key: cord-1041251-6f8908an
authors: Han, Li; Zhao, Jingyuan; Gao, Yuejing; Gu, Zhaolin; Xin, Kai; Zhang, Jianxin
title: Spatial Distribution Characteristics of PM(2.5) and PM(10) in Xi’an City Predicted by Land Use Regression Models
date: 2020-06-13
journal: Sustain Cities Soc
DOI: 10.1016/j.scs.2020.102329
sha: b6a7923b39b10f9b87160e77a005a3dd6cb9ab0e
doc_id: 1041251
cord_uid: 6f8908an

PM(2.5) and PM(10) could increase the risk for cardiovascular and respiratory diseases in the general public and severely limit the sustainable development in urban areas. Land use regression models are effective in predicting the spatial distribution of atmospheric pollutants, and have been widely used in many cities in Europe, North America and China. To reveal the spatial distribution characteristics of PM(2.5) and PM(10) in Xi'an during the heating seasons, the authors established two regression prediction models using PM(2.5) and PM(10) concentrations from 181 monitoring stations and 87 independent variables. The model results are as follows: for PM(2.5), R(2) = 0.713 and RMSE =8.355 μg/m(3); for PM(10), R(2) = 0.681 and RMSE = 14.842 μg/m(3). In addition to the traditional independent variables such as area of green space and road length, the models also include the numbers of pollutant discharging enterprises, restaurants, and bus stations. The prediction results reveal the spatial distribution characteristics of PM(2.5) and PM(10) in the heating seasons of Xi'an. These results also indicate that the spatial distribution of pollutants is closely related to the layout of industrial land and the location of enterprises that generate air pollution emissions. Green space can mitigate pollution, and the contribution of traffic emission is less than that of industrial emission. To our knowledge, this study is the first to apply land use regression models to the Fenwei Plain, a heavily polluted area in China. It provides a scientific foundation for urban planning, land use regulation, air pollution control, and public health policy making. It also establishes a basic model for population exposure assessment, and promotes the sustainability of urban environments.

land use layout, and adverse meteorological factors and are thereby likely to accumulate in cities (Jin et al. 2019 ). Long-term exposure to contaminated atmosphere increases the risk of cardiovascular and respiratory diseases (Feng et al. 2016 , Berman et al. 2019 , Barzeghar et al. 2020) , Properly planned commuting routes can reduce human exposure to pollution (Ahmed et al. 2020 , Pilla and Broderick 2015 , Qiu et al. 2017 . The variations in spatial distribution of urban air pollutants have become a widespread concern in several fields such as urban and rural planning, environmental science, and medicine (Yuan et al. 2019 , Son et al. 2018 , Zou et al. 2009 , Yang et al. 2020 .

Land use regression (LUR) models have proven to be an effective method for predicting the spatial distribution of pollutants (Jerrett et al. 2005) . LUR models work using pollutant concentration data collected at a limited number of monitoring stations in conjunction with characteristic variables such as land use information to evaluate pollutant concentrations in areas that lack monitoring stations. A main feature of LUR models is the correlation of land use characteristics to the concentrations of target pollutants, which can be used to determine the relationship between air pollutant concentrations and other geographic variables to simulate the spatial distribution of air pollutants in urban areas and identify the causes of pollutants to a certain extent (Henderson et al. 2007 , Hoek et al. 2008 . The interpretation of the relationship can guide urban land use adjustment. By adjusting the layout of industrial land, intensive land use is achieved to reduce pollutant emission, and ultimately accomplish sustainable urban land use.

