Content uploaded by Jiangying Xu
Author content
All content in this area was uploaded by Jiangying Xu on Aug 31, 2022
Content may be subject to copyright.
Atmospheric Pollution Research 13 (2022) 101539
Available online 27 August 2022
1309-1042/© 2022 Turkish National Committee for Air Pollution Research and Control. Production and hosting by Elsevier B.V. All rights reserved.
Spatial heterogeneity of river effects on PM
2.5
pollutants in waterfront
neighborhoods based on mobile monitoring
Jiangying Xu
a
,
b
, Mengyang Liu
a
,
b
, Hong Chen
a
,
b
,
*
a
School of Architecture & Urban Planning, Huazhong University of Science and Technology, Wuhan, Hubei, 430000, China
b
Hubei Engineering and Technology Research Center of Urbanization, Wuhan, Hubei, 430000, China
ARTICLE INFO
Keywords:
Ecological environment benet
Geographically weighted regression (GWR)
model
Water bodies
Waterfront neighborhoods
Street-level pollution
ABSTRACT
Concentrations of airborne particulate matter (PM) are inuenced by land cover types. Water bodies in cities can
inuence the spatial distribution of air pollution by altering the microclimate. However, the inuence of water
bodies on PM
2.5
concentrations is complex and requires further exploration, especially at the microscale. In this
study, a geographically weighted regression (GWR) model was constructed to explore the spatially heteroge-
neous effect of a river on the PM
2.5
concentrations in nearby riverside neighborhoods using mobile monitoring in
Wuhan. The results showed that the GWR model was applicable at the neighborhood scale and had a good
explanatory performance for PM
2.5
concentrations, with a higher R
2
(0.71). The river had multiple effects on the
PM
2.5
concentrations in the riverfront neighborhoods and strongly affected the air quality in neighborhoods
within an 800 m distance due to wind inltration, with the strongest effect at a 500–700 m distance. Moreover,
considering spatial nonstationarity, the effect of the large river on street-level air quality largely depended on its
effect on wind, and good ventilation conditions could amplify that effect. Commercial, road intersections,
second-level roads and parks were identied as sensitive environmental factors affecting the river’s inuence on
PM
2.5
concentrations. In addition, urban parks had a greater mitigating effect on PM
2.5
pollution than did water
bodies in this study. These results help to clarify the impact of rivers on air quality and provide a theoretical basis
for urban design to mitigate PM
2.5
pollution.
1. Introduction
Rapid urbanization has led to severe air pollution, and PM
2.5
(par-
ticles with aerodynamic diameters of less than 2.5
μ
m) has become the
main air pollutant in many Chinese cities (Fontes et al., 2017; Song et al.,
2017; Ye et al., 2018). The China Eco-environmental Status Bulletin for
2020 shows that days when PM
2.5
was the main pollutant accounted for
77.7% of the days with serious pollution in 136 cities that failed to meet
air quality standards (“Bulletin on China’s ecological environment 2020,
” n. d.). PM
2.5
has been found to be associated with a variety of health
issues, such as respiratory diseases and high blood pressure (Ferreira
et al., 2022; Lu et al., 2019; Zhang et al., 2022; Zhu et al., 2022). For
every 10
μ
g/m
3
reduction in the PM
2.5
concentration, there will be
approximately 2% fewer hospitalizations, 0.46% fewer hospitalizations
for acute exacerbation of COPD (AECOPD), 6–7% less annual mortality,
and a 2% reduction in mortality due to respiratory diseases (Lyu et al.,
2022; Schwartz, 2002; Wong et al., 2001; Wu et al., 2020). Thus, the
health threat due to PM
2.5
warrants urgent attention.
Urban development is a major factor in the occurrence of air pollu-
tion because it signicantly changes land use patterns (Lu et al., 2018;
Wu et al., 2015b; Xu et al., 2021). Research on the relationship between
air pollutants and the urban built environment focuses on two main
aspects, namely, physical space and ecological space. Studies on phys-
ical space focus on the inuence of land use and urban morphology on
the distribution of pollutant concentrations (Chen et al., 2021a; Liu
et al., 2021; Shi et al., 2018). Many studies have been conducted to
provide evidence of the impacts of urban land use and urban form on air
quality. The evidence from these studies suggests that different land use
patterns in cities have varied impacts on urban air pollution, and in-
dustrial areas with high emission intensities tend to cause higher levels
of pollution (de Hoogh et al., 2013; Habermann et al., 2015). Building
density, the standard deviation in building height, and porosity are the
main urban form elements that affect PM
2.5
dispersion (Chen et al.,
2021b; Liu et al., 2021; Shi et al., 2017, 2018).
Peer review under responsibility of Turkish National Committee for Air Pollution Research and Control.
* Corresponding author. School of Architecture & Urban Planning, Huazhong University of Science and Technology, Wuhan, Hubei, 430000, China.
E-mail address: Chhwh@hust.edu.cn (H. Chen).
Contents lists available at ScienceDirect
Atmospheric Pollution Research
journal homepage: www.elsevier.com/locate/apr
https://doi.org/10.1016/j.apr.2022.101539
Received 25 April 2022; Received in revised form 5 August 2022; Accepted 20 August 2022
Atmospheric Pollution Research 13 (2022) 101539
2
Research on ecological space focuses on green plants and water
bodies. Many studies have been conducted to provide evidence of the
impacts of water bodies on particulate matter (Liu et al., 2016; Xu et al.,
2021; Zhu and Zeng, 2018). Evidence from these studies suggests that,
on the one hand, cold and humid air at the riverside alters the local air
circulation, enhancing the ow of air away from the riverside and the
diffusion of particulate matter to areas of lower humidity (Liu et al.,
2016). On the other hand, the effect of water bodies on particulate
matter is greatly exacerbated by the hygroscopic growth and deposition
of particulate matter to the surface through humidity regulation (Kang
et al., 2015; Liu et al., 2016). However, the mechanism of the effect of
water bodies on PM
2.5
concentrations is complex, and water bodies also
showed a positive effect (Liu et al., 2016; Zhou et al., 2021) or no sig-
nicant effects (Wu et al., 2015a) on PM
2.5
concentrations in some
studies. There are still some knowledge gaps regarding how water
bodies affect PM
2.5
pollution that need to be further empirically
investigated.
Most empirical studies on urban air quality use statistical modeling
approaches to establish empirical relationships between air pollutant
concentrations and geographic predictors such as trafc and land use.
The land use regression (LUR) model is an effective method to estimate
urban pollution (Brauer et al., 2003; Liu et al., 2016; Ross et al., 2007)
and is mostly based on national xed-point monitoring. However, na-
tional xed-point monitoring stations are relatively sparse and this type
of monitoring has insufcient resolution to be used to study spatial
variations in air quality at the street level. Therefore, mobile monitoring
has been recently used for street-level air pollution studies because it can
capture more detailed spatial variability (Hankey and Marshall, 2015;
Liu et al., 2020; Minet et al., 2017). These studies mostly use multiple
linear regression (MLR) to analyze the relationship between morpho-
logical indicators and air pollutants (Liu et al., 2020; Minet et al., 2017).
However, due to the lack of consideration of the nonstationarity of
highly heterogeneous urban spaces, signicant relationships between
some specic built environments and particulate matter were not found
in some studies (Ho et al., 2015; Liu et al., 2016). The complexity of
street environments dictates the need to apply more complex models to
represent these phenomena and the underlying impact mechanisms.
In summary, some gaps exist in the knowledge regarding the effects
of water bodies on air quality. Previous studies on the effects of water
bodies on PM
2.5
focused on the area and morphology of the water bodies
(Zhu et al., 2016; Zhu and Zeng, 2018) and were mostly based on na-
tional monitoring sites to study the effects of water bodies on the
mesoscale environment (Wu et al., 2015a; Zhu and Zhou, 2019). How-
ever, the effect of water bodies on small-scale waterfront neighborhoods
was neglected and needs to be further explored. Mixed ndings on the
effects of water bodies on PM
2.5
concentrations suggest that perhaps
their effects are spatially heterogeneous.
