NAD 
< 
∼ 0 . 30
 Hanna and Chang, 2012 ). These criteria were used in the cur-
ent study to evaluate the performance of the model. However,
t should be noted that these proposed acceptance criteria are 
omewhat arbitrary ( Hanna and Chang, 2012 ). Thus, these cri- 
eria were used as a standard for discussion, not as an absolute 
est for accepting or rejecting the model. 
.7. Model sensitivity evaluation 
rior to estimation and cross-validation of f c and estima- 
ion of evaluation statistics, outliers were identified and re- 
oved from the data set to meet statistical assumptions 
 US EPA, 2000a ; Ott, 1990 ). However, for future applications it
ay not be possible to identify outliers due to smaller sample 
izes or a different sampler layout. Therefore, an evaluation 
f model sensitivity to outliers was conducted. The sensitiv- 
ty analysis included two parts. First, the effect of outliers on 
ariable selection and estimation of parameters for the em- 
irical equation of R p ( Eq. (5) ) was evaluated by comparing 
elected variables and regression coefficients with and with- 
ut the outliers. Second, the sensitivity of the modified model,
.e., AERMOD coupled with the dispersion correction factor 
as evaluated by comparing model performance by cross- 
alidation with and without the outliers. 
. Results and discussion 
.1. Meteorological conditions 
 summary of the meteorological conditions of the 11 sam- 
ling experiments are shown in Table 1 ; pressure, relative 
umidity, and temperature were similar among each experi- 
ent. Wind roses for each experiment are shown in Appendix 
 Fig. S1. Average wind speed was different among experi- 
ents and ranged from 1.00 m/sec for EXP00 up to 4.67 m/sec 
or EXP04. The mean and standard deviation of wind direc- 
ion were calculatedby Yamartino method ( Yamartino, 1984 ; 
S EPA, 2000b ). Average wind direction showed a large differ- 
nce among experiments, indicating that wind could come 
journal of environmental sciences 133 (2023) 8–22 13 
Table 1 – Summary statistics of meteorological parameters. Data are expressed as mean ± std. Mean and standard deviation 
of wind direction are calculated by Yamartino method ( Yamartino, 1984 ; US EPA, 2000b ). 
Pressure Relative humidity Temperature Wind speed Wind direction 
(Pa) (%) ( ◦C) (m/sec) (degrees from N) 
EXP00 100930 ± 130 45.4 ± 20.4 26.8 ± 4.20 1.00 ± 0.68 190 ± 63 
EXP01 100800 ± 130 46.1 ± 25.0 30.3 ± 5.38 1.31 ± 0.69 70 ± 54 
EXP02 100570 ± 100 43.0 ± 17.2 31.1 ± 5.20 3.15 ± 1.25 211 ± 30 
EXP03 100890 ± 90 44.9 ± 13.8 30.4 ± 4.33 3.44 ± 1.03 229 ± 29 
EXP04 100890 ± 160 44.6 ± 18.0 30.5 ± 4.14 4.67 ± 1.60 191 ± 11 
EXP05 100910 ± 50 58.9 ± 5.8 27.1 ± 1.58 3.94 ± 1.18 263 ± 34 
EXP06 100980 ± 170 49.3 ± 22.6 22.7 ± 4.40 1.62 ± 0.82 45 ± 39 
EXP07 100590 ± 70 64.8 ± 13.2 18.2 ± 2.63 4.39 ± 1.86 14 ± 10 
EXP08 100430 ± 90 50.8 ± 17.1 19.0 ± 4.22 3.72 ± 0.98 0.0 ± 17 
EXP09 100000 ± 130 49.8 ± 21.5 20.3 ± 4.99 1.54 ± 0.86 257 ± 58 
EXP10 100130 ± 120 46.1 ± 24.5 22.0 ± 5.17 1.28 ± 0.71 59 ± 48 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
from any direction on the site and that any direction could
be the downwind direction. 
The meteorological conditions of the experimental peri-
ods for the current study during September 2010 were also
compared with local meteorological records of the year 2010,
and the years of 2000-2020. The records of daily wind speed,
temperature, pressure, and relative humidity were retrieved
from St. Louis Lambert International Airport Station (KSTL),
and the scatter plot of the change in these parameters over
time were presented in Appendix A Fig. S2 ( WMO, 2008 ). The
average wind speed of the current study, 2.73 m/sec, was close
to daily average surface wind speed of year 2010, 2.94 m/sec,
and the daily average wind speed of years of 2000-2020, 3.09
m/sec at KSTL. Since wind is the main driving force of pol-
lutant transportation, we also evaluated the probability that
wind speed was within the range of the current study, i.e., 1.00
- 4.67 m/sec. Results showed that the probability that wind
speed measured at KSTL was within this range for year 2010
was 86% (Appendix A Fig. S2a), and 88% for years of 2000-2020
(Appendix A Fig. S2b). Thus, the measured wind conditions of
the current study were representative of local conditions. 
Temperature has a great seasonality, so the comparisons
were based on September only (Appendix A Fig. S2c, 2d). The
daily average temperature of year 2010 at KTSL was 21.7 °C,
and the daily average temperature was 21.9 °C for years of
2000-2020, which were slightly lower than average tempera-
ture of current study, 25.3 °C. This was because sampling time
was mostly during daytimes, which had a higher tempera-
ture. In addition, the average pressure and relative humidity
observed in the current study were also close to the daily
averages for 2010, and also the daily averages for 2000-2020
at KTSL (Appendix A Fig. S2e, 2f, 2g, 2h), and the difference
was mainly caused by the fact that sampling were conducted
during daytime. 
2.2. PM concentration measurements 
The observed PM concentrations from each experiment are
shown in Appendix A Table S5 and Appendix A Fig. S3, and
summary statistics are presented in Appendix A Table S6.
The results show that the average observed concentration
varied among experiments, which was the result of differ-
ent meteorological conditions and PM emissions variability
( Whitelock et al., 2013c ). 