One of the earliest LUR models was developed by Briggs et al. (1997) to predict NO2 concentrations in three European cities and plot pollution maps. Owing to the simplicity of model construction and ease of data acquisition as well as improvement in the modeling technology and J o u r n a l P r e -p r o o f content and higher diversity in the allowed variable types, LUR models have also been applied to the prediction of air pollutants such as NO2, PM10, and PM2.5 in Europe and North America (Allen et al. 2011 , Eeftens et al. 2012 , Moore et al. 2007 , Briggs et al. 2000 . Two early studies in 2010 (Chen et al. 2010a , Chen et al. 2010b ) predicted the concentrations of PM10, NO2, and SO2 in two Chinese cities (Tianjin and Jinan) and plotted pollutant distribution maps. Since 2013, China has begun to gradually establish and improve its air quality monitoring network, an action accompanied by a growing body of LUR-model-based studies investigating the spatial distribution of air pollutants in Chinese cities such as Chengdu, Changsha, Beijing, and Shanghai, with a particularly large number of studies in the latter two cities (Xiao et al. 2018 , Meng et al. 2016 , Meng et al. 2015 , Ji, Wang and Zhuang 2019 . Compared with the research results of other cities, this study is the first to apply land use regression models to the Fenwei Plain, a region in China with severe air pollution. The number of monitoring stations used in this study is the greatest so far. The dependent variables are selected specifically during the heating seasons. It explains the reasoning behind the selection of buffer zones. It applies comprehensively all aspects of regression model diagnosis, crossvalidation and verification of LUR applicability in different heating seasons. Hence, to a certain extent, it has enriched the application of land use regression models in the prediction of the spatial distribution of atmospheric pollutants.

Xi'an is an important western Chinese city located on the west of the rift basin of the Fenwei Plain, encompassing 11 cities; a plain subjected to smog in large areas and considered one of the most atmospherically polluted areas in China. The national air quality report released by the Ministry of Ecology and Environment of China revealed that between January and December 2018, Xi'an ranked 158 out of 169 cities in terms of air quality and was under severe air pollution. Xi'an has special natural J o u r n a l P r e -p r o o f 6 conditions (i.e., unique topography and unfavorable meteorological conditions) with a special land use status, economic development level, and industrial layout. Thus, the air pollution is further exacerbated (Song et al. 2015 . Therefore, a deep insight into the spatial distribution characteristics of the concentrations of PM2.5 and PM10 during the heating seasons of Xi'an is the key to supporting air pollution control.

The purpose of this study is to establish a regression prediction model for PM2.5 and PM10 in the heating seasons of Xi'an to test the applicability of the land use regression models, and to reveal the spatial distribution characteristics of these pollutants. Further, it aims to analyze the relationship between the spatial distributions of PM2.5 and PM10 and land use characteristics and therefore provides a scientific foundation for urban planning, land use regulation, air pollution control, and public health policy-making. It also establishes a basic model for population exposure assessment. The application of the land use regression models to areas requiring heating in winter and heavily polluted areas plays a positive role in achieving sustainable urban environment and promoting sustainable development in urban environments.

The air quality data used in this study were the daily mean concentrations of PM2.5 and PM10 at 181 air quality monitoring stations under the Xi'an Ecology and Environment Bureau. The concentration measurements were conducted during the winter heating season from November 15, 2018 to March 15, 2019, and the period-averaged concentrations of PM2.5 and PM10 obtained from each monitoring station were used as the dependent variables, as depicted in Fig. 1 . A total of 86 factors in the five categories of land use, road traffic facilities, socioeconomic development, emission source, and geospatial information were considered as independent variable candidates. For a particular monitoring station, independent variable candidates were extracted in two ways: (1) with the monitoring station as the center, circular buffer zones at various distances were delineated using GIS, and the independent variable candidates were the length, number, or area within the buffer zones, Fig. J o u r n a l P r e -p r o o f 7 2 shows the length of the roads in the 3000 m buffer zone around the New Software City station; (2) the independent variable candidates were the distances from the monitoring stations to notable objects or the characteristic values of monitoring stations, Fig. 3 displays the distance from the New Software City station to the nearest highways. After the extraction of variables, correlation analysis was performed between the independent and dependent variables using SPSS software. Variable screening was performed based on the magnitude of the Pearson correlation coefficients, and the selected independent variables were included in multiple stepwise regression analyses using R software. The established models were subjected to cross-validation to test their generalizability using R software.

The procedure followed in this study is depicted in Fig. 4 . producers as well as consumers. In particular, coal consumption is more centralized in the Fenwei Plain, accounting for nearly 90% of its total energy consumption, far greater than the national average of 60%. 

