Application of the Parametric Bootstrap Method for Confidence
Interval Estimation and Statistical Analysis of PM2.5 in Bangkok
BOONYARIT CHOOPRADIT*, RUJAPA PAITOON, NATTAWADEE SRINUAN, SATITA KWANKAEW
Department of Mathematics and Statistics, Faculty of Science and Technology,
Thammasat University, Pathumthani,
THAILAND
*Corresponding Author
Abstract: - Research in epidemiology and health science indicates that exposure to particles with an
aerodynamic diameter of less than 2.5 µm (PM2.5) causes harmful health consequences. Probability density
functions (pdf) are utilized to analyze the distribution of pollutant data and study the occurrence of high-
concentration occurrences. In this study, PM2.5 concentrations (in 3
μg m ) were recorded daily from January
2011 to December 2022 at 12 air quality monitoring locations in Bangkok. The study utilized two-parameter
distributions such as gamma, inverse Gaussian, lognormal, log-logistic, Weibull, and Pearson type V to identify
the most suitable statistical distribution model for PM2.5 in Bangkok. The Anderson-Darling test result
indicates that the inverse Gaussian and Pearson type V distributions are the most appropriate probability
density functions for the daily average PM2.5 concentration at stations in Bangkok. The projected 98th
percentile of daily PM2.5 levels at two locations is higher than the 24-hour threshold for daily PM2.5
concentrations in Thailand, posing significant health risks. Additionally, the two parametric bootstrap methods
used to estimate confidence intervals for the median, namely percentile bootstrap and simple bootstrap, indicate
that two stations have poor air quality for those with sensitive health conditions.
Key-Words: - PM2.5, Air pollution, Statistical analysis, Distribution, Parametric bootstrap, Bangkok
Received: June 25, 2023. Revised: March 19, 2024. Accepted: April 21, 2024. Published: May 22, 2024.
1 Introduction
Numerous epidemiological and toxicological studies
have consistently identified ambient fine particulate
matter (PM2.5) as a significant and detrimental
factor for human health. PM2.5 refers to particles
with a diameter of less than 2.5 µm, [1]. In 2019, air
pollution was classified as the fourth-most
significant risk factor in terms of its contribution to
premature mortality worldwide. Furthermore, in
terms of worldwide relevance, only high blood
pressure, tobacco consumption, and insufficient
eating habits exceeded it as contributing factors.
Countries situated in Asia, Africa, and the Middle
East persistently grapple with the most substantial
levels of atmospheric pollutant particle matter.
Based on the Health Effects Institute's 2020 study,
this harmful pattern underscores the ongoing and
substantial problems that these specific regions have
about air quality and the associated public health
risks, [2]. An estimation study by [3] found a
considerable number of deaths, more than 5500,
linked to PM2.5 pollution in the environment.
Thailand is rated 57th in the world for air quality
according to the 2022 World Air Quality Report by
[4], implying its significance as a nation with
environmental pollution issues. In 2022, Thailand
had an average concentration of PM2.5 particles
lowering to 18.1 3
μg m , meaning its air quality
increased. This indicates a sharp drop of 10.4%
from the amounts mentioned in 2021. The provinces
of Khon Kaen, Mae Hong Son, Chiang Mai,
Bangkok, Nonthaburi, and Nakhon Ratchasima
faced the highest density of PM2.5 concentrations
according to the geographical breakdown of air
pollution in Thailand. The information shows that
there are often geographical differences in the
quality of the air, with the provinces that are
emphasized having higher pollution readings. In
addition, Bangkok, the capital of Thailand, ranks
52nd in the world among capital cities in a global
study of capital cities. Indicating the trials faced by
a city when trying to resolve issues concerning air
quality at the urban level, Bangkok conveyed an
average PM2.5 concentration of 18
3
μg m in 2022.
According to [5], we have determined that the
average concentration of PM2.5 during a 24-hour
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.22 Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496 215 Volume 20, 2024
period >75
3
μg m
indicates that the air quality
level is harmful to people's health.
The study of statistical distributions is important
in the field of statistics as it reveals patterns of data,
[6], [7]. The statistical distribution model is utilized
to analyze the distribution of air pollutant data and
assess the occurrence of high-concentration
occurrences. By choosing a suitable statistical
distribution for air pollutants, the mean or median
concentration may be reliably predicted, [8], [9],
[10]. For this reason, a variety of statistical
distributions have been used in literature to describe
concentrations of PM2.5 distributions including
lognormal (LN), gamma, inverse Gaussian (IG),
log-logistic (LL), Weibull, and Pearson type V or
Inverse gamma (Pearson V) distributions, [9], [10],
[11].