A typical spatial distribution of observed concentrations
for PM 2.5 , PM 10 , and TSP are shown in Fig. 2 , where the great-
est observed concentrations occurred in the downwind direc-
tion from the source. In the case presented in Fig. 2 , the great-
est concentration of PM 2.5 and PM 10 appeared at I04 with 2.54
μg/m 3 and 22.70 μg/m 3 , respectively. Given that the US EPA
regulatory standards of PM 2.5 and PM 10 are 35 μg/m 3 and 150
μg/m 3 ( US EPA, 2018 ), and that WHO guideline values of PM 2.5
and PM 10 are 25 μg/m 3 and 50 μg/m 3 ( WHO, 2018 ), PM 2.5 and
PM 10 emitted from this cotton gin has a low potential health
risk to surrounding areas. Similarly, the greatest concentration
of TSP was found at I02 with 75.8 μg/m 3 , which is also lower
than US EPA regulatory standard, 260 μg/m 3 ( US EPA, 2018 ).
While observed PM 2.5 concentrations of the 11 experiments
are all below US EPA standard, some observations of PM 10 and
TSP were above the US EPA standards (Appendix A Fig. S3), in-
cluding O11 of EXP01 (PM 10 : 157.5 μg/m 3 ), O10 of EXP09 (PM 10 :
261.5 μg/m 3 ), I02 of EXP00 (TSP: 307.3 μg/m 3 ), O11 of EXP01
(582.8 μg/m 3 ), I10 of EXP06 (TSP: 290.2 μg/m 3 ), O10 of EXP09
(TSP: 596.59 μg/m 3 ), T01-01 and O10 of EXP10 (TSP: 260.5 μg/m 3
and 355.7 μg/m 3 ). In general, most of the monitored areas were
satisfied with WHO guidelines and US EPA standard, suggest-
ing a low impact to air quality and health risk potential. How-
ever, the concentration of PM 10 and TSP of some hotspots,
mostly inside the ginning plant area, exceeds the standard
and suggests potential risk to humans in the plant area. 
Samplers that had statistical outliers detected are sum-
marized in Appendix A Table S7, and the outliers were clas-
sified into low-outliers and high-outliers ( Cousineau and
Chartier, 2010 ). The table shows that outliers were much more
likely to appear on the lower samplers (1-m and 2-m heights)
suggesting influence of ground activities. The possible sources
that may contribute to high-outliers include movement of
cotton modules (Appendix A Fig. S4a); pneumatic cottonseed
loading (Appendix A Fig. S4b); and PM resuspension from road
and unpaved ground by traffic and wind (Appendix A Fig.
S4c, S4d, S4e). In addition, rice straw burning in distant fields
(Appendix A Fig. S4f) may also have led to large concentra-
tions at outer samplers (i.e., O09 and O10). These sources are
the possible reasons lead to PM concentrations above regula-
tory standards and guidelines mentioned above. On the other
14 journal of environmental sciences 133 (2023) 8–22 
Fig. 2 – Wind rose and hourly average concentration distribution (μg/m 3 ) of (b) PM 2.5 , (c) PM 10 and (d) TSP for EXP03. N/A 
means that insufficient particles were on the filter to derive a particle size distribution. 
h
c
t
A
s
r
2
a
P 
w
a
c
T 
A
w 
o
i 
P
d 
T
t
 
P
s
m
g
and, the possible sinks that can lower PM concentrations in- 
lude vegetation (Appendix A Fig. S4g) and water spraying in 
he industrial area (Appendix A Fig. S4h) ( Adrizal et al., 2008 ; 
zarov et al., 2017 ). All these activities are not part of atmo- 
pheric dispersion, and thus all the statistical outliers were 
emoved from further analysis. 
.3. PM concentration as a function of height, distance, 
nd wind 
M concentration changes as a function of height, distance,
ind speed, deviation from wind direction, and standard devi- 
tion of wind direction were investigated. Pearson correlation 
oefficients were calculated and are presented in Appendix A 
able S8 and in corresponding scatter plots Appendix A Fig. S5.
 negative correlation for PM 2.5 , PM 10 , and TSP concentration 
ith height ( p < 0.001 for PM 2.5 , p < 0.05 for PM 10 and TSP) was
bserved, i.e., lower concentrations were more likely to occur 
n the higher samplers. In addition, the concentration of PM 2.5 ,
M 10 , and TSP showed a significant negative correlation with 
istance from source ( p < 0.05 for PM 2.5 and PM 10 , p < 0.001 for
SP). This was most likely due to dispersion and deposition of 
he particles as the particles moved from the source. 
A significant positive correlation was found betweenPM 2.5 ,
M 10 , and TSP concentrations and average ambient wind 
peed ( p < 0.001). However, such positive correlation does not 
eet with theoretical expectation where lower wind speeds 
enerally correspond to lower dispersion and more deposi- 
journal of environmental sciences 133 (2023) 8–22 15 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
tion and lead to a higher concentration of pollutants near
the source ( Lu and Fang, 2002 ). Thus, the observed positive
correlation is probably due to particles from the surround-
ings, especially the bare ground and road being entrained or
resuspended by the increasing wind speed and contributing
particles to the total PM concentrations ( Gehrig and Buch-
mann, 2003 ; Elminir, 2005 ; Pérez et al., 2010 ; Nowak et al.,
2013 ). 
A negative correlation was found between PM 10 and TSP
concentrations and the deviation from wind direction ( p <
0.001); this pattern is consistent with the spatial distribution
of the concentrations (Appendix A Fig. S3). A lack of correla-
tion of deviation from wind direction with PM 2.5 was probably
caused by the larger dispersion coefficient due to its smaller
size ( Ounis and Ahmadi, 1990 ), making such a pattern insignif-
icant. Finally, a significant negative correlation was found be-
tween the standard deviation of wind direction and particle
concentrations for all sizes of pollutants ( p < 0.001), indicat-
ing that concentration of the pollutants was more likely to
decrease as wind direction become more variable during the
sampling period. This is consistent in that a multidirectional
wind causes greater mixing and stronger dilution which con-
tributes to lower concentrations ( Flagan and Seinfeld, 2012 ). 
2.4. PM size distribution as a function of height and 
distance 
Particulate size distribution changes as function of height
and distance were investigated because the effects on hu-
man health vary with PM size (Appendix A Table S8, Ap-
pendix A Fig. S6) ( Brown et al., 2013 ). Both the percent of PM 2.5
and the percent of PM 10 showed a significant decrease with
height ( p < 0.05), i.e., lower heights had more mass percent
of PM 2.5 and PM 10 . But this is contradictory to prior observa-
tions and theory where heavier particles deposit faster than
small particles leading to differential deposition of the parti-
cles and a decreased percentage of smaller particles at lower
heights ( Flagan and Seifeld, 2012 ; Yao et al., 2018 ). Our obser-
vations here may be due to the effects of the downdraft on the
low-altitude sources where smaller particles are more easily
pulled by the downward wind flow and/or the contribution of
entrained or resuspended particles from the ground as dis-
cussed above. 