The land use data used in this study were derived from the remote sensing monitoring data of J o u r n a l P r e -p r o o f

China's land use, built by the Chinese Academy of Sciences. The Landsat TM/ETM/OLI remote sensing images are the main data source. After going through processes such as image fusion, geometric correction, image enhancement and stitching etc., the land use types were classified into 6 first-level categories, 25 second-level categories and certain third-level categories according to the land use/cover classification system in China, via human-computer interactive visual interpretation. In this study, five types of land-use data were extracted, namely, those from farmland (paddy fields + dry land), green land (forest land + grassland), waterbodies (waterways + lakes + reservoirs + ponds), industrial and mining land (factories and mines + industrial areas + airports), and construction land (urban land + rural residential areas). The shape and size selection of a buffer zone was based on the diffusion range of the pollutants in the atmosphere and the impact of geographical elements on the pollutants. However, because of the complexity and uncertainty of atmospheric pollution diffusion, previous research results were usually referenced when determining a new buffer zone. In this study, the correlation coefficients of most independent variables increased with respect to pollutant concentrations as the buffer zones increased in size until the distance of 5000 m, a limiting distance beyond which the correlation coefficients would not increase. Thus, the maximum distance of buffer zones was set to 5000 m. GIS software was employed to establish station-centered circular buffer zones for each of the 181 monitoring stations at successive distances of 100, 300, 500, 1000, 2000, 3000, 4000, and 5000 m and then to extract the length or area data associated with each type of land use in each buffer zone.

Road traffic facility information referred to road networks, parking lots, and bus stops. In this study, five types of road network data were extracted from OpenStreetMap, namely, motorways, primary roads, secondary roads, tertiary roads, and trunk lines. The road network data in a buffer zone were extracted using two metrics: extraction of the length of motorways in the buffer zone or extraction of the total length of the five types of roads in the buffer zone. The parking lot data and bus stop data J o u r n a l P r e -p r o o f were obtained from Gaode map, i.e., the number of parking lots and bus stops in each buffer zone were calculated.

The socioeconomic information consisted of GDP and population data. The GDP data came from the kilogram grid dataset of the spatial distribution of China's GDP (GDP Grid China) constructed by the Chinese Academy of Sciences. It has a raster data form, with each raster representing the total GDP output value within the grid range (1 square kilometer) in the unit of ten thousand yuan/km 2 . The GDP value of each grid with a monitoring site was extracted from GIS data. The population data were derived from the sixth national census and used to calculate the population of each abovementioned sub-district or town.

The emission sources consisted of air-polluting enterprises, restaurants, and motorways. Airpolluting enterprise information was obtained from the Monitoring Information Release Platform of Key Pollutant-discharging Enterprises in Shaanxi Province and the List of Key Pollutant-discharging Units released by Xi'an Ecology and Environment Bureau. There were 37 pollutant-discharging enterprises. The distance between each monitoring station and the nearest pollutant-discharging enterprise from it was calculated. Also the number of pollutant-discharging enterprises in each buffer zone was counted. Restaurant data were obtained from Gaode map by calculating the number of restaurants in each buffer zone. Motorway data were obtained by calculating the distance from each monitoring station to its nearest motorway.

Geospatial information consisted of the elevation of each monitoring station and its distance to a water surface. Elevation data were obtained by extracting Digital Elevation Model (DEM) topographic data with a precision of 30 m. Considering the effect of water-land breeze on atmospheric pollution transport, the distance from each monitoring station to the nearest water bodies was calculated.

J o u r n a l P r e -p r o o f

A total of 87 independent variables were extracted. Bivariate correlation analysis was performed among the 87 independent variables and the dependent variables PM2.5 and PM10 using SPSS 17.0 to obtain the Pearson correlation coefficients of each independent variable with respect to PM2.5 and PM10.

Independent variables that are significantly correlated to the dependent variables (P <0.05) were selected. In each category of independent variables, the variables with the largest Pearson correlation coefficient were selected, which led to the selection of 13 independent variables for PM2.5 (Table 1) and 14 for PM10 (Table 2 ). Multiple stepwise regression of PM2.5 and PM10 was separately performed on these independent variables using R software to remove independent variables with a p-value > 0.01 while performing collinearity diagnostics to remove independent variables with Variance Inflation Factor (VIF) > 4.