A confidence interval (CI) is a statistical range of
values that is highly probable to encompass the
actual value of the population parameter of interest.
The CI for a parameter of interest is calculated by
removing or adding the product of the standard error
and a critical value. The computation assumes that
the estimate of the parameter conforms to an
approximately normal distribution. Regrettably,
when the assumption of normality is broken or the
maximum likelihood estimator cannot be easily
determined, it may be challenging to calculate these
CIs. In such cases, an additional method like the
bootstrap method can be employed, [12], [13], [14],
[15], [16].
Familiarity with the statistical distribution is a
crucial element in comprehending the statistical
characteristics of PM2.5 levels in Bangkok. Prior
studies have mostly examined the statistical
characteristics of PM2.5 levels and determined the
suitable distribution for PM2.5 values, [9], [10],
[11]. Therefore, this study analyzed the daily PM2.5
values recorded at twelve air quality monitoring
sites in Bangkok. These data were analyzed using
six two-parameter statistical distribution models to
determine the most suitable distribution for
capturing the daily PM2.5 values. Furthermore,
utilize a suitable distribution for estimating median
confidence intervals. The median values across all
stations were calculated with parametric bootstrap
confidence intervals. This is an advantage of this
study if there is no closed-form equation for the
median. The subsequent sections of this work are
structured in the following manner: Section 2
contains a detailed explanation of the materials and
techniques used in this study. Section 3 shows the
results. Finally, Section 4 includes the conclusion
and provides recommendations for future research.
2 Methods and Materials
2.1 Data
The current investigation acquired historical daily
concentrations of PM2.5 data (in
3
μg m
) from
January 2011 to December 2022 for the 12 air
quality monitoring sites in Bangkok. This data was
collected from the Pollution Control Department
under the Ministry of Natural Resources and
Environment, [17]. The station codes are displayed
below: X02T - Bansomdejchaopraya Rajabhat
University, X03T - Along Kanchanaphisek Road,
X05T - Meteorological Department (Bang Na),
X10T - Klong Chan Housing Community, X11T -
Huai Khwang Housing Community Stadium, X12T
- Nonsi Witthaya School, X50T - Chulalongkorn
Hospital, X52T - Metropolitan Electricity Authority
Thonburi Substation, X53T - Chokchai
Metropolitan Police Station, X54T - Din Daeng
Housing Community, X59T - The Government
Public Relations Department, X61T - Bodindecha
(Sing Singhaseni) School.
2.2 Distributions
This study primarily examines six two-parameter
distributions often employed to model daily
concentrations of PM2.5 data: gamma, inverse
Gaussian (IG), lognormal (LN), log-logistic (LL),
Weibull, and Pearson type V or Inverse gamma
(Pearson V) distributions. Table 1 summarizes the
probability density function (PDF), cumulative
distribution function (CDF), and distribution
parameters of these six distributions, [18], [19],
[20], [21], [22].
2.3 Maximum Likelihood Estimation
The specific characteristics of a theoretical
distribution are contingent upon the precise values
of its parameters. The most suitable parameter
values for the distributions were determined through
the utilization of the maximum likelihood estimation
(MLE) method. Let
12
, , , n
x x x
be a random sample
of size n drawn from a PDF of each
i
x
is
12
; , , ,
ik
fx
where
k
is an unknown k-
parameters. The likelihood function
L
associated with this random sample is defined as a
function of the unknown parameters in the following
manner:
12
1
L ; , , ,
n
ik
i
fx
where
12
, , , n
x x x
are the independent observations
from a random sample. Given the differentiability of
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
216
Volume 20, 2024
the likelihood function at
12
, , , k
, the
estimation of the parameters through the maximum
likelihood method involves computing partial
derivatives of concerning each parameter and
subsequently solving the resulting set of k equations
to find their zeros. Typically, computations involve
utilizing the logarithm of the likelihood function,
given that it is a strictly increasing function.
Consequently, the same parameter values will
optimize both the likelihood function and the log-
likelihood function. Thus, the MLE of
12
, , , k
is a solution of
ln L 0
k
.
The MLE for
12
, , , k
can be derived by
solving the resulting equations simultaneously
through a numerical procedure, such as the Newton-
Raphson method. In this study, the MLE estimates
of
12
ˆ ˆ ˆ
, , , k
are obtained using the "fitdist"
function from the "fitdistrplus" package in the R
software suite, [23].