The percent of PM 10 were positively correlated with dis-
tance from the emission source ( p < 0.001). This expected pos-
itive correlation is easily explained in that heavier particles
settle quickly during dry deposition resulting in an increase
in the mass percentage of smaller particles especially for a
longer distance from the source ( ∼180 m). The percent of PM 2.5
was also expected to have a positive correlation, however a
lack of correlation of the percent of PM 2.5 and distance from
the emission source could be due to interference from other
sources or resuspension from the surroundings ( Gehrig and
Buchmann, 2003 ; Elminir, 2005 ; Pérez et al., 2010 ; Nowak et al.,
2013 ). 
2.5. Performance of AERMOD 
AERMOD was used to predict the concentration of PM 2.5 , PM 10 ,
and TSP at all the sampling sites for each experiment, and the
results were presented in Appendix A Table S5 and Appendix
A Fig. S3. Spatial distributions of the model-predicted concen-
trations were consistent with observed concentrations (Ap-
pendix A Fig. S3). However, scatter plots of the observed con-
centrations ( C o ) versus model-predicted concentrations ( C p )
show that the model consistently over predicted the concen-
trations ( Fig. 3 a, 3 c, 3 e; Appendix A Fig. S7, Appendix A Table
S9). FAC2 values of the overall performance were only 0.10,
0.16, and 0.18 for PM 2.5 , PM 10 , and TSP, respectively, which
are substantially smaller than the acceptable value of 0.50
( Hanna and Chang, 2012 ). |FB|, NMSE, and NAD values also de-
viated from acceptance criteria. In addition, the R p values for
each sampling point in each experiment and for the overall
experiment were calculated (Appendix A Table S10). The re-
sults showed that concentrations were over-predicted by fac-
tors of 64.07, 6.97, and 7.44 on average for PM 2.5 , PM 10 , and
TSP, respectively. Clearly, pollutant concentrations were over-
predicted by AERMOD for such low-altitude emission sources,
and a dispersion correction factor is needed. 
The unsatisfactory performance of the model is con-
sistent with previous findings ( Zwicke, 1998 ; Fritz, 2003 ;
Venkatram et al., 2004 ; Pournazeri, 2012 ). AERMOD was orig-
inally designed for industrial situations with tall stacks and
long distances of up to 20 km from emission source to
sampling locations ( Perry et al., 2005 ). In addition, gaseous
pollutant datasets were used in the first model validations
( Cimorelli et al., 2005 ; Perry et al., 2005 ). Although AERMOD has
options for dry deposition within its simulation process, cot-
ton gin emissions contain particles from sticks, leaves, hulls,
and leaf materials ( Buser et al., 2013d ) which may lead to dif-
ferent aerodynamic behaviors and deposition rate. 
In addition, the stacks of ginning facilities are shorter than
stacks of the original AERMOD validation study, where the
wind and dispersion are less likely to be influenced by interac-
tions with the ground surface. In current study, particles could
be resuspended and obtained by surrounding samplers. Thus,
the influence of road and bare ground was further investigated
by separating the samplers into two groups: a group of sam-
plers that were close to road and bare ground (I03, I04, I05,
I07, O03, O04, O05, O09, O11, T02, T03, T04, T05, T06, T07, T10),
and a group of samplers that were distant from road and bare
ground (I02, I08, I10, I11, I12, O01, O02, O06, O07, O08, O10, O12,
T01, T08, T09, T11, T12). Then, the performance of each sam-
pler was calculated based on the 11 experiments (Appendix A
Table S11a, S11b, S11c), and the mean values of FB, MG, NMSE,
VG, NAD, FAC 2 of the two groups were compared. The results
showed significant differences for 5 out of 18 cases ( p < 0.05,
Appendix A Table S11d), while 13 out of 18 cases did not show
significant differences. For the 5 cases that had significant dif-
ferences, the group of samplers that were distant from the
road and bare ground consistently showed poorer model per-
formance, indicating that the ground surface could influence
model performance, but more investigations are needed to
confirm such land type effect. 
Since wind is an important factor for the transport of par-
ticles from the source, the correlation between model per-
formance measures ( FB, MG, NMSE, VG, NAD, FAC2 ) and wind
characteristics, namely average wind speed and the standard
deviation of wind direction, were investigated further (Ap-
pendix A Table S12). Wind speed was significantly and pos-
16 journal of environmental sciences 133 (2023) 8–22 
Fig. 3 – Scatter plot of original AERMOD model-predicted concentration ( C p ) versus observed concentration ( C o ) (a, c, e) and 
out-of-sample corrected model-predicted concentration ( C cp ) versus observed concentration ( C o ) (b, d, f). Solid lines refer to 
1:1 ratio and dotted lines refer to 1:2 and 1:0.5 ratios. DCF refers to dispersion correction factor; FB = fraction bias; 
NMSE = normalized mean square error; FAC2 = fraction of predictions within a factor of two of observations;NAD = normalized absolute difference. 
i
a
t 
i
M
i
c
e 
a
A 
m
c
c
l
tively correlated with FB, MG , and FAC2 for PM 10 and TSP 
nd with MG and FAC2 for PM 2.5 . Such correlations suggest 
hat systematic error decreased with increasing wind speed,
.e., the model performed better at greater wind speeds. AER- 
OD is a Gaussian plume-based model and assumes that 
f wind speed is sufficiently large then diffusive transport 
an be neglected in the wind direction ( Stockie, 2011 ). How- 
ver, if wind speed is small, this assumption does not hold,
nd systematic prediction errors may be introduced. Another 
ERMOD assumption is that wind direction is constant, i.e.,
odel performance improves if wind direction is relatively 
onstant ( Stockie, 2011 ). Thus, systematic errors will also in- 
rease as the standard deviation of wind direction becomes 
arger. 
journal of environmental sciences 133 (2023) 8–22 17 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2.6. Ratio of predicted to observed concentrations ( R p ) as 
a function of height, distance, and wind 
Several strategies, log transformation of data and use of cor-
rection factors, were considered to improve the predictive ca-
pability of AERMOD for cotton gin emissions. This was car-
ried out by using R p to examine the possible factors that may
lead to over-prediction or under-prediction. Since the proba-
bility distribution of R p was positively skewed (Appendix A Fig.
