Predicting PM2.5 in Urban and Suburban Beijing: A Comparative
Study of Random Forest and Linear Regression Models
Zhuoyang Zhou
a
Beijing Normal - Hong Kong Baptist University, Zhuhai, 519000, China
Keywords: PM2.5, Regression Model, Random Forest, Prediction Model.
Abstract: This study evaluates the performance of multiple linear regression (MLR) and random forest regression (RFR)
models in predicting PM2.5 concentrations across twelve air quality monitoring stations in Beijing, China,
using hourly meteorological and pollution data from 2013 to 2017. The analysis reveals that RFR significantly
outperforms MLR, with values improving from 0.11 to 0.22 (MLR) to 0.29–0.41 (RFR), demonstrating
better handling of non-linear interactions. However, both models exhibit critical limitations, particularly in
predicting extreme pollution events (PM2.5 > 300 µg/m³), where systematic underprediction occurs.
Geographical disparities in model accuracy are evident, with suburban stations (e.g., Dingling, Huairou)
exhibiting lower errors than urban-industrial sites (e.g., Dongsi, Aotizhongxin), likely due to the complexity
of emission sources and microclimates. Dew point temperature emerges as the most influential predictor,
while precipitation shows limited impact. These findings underscore the challenges in air quality forecasting
and advocate for localised, hybrid modelling approaches integrating real-time emission data to enhance
predictive reliability for public health applications.
1 INTRODUCTION
Air pollution, particularly that 2.5 microns or smaller
(PM2.5), seriously threatens the environment and
people's health worldwide. Air pollution, including
PM2.5, directly impacts people's health, which may
cause heart problems and breathing difficulties
(Brook et al., 2010). Additionally, it harms the setting
and the essence (Li et al., 2019). To better formulate
recommendations for cleaner environments and
lessen the damaging effects of air pollution, people
must be aware of how these issues impact PM2.5.
PM2.5 deposition in the air significantly impacts
weather conditions, such as temperature, humidity,
and wind speed. For instance, higher temperatures
can increase PM2.5, while higher humidity can make
it easier to create tiny antigens (Li et al., 2019; Perrino
et al., 2011). The drizzle of pollutants depends on
wind speed. PM2.5 particles are typically higher
because the air doesn't mix well when the wind blows
(Tai et al., 2010). PM2.5 costs may be accurately
predicted because of how connected these elements
are. Besides weather conditions, personal activities,
especially retreat-related, can drastically change
a
https://orcid.org/0009-0001-6098-6604
PM2.5. Individuals move more during holidays,
corporations may work differently, and more
situations take place, which may affect air quality.
For example, during major festivals like Chinese New
Year, there are fewer factory activities and cars on the
road, which makes the air fresh for a short time
(Wang et al., 2014; Wang et al., 2017).
On the other hand, trips that involve more travel
and tourism may produce more pollutants from
transportation and places to stay, leading to higher
levels of PM2.5 (Zhang et al., 2015). It is crucial to
understand how PM2.5 costs and air quality change.
Using this data, better air quality management
strategies can be created during the lively holidays.
Some studies have examined how PM2.5 prices,
weather conditions, and actions are connected in
various locations worldwide. In a study conducted in
Beijing, China, the wind's temperature, humidity, and
wind speed significantly impacted the amount of
PM2.5 provided. When the weather and the wind
were cooler, PM2.5 costs increased (Zhang et al.,
2017). A study in the United States found that
weather conditions played a key role in modifying
PM2.5 levels, with temperature and wind speed being
Zhou, Z.
Predicting PM2.5 in Urban and Suburban Beijing: A Comparative Study of Random Forest and Linear Regression Models.
DOI: 10.5220/0013825300004708
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 2nd International Conference on Innovations in Applied Mathematics, Physics, and Astronomy (IAMPA 2025), pages 331-336
ISBN: 978-989-758-774-0
Proceedings Copyright © 2025 by SCITEPRESS Science and Technology Publications, Lda.
331
the most important aspects (Perrino et al., 2011). Due
to the large number of people using lights and
increased traffic on the roads, a study in India found
that during the Diwali festival, the levels of harmful
PM2.5 in the air significantly increased (Guo et al.,
2014). When measuring PM2, these analyses
demonstrate that. 5, both weather conditions and
specific activities should be taken into account.