2.4 Assessment of Goodness-of-fit Test
The aforementioned six functions were compared
using various statistical approaches, such as the
Kolmogorov-Smirnov, Anderson-Darling (AD), and
Chi-Square goodness-of-fit tests. Nevertheless, the
empirical data provided by [24], indicates that the
AD goodness-of-fit test statistic, [25], is superior to
other tests when it comes to evaluating the
suitability of a positively skewed distribution.
Hence, it is advisable to utilize the AD test to assess
the adequacy of fit in this particular study, given this
discovery. The test statistic for the AD test may be
denoted by the following mathematical expression:
2
1
1
21
ln ln 1
n
i n i
i
i
A n F x F x
n


,
where F(.) is the expected CDF,
i
x
are the data
must be put in order
1n
xx
, and n is the
sample size. A lower AD statistic result indicates the
appropriateness of the statistical distribution being
examined, [7].
Table 1. Statistical properties of daily concentrations of PM2.5 distribution functions
Distribution
PDF
CDF
Parameter
gamma
1
( ; , ) exp
()
x
f x x


,
0x
F( ; , ) x
x

shape parameter
0
scale parameter
0
IG
2
12
32
; , exp
22
x
fx xx

 





,
0x
12
12
F( ; , ) 1
2
exp 1
x
xx
x
x



















shape parameter
0
mean
0
LN
2
2
2
ln x
1
; , exp 2
2
fx x







,
0x
2ln
F ; , x
x





location parameter
,
scale parameter
0
LL
2
;,
1
x
fx xx




,
0x
F ; , 1
x
xx

scale parameter
0
shape parameter
0
Weibull
1
( ; , ) exp
x
f x x






,
0x
F ; , 1 expxx


shape parameter
0
scale parameter
0
Pearson V
1
exp
( ; , ) x
fx x

,
0x
F( ; , ) 1 ;xx
shape parameter
0
scale parameter
0
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
217
Volume 20, 2024
2.5 Parametric Bootstrap Method for
Estimating Confidence Interval for the
Median
When dealing with skewed distributions, it is
important to note that the arithmetic mean,
sometimes known as the "mean," tends to be located
more toward the tail of the distribution compared to
the median. Therefore, the median is often preferred
as a measure of central tendency because the mean
does not always correspond to the center location in
the distribution, [26]. The determination of the
median (M) of a distribution depends on the specific
characteristics that define that distribution. To
calculate the median, one must use the cumulative
distribution function (CDF) of the particular
distribution being considered, denoted as CDF(x).
To get the median precisely, we need to solve the
equation CDF(M) = 0.5 using numerical techniques.
Regrettably, the Pearson type V distribution lacks a
straightforward mathematical equation to calculate
its median, unlike several other distributions.
Nevertheless, it is possible to approximate the
median by utilizing the quantile function of the
Pearson V distribution. Constructing a confidence
interval for the median of the Pearson V distribution
requires the use of statistical estimating techniques,
such as the bootstrap method, because there is no
closed-form equation for the median. The bootstrap
methodology was initially described by [27], who
provided a comprehensive explanation of the
fundamental principles underlying the basic
bootstrap approach. In [16] the authors presented a
comprehensive summary of the various bootstrap
methods. The authors classified and contrasted
several bootstrap approaches, demonstrating their
respective benefits. A detailed analysis, especially
comparing bootstrap confidence interval
approaches, can be found in the study conducted by
[15]. Their study examined the characteristics of
several bootstrap confidence interval methods.
Further analysis of the options for applying
bootstrap confidence intervals is provided by [14].
This resource offers practical advice on choosing
and implementing suitable methods for bootstrap
confidence intervals.
2.5.1 Parametric Percentile Bootstrap (PB)
Method
The procedure for constructing a parametric PB
confidence interval for the median of any
distribution may be summarized as follows:
1) Fit a parametric distribution to the
original sample data. Estimate the distribution
parameters.
2) Generate B bootstrap samples by
sampling randomly with replacement from the fitted
parametric distribution. The bootstrap sample size
equals the original sample size. This study uses B =
100,000.
3) Compute the median of each bootstrap
sample. This gives B estimates of the median.
4) Sort the B bootstrap median estimates in
ascending order.
5) The
2B
th and
12B
th values in
the sorted bootstrap medians provide the lower and
upper confidence limits of a
100 1 %
CI for the
true median.
6) Typically
= 0.05 is used for a 95% CI.
So the 2.5th and 97.5th percentile bootstrap medians
give the 95% CI endpoints.
