Impact of Land Use on PM2.5 Pollution in a Representative City of Middle China

Fine particulate matter (PM2.5) pollution has become one of the greatest urban issues in China. Studies have shown that PM2.5 pollution is strongly related to the land use pattern at the micro-scale and optimizing the land use pattern has been suggested as an approach to mitigate PM2.5 pollution. However, there are only a few researches analyzing the effect of land use on PM2.5 pollution. This paper employed land use regression (LUR) models and statistical analysis to explore the effect of land use on PM2.5 pollution in urban areas. Nanchang city, China, was taken as the study area. The LUR models were used to simulate the spatial variations of PM2.5 concentrations. Analysis of variance and multiple comparisons were employed to study the PM2.5 concentration variances among five different types of urban functional zones. Multiple linear regression was applied to explore the PM2.5 concentration variances among the same type of urban functional zone. The results indicate that the dominant factor affecting PM2.5 pollution in the Nanchang urban area was the traffic conditions. Significant variances of PM2.5 concentrations among different urban functional zones throughout the year suggest that land use types generated a significant impact on PM2.5 concentrations and the impact did not change as the seasons changed. Land use intensity indexes including the building volume rate, building density, and green coverage rate presented an insignificant or counter-intuitive impact on PM2.5 concentrations when studied at the spatial scale of urban functional zones. Our study demonstrates that land use can greatly affect the PM2.5 levels. Additionally, the urban functional zone was an appropriate spatial scale to investigate the impact of land use type on PM2.5 pollution in urban areas.


Introduction
In recent years, the air pollution problem generated by unprecedented urbanization and economic growth in China has become one of the greatest urban issues, particularly fine particulate matter (PM 2.5 ) pollution [1]. PM 2.5 , consisting of particles with aerodynamic diameters smaller than 2.5 µm, can absorb more hazardous substances than coarse particles and enter the human body by respiration, resulting in various respiratory and cardiovascular diseases [2]. Some epidemiological studies have confirmed that a long exposure to PM 2.5 will greatly increase rates of cardiopulmonary morbidity and mortality [3,4]. Therefore, gaining a better and clearer understanding of PM 2.5 pollution is of vital significance in preventing pollution and protecting public health.
Studies have shown that at the macro-scale, PM 2.5 pollution is significantly influenced by meteorological conditions [23][24][25][26]; at the micro-scale, PM 2.5 pollution is strongly related to the land use pattern [27][28][29][30]. Some researchers have suggested that optimizing the land use pattern may mitigate PM 2.5 pollution at a city or community level [31][32][33]. However, there are only a few researches analyzing the effect of land use on PM 2.5 pollution and the consensus about the exact nature of their relationship has not yet been reached [28,34]. Thus, exploring the effect of land use on PM 2.5 pollution seems to be urgent and significant.
To conduct research on the impact of land use on PM 2.5 pollution, available PM 2.5 data are critical. However, gaining access to enough PM 2.5 data creates a big challenge. Several approaches have been developed over the last decade to solve this challenge, including spatial interpolation (e.g., kriging and inverse distance weighing), air dispersion models, and land use regression (LUR) models. The interpolation of pollutant concentrations is based on dense monitoring sites, while the routine monitoring sites are often too sparse. Dispersion models simulating the fate of pollution and transport can be useful, but are often infeasible at a high spatial resolution and are extremely dependent on accurate and spatially resolved input data [35,36]. In recent years, LUR models have been proved to be a valid and cost-effective alternative to these conventional approaches [37]. LUR models are statistical regression models based on a Geographical Information System (GIS) platform. They can be used to predict the concentration of atmospheric pollutants at a given site by establishing a statistical relationship between pollutant measurements and potential predictor variables, e.g., land use, traffic, and physical characteristics, etc. [37]. This approach was initially applied to air pollution in the SAVIAH (Small Area Variations In Air quality and Health) study [38]. Since then, it has gained an increasing amount of attention all over the world. This paper therefore aims to employ LUR models and statistical analysis to explore the effect of land use on PM 2.5 pollution in urban areas. Nanchang, the capital city of the Jiangxi province, was selected as a case study. It is a representative city of central China, but has been facing a serious PM 2.5 pollution problem due to ongoing construction and heavy traffic. We applied LUR models to simulate the spatial variations of PM 2.5 concentrations in the Nanchang urban area, analysis of variance and multiple comparisons to study the PM 2.5 concentration variances among different types of urban functional zones, and multiple linear regression to investigate PM 2.5 concentration variances among the same type of urban functional zones. The research results could help correctly understand the PM 2.5 pollution pattern in urban areas. More importantly, they could provide a theoretical basis for urban PM 2.5 pollution control.