Despite numerous studies, People still aren't
aware of the relationship between PM2.5 levels,
temperature, and holidays. Most studies focused on
PM2's impact on a single cultural problem (Pope and
Dockery, 2006). Not many studies have examined
how various weather conditions interact with one
another to alter it (Pope & Dockery, 2006).
Moreover, people don't understand how weather
conditions and breaks affect PM2.5 (Wang et al.,
2017).
To remove these deficiencies, this study will
examine how PM2.5 charges relate to weather
conditions and falls in a particular area. The
assessment uses information on air quality and
weather conditions to observe how PM2.5 (a type of
air pollution) rates change over weather changes. To
improve air quality management and develop better
rules and regulations, the research may utilise
multiple linear regression analysis to examine the
effects of weather conditions.
Complex components, like the environment and
the lives of women, are affected by PM2.5 exposure.
Knowing how these factors affect PM2.5 rates is
crucial to making effective air quality management
strategies. This research hopes to raise the
consciousness by understanding how PM2.5 rates,
temperature, and holidays relate to a specific area.
Policymakers and professionals can comprehend
these issues because of this (Tai et al., 2010; Zhang et
al., 2015).
2 METHODOLOGY
2.1 Data Source and Description
The dataset used in this study was obtained from
Kaggle, comprising hourly air quality measurements
from twelve monitoring stations in Beijing, China,
spanning from March 1st, 2013, to February 28th,
2017.
2.2 Indicator Selection and Description
Meteorological variables-TEMP, DEWP, PRES, and
RAIN-were chosen as independent variables for their
established influence on PM2.5 dispersion and
formation (as showing in Table 1). For instance,
PM2.5 concentrations exhibit considerable
variability, with hourly readings ranging from 3 to
500 µg/m³, highlighting the severity of pollution
episodes. Temperature and dew point display
seasonal trends, while precipitation events are
sporadic but critical for pollutant scavenging. Table 1
lists all the variable names and their descriptions, and
ranges.
Table 1: Descriptions and ranges of variables
Variable Description Range
PM2.5 Fine particulate matter
concentration
(µg
/m³
)
2.0 to 999.0
TEMP Temperature (°C) -19.9 to 42.6
DEWP Dew point temperature (°C) -43.3 to 29.1
PRES Atmospheric pressure (hPa) 982.4 to 1042.8
RAIN Precipitation (mm) 0.0 to 72.5
2.3 Methodology
The analysis employs two regression techniques:
multiple linear regression (MLR) and random forest
regression (RFR).
Multiple Linear Regression (MLR): MLR
establishes baseline relationships between PM2.5 and
meteorological factors, providing interpretable
coefficients for each predictor. The model is
formulated as: PM2.5 = 𝛽
+ 𝛽
TEMP + 𝛽
DEWP + 𝛽
PRES + 𝛽
RAIN + 𝜖 , where 𝛽
is
the intercept, 𝛽
to 𝛽
are coefficients, and 𝜖 is the
error term.
Random Forest Regression (RFR): RFR, a
machine learning approach, captures non-linear
interactions and improves predictive accuracy. The
model uses bootstrap aggregation and random feature
selection, with hyperparameters (e.g., ntree = 500)
tuned via cross-validation to prevent overfitting. Key
advantages include handling non-linearity and
robustness to outliers. Normalisation was included to
address scale differences and remove missing values
(na.omit).
The analytical workflow began with a stratified
data approach. Each station's dataset was divided into
training (70%) and testing (30%) subsets. Both MLR
and RFR models were then trained on the training
subsets. This study used three evaluation indices (the
coefficient of determination (R²), root mean squared
error (RMSE) and mean absolute error (MAE)) to
compare the performance of MLR and RFR models
in predicting PM2.5 concentrations across Beijing's
monitoring stations.
IAMPA 2025 - The International Conference on Innovations in Applied Mathematics, Physics, and Astronomy
332
3 RESULTS AND DISCUSSION
3.1 Multiple Linear Regression
By using the R code, this paper constructs the
multiple linear regression model of the 12 stations.
The regression coefficients are shown in Table 2.
Table 2: Multiple Linear Regression Coefficients by
Station
Statio
n
Interc
e
p
t
PRES TEMP DEW
P
RAIN
Aotizh
ongxi
n
1407.
088
-1.243 -5.819 4.005 -3.686
Chang
p
in
g
1039.
030
-0.903 -4.621 3.260 -4.126
Dingli
ng
1016.
191
-0.885 -4.611 3.342 -3.747
Dongs
i
2048.