2.5.2 Parametric Simple Bootstrap (SB) Method
The procedure for constructing a parametric SB
confidence interval for the median of any
distribution may be summarized as follows:
1) Fit a parametric distribution to the
original sample data. Estimate the distribution
parameters. Find the median (m) from the original
sample data.
2) Generate B bootstrap samples by
sampling randomly with replacement from the fitted
parametric distribution. The bootstrap sample size
equals the original sample size. This study uses B =
100,000.
3) Compute the median of each bootstrap
sample. This gives B estimates of the median.
4) Sort the B bootstrap median estimates in
ascending order.
5) The
2B
th and
12B
th values in
the sorted bootstrap medians represent the lower (L)
and upper (U) limits.
6) The 2m - U and 2m - L values provide the
lower and upper confidence limits of a
100 1 %
CI for the true median.
3 Results and Discussion
Section 3.1 presents a summary of the statistical
properties of the daily PM2.5 concentrations
measured at the 12 chosen stations in Bangkok.
Subsequently, in Section 3.2, the suitable
distributions and estimated parameters for daily
PM2.5 levels at each station are determined.
Calculations are conducted to determine the
percentiles and exceedance probability for each
station indicated in Section 3.3. Section 3.4
calculates confidence intervals for the median at
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
218
Volume 20, 2024
each station using the parametric bootstrap
technique.
Fig. 1: Boxplots displaying the daily measurements of PM2.5 levels at the specified station in Bangkok
Table 2. Descriptive summary of daily PM2.5 levels (in
3
μg m
) at the specified station in Bangkok
Station
na
Mean
Median
Mode
SDb
Min.
Max.
Q1
Q2
Q3
Skewness
Kurtosis
X02T
1221
23.91
20.00
14.00
13.45
5
102
14.00
20.00
30.00
1.53
3.10
X03T
1524
31.63
28.00
21.00
14.71
11
131
21.00
28.00
38.00
1.77
4.56
X05T
2117
21.65
17.00
13.00
13.83
4
100
12.00
17.00
28.00
1.47
2.42
X10T
1477
20.55
17.00
11.00
11.86
4
84
12.00
17.00
26.00
1.43
2.29
X11T
1528
21.84
18.00
10.00
12.29
4
81
12.25
18.00
28.00
1.34
1.88
X12T
1217
21.41
19.00
12.00
11.12
5
78
13.00
19.00
27.00
1.36
2.16
X50T
2007
26.04
22.00
16.00
12.63
8
92
17.00
22.00
32.00
1.55
2.84
X52T
2160
25.60
21.00
15.00
15.03
5
105
15.00
21.00
32.00
1.63
3.12
X53T
1972
23.58
20.00
14.00
13.79
4
89
13.25
20.00
30.00
1.43
2.31
X54T
1607
32.56
29.00
22.00
13.78
13
112
22.00
29.00
39.00
1.60
3.60
X59T
2088
19.92
17.00
10.00
12.18
3
97
11.00
17.00
25.00
1.49
2.74
X61T
2088
23.11
19.00
14.00
12.42
7
94
15.00
19.00
28.00
1.79
3.79
an=sample size
bSD=standard deviation
3.1 Descriptive Statistics
Prior to finding appropriate distributions, a thorough
analysis of the descriptive statistics was performed
on the daily concentrations of PM2.5 measured at
the designated 12 sites in Bangkok. The findings of
this examination are shown and condensed in Figure
1 and Table 2. Figure 1 presents the boxplots that
depict the daily PM2.5 values measured at the
specified station in Bangkok. The graphic depiction
clearly shows that the distribution of daily PM2.5
values observed at all stations has a positive
skewness. Table 2 reveals that station X54T had the
highest average daily concentrations of PM2.5, with
a maximum value of 32.56
3
μg m
. In contrast,
station X59T registered the minimum average daily
concentrations of PM2.5, measuring 19.92
3
μg m
.
Furthermore, station X54T had the greatest median
daily concentrations of PM2.5 at 29.00
3
μg m
,
whereas stations X05T, X10T, and X59T had the
lowest median daily concentrations of PM2.5 of
17.00
3
μg m
. Based on the remaining output
presented in Table 2, it clearly shows that the
distribution of daily PM2.5 values observed at all
sites has a noticeable positive skewness.