The models were subjected to regression diagnostics using R software. For regression diagnostics, the results of model fitting were presented in four plots: (1) a Residual-versus-Fitted plot to test the assumption that the independent variable in question was linearly correlated to the dependent variable in question;

(2) a Normal Q-Q plot to test the normality of residuals; (3) a Scale-Location plot, intended to test the homoscedasticity assumption; and (4) a Residual-versus-Leverage plot to identify outliers, high-impact points, and high-leverage points.

The models were subjected to cross-validation using R software. Model validation was performed using the 10-fold cross-validation method aimed at testing the generalizability of the models. Thus, the samples in question were divided into 10 equal-sized subsamples, of which one subsample was retained as the validation group, whereas the remaining nine subsamples were used as the training group; repeating this process in turn for each subsample as the validation group finally led to a total of 10 prediction equations, whose R 2 values and RMSEs were averaged. The equations for RMSE and R 2 calculations are listed as follows:

In the RMSE equation 

The results section analyzed the relationship between the industrial land layout of the study area and the Regulatory Management Zone (RMZ) and the spatial distributions of PM2.5 and PM10 in the heating seasons in Xi'an. The definition of industrial land in urban and rural planning is applied here, which refers to land for production shops, warehouses and auxiliary facilities of industrial and mining enterprises, including land for special railways, docks and auxiliary roads, and parking lots. The information of industrial land layout was from the central urban area land use plan map from Xi'an's Urban Master Plan (2008-2020) and the remote sensing monitoring data of China's land use.

Bivariate correlation analysis indicated that 44 of the 87 independent variables were significantly correlated to PM2.5 and 42 to PM10. Among these 44 and 42 independent variables, those with the largest correlation coefficients in the respective categories were selected, ultimately leading to 13 and J o u r n a l P r e -p r o o f 14 independent variables selected for PM2.5 and PM10, respectively, which included 10 in common, as presented in Table 1 and Table 2 . coefficients with respect to PM2.5. The negative correlation coefficient r of GS-5000 with respect to PM10 was smaller than that of only EL, indicating that larger the green spaces in buffer zones, smaller were the concentrations of PM2.5 and PM10 at the stations. The negative correlation coefficient r of EL with respect to PM2.5 and PM10 was the second largest (after that of GS-4000) and the largest, respectively, which was attributed to the special terrain of Xi'an, i.e., stations with higher EL had larger green spaces. The correlation coefficient r of DIS-PE and DIS-MO with respect to the pollutants was large and significantly negative, indicating that closer the air-polluting enterprises and motorways, higher were the concentrations of PM2.5 and PM10. 

Bivariate correlation analysis confirmed that three independent variables, GS-4000, RE-5000,

and PE-5000, should be included in the PM2.5 LUR model. The regression coefficient was significant at the p < 0.05 level, and the adjusted R 2 value (Adj-R 2 ) was 0.713, indicating that 71.3% of the J o u r n a l P r e -p r o o f variance of the concentration of PM2.5 was accounted for by the model. The model RMSE was 8.355 μg/m 3 and the VIF of each variable was less than 4, indicating that there was no multicollinearity among the three independent variables (Table 3) . The observed and predicted concentrations of PM2.5

were compared (Fig.10) .

For the PM10 LUR model, bivariate correlation analysis confirmed that four independent variables, GS-5000, MO-5000, BS-5000, and PE-5000, should be included in the model. The regression coefficient was significant at the p < 0.05 level, and the Adj-R 2 was 0.681, indicating that 68.1% of the variance of the concentration of PM10 was accounted for by the model. The model RMSE was 14.842 μg/m 3 , and the VIF of each variable was less than 4, indicating that there was no multicollinearity among the four independent variables (Table 4 ). The observed and predicted concentrations of PM10

were compared (Fig.11) . Note: ***, ** and * indicate significant levels of significance at 0, 0.001, and 0.01 respectively. Note: ***, ** and * indicate significant levels of significance at 0, 0.001, and 0.01 respectively. 