Study Area
Nanchang City (28 • 09 N-29 • 11 N, 115 • 27 E-116 • 35 E), the capital of the Jiangxi Province, China, is located in the southwest of Poyang Lake and the middle-and-lower reaches of the Yangtze River. It belongs to a subtropical monsoon climate zone, with an average annual temperature ranging from 17 to 17.7 • C and an annual precipitation value of 1600-1700 mm. Nanchang is an important transportation and shipping center in central China. Many highways and railways traverse this region. The city has experienced a rapid population growth and increase in vehicles in the past decade. By the end of year 2014, the residential population of Nanchang city was 5.24 million and the number of vehicles reached 618,100. All of these factors contribute to the tremendous flow of vehicles per day and the significant amount of pollutants such as PM 2.5 . The study was conducted in the Nanchang urban area that has been defined by the Land Use Planning, which covers an area of 562.46 km 2 . There are nine nation-standard PM 2.5 monitoring sites defined by the China Environmental Monitoring Center (CEMC) reporting monitor data in the city on an hourly basis, and eight of them are located within the study area ( Figure 1). The eight monitoring sites are located in different urban functional zones. S1 and S3 are located in the residential zones, S2 and S7 are in the educational zone, S4 and S6 are in the industrial zones, S8 is in the commercial zone, and S5 is in the control functional zone.

LUR Model Setting
The equation of the LUR models is expressed as follows: where the dependent variable y is the pollutant concentrations, independent variables X1...Xn are the potential variables, β1...βn are the associated coefficients, and ε is the constant intercept.

Dependent Variable and Independent Variables
The monthly mean values of PM2.5 for the eight monitoring sites in 2014 were collected from the Nanchang Environmental Monitor Center (Table 1), and the specified monitoring site locations were also provided by the Monitor Center. The independent variables could be categorized into four classes: meteorological factors, traffic-related factors, land use factors, and population density. Circular buffers were created for 0.3, 0.6, 0.9, 1.2, 2.4, and 4.8 km radii using ArcGIS 10.2 (ESRI, Redlands, CA, USA). In total, 42 variables were used to build the LUR models. Each independent variable was explained as follows. A description of the independent variables is reported in Table 2.
Five meteorological variables were employed to characterize the weather conditions. They were relative humidity, air pressure, water vapor pressure, temperature, and wind speed. The monthly average values of the meteorological variables in 2014 were obtained from the Chinese Meteorological Data Share Service System (http://data.cma.cn/).

LUR Model Setting
The equation of the LUR models is expressed as follows: where the dependent variable y is the pollutant concentrations, independent variables X 1 ...X n are the potential variables, β 1 ...β n are the associated coefficients, and ε is the constant intercept.

Dependent Variable and Independent Variables
The monthly mean values of PM 2.5 for the eight monitoring sites in 2014 were collected from the Nanchang Environmental Monitor Center (Table 1), and the specified monitoring site locations were also provided by the Monitor Center. The independent variables could be categorized into four classes: meteorological factors, traffic-related factors, land use factors, and population density. Circular buffers were created for 0.3, 0.6, 0.9, 1.2, 2.4, and 4.8 km radii using ArcGIS 10.2 (ESRI, Redlands, CA, USA). In total, 42 variables were used to build the LUR models. Each independent variable was explained as follows. A description of the independent variables is reported in Table 2.
Five meteorological variables were employed to characterize the weather conditions. They were relative humidity, air pressure, water vapor pressure, temperature, and wind speed. The monthly average values of the meteorological variables in 2014 were obtained from the Chinese Meteorological Data Share Service System (http://data.cma.cn/).
The traffic-related variables included three subclasses: the intensity of main roads, intensity of secondary roads, and intensity of all roads. The road intensity was used to reflect the traffic conditions due to the unavailability of accurate traffic intensity data. Road intensity was computed by dividing the buffer area by the sum of road segments within the buffer. The data were collected from the transportation map of Nanchang urban master planning from 2011. Three subclasses of variables including the ecological land proportion (green spaces, rivers, and lakes), industrial land proportion, and distance to large ecological space were used to describe the land use situation. The ecological land or industrial land in every buffer zone was calculated to obtain the values of the ecological land proportion or industrial land proportion. The straight-line distance of the monitoring site to the nearest large ecological space (Gan River, Qinshan Lake, Huangjia Lake, Yao Lake, Xiang Lake, Qian Lake, Aixi Lake, Diezi Lake, and Meiling Forest) was measured to describe the distance to a large ecological space. The data were derived from the Nanchang land use map of 2014 and satellite images from 2014.
The residential land proportion was used to describe the population density as the population density was only available at a district level in Nanchang. The data were derived from the Nanchang land use map of 2014.