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
219
Volume 20, 2024
Fig. 2: Comparison of histograms and theoretical densities for statistical distributions evaluated on daily
PM2.5 values observed at the Bangkok station
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
220
Volume 20, 2024
Table 3. The AD goodness-of-fit test result for six distribution functions of daily PM2.5 values in Bangkok
Station
Distribution
gamma
IG
LN
LL
Weibull
Pearson V
X02T
14.631
6.083
7.156
8.875
23.628
4.213
X03T
21.523
11.581
11.541
11.852
42.110
6.258
X05T
18.310
3.440
5.179
7.850
32.483
5.388
X10T
11.488
3.165
4.219
6.265
22.027
4.004
X11T
14.434
5.918
7.229
9.661
24.191
6.028
X12T
8.421
2.152
2.772
4.229
18.537
2.255
X50T
28.853
13.992
14.442
14.562
51.731
6.665
X52T
27.031
9.046
10.399
11.791
46.220
4.032
X53T
12.935
2.425
3.577
6.458
26.899
4.716
X54T
18.762
9.583
9.665
10.154
40.511
4.830
X59T
12.921
2.353
3.508
6.389
27.071
6.449
X61T
43.266
23.075
22.877
20.196
69.617
11.095
Table 4. Estimated parameters of the fitted distributions of daily PM2.5 for the indicated station in Bangkok
Station
Fitted
Distribution
Estimated
parameters
Station
Fitted
Distribution
Estimated
parameters
X02T
Pearson V
ˆ
4.19782,
ˆ
77.40105
X50T
Pearson V
ˆ
5.94554,
ˆ
128.68040
X03T
Pearson V
ˆ
6.66742,
ˆ
178.79450
X52T
Pearson V
ˆ
4.03954,
ˆ
78.71426
X05T
IG
ˆ
21.65233,
ˆ
50.79283
X53T
IG
ˆ
23.57180,
ˆ
66.94004
X10T
IG
ˆ
20.54397,
ˆ
62.31721
X54T
Pearson V
ˆ
7.66919,
ˆ
216.63070
X11T
IG
ˆ
21.84124,
ˆ
69.83626
X59T
IG
ˆ
19.92205,
ˆ
51.45492
X12T
IG
ˆ
21.40408,
ˆ
81.00040
X61T
Pearson V
ˆ
5.37718,
ˆ
100.73330
3.2 Distributions of Daily Concentrations of
PM 2.5
Figure 2 displays histograms that show daily
concentrations of PM2.5 measured at 12 sites. The
histograms also include the fitted distributions for
each of the six models that were investigated. The
IG, LL, LN, and Pearson V distributions show a
clear and high agreement with the data histogram of
all stations. Table 3 presents the AD goodness-of-fit
test statistic for six distributions. The smaller values
in the table imply a better match with the actual
daily concentrations of PM2.5 data. The daily
PM2.5 values at the monitored stations, namely
X02T, X03T, X50T, X52T, X54T, and X61T,
showed a strong correlation with the Pearson V
distribution. The IG distribution was determined to
be the most suitable statistical distribution for the
daily concentrations of PM2.5 at stations X05T,
X10T, X11T, X12T, X53T, and X59T. The
maximum likelihood technique is used to estimate
the parameters that describe the distributions of
daily PM2.5 values. This estimation is done using
the "fitdist" function from the "fitdistrplus" package
in the R language. The estimated parameters for the
distributions of PM2.5 values fitted to each station
are displayed in Table 4.
3.3 Percentiles and Exceedance Probabilities
The administrative targets for air pollution
management typically range from the 98.0th to
99.9th percentile, [10], [11]. Therefore, it is crucial
to carefully determine the distribution of daily
PM2.5 levels in the higher range. Table 5 presents
estimated values for the 98th percentile of PM2.5
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
221
Volume 20, 2024
levels, as well as the probabilities of detecting levels
higher than 75
3
μg m
at certain sites. The
predicted 98th percentile of daily PM2.5 levels for
stations X03T and X52T are 76.03 and 76.09
3
μg m
, respectively. These values exceed the 24-
hour threshold for daily concentrations of PM2.5 in
Thailand, which is set at 75
3
μg m
. The daily
concentrations of PM2.5 at all stations have
predicted 98th percentile values that exceed the
USA EPA's 24-hour standard of 35
3
μg m
, [28].
Consequently, the threshold of 75
3
μg m
(in
Thailand) and 35
3
μg m
(in the USA) was
surpassed by the top two percent of the projected
one-year data. This percentile is the equivalent of 7
days out of a year.
To analyze the Thailand 24-hour standard, we
calculate the probability of finding levels higher
than 75
3
μg m
, denoted as P(PM2.5>75), based on
the Thailand air quality criteria provided in [5] as
the average concentrations of PM2.5 during 24
hours. The probability of exceeding certain
thresholds is displayed in Table 5. The data
indicates that station X12T has the lowest
probability, specifically 0.0019, while station X52T
has the highest probability, specifically 0.0209.