S8), log-transformation was applied to standardize the skewed
data to approach a normal distribution. Pearson correlation
calculations were performed between the possible factors and
the log-transformed R p (Appendix A Table S13 and Appendix
A Fig. S9). Log-transformed R p was positively correlated with
height for PM 2.5 and PM 10 ( p < 0.05), but not for TSP. On the
other hand, the extent of over-prediction decreased with in-
creasing distance from the source as a negative correlation
was observed between log-transformed R p and distance from
the source, but was significant only for TSP ( p < 0.001). 
Linear correlations were found between log-transformed
R p and wind speed, deviation from wind direction, and stan-
dard deviation of wind direction ( p < 0.001). The result
showed that as wind speed increased, the extent of over-
prediction decreased, meaning that over-prediction was more
severe for low wind conditions. Again, low wind speeds vio-
late one of the Gaussian model assumptions which can re-
sult in unsatisfied simulations. The negative correlation be-
tween log-transformed R p and deviation from wind direc-
tion suggests that concentrations for samplers in the down-
wind direction from the source were more severely overesti-
mated than other samplers, and as the angle difference in-
creases, the extent of overestimation decreased. Such a phe-
nomenon was also observed by previous air dispersion stud-
ies ( Venkatram et al., 2004 ; Isakov et al., 2004 ), and it was
probably related to errors introduced by the Plume Rise Model
Enhancement (PRIME) building downwash algorithm in AER-
MOD ( Petersen et al., 2017 ). Furthermore, a significant correla-
tion was found between log-transformed R p and standard de-
viation of wind direction, i.e., overestimation is more severe
when multidirectional wind occurs during a sampling period.
This is again likely due to the assumption of the Gaussian
plume model that wind direction is constant in the modelling
domain. 
2.7. Dispersion correction factor and model validation 
Stepwise ordinary least square regression showed that log-
transformed R p of PM 2.5 , PM 10 and TSP could be estimated by
the following equations: 
P M 2 . 5 : R p = exp ( b 0 + b 1 h + b 4 σ + b 5 θ) (13)
P M 10 : R p = exp ( b 0 + b 1 h + b 4 σ + b 5 θ) (14)
T SP : R p = exp ( b 0 + b 1 h + b 2 d + b 4 σ + b 5 θ) (15)
where the b i are estimated separately for each pollutant. The
dispersion correction factor, f c , was then used with the origi-
nal AERMOD prediction, C p , to calculate a corrected prediction,
C cp , by Eq. (3) and Eq. (5) . The dispersion correction factor to-
gether with AERMOD was considered as a modified model, and
this modified model was validated by 11-fold cross-validation
as described in Section 2.5 . ( Fig. 3 b, 3 d, 3 f; for results from an
individual experiment, see Appendix A Fig. S10). 
Scatter plots of the out-of-sample corrected predicted con-
centration, C cp , versus the observed concentrations ( Fig. 3 b,
3 d, 3 f) showed that the data points were more grouped and
closer to a 1:1 ratio line than the original AERMOD prediction
( Fig. 3 a, 3 c, 3 e), indicating that the correction factor improved
model performance. Additionally, the average R p decreased
from 64.07 to 3.75, from 6.97 to 1.52, and from 7.44 to 1.44, for
PM 2.5 , PM 10 , and TSP, respectively, suggesting that on average
the overprediction was mitigated significantly. 
The summary performance statistics of the modified
model (Appendix A Table S14) were greatly improved over
those of the original model. The | FB | values decreased sig-
nificantly for all pollutants, suggesting that model bias was
greatly mitigated. Although only | FB | of PM 10 was smaller than
0.30, the acceptance criterion, values of | FB | of PM 2.5 and TSP
were closer to 0.30 (0.38 and 0.39 for PM 2.5 and TSP, respec-
tively) than with the original model. NMSE was also reduced
for all pollutants, indicating that prediction accuracy was im-
proved. The NMSE of PM 10 was smaller than 3.0, satisfying the
acceptance criterion, and the NMSE of TSP was much closer to
3.0 than in the original AERMOD prediction (4.06 versus 11.4).
However, for PM 2.5 , the NMSE was still much greater than 3.0,
but it reduced from an original 15.2 to 12.2. The NAD decreased
for all pollutants; NAD values of PM 10 and TSP were closer to
the proposed acceptance criterion, 0.30 (0.41 for PM 10 and 0.39
for TSP), while NAD of PM 2.5 reduced from 0.89 to 0.64. The
overall FAC2 values increased from 0.10 to 0.36 for PM 2.5 , from
0.16 to 0.54 for PM 10 , and from 0.18 to 0.69 for TSP, which shows
more than three-fold increasement for all pollutants. Similar
improvements occurred for most individual experiments (Ap-
pendix A Table S9, S13). Although performance of the modified
model did not completely satisfy the proposed acceptance cri-
teria, prediction accuracy was greatly improved. 
In summary, cross-validation showed that the developed
dispersion correction factor model coupled with AERMOD out-
performed the original AERMOD. This suggests that informa-
tion of meteorological conditions and geometric relations be-
tween source and receptor can be used to modify model pre-
diction and improve model performance, and such modifi-
cation can be achieved by a simple, easy-to-use and cost-
effective method. 
The regression coefficients calculated for Eqs. (13) - (15)
based on all 11 experiments are listed in Appendix A Table S15
These empirical equations can be used to estimate R p and to
calculate f c for current cotton gin. In addition, for situations
without 5-min wind direction data, an alternative regression
was conducted by excluding standard deviation of wind direc-
tion, and regression results are also listed in Appendix A Table
S15. Although these empirical parameters and equations are
derived based on the current site and they are only applica-
ble to PM emitted by this cotton gin, the observed phenom-
ena, proposed dispersion correction factor and general ap-
proach of this study may be tested for other sites and emission
sources. Thus, the dispersion correction factor approach is
recommended for other cotton gins and low-altitude emission
18 journal of environmental sciences 133 (2023) 8–22 
Fig. 4 – Spatial distribution of maximum of 24-houraverage concentration of AERMOD original prediction and corrected 
AERMOD prediction for PM 2.5 (a, b), PM 10 (c, d), and TSP (e, f). Unit is μg/m 3 . refers to emission source. and ◦ refer to 
buildings. Red solid line refers to the boundary that exceeds regulatory standard, and dashed line indicates the higher 
concentration side. For map details see Fig. 1 Note that plots of each pollutant are on the same scale. 
s
c
f
2
T
a
t 
T
P
d 
T
e
o 
P
ources, where some simplified onsite measurements can be 
onducted to calibrate the site-specific dispersion correction 
actors. 