Therefore, the aim of this study is to comprehensively clarify the
inuence of the Yangtze River on the spatial distribution of PM
2.5
pol-
lutants in riverside neighborhoods and the spatial heterogeneity of the
impact effect. In this study, we introduce a geographically weighted
regression (GWR) model considering spatial nonstationarity (Zhou and
Lin, 2019) to assess the river’s contribution to the distribution of PM
2.5
pollutants in the waterfront neighborhoods based on mobile monitoring.
In this study, our goals are to (1) develop a GWR model based on land
use and urban morphology variables to better model the spatial distri-
bution of PM
2.5
concentrations in the neighborhood; (2) interpret the
spatial distribution of the coefcients of the “distance to the river”
variable in the GWR model results and analyze the multiple effects of
rivers on PM
2.5
concentrations; and (3) elaborate the spatial heteroge-
neity of the effect of rivers on PM
2.5
pollution by comparing the co-
efcients of different geographical locations and sensitive
environmental factors to provide a basis for urban design strategies to
improve air quality.
2. Materials and methodology
2.1. PM
2.5
mobile monitoring and preprocessing
2.1.1. Monitoring plan
Wuhan, China, which is located at a latitude of 29◦58′–31◦22′north
and a longitude of 113◦41′–115◦05′east, has a humid subtropical
climate. The Yangtze River crosses the central area of the city. In this
study, several communities on the west side of the Yangtze River were
selected for evaluation. These communities were located in the old city
of Wuhan, where there are no factories within 3 km and PM
2.5
pollution
mainly comes from vehicle exhaust emissions. The specic mobile
observation routes are shown in Fig. 1(A). These routes were randomly
selected by the principle of diverse land use, covering commercial,
residential and park land use types and containing two types of neigh-
borhoods, namely, large street districts and small neighborhoods. To
avoid confounding factors such as daily weather characteristics in this
study, repeated multiday monitoring was carried out along these xed
routes. Airborne PM
2.5
concentrations were collected using mobile ob-
servations. At the same time, a xed monitoring location was established
to record the background concentrations at Wufu Square at Hankou
Jiangtan near the Yangtze River. This location had no tree shading and
was less than 1 km away from the national measurement site at the river,
which facilitated comparison and correction of the observed data based
on the national measurements.
During mobile monitoring, PM
2.5
concentrations were measured
using a TSI AM520 particulate dust meter with a recording interval of 5
s. In addition, an SDL500 Extech temperature and humidity meter and
an Etrex 201x GPS were used to acquire temperature and relative hu-
midity data and location information, respectively, at the same time
interval. A small weather station (PC-8 environmental monitor) was
used to record background PM
2.5
concentrations, temperature, humid-
ity, wind speed and wind direction at 1 min intervals. All the in-
struments were calibrated before the ofcial experiment to ensure
accuracy. Details on these measurement instruments are provided in
Table S1 of the supplementary material.
The mobile monitoring data were collected during four days in 2020,
on October 30 and November 3, 4 and 10. No rain occurred on any of
these days. Two rides per day were conducted to collect PM
2.5
concen-
trations, GPS locations and meteorological data (09:30–11:30 and
16:30–18:30), with each mobile monitoring survey lasting approxi-
mately 2 h. The TSI AM520 dust meter and SDL500 Extech temperature
and humidity meter were mounted on an e-bike and secured with
cushioning foam, and the e-bike speed was maintained at 10–15 km/h to
prevent excessive noise in the data. The cyclist carried a GPS to record
location information during the ride.
2.1.2. Data preprocessing
The data processing mainly consisted of ve steps: (1) wind direction
check at a background point; (2) calibration of the background con-
centration with the concentrations at the national measurement point;
(3) anomaly processing of the mobile observation data and humidity
correction; (4) background concentration correction of the mobile
observation data; and (5) aggregation of the mobile observation data.
First, we processed the real-time wind data from the background
points along the river bank that were obtained from the same 4 day as
the actual measurements to produce a wind rose diagram. The wind
directions on these dates were examined and found to be mainly easterly
and southeasterly, in accordance with the requirements of this experi-
ment (Fig. 2). The wind direction determines the diffusion direction of
pollutants, and east and southeast winds from the river have a purifying
effect on the riverfront area, which is conducive to the diffusion of
pollutants.
Second, the hourly data released from the national measurement
point of Hankou Jiangtan were compared with the actual data at the
background point to verify the reliability of the background
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
3
concentrations at the actual measurement site. Background concentra-
tions were obtained from the xed point at Wufu Square near the
Jiangtan national xed point (approximately 0.5–1 km away), which
had no pollution sources within 500 m. The nal background point data
needed to be smoothed to obtain 5 s interval data for the next step (the
background concentration correction of the mobile observation data).
Thus, the regression analysis was performed after converting the na-
tional measurement point data and the background point data into
uniform 5 s datasets by linear interpolation. The results showed that the
background and national data showed high agreement, and the R
2
values of the regression models were above 0.8 (Fig. 3).
Third, when barbecue smoke and diesel generators are passed during
mobile observations, abnormally high pollutant concentrations can be
detected, which can interfere with the analysis results. Boxplot analysis
was used to identify outliers, and an IDV SQ9 mini motion video camera
was used to recheck the results. Data close to locations such as con-
struction areas or sprinklers were excluded. Moreover, we calibrated the
mobile PM
2.5
measurements from the portable aerosol monitor by
introducing an empirical correction factor that was used in previous
studies (Shi et al., 2016, 2019). High humidity leads to overestimations
when using light-scattering laser photometer-based aerosol monitors
(Spinazz`
e et al., 2017).
Fourth, the effect of background concentrations was minimized ac-
cording to the methods of a previous study (Lim et al., 2019; Tessum
et al., 2018). Because the meteorological data vary considerably from
day to day during the actual measurements, the PM
2.5
concentrations
depend on the background concentrations on that day. The previous
moving sample survey adjusted for potential temporal bias by sub-
tracting the daily fth percentile (Tessum et al., 2018) or used the ratio
of the average concentration level at nearby monitors during the sam-
pling period to the corresponding hourly monitored concentration (Lim
et al., 2019). We adjusted the daytime temporal trend by subtracting the
background point concentration value at the corresponding time from
all the concentration values obtained from mobile monitoring:
PMi,adjusted =PMi−PMbackground,i(1)
Fifth, the data from the mobile measurements were aggregated.
Since the GPS points of the investigator’s daily movements did not
exactly overlap (Fig. 1(C)), more than 1000 data points existed in the
original data that were obtained from each mobile monitor. It was
necessary to cluster the original data by inputting the data with equi-
distant aggregation points along the measurement route. The radius was
set to 100 m based on the investigator’s cycling speed and a 5 s data
collection interval. A total of 150 aggregation points were set up along
the mobile monitoring route (Fig. 1(B)). The average concentrations of
PM
2.5
pollutants were input to the aggregation points. The spatial dis-
tributions of the aggregated pollutant concentrations are shown in
Fig. 4. Low values of PM
2.5
concentrations occurred near the park, and
large parks played a signicant role in reducing PM
2.5
concentrations.
2.2. Setting predictor variables
Variables such as urban land use and urban morphology were
extracted in a Quantum geographic information system (QGIS). The land
use map in this study was drawn by using geographic information pro-
cessing tools combined with a Wuhan city planning map and Open-
StreetMap. In addition, variables related to trafc conditions, such as the
numbers of intersections, metro stations, bus stops, and the density of
rst-level (30–45 m) and second-level (12–30 m) roads, were included
in the land use elements. First-level roads are the main vehicle routes,
and second-level roads are the lower level roads for people and vehicles,
with fewer vehicles and a lower pollution emission intensity compared
to primary roads.
The building density characterizes the density of the ground cover
within a given buffer zone. The plot ratio represents the three-
dimensional building volume density within each buffer plot (Shi
Fig. 1. (A) Mobile monitoring route; (B) aggregation points along the mobile monitoring route; and (C) the aggregation process of the GPS data points.