Table 5. Estimated 98th percentile for PM2.5 levels (in
3
μg m
) and the probabilities for detecting levels
greater than 75
3
μg m
at the indicated stations and using the indicated distribution
Station
Fitted Distribution
PM2.5 (in
3
μg m
)
P(PM2.5>75)
Measured
Predicted
X02T
Pearson V
60.00
67.81
0.0155
X03T
Pearson V
74.00
76.03
0.0166
X05T
IG
61.00
62.44
0.0083
X10T
IG
54.00
52.57
0.0030
X11T
IG
56.92
55.29
0.0038
X12T
IG
52.00
52.53
0.0019
X50T
Pearson V
61.00
65.85
0.0090
X52T
Pearson V
70.00
76.09
0.0209
X53T
IG
63.00
63.22
0.0082
X54T
Pearson V
71.76
73.29
0.0139
X59T
IG
55.00
55.22
0.0041
X61T
Pearson V
60.00
58.19
0.0069
Table 6. The 95% confidence intervals for the median of daily PM2.5 at the specified 12 stations in Bangkok
Station
PB confidence intervals
SB confidence intervals
X02T
(19.30765, 20.72887)
(19.27113, 20.69235)
X03T
(27.52335, 28.92723)
(27.07277, 28.47665)
X05T
(17.34427, 18.52679)
(15.47321, 16.65573)
X10T
(17.07891, 18.32170)
(15.67830, 16.92109)
X11T
(18.31270, 19.58816)
(16.41184, 17.68730)
X12T
(18.29317, 19.61972)
(18.38028, 19.70683)
X50T
(22.39272, 23.44950)
(20.55050, 21.60728)
X52T
(20.63952, 21.79127)
(20.20873, 21.36048)
X53T
(19.48881, 20.74965)
(19.25035, 20.51119)
X54T
(28.86097, 30.19331)
(27.80669, 29.13903)
X59T
(16.22465, 17.29611)
(16.70389, 17.77535)
X61T
(19.48698, 20.43835)
(17.56165, 18.51302)
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
222
Volume 20, 2024
3.4 Confidence Intervals for the Median of
Daily PM2.5 Levels
The confidence intervals for the median of daily
PM2.5 readings at the 12 designated stations in
Bangkok are displayed in Table 6. The procedure
for constructing parametric PB and SB confidence
intervals for the median is outlined in the preceding
section. The findings indicate that the majority of
stations, except X03T and X54T, exhibit air quality
that is deemed acceptable. However, it is important
to note that certain individuals, particularly those
who are very susceptible to air pollution, may still
face potential health risks. Stations X03T and X54T
have unhealthy air quality for sensitive groups.
Individuals who are part of sensitive groups may
have adverse health impacts, whereas the general
public is less likely to be impacted.
4 Conclusion
Within this paper, an analysis was conducted on
daily concentrations of PM2.5 values at twelve air
quality monitoring sites in Bangkok. Six two-
parameter statistical distribution models were used
to identify the most appropriate distribution for
accurately representing daily concentrations of
PM2.5 data. The confidence intervals for the median
of daily PM2.5 at each station have been computed.
The daily concentrations of PM2.5 data at stations
X02T, X03T, X50T, X52T, X54T, and X61T
followed a Pearson V distribution. After careful
analysis, it was concluded that the IG distribution is
the most appropriate statistical distribution for
stations X05T, X10T, X11T, X12T, X53T, and
X59T. The PM2.5 data at all sites have been
predicted to surpass the 98th percentile values of the
USA EPA's 24-hour threshold of 35
3
μg m
.
Moreover, the projected 98th percentile values
above the 24-hour threshold of 75
3
μg m
for daily
concentrations of PM2.5 in Thailand at stations
X03T and X52T. Confidence intervals indicate that
stations X03T and X54T have air quality that is
detrimental to sensitive populations and may have
negative health effects based on the median of daily
PM2.5 results.
This paper specifically focuses on the limitations
of two-parameter statistical distributions. Future
research should focus on analyzing distributions
with more than two parameters and prioritize
investigating the correlation between PM2.5 and
meteorological factors. Furthermore, it is important
to examine the annual mortality rates related to
PM2.5 in significantly affected areas like Bangkok
or other regions in Thailand that are experiencing
significant PM2.5 issues.
Acknowledgement:
We extend our gratitude to the referees for their
valuable suggestions on the manuscript and
appreciate the helpful guidance provided by
Associate Professor Patchanok Srisuradetchai. The
authors gratefully acknowledge the financial support
provided by the Faculty of Science and Technology,
Thammasat University, Contract No. 1/2566.