The model evaluation measures included regression diagnostics, cross-validation, and the spatial autocorrelation test. Regression diagnostics were performed by checking four plots, namely, a residual- Adj-R 2 , and RMSE were all less than the respective averages in the 10-fold cross-validation, which indicated that there was no overfitting and underfitting and the LUR models had good generalizability.

For the spatial autocorrelation test of PM2.5 residuals, the p-value was greater than 0.05 (95% J o u r n a l P r e -p r o o f 20 confidence level) and the z-score was above the threshold of -1.65, indicating that the PM2.5 residuals were randomly distributed in space without spatial clustering. For the PM10 residuals, the p-value of the spatial autocorrelation test was greater than 0.05 (95% confidence level) and the z-score did not exceed the threshold of 1.65, indicating that the PM10 residuals were also randomly distributed in space without spatial clustering (Table 5) . 

To Table 6 . Table 7 shows that the independent variables of the PM10 regression model of the recent heating season are the same as those of the previous heating season. Adjustable R 2 = 0.639, lower than 0.681, the value of the previous heating season; RMSE = 12.704μg /m 3 , also lower than the value from the previous heating season at 14.842 μg/m 3 . The PM2.5 and PM10 land use regression models for the recent heating season exhibit reduced accuracy in fitting and smaller errors, but they J o u r n a l P r e -p r o o f still have good prediction capabilities, proving the applicability of the land use regression models in different heating seasons. Note: ***, ** and * indicate significant levels of significance at 0, 0.001, and 0.01 respectively. Note: ***, ** and * indicate significant levels of significance at 0, 0.001, and 0.01 respectively.

Xi'an was divided into a grid of 10,525 cells using ArcGIS, each of 1 km × 1 km area, and regression mapping was performed using the PM2.5 and PM10 regression models. During regression mapping, the independent variables associated with a cell were extracted at the centroid of the cell and then substituted in the PM2.5 and PM10 regression models to predict the concentrations of PM2.5 and PM10 for the cell. The spatial distribution maps (Figs.14 and 15) of the concentrations of PM2.5 and PM10 in Xi'an were generated using the visualization function of ArcGIS, which presented significant differences between the spatial distributions of the concentrations of PM2.5 and PM10 in Xi'an.

J o u r n a l P r e -p r o o f 4. Discussion

The correlation coefficients of the green-space area with respect to the concentrations of PM2.5 and PM10 were the largest in each buffer zone, and the coefficients increased with the increase in the area of the buffer zone, which indicated that with increase in green space, the negative correlation between the pollutant concentrations and the green spaces was more significant. The independent variable shared by the PM2.5 and PM10 land use regression prediction models of Xi'an is the number of polluting enterprises with a buffer area of 5000 m. Moreover, the results indicated that while green-J o u r n a l P r e -p r o o f space area was the most critical independent variable in both the models, it exerted its largest impact at various buffer-zone distances. In addition to the green-space area, the PM2.5 LUR model also included the number of restaurants in the 5000-m-distance buffer zone as an independent variable, whereas the PM10 LUR model included the number of bus stops and the length of motorways in the 5000-m-distance buffer zone as two independent variables. The reason green-space area was the most critical independent variable in both models was that Xi'an has a unique terrain and urban morphology.

That is to say, the forest coverage of Xi'an is as high as 48%, with a large area of green spaces and vegetation in the Qinling and Li Mountains. There is a strong negative correlation between the concentration of pollutants and the green-space area of the buffer zone and a significant positive correlation between the green-space area of the buffer zone and its elevation with a correlation coefficient of 0.86. The green-space area in a buffer zone around a station with a high elevation tends to be big. It indirectly reflects the strong negative correlation between the elevation of the monitoring station and the concentration of pollutants. Comparison with the studies of other cities in Table 8 indicates that there are similarities among Beijing, Czech-Poland and Xi'an in that a great number of mountains and green-space areas exist in the territories of these study areas. These researches also contain the independent variable of green-space area, which shows that the pollutant distributions are all impacted by the mountains and green space , Ji et al. 2019 , Bitta et al. 2018 . Those studies, except for that in Beijing by Wu et al. (2015) did not include industry-related independent variables. Other studies included the independent variable of pollution emission or industrial land area.