266
-1.858 -6.647 4.300 -6.038
Guany
uan
1335.
387
-1.172 -5.770 4.047 -5.724
Guche
ng
1373.
740
-1.208 -5.801 3.843 -6.324
Huair
ou
490.3
54
-0.379 -3.600 2,885 -3.180
Nongz
hangu
an
1886.
982
-1.698 -6.840 4.300 -6.392
Shuny
i
1258.
115
-1.100 -5.488 3.907 -4.653
Tianta
n
1935.
161
-1.755 -6.278 3.960 -5.063
Wanli
u
1341.
457
-1.182 -5.534 3.722 -2.901
Wans
houxi
g
on
g
1889.
810
-1.705 -6.570 3.939 -6.647
From the results provided by the R code, this paper
can summarise the coefficient ranges and offer some
possible explanations for these results. The summary
will be shown in Table 3 below.
As table 3 shows, RAIN (mm) -6.65 to -3.18
Rainfall significantly reduces PM2.5, with urban
stations (e.g., Wanshouxigong) showing stronger
effects. This is possibly due to rain efficiently
depositing PM2.5.
Urban stations (e.g., Dongsi, Nongzhanguan)
exhibit larger coefficients for PRES, TEMP, and
RAIN, suggesting that meteorological factors play a
more pronounced role in PM2.5 variability in densely
populated areas. Suburban stations (e.g., Huairou,
Dingling) show weaker relationships, possibly due to
fewer local emissions and greater influence of
regional transport. Huairou Station has the smallest
coefficients (e.g., PRES: -0.38, TEMP: -3.60). This is
possibly because of its location in a rural,
mountainous area, which reduces the sensitivity of
PM2.5 to local weather.
Table 3. MLR Coefficient Ranges and Interpretations
Variables Range Explanation
Intercept 490.35
to
2048.27
Intercepts of PM2.5 are
significantly different from
station to station. Urban sites
(e.g., Dongsi, Nongzhanguan)
have higher intercepts, likely
due to more substantial local
emissions.
PRES
(hPa)
-1.86 to
-0.38
Higher atmospheric pressure
will lead to lower PM2.5. This
is possible because stable
weather conditions suppress
vertical spread. But the effect is
weaker in suburban stations.
TEMP
(°C)
-6.84 to
-3.60
Temperature has a negative
effect. This is possible because
warmer conditions enhance
atmospheric mixing, and
seasons have higher
temperatures, like summer
reduce coal heatin
g
emissions.
DEWP
(°C)
2.89 to
4.30
Higher dew point strongly
increases PM2.5. This is
possibly caused by moisture-
enhanced secondary aerosol
formation and stagnant air
masses.
RAIN
(mm)
-6.65 to
-3.18
Rainfall significantly reduces
PM2.5, with urban stations
(e.g., Wanshouxigong) showing
stronger effects. This is
possibly due to rain efficiently
depositing PM2.5.
Dongsi Station shows the strongest negative effect of
TEMP (-6.647), potentially linked to its central urban
setting, where temperature inversions trap pollutants.
The negative RAIN coefficients align with Beijing’s
observed "post-rain blue sky" phenomenon, where
precipitation removes particulate matter. The stronger
effect at urban stations (e.g., coefficient of -6.65 at
Wanshouxigong) may reflect higher initial PM2.5
concentrations available for wet deposition.
3.2 Random Forest Regression
This study also uses R code to construct a random
forest regression model (RFR) and calculate the
importance score (IncMSE) of these variables.