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
4
Fig. 2. The wind direction and speed during 9:30–11:30 (am) & 16:30–18:30 (pm) on (A) October 30 (am); (B) November 3 (am); (C) November 4 (am); (D)
November 10 (am); (E) October 30 (pm); (F) November 3 (pm); (G) November 4 (pm); and (H) November 10 (pm).
Fig. 3. Comparison between the xed equipment and national stationary stations (mg/m
3
).
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
5
et al., 2016). Height characterizes the intensity of perpendicular
development within a certain area of the city (Salamanca et al., 2011).
Within a given buffer zone, the variation in building height is reected
by the standard deviation of the height.
To investigate the extent of the Yangtze River’s inuence on water-
front neighborhoods, all aggregation points were calculated with the
distance to the river values. The distance to the river value represents
the perpendicular distance from each aggregation point to the river
shoreline along the mobile observation route.
Two groups of variables considered in this study, urban morphology
and land use features, are shown in Table 1. Following the methods of
previous research (Liu et al., 2021), the buffer sizes were set to four
levels: 50, 100, 150, and 200 m. The same metric calculated using two
different buffers was used as two separate predictor variables in the
correlation analysis, which meant that these metrics needed to be
calculated four times for each predictor variable type listed in Table 1
except for the distance to the river variable. In total, 49 variables were
considered in the process of optimal buffer identication. The values of
these predicator variables were entered into the aggregation points to
describe the effect of the considered factors on air quality. The corre-
sponding variables were the PM
2.5
concentrations. In this modeling
process, the daily morning and afternoon PM
2.5
concentrations were
used as two separate corresponding variables.
2.3. GWR and MLR model construction and validation
2.3.1. Identifying the critical buffer
First, the critical buffer radius for each predictor variable was iden-
tied. To determine the critical buffer radius of the predictor variables
on PM
2.5
concentrations, Pearson’s correlation analysis was performed
for each predictor variable with PM
2.5
concentrations. Only the buffer-
based variable with the highest |r| was selected as the predictor vari-
able. Additionally, to ensure the accuracy of the model predictions,
variables with nonsignicant correlations with PM
2.5
(p value >0.05) in
both the morning and afternoon were excluded, such as SDH, FR, and
AH.
2.3.2. Model construction and validation
GWR is a local linear regression method that considers the spatial
heterogeneity of variables. The spatial location is embedded in the local
smoothing idea represented by the regression parameters, and sampling
points are modeled by local weighting (Brunsdon et al., 2010). GWR
models relate the explanatory variables to geographical locations, con-
structing a local original least square (OLS) model for each sampling
position.
The formula for the GWR can be represented as Equation (1), where
(ui,vi)represents the geographic sampling position, βi(ui,vi)is the
interception of (ui,vi), n is the number of independent variables, βi(ui,vi)
are the corresponding parameters of (ui,vi)and
ε
i(ui,vi)is the corre-
sponding error term. The bandwidth of the GWR model is an important
parameter, and the local model training includes only samples within
the bandwidth range (Brunsdon et al., 2010; Feuillet et al., 2014). To
obtain a better prediction performance, an adaptive bandwidth strategy
was used in this study.
Fig. 4. The distributions of the nal PM
2.5
concentrations of aggregated points (mg/m
3
) in the mornings of (A) October 30; (B) November 3; (C) November 4; and (D)
November 10.
Table 1
Information on predictor variables data.
Predictor variable Name
abbreviation
Unit Buffers
Urban form
Building density BD % 50, 100, 150, 200
m
Plot ratio PR % 50, 100, 150, 200
m
Average height AH m 50, 100, 150, 200
m
Standard deviation of height SDH M 50, 100, 150, 200
m
Land use features
Parks PARK % 50, 100, 150, 200
m
Residences RESI % 50, 100, 150, 200
m
Commercial buildings COMM % 50, 100, 150, 200
m
Distance to the Yangtze
River
DIST M –
First-level road density FR m/
m
2
50, 100, 150, 200
m
Second-level road density SR m/
m
2
50, 100, 150, 200
m
Bus stops numbers BUS n/a 50, 100, 150, 200
m
Metro stations numbers MET n/a 50, 100, 150, 200
m
Intersections numbers INT n/a 50, 100, 150, 200
m
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
6
Yi=βi(ui,vi) + ∑
n
k=0
βk(ui,vi)xik +
ε
i(ui,vi)(2)
Previous studies have shown that GWR models have a better pre-
diction performance than do MLR models for PM
2.5
concentrations at a
larger scale because of the spatial heterogeneity consideration (Shi et al.,
2019); in the discussion section of this paper, we rst compared the
GWR and MLR to verify the performance of the GWR at the neighbor-
hood scale.
Before the MLR and GWR, a multicollinearity test was performed. A
strong linear correlation between two predictor variables is called
multicollinearity, which may cause bias in the interpretation of the ef-
fects of other predictor variables (Masiol et al., 2018). To eliminate this
phenomenon, variance ination factor (VIF) values were used as a
measure, and all predictor variables were excluded if they had VIF
values greater than 5.
All aggregated point data were randomly divided into training (70%)
and testing (30%) sets, and the process worked as follows: (i) the model
was developed using the training data set; (ii) the test data set was
applied to the model to predict observations from the new data; and (iii)
the prediction error was quantied as the mean squared difference be-
tween the observed and predicted outcome values.
3. Results
3.1. The critical buffer radius
The buffer size with the highest |r| on PM
2.5
for each variable was
identied by retaining only the buffer variable with the highest corre-
lation to the buffer. Additionally, the correlation of these variables with
PM
2.5
was not signicant (p value >0.05) in neither the morning nor the
afternoon and thus these variables were excluded and not used as inputs
in the MLR and GWR models. The detailed results of the correlation
analyses of the optimal buffer variables are presented in Tables S2 and
S3 of the supplementary material. As a result, only 16 variables
(including only predictor variables) remained to be used for regression
modeling (Fig. 5), including DIST.
3.2. The performance of the GWR and MLR models
We developed separate models for the morning and afternoon sur-
veys, but the distance to river element was excluded from the afternoon
model. Wind direction allows the distance to the river to be excluded
from the model. The main wind direction in the afternoon on three of the
days was north or south, which was not conducive to river breeze
inltration into the riverfront neighborhoods (Fig. 2(E, G, H)). The main
wind direction in the morning on all four days was southeast, which
happened to be the wind direction that favored the inltration of the
river breeze. This paper mainly investigated the effect of distance from
the river on PM
2.5
. Therefore, the following analyses in this paper were
based on the results of the morning model.
The Akaike information criterion (AICc), R squared (R
2
), adjusted R
squared (Adj R
2
), cross validation R squared (CV R
2
), root mean squared
error (RMSE) and mean absolute error (MAE) were calculated for the
GWR and MLR models to compare model quality. Except for the R
2
, Adj
R
2
and CV R
2
, a lower value indicated a more efcient model. The GWR
model accounted for the effect of spatial heterogeneity and better
captured the effect of each element at different locations. A linear model
was built for each location point separately, thus achieving better pre-
diction results. As shown in Table 2, the GWR outperformed the MLR in
terms of model t, indicating that the GWR better explained the spatial
data. The R
2
values increased from 0.429 in the MLR model to 0.710 in
the GWR model. The explanatory power increased signicantly when
spatial nonstationarity was considered in the model.
In addition, based on the GWR model, the spatial variability in the
model’s local R
2
could be delineated. Fig. 6(A) shows that the GWR
model had a higher explanatory power in the south than in the north. In
Fig. 5. Correlation matrix of the predictor variables with PM
2.5
pollutants in the morning (A) and afternoon (B).
Table 2
Comparison results of the MLR and GWR models.