References:
[1] A. Garcia, E. Santa-Helena, A. De Falco, J.
de Paula Ribeiro, A. Gioda, and C. R. Gioda,
“Toxicological Effects of Fine Particulate
Matter (PM2.5): Health Risks and
Associated Systemic Injuries—Systematic
Review,” Water Air Soil Pollut., vol. 234,
no. 6, p. 346, 2023, doi:
https://doi.org/10.1007/s11270-023-06278-9.
[2] Health Effects Institute, “State of Global Air
2019,” Boston, 2019, [Online].
https://www.stateofglobalair.org/sites/default
/files/soga_2019_report.pdf (Accessed Date:
July 13, 2023).
[3] N. R. Fold, M. R. Allison, B. C. Wood, P. T.
B. Thao, S. Bonnet, S. Garivait, R. Kamens,
and S. Pengjan, “An Assessment of Annual
Mortality Attributable to Ambient PM2.5 in
Bangkok, Thailand,” Int J Environ. Res.
Public Health, vol. 17, no. 19, p. 7298, Oct.
2020, doi:
https://doi.org/10.3390/ijerph17197298.
[4] IQAir, “The 2022 World Air Quality Report:
Region & City PM2.5 Ranking,” 2022,
[Online].
https://www.greenpeace.org/static/planet4-
thailand-stateless/2023/03/d1d69c24-
2022_world_air_quality_report_en.pdf
(Accessed Date: July 13, 2023).
[5] P. Suraswadi, “Pollution Control Department
announces Thailand’s 2023 Air Quality
Index,” [Online].
https://ratchakitcha.soc.go.th/documents/140
D157S0000000000300.pdf?fbclid=IwAR2bJ
Gm_x9Df4ZBCOu-
T9_RqDUdd27EhzGBL1jjhuU1lb-
Wo6k6t4YKXsxM (Accessed Date: January
25, 2024).
[6] S. Wasinrat and B. Choopradit, “The Poisson
Inverse Pareto Distribution and Its
Application,” Thailand Statistician, vol. 21,
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
223
Volume 20, 2024
no. 1, pp. 110–124, 2023, [Online].
https://ph02.tci-
thaijo.org/index.php/thaistat/article/view/248
027 (Accessed Date: June 25, 2023).
[7] B. Choopradit and S. Wasinrat, “The
Poisson-Paralogistic Distribution: Properties
and Applications,” Journal of Applied
Science and Emerging Technology, vol. 22,
no. 1, p. e249656, Apr. 2023, [Online].
https://ph01.tci-
thaijo.org/index.php/JASCI/article/view/249
656 (Accessed Date: June 25, 2023).
[8] W. E. Wilson, J. C. Chow, C. Claiborn, W.
Fusheng, J. Engelbrecht, and J. G. Watson,
“Monitoring of particulate matter outdoors,”
Chemosphere, vol. 49, no. 9, pp. 1009-1043,
2002, doi: https://doi.org/10.1016/S0045-
6535(02)00270-9.
[9] H.-C. Lu and G.-C. Fang, “Estimating the
frequency distributions of PM10 and PM2.5
by the statistics of wind speed at Sha-Lu,
Taiwan,” Science of The Total Environment,
vol. 298, no. 1, pp. 119–130, 2002, doi:
https://doi.org/10.1016/S0048-
9697(02)00164-X.
[10] F. Karaca, O. Alagha, and F. Ertürk,
“Statistical characterization of atmospheric
PM10 and PM2.5 concentrations at a non-
impacted suburban site of Istanbul, Turkey,”
Chemosphere, vol. 59, no. 8, pp. 1183–1190,
2005, doi:
https://doi.org/10.1016/j.chemosphere.2004.
11.062.
[11] C. Marchant, V. Leiva, M. F. Cavieres, and
A. Sanhueza, “Air contaminant statistical
distributions with application to PM10 in
Santiago, Chile,” in Reviews of
Environmental Contamination and
Toxicology Volume 223, in Reviews of
Environmental Contamination and
Toxicology. , Springer Science and Business
Media, LLC, 2013, pp. 1–31. doi:
https://doi.org/10.1007/978-1-4614-5577-
6_1.
[12] W. Panichkitkosolkul and P. Srisuradetchai,
“Bootstrap Confidence Intervals for the
Parameter of Zero-truncated Poisson-Ishita
Distribution,” Thailand Statistician, vol. 20,
no. 4, pp. 918–927, Sep. 2022, [Online].
https://ph02.tci-
thaijo.org/index.php/thaistat/article/view/247
474 (Accessed Date: July 5, 2023).
[13] M.-T. Puth, M. Neuhäuser, and G. D.