Model Development and Evaluation
In our study, twelve months were divided into: spring (March to May), summer (June to August), autumn (September to November), and winter (December to February). The LUR models of four seasons were developed, respectively, with SPSS Statistics 19.0 (IBM Corp., Armonk, NY, USA). The 24 samples of every season were randomly divided into two groups: a training data set and a test data set. A total of 75% of samples were used to develop the model and the remaining 25% were used for the model evaluation. The backward model-building algorithm proposed by Henderson et al. (2007) was introduced [35]. The steps were as follows: (1) correlation between PM 2.5 and each independent variable was calculated through an individual univariate regression model; (2) variables that had a counter-intuitive correlation with PM 2.5 were eliminated (e.g., traffic-related variables had negative coefficients and the ecological land proportion had a positive coefficient); (3) the highest-ranking variable in each subclass was identified and other subclass variables with a correlation of more than 0.6 with the highest-ranking variable were eliminated; (4) all remaining variables were entered into a stepwise linear regression; (5) the variables that had insignificant t-statistics (0.1) were removed (the t-statistics were lowered from 0.05 to 0.1 to control the meteorological variables); and (6) steps 4 and 5 were repeated until convergence was attained and variables that contributed less than 1% to the R 2 value of the final model were removed. The entire procedure was repeated three times for every season, and thus, three LUR models were developed for every season and the best fitting one was used as the final LUR model. In this way, the a priori division of samples could be avoided. The final LUR models were evaluated by comparing predicted PM 2.5 concentrations with measured PM 2.5 concentrations from the test data set.

Selection of Urban Functional Zones
Five types of urban functional zones, including commercial, industrial, residential, educational, and control functional zones, were selected in the study area based on the Nanchang urban cadastral survey map and the Nanchang urban master planning map. When choosing urban functional zones, two rules were followed: (1) maintaining integral land parcels; (2) maintaining the evident land use.
In particular, the residential land accounted for more than 50% of the total residential functional zone area; the commercial land accounted for more than 60% of the total commercial functional zone area; and the industrial land (land for high-tech industry and storage included) accounted for more than 40% of the total industrial functional zone area. The land used for universities was chosen as an educational functional zone and one university was usually contained in an educational functional zone. Control functional zones included land use types, e.g., forest, water body, and farmland, and the area of these land use types accounted for more than 80% of the total control functional zone area.

Statistical Analysis
Once the PM 2.5 concentrations in the urban functional zones had been estimated, the analysis of variance and multiple comparisons test were carried out under the assumption of equal variances (homoscedasticity) and normal distribution. The statistical analysis was accomplished using SPSS Statistics 19.0. The analysis of variance can be used to test the null hypothesis H 0 , in which the PM 2.5 concentrations in all functional zones have the same mean values, against the alternative hypothesis H 1 , where the mean values µ i of k groups are not the same. This can be written formally as follows [39].
The F-ratio and probability value (p-value) were obtained through a one-factor analysis of variance command. If F > F (α, k − 1, N − k), then H 1 can be accepted. Additionally, a multiple comparison test is necessary to determine which group pairs' mean values are significantly different. The least significant difference (LSD) test at a 0.05 level of probability was used to perform multiple comparisons. Using this method, the pairs of functional zones for which the PM 2.5 concentrations are significantly different from each other can be identified.