AICc R
2
Adj R
2
CV R
2
RMSE MAE
MLR 222.600 0.429 0.397 0.393 0.485 0.387
GWR 155.256 0.710 0.648 0.615 0.344 0.283
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
7
comparison, the standard deviation distribution of the GWR model was
more uniform (Fig. 6(B)). Many observations had high local R
2
values,
reaching 0.6. This was particularly evident in high-density commercial
and high-density residential areas where a large number of human ac-
tivities occur.
3.3. Contributing factor analysis for PM
2.5
concentrations
The detailed results of the predictor variables in the MLR and GWR
model are shown in Tables 3 and 4. The results showed a strong rela-
tionship between PM
2.5
concentrations and built environment factors.
Urban morphology and land use features were included as built envi-
ronment factors. The coefcients of building density were positive at
most of the sampling points. As shown in Table 4, the mean building
density within the 200 m buffer was 0.36, which showed that each 1%
increase in building density increases the average PM
2.5
concentrations
by 0.36%. In addition, the plot ratio had a similar inuencing mecha-
nism on PM
2.5
as building density, except that the effect was weaker
than that of building density.
Among the factors related to land use, the effect of parks on the
concentrations of pollutants was always negative (Fig. 7(C)). In Table 4,
the average coefcient of parks was negatively correlated with the
spatial distribution of PM
2.5
concentrations. Generally, a 1% increase in
the park area induced an average decrease in PM
2.5
concentrations of
0.000005%. Even though the coefcient for residence in the GWR model
was positive, there was still a downward trend in several areas (Fig. 7
(D)). As Fig. 7(D) shows, negative coefcients from residential areas
were observed in some external parts of the old city rather than in the
central areas. According to Fig. 7(F) results, the higher density of
second-level roads far from the Yangtze River tended to decrease the
PM
2.5
concentrations. Furthermore, positive coefcients of the second-
level roads near the Yangtze River were also observed.
Fig. 6. Spatial distribution of the local R
2
(A) and standard residual (B) according to the GWR model.
Table 3
Results of the multiple linear regression (MLR) model.
Predictor variables Estimate Std. Error t value Prob (>|t|) VIF
(Intercept) 2.203e-01 2.029e-01 1.086 0.27932 n/a
DISTANCE −4.281e-04 9.807e-05 −4.365 2.44e-05 *** 2.187252
BD_200 1.671e+00 1.202e+00 1.390 0.16668 2.215967
PR_200 2.326e-02 4.128e-02 0.563 0.57407 1.876790
SR_200 4.598e-06 8.535e-05 0.054 0.95711 2.206608
BUS_200 2.708e-02 4.530e-02 0.598 0.55089 1.179651
INT_200 −4.096e-03 1.621e-02 −0.253 0.80090 1.511264
PK_200 −6.036e-06 2.296e-06 −2.629 0.00951 ** 2.329497
RESI_200 2.305e-06 1.336e-06 1.725 0.08679 . 2.401720
Signicance codes: 0 =***, 0.001 =**, 0.01 =*, 0.05 =‘.‘, 0.1 =No code.
Table 4
Results of the geographically weighted regression (GWR) model.
Predictor variable Min Mean Max STD
Urban form
Building density −2.815099 0.358639 3.069797 1.873746
Plot ratio −0.063613 0.013864 0.058247 0.028097
Land use features
Park −0.000014 −0.000005 −0.000002 0.000004
Residences −0.000003 0.000004 0.000009 0.000004
Distance to Yangtze River −0.000831 −0.000426 0.000429 0.00032
Second-level road length −0.000582 −0.000091 0.000301 0.000319
Bus stops −0.123282 0.005246 0.099782 0.065289
Intersections −0.066775 −0.021351 0.016600 0.062266
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
8
The effect of distance from the river on PM
2.5
was mixed, with pos-
itive DIST coefcients in commercial areas and negative ones in resi-
dential areas as well as near parks. The absolute value of the negative
DIST coefcient near the park was much larger than the absolute value
of the positive DIST coefcient in the commercial area.
4. Discussion
4.1. Comparison of the GWR and MLR models at the microscale
A comparison with the traditional model (MLR) was conducted to
observe the performance of the GWR model on the same data set. The
performance of the GWR in this study conrmed that this model type
could be used to estimate PM
2.5
concentrations at the street level. The R
2
value of the GWR model was 0.71, which was 0.28 higher than that of
the MLR model. The advantage of the GWR model was consistent with
the results of recent articles on mobile monitoring based on different
instruments. Hankey and Marshall (2015) constructed an LUR model of
PM
2.5
concentrations using bicycle-based mobile observations with
adjusted R
2
values between 0.3 and 0.49. Shi et al. (2016) created an
MLR model of PM
2.5
using vehicle-based mobile monitoring integrating
urban form features as independent variables with an adjusted R
2
value
of 0.35 (Shi et al., 2018). These results indicated that the GWR has
advantages in predicting the concentrations of street pollutants.
4.2. The inuence of rivers on PM
2.5
concentrations
4.2.1. Identifying the multiple effects of rivers on the air quality of
riverfront areas
Multiple effects of rivers were found on the air quality of riverfront
neighborhoods with different geographical patterns from the perspec-
tive of spatial heterogeneity. As shown in Fig. 8(B), the river had
different effects on PM
2.5
concentrations among different geographical
locations in the riverfront area. In the commercial area, the DIST coef-
cient of the GWR model had a positive value, i.e., the farther the dis-
tance to the river was, the higher the pollution concentrations. In the
residential area as well as near the park, the coefcient of the distance
from the river in the GWR model had a negative value, i.e., the farther
the distance from the river was, the lower the pollution concentrations.
These two different impact effects can be understood in two ways.
The positive correlation between the DIST and PM
2.5
concentrations
suggests that the river has a positive effect on the abatement of PM
2.5
pollutant concentrations in the riverside area. On the one hand, this may
be because large bodies of water have no obstructive effect on natural
winds, which provides favorable conditions for the diffusion of pollut-
ants (Crosman and Horel, 2016); the river winds increase in strength
Fig. 7. Local coefcient spatial distribution of the GWR model: (A) building density; (B) plot ratio; (C) parks; (D) residences; (E) distance to the Yangtze River; (F)
second-level road length; (G) building density; and (H) intersections.
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
9
with decreasing proximity to the river, accelerating the diffusion of
pollutants, and thus, the closer the distance to the river is, the lower the
pollutant concentrations are. On the other hand, this may be due to a
higher humidity in the riverside streets than elsewhere; when humidity
reaches a critical value, particle deposition is accelerated (Zhu et al.,
2016; Zhu and Zeng, 2018). In addition, the water body itself takes the
place of a potential pollutant source and is a cleaner alternative
compared to the road (Wu et al., 2015a).
The negative correlation between the DIST and PM
2.5
concentrations
implies that the river may be responsible for the elevated PM
2.5
con-
centrations. On the one hand, PM
2.5
may come from pollution sources
near the river, such as Riverside Avenue (Alshetty and Nagendra, 2022;
Weinbruch et al., 2014); in addition, with the dense distribution of trees
on the riverfront side of the riverwalk, particulate matter collected by
the trees may oat again (Nowak et al., 2014). On the other hand, the
relative humidity conditions of the selected dates were almost under
50% in this study (Fig. S1 of the supplementary material). When the
humidity does not reach the critical value that enables particulate
matter to settle, the PM
2.5
surface adsorption capacity is weak, and it is
not easy to absorb moisture, coalesce and settle, which may cause the
suspension of pollutants. Meanwhile, gaseous precursor pollutants are
more likely to form new ne particulate matter through chemical re-
actions under relatively high humidity adsorption conditions (Zhu et al.,
2016).
4.2.2. Identifying the distance to the river’s impact on PM
2.5
pollution in the
riverfront area
According to the sensitivity test, the inuence of the river’s wind
inltration on the air quality of nearby waterfront neighborhoods has a
range of approximately 800 m, i.e., the river mainly and signicantly
inuences the air quality of the waterfront area within 800 m of its
periphery by wind inltration Fig. 8(A). This is consistent with Zhu and
Zeng’s (2018) eld study results for lakes but larger than the inuence
range they found for lake wetlands (500 m). This suggests that large
water bodies such as the Yangtze River have a greater range of river
wind inltration effects on the air quality of riverfront neighborhoods.