.8. Sensitivity to outliers 
he dispersion correction factor model was evaluated with 
nd without statistical outliers to evaluate the sensitivity to 
he log-normal assumption of observed concentration data.
he selected predictors for estimating R p were the same for 
M 2.5 , PM 10 , and TSP, and the regression coefficients were in- 
istinguishable within uncertainty (Appendix A Table S16).
his result suggests that the dispersion correction factor mod- 
ls and the derived empirical equations were not sensitive to 
utliers. Then, dispersion correction factor models of PM 2.5 ,
M 10 , and TSP were used to estimate out-of-sample R p , cal- 
journal of environmental sciences 133 (2023) 8–22 19 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
culate f c and C cp by 11-fold cross-validation. Results showed
that there were negligible differences in the summary statis-
tics between the corrected model with and without outliers
(Appendix A Tables S14, S16). 
2.9. Application of the dispersion correction factor for risk
assessment 
The developed and calibrated dispersion correction factor was
applied for prediction of PM 2.5 , PM 10 , and TSP concentration
and risk assessment of the current cotton ginning plant by
using available onsite meteorological data measured from
September 12-31, 2010. Model configuration and model inputs
were the same as previously described (Appendix A Table S2,
S3). Maximum 24-hour average concentration was calculated
by AERMOD based on 101 × 101 × 5 m uniform Cartesian grids
with 1.8 m height (i.e., 10201 receptors in total). Ginning facil-
ities were assumed to run 24 hours a day to reflect the worst-
case scenario. Then, dispersion correction factors were calcu-
lated based on Table 1 and Eq. (5) for each of the pollutants
and applied using Eq. (3) to the original AERMOD output to de-
termine the corrected AERMOD prediction. The risk quotient
( RQ ), which is defined as the ratio of exposure to the regula-
tory standard, was calculated using 35 μg/m 3 , 150 μg/m 3 , and
260 μg/m 3 for the regulatory standards of PM 2.5 , PM 10 , and TSP,
respectively ( US EPA, 2018 ). 
Original and corrected AERMOD predict concentrations of
the pollutants are shown in Fig. 4 , histogram and descriptive
statistical summary of pollutant concentrations of the 10201
receptors are shown in Appendix A Fig. S11, and the corre-
sponding RQ are shown in Appendix A Fig. S12. PM 2.5 and PM 10
concentrations in the simulation domain of both original AER-
MOD predictions and corrected predictions were smaller than
regulatory permitted levels. The corrected pollutant concen-
trations around the ginning facilities were effectively lowered
from the original AERMOD prediction, which was reflected by
the smaller yellowish area on the map ( Fig. 4 a, 4 b, 4 c, 4 d). The
average 24-hr average concentration of PM 2.5 in the simulation
domain was 5.53 μg/m 3 for the original AERMOD prediction
and 0.81 μg/m 3 for the corrected prediction. The average 24-hr
average concentration of PM 10 was 32.2 μg/m 3 for the original
AERMOD prediction and 17.8 μg/m 3 for the corrected predic-
tion. TSP concentrations of some locations exceeded the reg-
ulatory standards, giving RQ a greater than 1 value (Appendix
A Fig. S12e, S12f). However, the area of corrected model that
exceeds regulatory standards was much smaller than the vio-
lation area of the original model ( Fig. 4 e, 4 f). The average orig-
inal prediction in the domain was 313 μg/m 3 with 46% of the
receptors exceeding regulatory standards while the average
corrected prediction was 140 μg/m 3 with 14% of the receptors
exceeding regulatory standards. 
2.10. Limitations of the current study 
There are some limitations with the current study. First, the
monitored data are based on one cotton gin, so some of the re-
sults are site-specific, e.g., empirical dispersion factors should
vary site by site. However, the general conclusions of the in-
fluential factors should be held for other cotton gins or low-
altitude emission sources. Second, the background concen-
tration was not measured. However, due to multidirectional
wind during sampling periods and spatial heterogeneity of
the concentration ( Zhu et al., 2002 ), it was difficult to have a
single onsite background measurement. The influence of the
background concentration was taken into account when cali-
brating dispersion correction factors and cross-validating the
model. However, further work to better correct for background
pollutant concentration and multidirectional winds should
further reduce cross-validated model uncertainty. In addition,
other possible variables may also influence on the disper-
sion correction factors, such as the influence of stacks, roads,
buildings and production activities, and thus these factors
may need further investigation and consideration. In princi-
ple, we might expect the same systematic dependencies on
site conditions could be identified for high altitude emissions
for which AERMOD was originally designed. If this is true, then
cross-validation of appropriately calibrated correction factors
should produce smaller simulation errors. 
3. Conclusions 
Field samples of PM 2.5 , PM 10 , and TSP from a low-altitude
emission source were collected in 11 experiments at a typi-
cal cotton gin. The concentrations and dispersion of the air
pollutants were assessed. Onsite monitored data shows that
the ginning plant has a low potential risk to the surround-
ing areas, but there were some hotspots exceeding US EPA
standards and WHO guidelines inside the plant area. Pollu-
tant concentrations were overestimated by the regulatory rec-
ommended model AERMOD, but in systematic ways linked to
height, distance from source, and wind direction. A disper-
sion correction factor method was proposed, applied, cross-
validated and is recommended for regulatory and practical
use of AERMOD under similar conditions. Based on the results,
the modified model reduced overestimation factors from 64.7
to 3.75 for PM 2.5 , from 6.97 to 1.52 for PM 10 , and from 7.44 to
1.44 for TSP. Site-specific empirical dispersion correction fac-
tors were also derived and applied for risk assessment as a
case study. 
An improved model, that successfully predicts both low
and high-altitude emissions using physical processes, would
be desirable. In the meantime, DCFs applied to the AERMOD
provide the means by which large systematic errors in low-
altitude emission scenario may be reduced. The same method
may also be useful for developing and evaluating dispersion
models for other low-altitude emission sources. More gener-
ally, the correction permits the adjustment and application of
regulatory models for risk assessment ( Buser et al., 2009 ) and
developing remediation techniques ( US EPA, 2020 ). The same
approach may also be taken for evaluation and remediation of
a range of other air pollutants and environmental conditions.