LUR Models
The final LUR models are reported in Table 3. Four variables were entered into the final LUR models after normalization, including meteorological factors, traffic-related factors, and land use factors. The variable of relative humidity was entered into the LUR models of spring, summer, and autumn (p < 0.01), and the variable of temperature was entered into the LUR models of spring and winter (p < 0.01). The intensity of the main roads within a 300 m buffer was found to be the dominant variable affecting PM 2.5 pollution, because it was the only variable that entered all of the LUR models (p < 0.01). Land use factors including industrial land proportion and ecological land proportion also greatly impacted the PM 2.5 concentrations, since they were entered into three LUR models (p < 0.1). The final models explained 76.4%, 89.9%, 94.1%, and 96.1% of the spatial variability of quarterly PM 2.5 concentrations, respectively. To evaluate the performance of the final LUR models, the equations were applied to the test data set and the R 2 value between the predicted and measured PM 2.5 concentrations was calculated. The R 2 value was 0.764. In addition, predicted data were plotted against measured data for validation ( Figure 2). The figure shows that the predicted PM 2.5 concentrations were well correlated with the measured concentrations. To evaluate the performance of the final LUR models, the equations were applied to the test data set and the R 2 value between the predicted and measured PM2.5 concentrations was calculated. The R 2 value was 0.764. In addition, predicted data were plotted against measured data for validation ( Figure 2). The figure shows that the predicted PM2.5 concentrations were well correlated with the measured concentrations. Grids with a dimension of 1 km  1 km were created in the whole study area and the seasonal PM2.5 concentrations were calculated at each intersection using the final LUR models. We assumed there was no trend in the data and a spatially homogenous variation, and the seasonal spatial distributions of PM2.5 were then interpolated using the Ordinary Kriging approach. As shown in Figure 3, PM2.5 concentrations demonstrated a discernible spatial variation. High concentration areas occurred in the centre of the study area, while low concentration areas were mainly distributed on city borders. The northwest and southwest were low concentration areas throughout the year. The figure also discloses that the PM2.5 concentrations of most of the Nanchang urban area met the legislated 24-h average value, but exceeded the annual mean value, which are 75 μg/m 3 and 35 μg/m 3 in China, respectively. Grids with a dimension of 1 km × 1 km were created in the whole study area and the seasonal PM 2.5 concentrations were calculated at each intersection using the final LUR models. We assumed there was no trend in the data and a spatially homogenous variation, and the seasonal spatial distributions of PM 2.5 were then interpolated using the Ordinary Kriging approach. As shown in Figure 3, PM 2.5 concentrations demonstrated a discernible spatial variation. High concentration areas occurred in the centre of the study area, while low concentration areas were mainly distributed on city borders. The northwest and southwest were low concentration areas throughout the year. The figure also discloses that the PM 2.5 concentrations of most of the Nanchang urban area met the legislated 24-h average value, but exceeded the annual mean value, which are 75 µg/m 3 and 35 µg/m 3 in China, respectively.

Statistic Analysis of PM2.5 ConcentrationVariances among Different Types of Urban Functional Zones
Five types and a total of 25 urban functional zones were selected in the study area to analyze the PM2.5 concentration variances among different types of urban functional zones, as shown in Figure 4. The PM2.5 concentrations in the four seasons of these functional zones are shown in Table 4.

Statistic Analysis of PM 2.5 ConcentrationVariances among Different Types of Urban Functional Zones
Five types and a total of 25 urban functional zones were selected in the study area to analyze the PM 2.5 concentration variances among different types of urban functional zones, as shown in Figure 4. The PM 2.5 concentrations in the four seasons of these functional zones are shown in Table 4.

Statistic Analysis of PM2.5 ConcentrationVariances among Different Types of Urban Functional Zones
Five types and a total of 25 urban functional zones were selected in the study area to analyze the PM2.5 concentration variances among different types of urban functional zones, as shown in Figure 4. The PM2.5 concentrations in the four seasons of these functional zones are shown in Table 4.   Analyses of the PM 2.5 concentration variances in the urban functional zones were conducted after a normal distribution test and variance homogeneity test. Table 5 shows the one-factor variance analysis results. In spring, the F-ratio of 18.062 (p < 0.01) indicates that the PM 2.5 concentration variances among different types of urban functional zones were significant. We can also conclude the same rule in summer, autumn, and winter.  Table 4 also expresses the multiple comparison results. In the four seasons, the multiple comparison results among different types of urban functional zones were the same. The results show that the PM 2.5 concentration variances between the control and other four types of urban functional zones were significant. The PM 2.5 concentration variances between the industrial or commercial functional zones and the residential or educational functional zones were also significant. However, there were no statistically significant PM 2.5 concentration variances between the industrial and commercial functional zones, and the educational and residential functional zones.