Within 800 m of the inuence range of the river, the inuence of the
DIST on PM
2.5
concentrations shows a strong positive correlation Fig. 8
(A), i.e., the further the distance from the river is, the higher the PM
2.5
concentration. As shown in Fig. 8(B), within area A, the DIST coefcient
in the GWR model is signicantly positive. The riverfront of area A has
an open view, with no tall buildings nor dense greenery, which con-
tributes to the effect of river wind inltration. Lower PM
2.5
concentra-
tions are also detected in area A (Fig. 8(C–F)); river wind inltration
accelerates the diffusion of pollutants in this area.
Between 800 and 1500 m from the river, the effect of the river on the
Fig. 8. Relationship between rivers and PM
2.5
pollutant concentrations: (A) the local coefcients of local DIST coefcients of the GWR model; and (B) the r plot
between the DIST and PM
2.5
concentrations; the distribution of row PM
2.5
concentrations (mg/m
3
) on (C) October 30; (D) November 3; (E) November 4; and (F)
November 10.
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
10
PM
2.5
concentrations shows a weak inuence Fig. 8(A). Accordingly,
high and low PM
2.5
concentrations are detected in this region, which
may be the reason for the weak correlation coefcient. Beyond
1500–2000 m, the effect of the river on the PM
2.5
concentrations shows a
sharply strengthened negative correlation, which is mainly due to the
interference caused by the effect of urban parks on the PM
2.5
concen-
trations. As shown in Fig. 8(B), the park is located more than 1500 m
away from the river. Lower PM
2.5
concentrations are also detected in
areas near the park, which validates the PM
2.5
deposition effect of urban
parks (Chen et al., 2019; Shi et al., 2016; Wu et al., 2018).
It is noteworthy that area B is in the 800 m range and area D is in the
greater than 1500 m range. Although areas B and D are located at the
same distances from the river as areas A and C, respectively, they present
different DIST coefcients from A and C. Comparing the built
environment of areas A and B, both areas are found to have small
neighborhoods with dense road networks, with the difference between
them being that area A has no riverfront greenery and a more open view
between it and the Yangtze River, while area B has denser riverfront
greenery between it and the Yangtze River. Studies have shown that
dense trees along streets may lead to higher PM
2.5
concentrations
instead of diffusing pollutants within the street (Baldauf, 2017; Janh¨
all,
2015; Jeanjean et al., 2017; Wania et al., 2012). Therefore, the dense
greenery along the riverfront may be the reason for the different DIST
coefcients between the A and B areas. Comparing the C and D areas, the
greatest difference between the two areas is in their function. Area C
functions as a green park area, while area D functions as a residential
area, which is the reason for the difference in the DIST values between
these areas. Such results also validate the contribution of urban parks to
Fig. 9. The spatial nonstationarity in the DIST coefcients of the GWR model with different built environments: (A) commercial density; (B) second-level road
density; (C) intersection number; and (D) park density. Regression analysis of the built environment and the DIST coefcients of the GWR model:
€
commercial
density; (F) second-level road density; (G) intersection number; and (H) park density.
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
11
PM
2.5
deposition.
4.2.3. Identifying the spatial heterogeneity of the river’s impact on PM
2.5
pollution in the riverfront region
Spatial heterogeneity analysis of the DIST coefcients from the GWR
model revealed that commercial, second-level roads, road intersections
and parks are the key factors that lead to the different inuence effects of
the DIST. As shown in Fig. 9, in the distribution of the distance from the
river coefcients with commercial density, number of road in-
tersections, park density, road density and building density as the base
map, the DIST coefcients showed similar spatial regularity as the base
map. Spatial heterogeneity analysis was conducted for other factors
using the same method, and no signicant evidence was found that led
to spatial variations in the distance to the river coefcients. The detailed
results of these analyses are shown in Fig. S2 of the supplementary
material.
The correlation and regression statistical analyses of the DIST coef-
cient of the GWR model with the COMM, SR, INT and PARK variables
conrm the spatial distribution described above (Fig. 9). In terms of the
spatial distribution, the DIST coefcient increases with increasing
COMM, SR and INT and decreases with increasing PARK. Among these
variables, the effect of COMM on the DIST coefcients is the largest, with
a Pearson correlation coefcient of 0.664 and R
2
of the regression model
of 0.441. The effects of SR and INT on the DIST coefcients have Pear-
son’s correlation coefcients of 0.554 and 0.562 and R
2
values of 0.308
and 0.316, respectively. Although PARK has the weakest effect on the
DIST coefcient, the Pearson correlation coefcient between the two
variables reaches −0.418, which is a statistically signicant correlation
(P value <0.01). Such a result can be understood in two ways:
(1) The effects of COMM, SR, and INT on the DIST coefcients have
the same trend. As Fig. 9(A-C) show, the DIST coefcients in areas
with dense distributions of COMM, SR and INT show positive
values, i.e., the river has a positive effect on reducing PM
2.5
pollutant concentrations in the riverside area. This indicates that
trafc pollution from the roads is the main source of pollution in
the area, implying that a higher road density may mean more
pollution emissions. However, in the riparian area, the road
system is a small-block, dense road network that supports the
natural dispersion of pollutants and reduces the effect of relative
humidity on the condensation of gaseous precursors. That is, this
small-block, dense road network reduces the nal PM
2.5
con-
centration by both increasing PM
2.5
dispersion and reducing the
condensation of gas precursors. In areas where COMM, SR and
INT are densely distributed, the road pattern is a small-block,
dense road network that provides the necessary conditions for
river wind inltration. Roads in cities are important ventilation
channels, and small-block, dense road networks form ventilation
channels that provide suitable conditions for river wind inltra-
tion (Yuan, 2019). The higher the road network density is, the
higher INT, which is more open and provides favorable condi-
tions for river wind inltration (Guo et al., 2022). Trafc pollu-
tion generated by the roads themselves does not accumulate in
the area and is carried away by the river breeze that inltrates
through the ventilation channels formed by the roads. At the
same time, the dense road network also characterizes more
human activity, and the high level of human activity discharges
heat and reduces the humidity that is brought by the river (Teufel
et al., 2021), which to a certain extent reduces the negative
impact due to humidity.
(2) The black line in the gure is a linear regression line, and the red
line is the locally weighted scatterplot smoothing (LOWESS)
curve. If the two are close, it indicates that the two elements used
for correlation analysis are more in line with the linear rela-
tionship; if they are far apart, it indicates that there is not a simple
linear relationship between the two elements, and there may be a
nonlinear relationship or an inuence by other elements. As
shown in Fig. 9(H), the red line is far from the black line, indi-
cating that the effect of the park on the DIST coefcient may be
nonlinear. The negative correlation between PARK and the DIST
coefcient indicates that the reduction effect of urban parks on
PM
2.5
concentrations is greater than the effect of water bodies on
PM
2.5
concentrations. Such results are consistent with the results
of Xu et al. (2021), who found that the effects of large cultivated
areas and forests on PM
2.5
were greater than the effect of water
bodies, while this paper also nds differences in the effects of
urban parks and water bodies on PM
2.5
concentrations. The
reduction effect of urban parks on particulate matter concentra-
tions has been demonstrated in most studies (Chen et al., 2019;
Shi et al., 2016; Wu et al., 2018); around the parks, the strong
negative correlation between the DIST and pollutants proves this
effect, and the negative correlation is stronger in areas closer to
the parks. As shown in Fig. 9(D), the absolute value of the
negative coefcient of the DIST near the park is approximately
twice the positive coefcient of the DIST in the area closest to the
river, which is because the reduction effect of urban parks on
PM
2.5
concentrations is greater than the effect of rivers on PM
2.5
concentrations. It is worth noting that this conclusion is drawn
from this specic case, in which the park is much closer to the
considered point than is the river. Whether this conclusion is
generalizable needs to be further veried.