Disclaimer 
Mention of trade names or commercial products in this pub-
lication is solely for the purpose of providing specific infor-
mation and does not imply recommendation or endorsement
by the University of Maryland (UMD) or U.S. Department of
20 journalof environmental sciences 133 (2023) 8–22 
A
p
D
T
n
p
A
T
f
t
g
A
A
S
f
r
A
A
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
C
C
C
D
D
Đ
E
F
F
F
G
griculture (USDA). UMD and USDA are equal opportunity 
roviders and employers. 
eclaration of Competing Interest 
he authors declare that they have no known competing fi- 
ancial interests or personal relationships that could have ap- 
eared to influence the work reported in this paper. 
cknowledgment 
he authors wish to acknowledge the dedicated contributions 
rom numerous field and technical staff, students, and volun- 
eers. Funding for this project was provided by National Pro- 
rams 306, Product Quality and New Uses, and 212, Soil and 
ir. 
ppendix A Supplementary data 
upplementary material associated with this article can be 
ound, in the online version, at doi: 10.1016/j.jes.2022.03.048 . 
eferences 
drizal, A., Paul, H.P., Ralph, M.H., Ricky, M.B., Myers, C.A.B.,
Martin, G.P., et al., 2008. Vegetative buffers for fan emissions 
from poultry farms: 2. ammonia, dust and foliar nitrogen. J. 
Environ. Sci. Health, Part B 43, 96–103 no. 12008 .
zarov, A.V., Zhukova, N.S., Antonov, F.G., 2017. Water-spray 
systems reducing negative effects of fine-dispersion dust at 
operator’s workplaces of machine-building industries. 
Procedia Eng. 206, 1407–1414 .
aawain, M., Al-Mamun, A., Omidvarborna, H., Al-Jabri, A., 2017. 
Assessment of hydrogen sulfide emission from a sewage 
treatment plant using AERMOD. Environ. Monit. Assess. 189 
(6) p.263 .
airy, J., Parnell, C.B., Jr, McGee, R.O., Steubing, K., 2012. AERMOD 
dispersion modeling of particulate matter emissions from 
cattle feedyards and cotton gins. In: 2012 Beltwide Cotton 
Conferences, Florida, USA. January 3-6, 2012.
ehnami, A., Benis, K.Z., Shakerkhatibi, M., Derafshi, S.,
Chavoshbashi, M.M., 2019. A systematic approach for 
selecting an optimal strategy for controlling VOCs emissions 
in a petrochemical wastewater treatment plant. Stoch. 
Environ. Res. Risk Assess. 33 (1), 13–29 .
onifacio, H.F., Maghirang, R.G., Razote, E.B., Trabue, S.L.,
Prueger, J.H., 2013. Comparison of AERMOD and WindTrax 
dispersion models in determining PM 10 emission rates from a 
beef cattle feedlot. J. Air Waste Manag. Assoc. 63 (5), 545–556 .
oykin, J.C., Buser, M.D., Whitelock, D.P., Holt, G.A., 2013a. 
Combined mote system PM2. 5 emission factors and rates for 
cotton gins: Method 201A combination PM10 and PM2. 5 
sizing cyclones. J. Cotton Sci. 17 (4), 447–456 .
oykin, J.C., Buser, M.D., Whitelock, D.P., Holt, G.A., 2013b. Second 
stage seed-cotton cleaning system PM2. 5 emission factors 
and rates for cotton gins: Method 201A combination PM10 
and PM2. 5 sizing cyclones. J. Cotton Sci. 17 (4), 333–345 .
rown, J.S., Gordon, T., Price, O., Asgharian, B., 2013. Thoracic and 
respirable particle definitions for human health risk 
assessment. Part. Fibre Toxicol. 10 (1), 12 .
rugge, D., Durant, J.L., Rioux, C., 2007. Near-highway pollutants in 
motor vehicle exhaust: a review of epidemiologic evidence of 
cardiac and pulmonary health risks. Environ. Health 6 (1), 23 .
user, M.D., Parnell, C.B., Shaw, B.W., Lacey, R.E., 2007a. Particulate 
matter sampler errors due to the interaction of particle size 
and sampler performance characteristics: background and 
theory. Trans. ASABE 50 (1), 221–228 .
user, M.D., Parnell, C.B., Shaw, B.W., Lacey, R.E., 2007b. Particulate 
matter sampler errors due to the interaction of particle size 
and sampler performance characteristics: ambient PM 2. 5 
samplers. Trans. ASABE 50 (1), 241–254 .
user, M.D., Parnell, C.B., Shaw, B.W., Lacey, R.E., 2007c. Particulate 
matter sampler errors due to the interaction of particle size 
and sampler performance characteristics: ambient PM 10 
samplers. Trans. ASABE 50 (1), 229–240 .
user, M.D., Whitelock, D.P., Park, M., Boykin, N.J.C., 2009. 
Characterization of cotton gin particulate matter 
emissions–project plan. J. Cotton Sci. 16, 105–116 .
user, M.D., Whitelock, D.P., Boykin, J.C., Holt, G.A., 2013a. 
Unloading system PM 2.5 emission factors and rates for cotton 
gins: Method 201A combination PM 10 and PM 2.5 sizing 
cyclones. J. Cotton Sci. 17 (4), 309–319 .
user, M.D., Moore, T.W., Whitelock, D.P., 2013b. Mote trash system 
PM 2. 5 , PM 10 and Total PM emission factors for cotton gin B 
using method 201a with a PM 2.5 cyclone. J. Cotton Sci. 17 (4), 
479–488 .
user, M.D., Whitelock, D.P., Boykin, J.C., Holt, G.A., 2013c. 
Combined lint cleaning system PM 2. 5 emission factors and 
rates for cotton gins: Method 201A combination PM 10 and 
PM 2.5 sizing cyclones. J. Cotton Sci. 17 (4), 391–401 .
user, M.D., Whitelock, D.P., Boykin, J.C., Holt, G.A., 2013d. Third 
stage seed-cotton cleaning system PM 2.5 emission factors and 
rates for cotton gins: method 201A combination PM 10 and 
PM 2.5 sizing cyclones. J. Cotton Sci. 17 (4), 346–356 .
hen, H., Bai, S., Eisinger, D., Niemeier, D., Claggett, M., 2009. 