Statistic Analysis of PM 2.5 Concentration Variances among the Same Type of Urban Functional Zones
Since residential land occupies the highest proportion of the urban area, the residential zone was selected as the typical functional zone to analyze PM 2.5 concentration variances among the same type of urban functional zone. Another 15 residential functional zones were added to the original residential zone sample. Variables of intensity of the main roads, building volume rate, building density, and green coverage rate were used to build the multiple linear regression model for the annual PM 2.5 prediction. As Table 6 shows, the model had a low fitting degree (adjusted R 2 = 0.363). The intensity of the main roads positively correlated with PM 2.5 concentrations and was the primary influencing variable in PM 2.5 prediction (p < 0.01). The building volume rate was positively correlated with PM 2.5 concentrations (p > 0.1) and the green coverage rate was negatively correlated with PM 2.5 concentrations (p > 0.1). The building density showed a negative correlation with PM 2.5 concentrations, which was counter-intuitive (p > 0.1).

LUR Models
We developed LUR models incorporating meteorological factors for predicting quarterly PM 2.5 concentrations in the Nanchang urban area, China. The adjusted R 2 values of the seasonal LUR models were 0.764, 0.899, 0.941, 0.961, respectively, explaining the spatial variability of the pollutant concentrations. In previous studies, the adjusted R 2 values of the LUR models ranged from 0.36 to 0.94 for PM 2.5 [40,41]. The good performance of our models may be attributed to the combination of meteorological factors. Few LUR models include meteorological variables, although many studies have demonstrated that meteorology can significantly influence the pollutant concentration [23][24][25][26], possibly due to the lack of enough data or an appropriate methodology. Obtaining the meteorological conditions at each monitoring site is costly and time-consuming. In this study, we presupposed an identical meteorological condition at every site, as the study area was not very large. Different meteorological factors were entered into the LUR models of different seasons, demonstrating that the influence of meteorological factors on PM 2.5 concentrations varied as the seasons changed.
Among all the independent variables, the intensity of main roads within a 300 m buffer was the dominant variable affecting PM 2.5 concentrations, indicating that PM 2.5 concentrations are closely related to traffic conditions. Some studies used vehicle intensity, while other studies used road length or road intensity to represent traffic conditions [26,[42][43][44][45]. Compared to road length or road intensity, vehicle intensity is more representative of vehicle exhaust, but the data are often unavailable for researchers because of the high cost of vehicle monitoring. Studies have also proved that the performance of LUR models developed with road length or road intensity didn't differ from those developed with vehicle intensity [35,46]. Therefore, road intensity was used in our models in the absence of vehicle intensity. The independent variable of industrial land proportion increasing PM 2.5 pollution in other studies was also included in the LUR models [33,42]. The variables of road intensity and industrial land proportion implying sources of PM 2.5 in Nanchang are mainly local transportation and major industries. The independent variable of ecological land proportion decreasing PM 2.5 pollution in other Chinese cities was also included in the LUR models [33,47], suggesting that the function of natural spaces in removing pollutants is evident. It should be noticed that the independent variable of population density increasing PM 2.5 concentration in other Chinese cities was not included in our models [33,47]. The reason for this is that the spatial resolution of the variable was not good enough in our study.
The number of monitoring sites might be an important factor influencing the accuracy of LUR models. However, at present, there is no rigorous methodology to determine the number of required monitoring sites. The population and size of cities are generally thought to be taken into account when determining the actual number of monitoring sites [40]. In our study, there were eight monitoring sites and the coverage area was 562.46 km 2 , resulting in a monitoring density of one site for every 70 km 2 .
Although it was a small number of monitoring sites, the spatial coverage was comparable to other LUR models reported in the literature [33,35,37,42,46].