4.3. Implications for urban planning and design
This study further claries the relationship between rivers and the
variation in PM
2.5
concentrations in riverfront neighborhoods by
considering spatial nonstationarity to some extent. The results of this
study shed light on the mechanism of a river’s inuence on PM
2.5
con-
centrations and provide the following insights for urban planning and
design for the purpose of reducing PM
2.5
concentrations.
Road network density in large waterfront neighborhoods is critical;
in other words, ensuring that the riverfront neighborhoods are well
ventilated is important for taking advantage of the river breeze to
improve air quality. For urban planning and design, a high density of
riverfront neighborhoods in waterfront urban centers is critical for
optimal land use. However, dense buildings in riverfront neighborhoods
mean less open space and road networks, which can impede airow and
diminish the effect of river breezes on the air environment of the
riverfront neighborhoods. Urban development patterns consisting of
large blocks with sparse road networks can potentially lead to increased
health risks in the absence of adequate, evidence-based design guidance.
In this study, the abatement effect of large water bodies on PM
2.5
con-
centrations in waterfront neighborhoods was stronger in small neigh-
borhoods with dense road networks than in large blocks with sparse road
networks. On the one hand, large water bodies have no obstructive effect
on natural winds and provide favorable conditions for pollutant
dispersion (Crosman and Horel, 2016); on the other hand, dense road
networks form ventilation channels, which provide suitable conditions
for river breeze inltration (Yuan, 2019). The different effects of
different road network patterns on river-inuenced PM
2.5
concentra-
tions in riverfront neighborhoods suggest that in the street planning and
design of riverfront neighborhoods, priority should be given to
small-block, dense road networks and the residential model of large
street districts should be minimized.
Ensuring larger areas of concentrated green space and water bodies
in cities is important, as both features contribute to PM
2.5
reduction. We
also found differences in the effects of water bodies and parks on PM
2.5
concentrations, with urban parks contributing more to PM
2.5
deposition
than do water bodies; this suggests that in urban planning and design
practices, high priority should be given to increasing urban park
coverage (the area of the urban park in this study is approximately 1
km
2
), fully utilizing open space, and increasing the number of urban
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
12
parks.
4.4. Limitations
Although this study provides useful information, some limitations
still exist. First, the presence of a long river bank with greenery in the
area selected for this study caused some interference in studying the
effect of the river on air quality in the riverfront neighborhood, which
may be the reason for the negative correlation between the distance
from the river and the pollutant concentrations in the study results; care
should be taken to avoid greenery in the experimental design of the
subsequent study. Second, urban parks were found to have a greater
mitigating effect on PM
2.5
pollution than did water bodies in this specic
case, in which the park was much closer to the considered point than
was the river. Whether this conclusion is generalizable needs to be
further veried. In a future study, we will site the actual measurement
sites at equal distances from parks and rivers for further comparison. In
addition, this study considered only spatial nonstationarity and did not
explore temporal nonstationarity. Interactions between land use and
urban morphology and PM
2.5
concentrations need to be investigated in
depth in subsequent studies.
5. Conclusions
In this study, the spatially heterogeneous inuence of a river on the
variation in PM
2.5
concentrations in riverside neighborhoods was
investigated through GWR. Multiple effects of the river on PM
2.5
con-
centrations were found considering spatial nonstationarity. A sensitivity
analysis between the DIST and PM
2.5
concentrations identied 800 m as
the extent to which the river strongly inuenced the air quality of its
surrounding neighborhoods, and the strongest effect was found at
500–700 m.
The focus of this study was to identify areas that are sensitive to the
effect of a river in terms of air quality and to provide a theoretical basis
for the planning and design of riverfront neighborhoods. By introducing
the GWR model, the spatial heterogeneity of the effect of rivers on air
quality was found. The effect of rivers on air quality mainly depends on
the ventilation conditions of the waterfront area, and good ventilation
conditions can amplify the effect of rivers in terms of air quality
improvement. Commercial, road intersections, secondary roads and
parks were identied as sensitive environmental factors affecting the
effect of rivers on regulating PM
2.5
pollution. The heterogeneity analysis
also found that urban parks had a greater effect on air quality than did
rivers. Considering the results of our assessment of the river’s impact on
air quality in the riverfront neighborhoods, the planning and design of
riverfront streets should highly prioritize small blocks with a dense road
network. In addition, urban parks (the area of urban parks in this study
is approximately 1 km
2
) should be prioritized over rivers. This study
provides a step toward understanding the relationship between street-
level air pollution and various factors of the built environment consid-
ering spatial nonstationarity. The results of this study provide an intu-
itive and accurate quantitative reference for urban planners and policy
makers regarding urban renewal in the era of standardized
management.
Author statement
Jiangying XU: Conceptualization, Data curation, Methodology,
Visualization, Writing – original draft. Mengyang LIU: Conceptualiza-
tion, Software, Writing – review & editing. Hong CHEN: Supervision,
Funding acquisition.
Declaration of competing interest
The authors declare that they have no known competing nancial
interests or personal relationships that could have appeared to inuence
the work reported in this paper.
Acknowledgement
This study was supported the National Natural Science Foundation of
China (No. 51778251).
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.apr.2022.101539.
References
Alshetty, D., Nagendra, S.M.S., 2022. Impact of vehicular movement on road dust
resuspension and spatiotemporal distribution of particulate matter during
construction activities. Atmos. Pollut. Res. 13, 101256 https://doi.org/10.1016/j.
apr.2021.101256.
Baldauf, R., 2017. Roadside vegetation design characteristics that can improve local,
near-road air quality. Transport. Res. Part Transp. Environ. 52, 354–361. https://
doi.org/10.1016/j.trd.2017.03.013.
Brauer, M., Hoek, G., Van Vliet, P., Meliefste, K., Fischer, P., Gehring, U., Heinrich, J.,
Cyrys, J., Bellander, T., Lewne, M., Brunekreef, B., 2003. Estimating lonq-term
average particulate air pollution concentrations: application of trafc indicators and
geographic information systems. Epidemiology 14, 228–239. https://doi.org/
10.1097/00001648-200303000-00019.
Brunsdon, C., Fotheringham, A.S., Charlton, M.E., 2010. Geographically weighted
regression: a method for exploring spatial nonstationarity. Geogr. Anal. 28, 281–298.
https://doi.org/10.1111/j.1538-4632.1996.tb00936.x.
Bulletin on China’s Bulletin on China’s Ecological Environment 2020, n.D.
Chen, M., Bai, J., Zhu, S., Yang, B., Dai, F., 2021a. The inuence of neighborhood-level
urban morphology on PM
2.5
variation based on random forest regression. Atmos.
Pollut. Res. 12, 101147 https://doi.org/10.1016/j.apr.2021.101147.
Chen, M., Bai, J., Zhu, S., Yang, B., Dai, F., 2021b. The inuence of neighborhood-level
urban morphology on PM
2.5
variation based on random forest regression. Atmos.
Pollut. Res. 12, 101147 https://doi.org/10.1016/j.apr.2021.101147.
Chen, M., Dai, F., Yang, B., Zhu, S., 2019. Effects of urban green space morphological
pattern on variation of PM
2.5
concentration in the neighborhoods of ve Chinese
megacities. Build. Environ. 158, 1–15. https://doi.org/10.1016/j.
buildenv.2019.04.058.
Crosman, E.T., Horel, J.D., 2016. Winter lake breezes near the great salt lake. Bound.