Predicting near-road PM 2.5 concentrations: comparative 
assessment of CALINE4, CAL3QHC, and AERMOD. Transp. Res. 
Rec. 2123 (1), 26–37 .
imorelli, A.J., Steven, G.P., Akula, V., Jeffrey, C.W., Robert, J.P.,
Robert, B.W., et al., 2005. AERMOD: A dispersion model for 
industrial source applications. Part I: General model 
formulation and boundary layer characterization. J. Appl. 
Meteorol. 44 (5), 682–693 .
ousineau, D., Chartier, S., 2010. Outliers detection and 
treatment: a review. Int. J. Psychol. Res. 3 (1), 58–67 .
ai, P., Shen, D., Tang, Q., Huang, K., Li, C., 2020. PM 2. 5 from a 
broiler breeding production system: The characteristics 
and microbial community analysis. Environ. Pollut. 256 
p.113368 .
amuchali, A.M., Guo, H., 2020. Developing an odour emission 
factor for an oil refinery plant using reverse dispersion 
modeling. Atmos. Environ. 222 p.117167 .
or đevi ́c, D.S., Šolevi ́c, T.M., 2008. The contributions of high-and 
low altitude emission sources to the near ground 
concentrations of air pollutants. Atmos. Res. 87 (2), 170–182 .
lminir, H.K., 2005. Dependence of urban air pollutants on 
meteorology. Sci. Total Environ. 350 (1-3), 225–237 .
lagan, R.C., Seinfeld, J.H., 2012. Fundamentals of air pollution 
engineering. Courier Corporation .
ederal Register, 2013. National ambient air quality standards for 
particulate matter; final rule. Federal Register 78 (10), 
3086–3287 .
ritz, B.K., 2003. Dispersion modeling of particulate emissions 
from low-level point sources. Texas A&M University, Texas, 
USA PhD thesis .
ehrig, R., Buchmann, B., 2003. Characterising seasonal variations 
and spatial distribution of ambient PM 10 and PM 2. 5 
concentrations based on long-term Swiss monitoring data. 
Atmos. Environ. 37 (19), 2571–2580 .
https://doi.org/10.1016/j.jes.2022.03.048
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0001
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0002
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0002
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0002
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0002
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0003
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0003
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0003
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0003
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0003
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0005
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0005
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0005
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0005http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0005
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0005
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0006
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0006
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0006
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0006
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0006
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0006
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0007
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0007
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0007
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0007
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0007
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0008
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0008
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0008
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0008
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0008
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0009
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0009
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0009
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0009
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0009
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0010
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0010
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0010
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0010
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0011
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0011
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0011
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0011
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0011
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0012
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0012
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0012
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0012
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0012
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0013
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0013
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0013
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0013
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0013
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0014
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0014
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0014
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0014
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0014
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0015
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0015
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0015
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0015
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0015
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0016
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0016
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0016
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0016
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0017
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0017
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0017
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0017
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0017
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0018
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0018
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0018
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0018
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0018
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0019
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0019
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0019
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0019
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0019
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0019
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0020
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0021
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0021
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0021
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0022
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0022
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0022
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0022
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0022
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0022
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0023
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0023
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0023
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0024
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0024
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0024
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0025
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0025
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0026
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0026
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0026
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0027
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0028
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0028
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0029
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0029
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0029
journal of environmental sciences 133 (2023) 8–22 21 
 
 
 
 
 
 
 
 
Gelman, A., Hwang, J., Vehtari, A., 2014. Understanding predictive 
information criteria for Bayesian models. Stat. Comput. 24 (6), 
997–1016 .
Gibson, M.D., Kundu, S., Satish, M., 2013. Dispersion model 
evaluation of PM 2. 5 , NO x and SO 2 from point and major line 
sources in Nova Scotia, Canada using AERMOD Gaussian 
plume air dispersion model. Atmos. Pollut. Res. 4 (2), 
157–167 .
Hadlocon, L.S., Zhao, L.Y., Bohrer, G., Kenny, W., Garrity, S.R.,
Wang, J., Wyslouzil, B., et al., 2015. Modeling of particulate 
matter dispersion from a poultry facility using AERMOD. J. Air 
Waste Manag. Assoc. 65 (2), 206–217 .
Hanna, S., Chang, J., 2012. Acceptance criteria for urban dispersion
model evaluation. Adv. Atmos. Sci. 116 (3-4), 133–146 .
Heath, G.A., Granvold, P.W., Hoats, A.S., Nazaroff, W.W., 2005. 
Quantifying the air pollution exposure consequences of 
distributed electricity generation. University of California, 
Berkeley, USA Report .
Hooten, M.B., Hobbs, N.T., 2015. A guide to Bayesian model 
selection for ecologists. Ecol. Monogr. 85 (1), 3–28 .
Isakov, V., Sax, T., Venkatram, A., Pankratz, D., Heumann, J., Fitz, D.,
2004. Near-field dispersion modeling for regulatory 
applications. J. Air Waste. Manag. Assoc. 54 (4), 473–482 .
Jenkins, S.F., Barsotti, S., Hincks, T.K., Neri, A., Phillips, J.C.,
Sparks, R.S.J., et al., 2015. Rapid emergency assessment of ash 
and gas hazard for future eruptions at Santorini Volcano. 
Greece. J. Appl. Volcanol. 4 (1), 1–22 .
Kelleghan, D.B., Hayes, E.T., Everard, M., Curran, T.P., 2021. 
Predicting atmospheric ammonia dispersion and potential 
ecological effects using monitored emission rates from an 
intensive laying hen facility in Ireland.Atmos. Environ. 247 
p.118214 .
Kim, K.H., Kabir, E., Kabir, S., 2015. A review on the human health 
impact of airborne particulate matter. Environ. Int. 74, 
136–143 .
Liu, H., Kim, D., 2019. Simulating the uncertain environmental 
impact of freight truck shifting programs. Atmos. Environ. 214 
p.116847 .
Löndahl, J., Massling, A., Pagels, J., Swietlicki, E., Vaclavik, E.,
Loft, S., 2007. Size-resolved respiratory-tract deposition 
of fine and ultrafine hydrophobic and hygroscopic aerosol 
particles during rest and exercise. Inhal. Toxicol. 19 (2), 
109–116 .