In this study, two industry-related independent variables were selected, namely the area of industrial land and the distance from air-polluting enterprise, and the distance from air-polluting enterprises was finally included in the model. Moreover, studies in Beijing and the other Chinese cities of Shanghai J o u r n a l P r e -p r o o f 26 and Tianjin also included road length as an independent variable , Meng et al. 2016 , Chen et al. 2010a , consistent with the present study. The Adj-R 2 values of the PM2.5 and PM10 LUR models of Xi'an were 0.713 and 0.681, respectively, lower than the Adj-R 2 values of 0.877 and 0.81, respectively, obtained in LUR models of Shanghai and Beijing (Ji et al. 2019 ) but higher than 0.43-0.65 and 0.54, respectively, for Beijing in another study and Hong Kong (Lee et al. 2017) . Therefore, the Adj-R 2 value obtained in this study was considered to represent a moderate level compared with other reports. This study used data observed at 181 monitoring stations-the largest number and density of monitoring stations among studies of this type.

However, the higher density of monitoring stations in this study did not lead to improvement in fitting accuracy compared with other studies, suggesting that fitting accuracy may not be simply dependent on the number and density of monitoring stations. 

The (YLDAI), respectively, with significantly higher concentrations of PM2.5 and PM10 compared with other areas. Huang et al. (2015) confirmed that the accumulation of PM2.5 in Xi'an was closely related to industrial production. Wang et al. (2014) reported that as high as 58% of the concentrations of PM2.5

in Xi'an in the heavy pollution months were caused by industrial activities. Song et al. (2015) revealed that transportation and industrial emissions were the main sources of PM2.5 in Xi'an. However, the correlation coefficients between the areas of industrial and mining land use and the concentrations of PM2.5 and PM10 were quite small, and the correlation was not statistically significant, which may be attributed to the possibility that the information collected on industrial and mining land use was not truly reflective of the actual situation. 

The spatial distribution characteristics of PM2.5 and PM10 were investigated in the RMZ, which is larger than the main urban district. Fig. 17 presents the layout of industrial land use, water systems, green spaces, and roads, all of which were factors related to the distribution of the concentrations of PM2.5 and PM10. The highest concentrations of PM2.5 and PM10 were observed in the area named WREWTR, which lies between the West Urban Ring Motorway and the West Third Ring Road, and in the HIDZ area. Both of these areas and their adjacent spaces contain a large industrial land and a large number of air-polluting enterprises, which also exist in the area north of the RMZ (NRMZ).

However, there was no accumulation of high concentrations of PM2.5 and PM10 in NRMZ, which may be attributed to the presence of large water bodies and green spaces around the area due to its proximity to the Wei River. On one hand, the large water bodies can generate water-land breeze through a mechanism similar to sea-land breeze ( 

Owing to the limited time available for data collection, this study focused on the heating season of Xi'an for investigating the spatial distribution of PM2.5 and PM10, thus being unable to address the distribution in other seasons and, thereby, making it impossible to understand inter-seasonal differences in the spatial distribution of PM2.5 and PM10. The land-use data employed in this study were mainly retrieved from remote sensing images as the main data source. However, these images were generated in an earlier time period than the pollution data, and, therefore, a certain degree of J o u r n a l P r e -p r o o f inter-period differences were noticed in land use, possibly introducing errors into the results of correlation analysis between the independent and dependent variables. Owing to the lack of corresponding meteorological monitoring data at the monitoring stations, this study did not consider the impact of meteorological factors such as wind direction, wind speed, air temperature, and humidity on the concentrations of PM2.5 and PM10. Further, the accuracy of model fitting can be improved using pollutant concentration data observed under lower wind speeds.

This study is the first to apply land use regression models to the Fenwei Plain, a heavily polluted area in China, and analyzes 87 factors from five categories of information including land use information, road traffic facility information, socioeconomic information, geospatial information, and emission source information in Xi'an. The correlation between the factors and the PM2. Conflict of Interest:All authors declare no conflict of interest.