Layer Meteorol 159, 439–464. https://doi.org/10.1007/s10546-015-0117-6.
de Hoogh, K., Wang, M., Adam, M., Badaloni, C., Beelen, R., Birk, M., Cesaroni, G.,
Cirach, M., Declercq, C., D˙
edel˙
e, A., Dons, E., de Nazelle, A., Eeftens, M., Eriksen, K.,
Eriksson, C., Fischer, P., Graˇ
zuleviˇ
cien˙
e, R., Gryparis, A., Hoffmann, B., Jerrett, M.,
Katsouyanni, K., Iakovides, M., Lanki, T., Lindley, S., Madsen, C., M¨
olter, A.,
Mosler, G., N´
ador, G., Nieuwenhuijsen, M., Pershagen, G., Peters, A., Phuleria, H.,
Probst-Hensch, N., Raaschou-Nielsen, O., Quass, U., Ranzi, A., Stephanou, E.,
Sugiri, D., Schwarze, P., Tsai, M.-Y., Yli-Tuomi, T., Varr´
o, M.J., Vienneau, D.,
Weinmayr, G., Brunekreef, B., Hoek, G., 2013. Development of land use regression
models for particle composition in twenty study areas in europe. Environ. Sci.
Technol. 47, 5778–5786. https://doi.org/10.1021/es400156t.
Ferreira, A.P.S., Ramos, J.M.O., Gamaro, G.D., Gioda, A., Gioda, C.R., Souza, I.C.C.,
2022. Experimental rodent models exposed to ne particulate matter (PM
2.5
)
highlighting the injuries in the central nervous system: a systematic review. Atmos.
Pollut. Res. 13, 101407 https://doi.org/10.1016/j.apr.2022.101407.
Feuillet, T., Coquin, J., Mercier, D., Cossart, E., Decaulne, A., J´
onsson, H.P.,
Sæmundsson, þorsteinn, 2014. Focusing on the spatial non-stationarity of landslide
predisposing factors in northern Iceland: do paraglacial factors vary over space?
Prog. Phys. Geogr. Earth Environ. 38, 354–377. https://doi.org/10.1177/
0309133314528944.
Fontes, T., Li, P., Barros, N., Zhao, P., 2017. Trends of PM
2.5
concentrations in China: a
long term approach. J. Environ. Manag. 196, 719–732. https://doi.org/10.1016/j.
jenvman.2017.03.074.
Guo, X., Buccolieri, R., Gao, Z., Zhang, M., Lyu, T., Rui, L., Shen, J., 2022. On the effects
of urban-like intersections on ventilation and pollutant dispersion. Build. Simulat.
15, 419–433. https://doi.org/10.1007/s12273-021-0816-5.
Habermann, M., Billger, M., Haeger-Eugensson, M., 2015. Land use regression as method
to model air pollution. Previous results for gothenburg/Sweden. Procedia Eng. 115,
21–28. https://doi.org/10.1016/j.proeng.2015.07.350.
Hankey, S., Marshall, J.D., 2015. Land use regression models of on-road particulate air
pollution (particle number, black carbon, PM
2.5
, particle size) using mobile
monitoring. Environ. Sci. Technol. 49, 9194–9202. https://doi.org/10.1021/acs.
est.5b01209.
Ho, C.-C., Chan, C.-C., Cho, C.-W., Lin, H.-I., Lee, J.-H., Wu, C.-F., 2015. Land use
regression modeling with vertical distribution measurements for ne particulate
matter and elements in an urban area. Atmos. Environ. 104, 256–263. https://doi.
org/10.1016/j.atmosenv.2015.01.024.
Janh¨
all, S., 2015. Review on urban vegetation and particle air pollution – deposition and
dispersion. Atmos. Environ. 105, 130–137. https://doi.org/10.1016/j.
atmosenv.2015.01.052.
J. Xu et al.
Atmospheric Pollution Research 13 (2022) 101539
13
Jeanjean, A.P.R., Buccolieri, R., Eddy, J., Monks, P.S., Leigh, R.J., 2017. Air quality
affected by trees in real street canyons: the case of Marylebone neighbourhood in
central London. Urban For. Urban Green. 22, 41–53. https://doi.org/10.1016/j.
ufug.2017.01.009.
Kang, X.-M., Cui, L.-J., Zhao, X.-S., Li, W., Zhang, M.-Y., Wei, Y.-Y., Lei, Y.-R., Ma, M.-Y.,
2015. Effect of wetlands on reducing atmospheric ne particles PM
2.5
in Beijing.
Chin. J. Ecol. 34, 2807–2813.
Lim, C.C., Kim, H., Vilcassim, M.J.R., Thurston, G.D., Gordon, T., Chen, L.-C., Lee, K.,
Heimbinder, M., Kim, S.-Y., 2019. Mapping urban air quality using mobile sampling
with low-cost sensors and machine learning in Seoul, South Korea. Environ. Int. 131,
105022 https://doi.org/10.1016/j.envint.2019.105022.
Liu, C., Henderson, B.H., Wang, D., Yang, X., Peng, Z., 2016. A land use regression
application into assessing spatial variation of intra-urban ne particulate matter
(PM
2.5
) and nitrogen dioxide (NO2) concentrations in City of Shanghai, China. Sci.
Total Environ. 565, 607–615. https://doi.org/10.1016/j.scitotenv.2016.03.189.
Liu, J., Cai, W., Zhu, S., Dai, F., 2020. Impacts of vehicle emission from a major road on
spatiotemporal variations of neighborhood particulate pollution—a case study in a
university campus. Sustain. Cities Soc. 53, 101917 https://doi.org/10.1016/j.
scs.2019.101917.
Liu, M., Chen, H., Wei, D., Wu, Y., Li, C., 2021. Nonlinear relationship between urban
form and street-level PM
2.5
and CO based on mobile measurements and gradient
boosting decision tree models. Build. Environ. 205, 108265 https://doi.org/
10.1016/j.buildenv.2021.108265.
Lu, D., Mao, W., Yang, D., Zhao, J., Xu, J., 2018. Effects of land use and landscape pattern
on PM
2.5
in Yangtze River Delta, China. Atmos. Pollut. Res. 9, 705–713. https://doi.
org/10.1016/j.apr.2018.01.012.
Lu, X., Lin, C., Li, W., Chen, Y., Huang, Y., Fung, J.C.H., Lau, A.K.H., 2019. Analysis of
the adverse health effects of PM
2.5
from 2001 to 2017 in China and the role of
urbanization in aggravating the health burden. Sci. Total Environ. 652, 683–695.
https://doi.org/10.1016/j.scitotenv.2018.10.140.
Lyu, B., Cai, Y., Sun, Z., Li, J., Liang, L., 2022. Evaluating temporally decomposed
associations between PM
2.5
and hospitalisation risks of AECOPD: a case study in
Beijing from 2010 to 2019. Atmos. Pollut. Res. 13, 101356 https://doi.org/10.1016/
j.apr.2022.101356.
Masiol, M., Zíkov´
a, N., Chalupa, D.C., Rich, D.Q., Ferro, A.R., Hopke, P.K., 2018. Hourly
land-use regression models based on low-cost PM monitor data. Environ. Res. 167,
7–14. https://doi.org/10.1016/j.envres.2018.06.052.
Minet, L., Gehr, R., Hatzopoulou, M., 2017. Capturing the sensitivity of land-use
regression models to short-term mobile monitoring campaigns using air pollution
micro-sensors. Environ. Pollut. 230, 280–290. https://doi.org/10.1016/j.
envpol.2017.06.071.
Nowak, D.J., Hirabayashi, S., Bodine, A., Greeneld, E., 2014. Tree and forest effects on
air quality and human health in the United States. Environ. Pollut. 193, 119–129.
https://doi.org/10.1016/j.envpol.2014.05.028.
Ross, Z., Jerrett, M., Ito, K., Tempalski, B., Thurston, G., 2007. A land use regression for
predicting ne particulate matter concentrations in the New York City region.
Atmos. Environ. 41, 2255–2269. https://doi.org/10.1016/j.atmosenv.2006.11.012.
Salamanca, F., Martilli, A., Tewari, M., Chen, F., 2011. A study of the urban boundary
layer using different urban parameterizations and high-resolution urban canopy
parameters with WRF. J. Appl. Meteorol. Climatol. 50, 1107–1128. https://doi.org/
10.1175/2010JAMC2538.1.