Lu, H.C., Fang, G.C., 2002. Estimating the frequency distributions 
of PM 10 and PM 2. 5 by the statistics of wind speed at Sha-Lu. 
Taiwan. Sci. Total Environ. 298 (1-3), 119–130 .
Mestl, H.E.S., Fang, J., 2003. Air quality estimates in Taiyan. Shanxi
Province, China. CICERO report. Available: 
https://pub.cicero.oslo.no/cicero-xmlui/bitstream/handle/ 
11250/192036/CICERO _ Report _ 2003-01.pdf?sequence=1 . 
Accessed June 11, 2021.
Mestl, H.E.S., Aunan, K., Fang, J., Seip, H.M., Skjelvik, J.M.,
Vennemo, H., 2005. Cleaner production as climate 
investment—integrated assessment in Taiyuan City. China. J. 
Clean Prod. 13 (1), 57–70 .
Mestl, H.E.S., Aunan, K., Seip, H.M., 2006. Potential health benefit 
of reducing household solid fuel use in Shanxi province. 
China. Sci. Total Environ. 372 (1), 120–132 .
Misra, A., Roorda, M.J., MacLean, H.L., 2013. An integrated 
modelling approach to estimate urban traffic emissions. 
Atmos. Environ. 73, 81–91 .
Nowak, D.J., Hirabayashi, S., Bodine, A., Hoehn, R., 2013. Modeled 
PM 2. 5 removal by trees in ten US cities and associated health 
effects. Environ. Pollut. 178, 395–402 .
Onay, T.T., Copty, N.K., Gökçe, H.B., Aydın-Sarıkurt, D., Mumcu, M.,
Arıo ̆glu, E., 2019. Air quality impact assessment for the Eurasia
Tunnel in Istanbul. Turkey. Environ. Monit. Assess. 191 (3), 
1–15 .
O’Shaughnessy, P.T., Altmaier, R., 2011. Use of AERMOD to 
determine a hydrogen sulfide emission factor for swine 
operations by inverse modeling. Atmos. Environ. 45 (27), 
4617–4625 .
Ott, W.R., 1990. A physical explanation of the lognormality of 
pollutant concentrations. J. Air Waste Manag. Assoc. 40 (10), 
1378–1383 .
Ounis, H., Ahmadi, G., 1990. A comparison of Brownian and 
turbulent diffusion. Aerosol Sci. Technol. 13 (1), 47–53 .
Petersen, R.L., Guerra, S.A., Bova, A.S., 2017. Critical review of the 
building downwash algorithms in AERMOD. J. Air Waste 
Manag. Assoc. 67 (8), 826–835 .
Pérez, N., Pey, J., Cusack, M., Reche, C., Querol, X., Alastuey, A.,
et al., 2010. Variability of particle number, black carbon, and 
PM 10 , PM 2. 5 , and PM1 levels and speciation: influence of road 
traffic emissions on urban air quality. Aerosol Sci. Technol. 44 
(7), 487–499 .
Perry, S.G., Cimorelli, A.J., Paine, R.J., Brode, R.W., Weil, J.C.,
Venkatram, A., et al., 2005. AERMOD: A dispersion model for 
industrial source applications. Part II: Model performance 
against 17 field study databases. J. Appl. Meteorol. Climatol. 44
(5), 694–708 .
Piwowar, A., Dziku ́c, M., 2019. Development of renewable energy 
sources in the context of threats resulting from low-altitude 
emissions in rural areas in Poland: a review. Energies 12 (18) 
p.3558 .
Pohl, H.R., Citra, M., Abadin, H.A., Szadkowska-Sta ́nczyk, I.,
Kozajda, A., Ingerman, L., et al., 2017. Modeling emissions from
CAFO poultry farms in Poland and evaluating potential risk to 
surrounding populations. Regul. Toxicol. Pharmacol. 84, 18–25 .
Pournazeri, S., Princevac, M., Venkatram, A., 2012. Rise of buoyant 
emissions from low-level sources in the presence of upstream
and downstream obstacles. Boundary Layer Meteorol 144 (2), 
287–308 .
Stockie, J.M., 2011. The mathematics of atmospheric dispersion 
modeling. SIAM Rev 53 (2), 349–372 .
Tomasi, E., Antonacci, G., Giovannini, L., Zardi, D., Ragazzi, M., 
2015. Atmospheric dispersion modelling with AERMOD for 
comparative impact assessment of different pollutant 
emission sources in an Alpine valley. WIT Trans on Ecol 
Environ 198, 431–442 .
Tong, R., Liu, J., Wang, W., Fang, Y., 2020. Health effects of PM 2. 5 
emissions from on-road vehicles during weekdays and 
weekends in Beijing, China. Atmos. Environ. 223 p.117258 .
Tyovenda, A.A., Ayua, T.J., Sombo, T., 2021. Modeling of gaseous 
pollutants (CO and NO 2 ) emission from an industrial 
stack in Kano city, northwestern Nigeria. Atmos. Environ. 253 
p.118356 .
US Code of Federal Regulations (CFR), 2006. National ambient air 
quality standards for particulate matter; final rule. 40 CFR, 
Part 50. U.S. Government Printing Office, Washington, D.C..
US EPA, 2000a. Guidance for data quality assessment. Practical 
methods for data analysis, Washington, DC .
US EPA, 2000b. Meteorological monitoring guidance for regulatory
modeling applications.
US EPA, 2004. AERMOD. Description of model formulation, 
Washington, DC .
US EPA, 2018. Table of historical particulate matter (PM) national 
ambient air quality standards (NAAQS). Available: 
https://www.epa.gov/pm-pollution/table-historical- 
particulate-matter-pm-national-ambient-air-quality- 
standards-naaqs. Accessed December 8, 
2019.
US EPA, 2019. Air Quality Dispersion Modeling - Preferred and 
Recommended Models. Available: 
https://www.epa.gov/scram/air-quality-dispersion-modeling- 
preferred-and-recommended-models. Accessed December 8, 
2019.
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0030
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0030
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0030
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0030
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0031
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0031
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0031
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0031
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0032
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0033
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0033
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0033
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0034
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0034
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0034
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0034
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0034
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0035
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0035
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0035
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0036
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0036
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0036
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0036
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0036
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0036
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0036
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0037
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0038
http://refhub.elsevier.com/S1001-0742(22)00163-2/sbref0038