Predicting PM2.5 in Urban and Suburban Beijing: A Comparative Study of Random Forest and Linear Regression Models
333
Table 4: Variable Importance Score (IncMSE)
Variable
RAIN
(%)
DEWP
(%)
TEMP
(%)
PRES
(%)
Aotizhongxin 42.5 24.5 18.6 14.4
Changping 33.9 29.1 19.2 17.7
Din
g
lin
g
39.9 30.2 16.6 13.3
Don
g
si 38.3 24.2 19.2 18.3
Guan
uan 39.6 23.3 17.9 19.2
Gucheng 36.2 25.1 18.7 20
Huairou 32.9 26.5 20.9 19.8
Nongzhanguan 39.1 21.9 18.2 20.7
Shun
y
i 33.9 24.7 19.1 22.3
Tiantan 31 29.7 22.3 17
Wanliu 44.9 23 17.9 14.2
Wanshouxigong 40.9 24.4 18.5 16.2
The variable importance scores (IncMSE) from
Random Forest Regression reveal distinct patterns in
how meteorological factors influence PM2.5
concentrations across Beijing's monitoring stations
(Table 4). The results highlight both consistent trends
and notable spatial variations in atmospheric
processes affecting air quality. RAIN emerges as the
most important predictor at all stations (31.0-44.9%
importance), particularly at Wanliu (44.9%) and
Aotizhongxin (42.5%). This reflects Beijing's
reliance on wet deposition for particulate removal,
where precipitation effectively scavenges aerosols
from the atmosphere. The stronger effect at urban
stations suggests higher initial PM2.5 loading
available for removal. DEWP shows moderate
importance (21.9-30.2%), peaking at Dingling
(30.2%) and Tiantan (29.7%). This importance likely
represents moisture-enhanced secondary aerosol
formation and stagnant conditions during high
humidity episodes. TEMP (16.6-22.3%) and PRES
(13.3-22.3%) show more variable importance across
stations. For example, Tiantan station shows
unusually high TEMP importance (22.3%), possibly
due to its location near parks where temperature
inversions may trap pollutants.
Urban stations (Dongsi, Nongzhanguan) show
balanced importance across all variables. Suburban
stations (Huairou, Shunyi) display elevated PRES
importance (19.8-22.3%), suggesting the greater
influence of synoptic weather patterns. Wanliu
Station shows extreme RAIN dominance (44.9%)
with low DEWP importance (23.0%), possibly due to
its location near water bodies enhancing rain effects.
Tiantan Station has unusually high TEMP importance
(22.3%), potentially reflecting the urban heat island
effect in this cultural landmark area. Huairou Station
demonstrates the most balanced distribution,
consistent with its rural location, where no single
factor dominates.
The strong RAIN importance suggests that
weather modification (e.g., cloud seeding) could be
particularly effective during pollution episodes. High
DEWP importance indicates that humidity control
measures might help reduce secondary aerosol
formation. Urban stations may benefit most from
emission controls before forecasted precipitation
events. Suburban stations require more attention to
pressure systems and temperature variations.
The IncMSE results demonstrate that while
rainfall universally dominates PM2.5 variability
across Beijing, the relative importance of other
factors varies substantially by location. This spatial
heterogeneity underscores the need for tailored air
quality management strategies that account for local
meteorological sensitivities. The outlier behaviour at
stations like Wanliu and Tiantan suggests that
microclimate effects may significantly modify
pollution-weather relationships in specific urban
contexts. Future work should incorporate finer-scale
topographic and land-use data to better explain these
station-level differences.
3.3 Model Performance Metrics
Table 5 lists all the R², RMSE and MAE of the 12
stations. The evaluation metrics reveal several key
patterns in the performance of MLR and RFR models
across Beijing's air quality monitoring stations.
Both models show limited predictive power
overall, with test R² values ranging from 0.112-0.225
for MLR and 0.287-0.411 for RFR, indicating that
meteorological factors alone explain less than half of
PM2.5 variability. This suggests that additional
predictors like wind patterns, emission sources, or
temporal factors may be necessary for improved
accuracy. The RFR models consistently outperform
MLR in training data (R² 0.425-0.509 vs 0.117-
0.220), but this advantage diminishes in test data,
revealing moderate overfitting, particularly at stations
like Aotizhongxin where the train-test gap exceeds
0.12. This overfitting likely stems from the RFR's
complexity of capturing noise in the training data.
Spatial patterns in model performance reflect
Beijing's air pollution dynamics. Urban stations
(Dongsi, Nongzhanguan) show the highest R² values
for both models, with Nongzhanguan's RFR
achieving the best test performance (R²=0.406). This
urban advantage may result from stronger, more
consistent relationships between meteorological
conditions and local emissions in built-up areas. In
contrast, suburban stations like Huairou demonstrate
the poorest performance (test R²=0.287 for RFR),
IAMPA 2025 - The International Conference on Innovations in Applied Mathematics, Physics, and Astronomy
334
likely because regional transport of pollutants
weakens local weather-PM2.5 correlations.
Table 5: Model Performance Metrics (R²/RMSE/MAE)
Stati
on
Mo
del
R2
_T
rai
n
R2
_T
est
RM
SE_
Trai
n
RM
SE_
Test
MA
E_T
rain
MA
E_T
est
Aoti
zho
ngxi
n
ML
R
0.1
88
0.1
78
74.0
21
74.5
37
53.9
44
53.