Schwartz, B.S., 2002. Neurobehavioural testing in workers occupationally exposed to
lead. Occup. Environ. Med. 59, 648–649. https://doi.org/10.1136/oem.59.9.648.
Shi, Y., Lau, K.K.-L., Ng, E., 2017. Incorporating wind availability into land use
regression modelling of air quality in mountainous high-density urban environment.
Environ. Res. 157, 17–29. https://doi.org/10.1016/j.envres.2017.05.007.
Shi, Y., Lau, K.K.-L., Ng, E., 2016. Developing street-level PM
2.5
and PM
10
land use
regression models in high-density Hong Kong with urban morphological factors.
Environ. Sci. Technol. 50, 8178–8187. https://doi.org/10.1021/acs.est.6b01807.
Shi, Y., Ren, C., Lau, K.K.-L., Ng, E., 2019. Investigating the inuence of urban land use
and landscape pattern on PM
2.5
spatial variation using mobile monitoring and
WUDAPT. Landsc. Urban Plann. 189, 15–26. https://doi.org/10.1016/j.
landurbplan.2019.04.004.
Shi, Y., Xie, X., Fung, J.C.-H., Ng, E., 2018. Identifying critical building morphological
design factors of street-level air pollution dispersion in high-density built
environment using mobile monitoring. Build. Environ. 128, 248–259. https://doi.
org/10.1016/j.buildenv.2017.11.043.
Song, C., Wu, L., Xie, Y., He, J., Chen, X., Wang, T., Lin, Y., Jin, T., Wang, A., Liu, Y.,
Dai, Q., Liu, B., Wang, Y., Mao, H., 2017. Air pollution in China: Status and
spatiotemporal variations. Environ. Pollut. 227, 334–347. https://doi.org/10.1016/
j.envpol.2017.04.075.
Spinazz`
e, A., Fanti, G., Borghi, F., Del Buono, L., Campagnolo, D., Rovelli, S.,
Cattaneo, A., Cavallo, D.M., 2017. Field comparison of instruments for exposure
assessment of airborne ultrane particles and particulate matter. Atmos. Environ.
154, 274–284. https://doi.org/10.1016/j.atmosenv.2017.01.054.
Tessum, M.W., Larson, T., Gould, T.R., Simpson, C.D., Yost, M.G., Vedal, S., 2018. Mobile
and xed-site measurements to identify spatial distributions of trafc-related
pollution sources in los angeles. Environ. Sci. Technol. 52, 2844–2853. https://doi.
org/10.1021/acs.est.7b04889.
Teufel, B., Sushama, L., Poitras, V., Dukhan, T., B´
elair, S., Miranda-Moreno, L., Sun, L.,
Sasmito, A.P., Bitsuamlak, G., 2021. Impact of COVID-19-related trafc slowdown
on urban heat characteristics. Atmosphere 12, 243. https://doi.org/10.3390/
atmos12020243.
Wania, A., Bruse, M., Blond, N., Weber, C., 2012. Analysing the inuence of different
street vegetation on trafc-induced particle dispersion using microscale simulations.
J. Environ. Manag. 94, 91–101. https://doi.org/10.1016/j.jenvman.2011.06.036.
Weinbruch, S., Worringen, A., Ebert, M., Scheuvens, D., Kandler, K., Pfeffer, U.,
Bruckmann, P., 2014. A quantitative estimation of the exhaust, abrasion and
resuspension components of particulate trafc emissions using electron microscopy.
Atmos. Environ. 99, 175–182. https://doi.org/10.1016/j.atmosenv.2014.09.075.
Wong, G.W.K., Ko, F.W.S., Lau, T.S., Li, S.T., Hui, D., Pang, S.W., Leung, R., Fok, T.F.,
Lai, C.K.W., 2001. Temporal relationship between air pollution and hospital
admissions for asthmatic children in Hong Kong: air pollution and hospital
admissions for asthmatic children. Clin. Exp. Allergy 31, 565–569. https://doi.org/
10.1046/j.1365-2222.2001.01063.x.
Wu, H., Yang, C., Chen, J., Yang, S., Lu, T., Lin, X., 2018. Effects of Green space
landscape patterns on particulate matter in Zhejiang Province, China. Atmos. Pollut.
Res. 9, 923–933. https://doi.org/10.1016/j.apr.2018.03.004.
Wu, J., Li, J., Peng, J., Li, W., Xu, G., Dong, C., 2015a. Applying land use regression
model to estimate spatial variation of PM
2.5
in Beijing, China. Environ. Sci. Pollut.
Res. 22, 7045–7061. https://doi.org/10.1007/s11356-014-3893-5.
Wu, J., Xie, W., Li, W., Li, J., 2015b. Effects of urban landscape pattern on PM
2.5
pollution—a beijing case study. PLoS One 10, e0142449. https://doi.org/10.1371/
journal.pone.0142449.
Wu, X., Braun, D., Schwartz, J., Kioumourtzoglou, M.A., Dominici, F., 2020. Evaluating
the impact of long-term exposure to ne particulate matter on mortality among the
elderly. Sci. Adv. 6, 1–9. https://doi.org/10.1126/sciadv.aba5692.
Xu, W., Jin, X., Liu, M., Ma, Z., Qian, Wang, Zhou, Y., 2021. Analysis of spatiotemporal
variation of PM
2.5
and its relationship to land use in China. Atmos. Pollut. Res. 12,
101151 https://doi.org/10.1016/j.apr.2021.101151.
Ye, W.-F., Ma, Z.-Y., Ha, X.-Z., 2018. Spatial-temporal patterns of PM
2.5
concentrations
for 338 Chinese cities. Sci. Total Environ. 631, 524–533. https://doi.org/10.1016/j.
scitotenv.2018.03.057.
Yuan, M., 2019. Exploring the association between the built environment and remotely
sensed PM
2.5
concentrations in urban areas. J. Clean. Prod. 10.
Zhang, L., Zhang, J., Chen, S., Tian, X., Zhao, Y., Liu, L., Tao, L., Wang, X., Guo, X.,
Luo, Y., 2022. Association between blood pressure and short-term exposure to
ambient air pollutants in Beijing, China. Atmos. Pollut. Res. 13, 101293 https://doi.
org/10.1016/j.apr.2021.101293.
Zhou, S., Lin, R., 2019. Spatial-temporal heterogeneity of air pollution: the relationship
between built environment and on-road PM
2.5
at micro scale. Transport. Res. Part
Transp. Environ. 76, 305–322. https://doi.org/10.1016/j.trd.2019.09.004.
Zhou, X., Zhang, S., Zhu, D., 2021. Impact of urban water networks on microclimate and
PM
2.5
distribution in downtown areas: a case study of wuhan. Build. Environ. 203,
108073 https://doi.org/10.1016/j.buildenv.2021.108073.
Zhu, C., Zeng, Y., 2018. Effects of urban lake wetlands on the spatial and temporal
distribution of air PM
10
and PM
2.5
in the spring in Wuhan. Urban For. Urban Green.
31, 142–156. https://doi.org/10.1016/j.ufug.2018.02.008.
Zhu, C., Zeng, Y., Chen, Y., Guo, H., 2016. Effects of urban lake wetland on air PM
10
and
PM
2.5
concentration—a case study of wuhan. Chin. Landsc. Archit. 32, 88–93.
Zhu, D., Zhou, X., 2019. Effect of urban water bodies on distribution characteristics of
particulate matters and NO
2
. Sustain. Cities Soc. 50, 101679 https://doi.org/
10.1016/j.scs.2019.101679.
Zhu, Q., Hou, J., Wang, G., Xu, T., Yin, W., Cheng, J., Wang, L., Yuan, J., 2022. Impaired
lung function related to microenvironmental exposure to PAHs mixture in PM
2.5
: a
repeated measurement study. Atmos. Pollut. Res. 13, 101273 https://doi.org/
10.1016/j.apr.2021.101273.
J. Xu et al.