Wets (2020) A route to school informational intervention for air pollution exposure reduction

The transferability of NO and NO2 land use regression models between cities and pollutants

Analyzing temporal changes in urban forest structure and the effect on air quality improvement

Long-term trend of ambient air PM10, PM2.5, and O3 and their health effects in Tabriz city

Acute Air Pollution Exposure and the Risk of Violent Behavior in the United States

Air Pollution Dispersion Modelling Using Spatial Analyses

Impact of a sea breeze event on air pollution at the Eastern Tunisian Coast

Mapping urban air pollution using GIS: a regression-based approach

A regression-based method for mapping traffic-related air pollution: application and testing in four contrasting urban environments

A land use regression for predicting NO2 and PM10 concentrations in different seasons in Tianjin region

Application of land use regression for estimating concentrations of major outdoor air pollutants in Jinan

Chemical nature of PM2.5 and PM10 in Xi'an, China: Insights into primary emissions and secondary particle formation

Development of Land Use Regression Models for PM2.5, PM2.5 Absorbance, PM10 and PMcoarse in 20 European Study Areas; Results of the ESCAPE Project

Scale-and Region-Dependence in Landscape-PM2.5 Correlation: Implications for Urban Planning. Remote Sensing

The health effects of ambient PM2.5 and potential mechanisms

Application of land use regression to estimate long-term concentrations of traffic-related nitrogen oxides and fine particulate matter

A review of land-use regression models to assess spatial variation of outdoor air pollution

Spatial and Temporal Distribution of PM2.5 Pollution in Xi'an City, China

A review and evaluation of intraurban air pollution exposure models

Spatial distribution differences in PM2.5 concentration between heating and nonheating seasons in Beijing

Using Bayesian spatio-temporal model to determine the socio-economic and meteorological factors influencing ambient PM2.5 levels in 109 Chinese cities

Land use regression modelling of air pollution in high density high rise cities: A case study in Hong Kong

A land use regression application into assessing spatial variation of intra-urban fine particulate matter (PM2.5) and nitrogen dioxide (NO2) concentrations in City of Shanghai

Land use regression models coupled with meteorology to model spatial and temporal variability of NO2 and PM10 in Changsha, China

Dynamic spatial spillover effect of urbanization on environmental pollution in China considering the inertia characteristics of environmental pollution

A land use regression model for estimating the NO2 concentration in

Estimating groundlevel PM10 in a Chinese city by combining satellite data, meteorological information and a land use regression model

Estimate annual and seasonal PM1, PM2.5 and PM10 concentrations using land use regression model

A land use regression model for predicting ambient fine particulate matter across Los Angeles, CA

The importance of local scale for assessing, monitoring and predicting of air quality in urban areas

A GIS model for personal exposure to PM10 for Dublin commuters

Pedestrian exposure to traffic PM on different types of urban roads: A case study of

Air pollution removal by trees in public green spaces in Strasbourg city

Land use regression models to assess air pollution exposure in Mexico City using finer spatial and temporal input parameters

Detecting urban land-use configuration effects on NO2 and NO variations using geographically weighted land use regression

Estimating PM2.5 Concentrations in Xi'an City Using a Generalized Additive Model with Multi-Source Monitoring Data

Source contributions to primary and secondary inorganic particulate matter during a severe wintertime PM2.5 pollution episode in

Source apportionment of environmentally persistent free radicals (EPFRs) in PM2.5 over

Applying land use regression model to estimate spatial variation of PM2.5 in Beijing

Spatiotemporal Characteristics of Air Pollutants (PM10, PM2.5, SO2, NO2, O3, and CO) in the Inland

Air pollution dispersal in high density urban areas: Research on the triadic relation of wind, air pollution, and urban form

A nonlinear model for estimating ground-level PM10 concentration in Xi'an using MODIS aerosol optical depth retrieval

Exploring the association between the built environment and remotely sensed PM2.5 concentrations in urban areas

Effects of urban lake wetlands on the spatial and temporal distribution of air PM10 and PM2.5 in the spring in Wuhan

Effect of urban water bodies on distribution characteristics of particulate matters and NO2. Sustainable Cities and Society

Spatially differentiated and source-specific population exposure to ambient urban air pollution

This study was supported by the National Natural Science Foundation of China (51678058); the Natural Science Foundation of Shaanxi Province (2019JM-475).