Impact of Land Use on PM 2.5 Pollution
The paper studied the impact of land use on PM 2.5 pollution from two aspects of land use type and land use intensity. The impact of land use type on PM 2.5 pollution was investigated by analyzing the PM 2.5 concentration variances among different types of urban functional zones. Through the analysis of PM 2.5 data from different types of functional zones, the same rule in four seasons was found. The highest PM 2.5 concentration was found in industrial and commercial functional zones, while the lowest occurred in control functional zones. The PM 2.5 concentration in residential and educational functional zones was in between these zone types. PM 2.5 pollution in the commercial zone was relatively high in comparison with industrial functional zones, and the residential zone was slightly higher than educational functional zones. The PM 2.5 concentrations in different types of functional zones have also been investigated through a sample survey and a similar pattern has been found [48], which confirms the high simulation accuracy of the final LUR models. Further, our results demonstrate that the PM 2.5 concentration variances among different urban functional zones were statistically significant. The significant PM 2.5 variances suggest that the PM 2.5 pollutants in the Nanchang urban area mainly come from local transportation and major industries, echoing the results demonstrated in the LUR models. We can also conclude that the urban functional zones which are characterized by a dominant land use type had a great impact on PM 2.5 pollution and the impact did not change as the seasons changed.
The impact of land use intensity on PM 2.5 pollution was investigated through predicting annual PM 2.5 concentrations with indexes including the building volume rate, building density, and green coverage rate. The concept of land use intensity is far from an innovative term and first appeared in David Ricardo's Land Rent Theory, which is similar to concepts of smart growth, compact city, Infill Development, and Urban Growth Boundary [49]. In China, land use intensity is considered as the national guideline to alleviate the demand for urban land driven by economic and population growth. However, studies have shown that a higher land use intensity leads to more prominent environmental problems, like noise, dust, and toxic pollutants, because a higher land use intensity increases the concentration of the urban activities [50]. In our paper, the multiple linear regression results showed insignificant t-statistics and inconsistent coefficients with a priori assumptions, illustrating that the indexes had an insignificant or counter-intuitive impact on PM 2.5 concentrations. This may due to the complex physical-chemical mechanism of PM 2.5 pollution or the improper study spatial scale.
In our paper, urban functional zones were used as the basic research unit to explore the effect of land use on PM 2.5 pollution. Some studies analyzed the effect of land use on PM 2.5 pollution through calculating the correlation between PM 2.5 pollution and land use/land cover types [34,51,52]. Compared to a single land use type, urban functional zones including a variety of land use types, but characterized by a dominant land use type, are more appropriate for urban areas. Scholars also believe that urban functional zones can better reflect the relationship between urban land use and air pollution as its specific social-economic function [30]. The results demonstrate that the urban functional zone was a proper spatial scale to investigate the impact of land use type on PM 2.5 pollution in urban areas.

Limitations
There are some limitations that need to be addressed. The first limitation of this study was the weakness related to applying the LUR model to a large area. As the study area in our paper is not very large, we presupposed identical meteorological conditions at every monitoring site. In large areas, the meteorological variables vary from one monitoring site to another and will show a differential influence at each site. Secondly, only a one-year period was considered in this paper due to the data access limitations. Using data from longer periods can help improve the prediction ability of an LUR model. Lastly, it should be noted that this research has explored the impact of land use on PM 2.5 pollution through analyzing the intra-urban spatial variability of PM 2.5 concentrations. Further research is needed to investigate the detailed mechanisms of how land use influences PM 2.5 concentration.

Conclusions
This paper attempted to use LUR models to simulate the variances of the PM 2.5 level in the Nanchang urban area and statistical analysis to explore the impact of land use on PM 2.5 pollution. The seasonal LUR models showed a good fit and could explain the spatial variability in PM 2.5 concentrations well. PM 2.5 exhibits a large variation in different seasons, with the highest pollution values in winter and the lowest in summer, due to the complicated influence of the meteorological factors of temperature and relative humidity [53,54]. Similar to many other studies, the dominant PM 2.5 impacting variable was the traffic conditions that were characterized by the road intensity in this paper [37,42,[55][56][57]. The analysis of variance and multiple comparison test shows statistically significant variances in PM 2.5 concentrations among different types of urban functional zones throughout the year, demonstrating that the land use types generated a great impact on PM 2.5 concentrations and the impact did not change as the seasons changed. The multiple linear regression results illustrate that the land use intensity indexes including the building volume rate, building density, and green coverage rate exhibited an insignificant or counter-intuitive impact on PM 2.5 concentrations. The study also concludes that the urban functional zone was a proper spatial scale to investigate the impact of land use type on PM 2.5 pollution in urban areas, but might not be a proper spatial scale to explore the impact of land use intensity on PM 2.5 pollution. A reasonable methodology and optimized spatial scale are still yet to be explored to further investigate how land use intensity affects PM 2.5 pollution.