836
RF
R
0.4
61
0.3
37
65.1
33
69.1
23
47.4
68
49.
970
Cha
ngpi
ng
ML
R
0.1
55
0.1
51
65.9
41
67.9
54
48.4
91
49.
303
RF
R
0.4
57
0.3
40
56.2
16
61.5
23
41.0
79
44.
279
Din
glin
g
ML
R
0.1
53
0.1
45
67.1
05
65.4
78
48.6
90
48.
192
RF
R
0.4
74
0.3
35
56.7
18
59.0
53
40.9
56
43.
165
Don
gsi
ML
R
0.2
13
0.2
19
76.7
42
76.7
22
55.8
33
55.
340
RF
R
0.5
01
0.4
11
67.2
63
70.5
84
48.9
59
51.
219
Gua
nyu
an
ML
R
0.1
91
0.1
85
73.0
75
72.5
07
52.9
45
52.
631
RF
R
0.4
63
0.3
66
64.9
45
67.0
30
47.1
60
49.
145
Guc
hen
g
ML
R
0.1
75
0.1
78
75.2
72
74.9
26
53.9
18
53.
742
RF
R
0.4
54
0.3
56
65.9
84
68.7
39
47.5
81
49.
774
Hua
irou
ML
R
0.1
17
0.1
12
67.2
77
66.2
98
49.2
68
48.
953
RF
R
0.4
32
0.2
87
57.3
53
60.3
62
41.6
64
44.
030
Non
gzha
ngu
an
ML
R
0.2
20
0.2
25
76.3
51
75.5
19
55.2
29
55.
291
RF
R
0.5
09
0.4
06
67.0
77
69.7
87
48.3
95
51.
138
Shu
nyi
ML
R
0.1
73
0.1
73
74.0
18
73.6
08
53.6
96
53.
561
RF
R
0.4
53
0.3
50
64.3
62
67.3
06
46.4
36
48.
655
Tian
tan
ML
R
0.2
10
0.2
06
72.2
70
71.3
82
52.2
63
52.
236
RF
R
0.4
88
0.3
74
63.7
72
66.2
89
46.4
57
48.
765
Wan
liu
ML
R
0.1
66
0.1
64
75.0
71
74.2
80
54.6
59
54.
362
RF
R
0.4
25
0.3
34
66.8
81
68.8
77
48.6
28
50.
350
Wan
shou
xigo
n
g
ML
R
0.2
01
0.2
10
76.7
75
76.6
97
54.9
94
55.
165
The models' relative performance varies spatially too-
at Dingling, RFR reduces test RMSE by 9.8%
compared to MLR, while at Wanshouxigong the
improvement is just 6.1%.
Notable anomalies include Dongsi station, where
MLR unexpectedly matches RFR's test performance
(R²=0.219 vs 0.411), suggesting linear relationships
may suffice at this urban location. Meanwhile,
Wanliu shows unusually poor RFR performance
despite its urban setting, possibly due to microclimate
effects from nearby water bodies. The consistent
MAE values (45-55 μg/m³ across stations) indicate
both models struggle with extreme PM2.5 events, a
critical limitation for pollution warning systems.
These results underscore that while RFR generally
outperforms MLR, its advantages are modest and
station-specific, highlighting the need for localised
model tuning in Beijing's heterogeneous airshed. The
persistent low values across all stations suggest
that effective PM2.5 forecasting requires
incorporating non-meteorological predictors like
real-time emission data.
4 CONCLUSION
This comprehensive evaluation of MLR and RFR
models for PM2.5 prediction across Beijing's
monitoring network yields several important insights
with significant implications for air quality
management. The analysis reveals that while both
models demonstrate limited predictive capability
using only meteorological variables, the Random
Forest approach consistently outperforms traditional
linear regression, albeit with notable spatial variations
in performance. The urban stations, particularly
Nongzhanguan and Dongsi, show relatively better
model performance (test up to 0.41), likely due to
a stronger coupling between local emissions and
meteorological conditions in densely populated areas.
In contrast, suburban stations like Huairou exhibit
poorer performance, suggesting that regional
pollutant transport and other non-local factors play a
more dominant role in these locations. The consistent
gap between training and test performance in RFR
models (average ΔR² of 0.12) indicates moderate
overfitting, emphasising the need for more robust
regularisation or inclusion of additional relevant
predictors.