Ruxton, “On the variety of methods for
calculating confidence intervals by
bootstrapping,” Journal of Animal Ecology,
vol. 84, no. 4, pp. 892–897, 2015, doi:
https://doi.org/10.1111/1365-2656.12382.
[14] J. Carpenter and J. Bithell, “Bootstrap
confidence intervals: when, which, what? A
practical guide for medical statisticians,” Stat
Med, vol. 19, no. 9, p. 1141—1164, May
2000, doi:
https://doi.org/10.1002/(SICI)1097-
0258(20000515)19:9<1141::AID-
SIM479>3.0.CO;2-F.
[15] T. J. DiCiccio and B. Efron, “Bootstrap
Confidence Intervals,” Statistical Science,
vol. 11, no. 3, pp. 189–212, 1996, [Online].
http://www.jstor.org/stable/2246110
(Accessed Date: July 23, 2023).
[16] B. Efron and R. J. Tibshirani, An
Introduction to the Bootstrap, 1st ed.
Chapman and Hall/CRC, 1994.
[17] Pollution Control Department, “Daily PM2.5
concentration data for the 12 air quality
monitoring stations in Bangkok,” [Online].
http://air4thai.pcd.go.th/webV3/#/History
(Accessed Date: May 31, 2023).
[18] S. Klugman, H. Panjer, and G. Willmot, Loss
Models: From Data to Decisions, 5th ed.
Wiley, 2019.
[19] M. Evans, N. Hastings, and B. Peacock,
Statistical distributions, 3rd ed. New York:
Wiley, 2000.
[20] G. E. Crooks, “The Amoroso Distribution,”
arXiv: Statistics Theory, 2010, [Online].
https://doi.org/10.48550/arXiv.1005.3274
(Accessed Date: June 09, 2023).
[21] R. Chhikara and L. Folks, The Inverse
Gaussian Distribution: Theory,
Methodology, and Applications, 1st ed. CRC
Press, 2019.
[22] J. Wang, L. Zhang, X. Niu, and Z. Liu,
“Effects of PM2.5 on health and economic
loss: Evidence from Beijing-Tianjin-Hebei
region of China,” J Clean Prod, vol. 257, p.
120605, 2020, doi:
https://doi.org/10.1016/j.jclepro.2020.12060
5.
[23] M. L. Delignette-Muller and C. Dutang,
“fitdistrplus: An R Package for Fitting
Distributions,” J Stat Softw, vol. 64, no. 4,
pp. 1–34, 2015, doi:
https://doi.org/10.18637/jss.v064.i04.
[24] M. A. Stephens, “EDF Statistics for
Goodness of Fit and Some Comparisons,” J
Am Stat Assoc, vol. 69, no. 347, pp. 730–
737, 1974, doi:
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
224
Volume 20, 2024
https://doi.org/10.1080/01621459.1974.1048
0196.
[25] T. W. Anderson and D. A. Darling,
“Asymptotic Theory of Certain ‘Goodness of
Fit’ Criteria Based on Stochastic Processes,”
The Annals of Mathematical Statistics, vol.
23, no. 2, pp. 193 – 212, 1952, doi:
https://doi.org/10.1214/aoms/1177729437.
[26] A. Agresti and B. Finlay, Statistical Methods
for the Social Sciences, 3rd ed. Upper Saddle
River, NJ: Prentice Hall, 1997.
[27] B. Efron, “Bootstrap Methods: Another Look
at the Jackknife,” The Annals of Statistics,
vol. 7, no. 1, pp. 1 – 26, 1979, doi:
https://doi.org/10.1214/aos/1176344552.
[28] U.S. Environmental Protection Agency,
“Final Reconsideration of the National
Ambient Air Quality Standards for
Particulate Matter (PM),” [Online].
https://www.epa.gov/system/files/documents
/2024-02/pm-naaqs-overview.pdf (Accessed
Date: February 09, 2024).
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
- Boonyarit Choopradit: Conceived the research,
provided an original idea of the study, provided
methodology, analyzed the data, programmed,
provided a description, interpreted the data, and
wrote the paper.
- Rujapa Paitoon, Nattawadee Srinuan, and Satita
Kwankaew: Selected research data, and reviewed
the paper.
All authors discussed the results and contributed to
the final manuscript.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
This article was funded by the Faculty of Science
and Technology, Thammasat University, Contract
No. 1/2566.
Conflict of Interest
The authors have no conflicts of interest to declare.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
Boonyarit Choopradit, Rujapa Paitoon,
Nattawadee Srinuan, Satita Kwankaew
E-ISSN: 2224-3496
225
Volume 20, 2024