The spatial patterns in model performance reflect
Beijing's complex air pollution dynamics, where
urban-scale processes appear more predictable than
regional-scale phenomena. The superior performance
of RFR models, particularly in urban settings,
Predicting PM2.5 in Urban and Suburban Beijing: A Comparative Study of Random Forest and Linear Regression Models
335
suggests that nonlinear relationships between
meteorological factors and PM2.5 concentrations are
important to capture. However, the modest absolute
performance levels (R² < 0.5 even for the best
models) strongly indicate that meteorological
variables alone are insufficient for accurate PM2.5
prediction. This limitation is particularly evident
during extreme pollution events, as shown by the
consistently high MAE values (45-55 μg/m³). The
station-specific variations in model performance,
such as the unexpectedly strong showing of MLR at
Dongsi or the poor RFR performance at Wanliu,
highlight the importance of localised model
development that accounts for microclimate effects
and unique station characteristics.
These findings have several important
implications for both research and air quality
management. First, they underscore the need to
incorporate additional predictors beyond basic
meteorological variables, particularly emission-
related indicators and wind pattern data. Second, they
suggest that different modelling approaches may be
warranted for different parts of the metropolitan area,
with more sophisticated techniques like RFR being
prioritised for urban core stations. Finally, the results
indicate that current models have limited capability in
predicting extreme pollution events, which should be
a focus area for future model improvement. Future
research directions should include testing more
advanced machine learning architectures,
incorporating real-time emission data, and
developing ensemble approaches that combine the
strengths of different modelling paradigms.
Ultimately, while meteorological factors provide a
useful foundation for PM2.5 prediction in Beijing,
significant improvements in forecasting accuracy will
require a more comprehensive approach that accounts
for the full range of physical and chemical processes
governing air pollution in this complex urban
environment.
REFERENCES
Brook, R. D., Rajagopalan, S., Pope, C. A., Brook, J. R.,
Bhatnagar, A., Diez-Roux, A. V., Holguin, F., Hong,
Y., Luepker, R. V., Mittleman, M. A., Peters, A.,
Siscovick, D., Smith, S. C., Whitsel, L., Kaufman, J. D.
2010. Particulate matter air pollution and
cardiovascular disease: An update to the scientific
statement from the American Heart Association.
Circulation, 121(21), 2331-2378.
Guo, H., Zhang, Q., Cheng, Y., He, K. 2014. The impact of
the Chinese New Year on air quality in China: A
review. Atmospheric Environment, 98, 647-656.
Li, J., Zhang, Y., Wang, X., Li, Z. 2019. The impact of the
Spring Festival on air quality in China: A review.
Environmental Pollution, 245, 707-718.
Perrino, C., Tiwari, S., Catrambone, M., Dalla Torre, S.,
Rantica, E., Canepari, S. 2011. Chemical
characterization of PM2.5 during the Diwali festival in
Delhi, India. Atmospheric Environment, 45(34), 6123-
6130.
Pope, C. A., Dockery, D. W. 2006. Health effects of fine
particulate air pollution: Lines that connect. Journal of
the Air & Waste Management Association, 56(6), 709-
742.
Tai, A. P. K., Mickley, L. J., Jacob, D. J. 2010. Correlations
between fine particulate matter (PM2.5) and
meteorological variables in the United States:
Implications for the sensitivity of PM2.5 to climate
change. Atmospheric Environment, 44(32), 3976-3984.
Wang, G., Zhang, R., Wang, L., Li, Z. 2014. Impact of
meteorological conditions on PM2.5 concentrations in
Beijing, China. Atmospheric Chemistry and Physics,
14(22), 12307-12322.
Wang, Y., Zhang, Q., Wang, L., Zhang, R., Li, Z. 2017. The
impact of meteorological factors on PM2.5
concentrations in urban areas: A case study in Beijing,
China. Atmospheric Environment, 165, 52-63.
Zhang, Q., Quan, J., Tie, X., Cao, J., Han, S., Wang, Z.,
Zhao, D., Li, J., Liu, X. 2015. The impact of
meteorological conditions on PM2.5 concentrations in
China: A review. Atmospheric Environment, 122, 823-
833.
Zhang, R., Wang, G., Zhang, Q., Li, Z. 2017. The impact of
holidays on air quality in China: A review.
Environmental Pollution, 231, 1234-1245.
IAMPA 2025 - The International Conference on Innovations in Applied Mathematics, Physics, and Astronomy
336