2025 Volume 13 Issue 1 Pages 4-21
Air pollution caused by inhalable particulate matter seriously threatens public health and the sustainable development of cities. Previous studies have shown that features of the built environment are closely related to concentrations of inhalable particulate matter, but studies employing different scales and zoning methods have yielded inconsistent findings, and the impacts of possible modifiable areal unit problems (MAUP) have seldom been considered. Here, we evaluated the influence of various built environment parameters, including land use indicators and building form indicators, on air quality in the Gulou District of Nanjing, China. We used a multiple linear regression approach to analyze high-precision urban PM2.5 and PM10 concentration data collected using hand-held instruments. By controlling scale and zoning methods from the block scale of 500 meters to 3000 meters, we explored scale and zoning effects and their interactions. We found: 1) As scale increases, the number of built environment parameters that has a significant impact on pollutant concentrations decreases, and the degree of model fitting first increases before decreasing. Zoning method has a relatively small impact on the model; 2) The proportion of land area used for green space and water systems, roads, public service facilities, and municipal facilities and average building height and density are six important factors that affect changes in PM2.5 and PM10 concentrations.
Rapid economic development and accelerating urbanization have exacerbated urban environmental problems. Worsening air pollution is an especially large problem, causing a suite of social and economic problems and seriously threatening both public health and the sustainable development of cities. According to the United Nations Environment Programme, air pollution has caused more than 1 million prenatal deaths (Wania, Bruse et al., 2012) and incalculable economic losses. Therefore, improving understanding of the factors that influence air pollution and the mechanisms that drive their effect is of great significance for regional air pollution management, planning, and regulation (Cao, Yang et al., 2024).
PM2.5 and PM10 are the primary air pollutants in most Chinese cities. Existing studies tend to focus on the sources and transmission channels (Bao, Feng et al., 2010; Wei, F., Teng et al., 2001; Xiao, Bi et al., 2012), spatial distribution characteristics (Song, Tang et al., 2002; Wei, Y., Yin et al., 2009; Zhao, Wang et al., 2014), and influencing factors (Aini, Shen et al., 2023; Buseck and Pósfai, 1999; Tsai, Kuo et al., 2007; Zhan, Gao et al., 2020; Zhu, 2019) of PM2.5 and PM10. Various types of urban land use types have significant impacts on air quality, including transportation land use, which forms an important source of PM2.5 and PM10. Urban building form can also have a significant effect on ventilation and pollutant dispersion, which in turn affects air quality (Ewing, Pendall et al., 2003).
However, studies focusing on the effects of land use type and building form on PM have reported inconsistent or even contradictory findings. For example, most research has found that green space and water systems reduce PM concentrations significantly due to their effects on traffic emissions. However, Yang (2019) found that air pollution parameters were significantly and negatively correlated with distance from water areas but significantly and positively correlated with distance from main urban roads. Research regarding the effects of building morphology is also inconclusive, with findings by Wu, N. (2019) suggesting that the effects of building height, floor:area ratio, building density, and other building morphological parameters on air pollutants are variable. Our analysis of these findings suggests that inconsistencies are caused by two sources: 1) the scale of study areas and 2) differences in how research areas are divided.
To investigate how study area scale impacts research on the effects of urban land cover type, we explored the influence of various built environment parameters, including land use and building morphology, on air pollutant levels at the block scale. Additionally, we examine scale and zoning effects between these parameters by comparing differences in study scales and zoning modes. Exiting studies mainly obtain urban air quality data from monitoring stations. Because there are only a small number of these stations, sample data does not capture small-scale differences urban air quality. To avoid this problem, we used portable instruments to collect measurements with sufficient spatial resolution to accurately assess air quality at the block scale.
In Section 2, we review literature regarding the effects of the built environment on air pollutant concentrations and the modifiable areal unit problem (MAUP). Section 3 outlines the data sources and the main methodology employed in this study. Section 4, which is the focus of this study, analyzes empirical findings on the effects of the built environment on PM2.5 and PM10 concentrations at different scales and elucidates how MAUP can influence research outcomes at different scales and under different zoning approaches. In Section 5, conclusions derived from the empirical model are discussed and explained, and the main contributions and implications of this work are summarized. Finally, in Section 6, the main conclusions of the paper are outlined and the shortcomings and future directions of this research are discussed.
Existing research on the relationship between the built environment and air quality focuses on two types of indicators: land use and building form.
Our exploration of land use indicators focuses on the impact of urban functions and land use expansion. The main indicators used in these types of studies are land use ratios and distance from various types of land use features. For the most part, this research has been conducted at the city level. As early as the 1970s, Ewing, Pendall et al. (2003) pointed out that the car-oriented land use development model reduces urban air quality due to the larger traffic volume and associated pollutant emissions and proposed that models emphasizing public transportation should be developed. Johnson, Isakov et al. (2010) found that urban land used for transportation and development in surrounding areas has a significant negative effect on air quality in Portland, Oregon, USA using estimates generated by the land use model (LUR). Based on data from air quality monitoring sites in Jiangsu Province, China, Li, Y, Liu et al. (2016) found that arable land and water reduce PM2.5 concentrations, while construction increases them. Peng (2016) used PM, meteorological, and remote sensing data from Zhengzhou, China, to demonstrate that pollutant concentration is negatively correlated with urban green, agricultural, and non-construction land use but is positively correlated with construction land. Yang (2019) constructed a binary logit model to quantitatively study the influence of the built-up urban environment on air quality in Wuhan City, China. This work found that air pollution parameters are significantly and negatively correlated with distance to water, airport, and industrial land and significantly and positively correlated with distance to road, central city, residential, and public service facility land. Using data from meteorological monitoring stations, Xie (2017) demonstrated that PM2.5 concentration is significantly negatively correlated with the spatial distribution of forest, arable land, greenery, and water bodies and is positively correlated with construction and bare land.
Here, our investigation of building form indicators focuses on the relationship between urban street space, building morphology, block coherence, and air pollutant concentration. Indicators commonly used in this research include average building height, building density, floor:area ratio, urban agglomeration, and dispersion. The first three indicators describe urban building form from vertical, horizontal, and overall perspectives, while urban agglomeration and dispersion comprehensively assess overall urban form in different areas. The building morphology index assesses urban air pollution particle concentrations by estimating the effects of population clustering and urban ventilation. McCarty and Kaza (2015) investigated the relationship between urban form and air quality in the United States and, after controlling for population factors and urbanization levels, concluded that fragmented and disordered urban form promotes air quality. Yuan, Song et al. (2017) used an integrated land use-transport-emissions model to explore the effects of compact urban form on air pollution exposure in Xiamen, China, a high-density city.They found that compact urban form may significantly reduce regional transportation emissions but may also increase population exposure to air pollution and argue that tight growth strategies should be implemented in existing high-density cities. Huang (2019) found that three-dimensional indicators of building height and PM2.5 concentrations are closely related in the Xiasha area of Hagzhou, China. Wu, N. (2019) demonstrated a link between building morphological factors, such as building height, volume ratio, and density, and the factors regulating the spatial distribution of PM2.5 at multiple scales, from 500-1500m, in Wuhan, China. At the 500 m scale, building height was significantly and negatively correlated with PM2.5 concentration. In addition, average height, building bulk density, and maximum height did not have a significant effect on PM2.5 at scales between 500-1500 m, indicating that there are scale differences in the effects of building form factors on PM2.5 concentrations.
Although much work has been done to advance understanding of the effects of land use and building form on air quality, several key details have not been fully elucidated. First, most research has been conducted at the macro-scale, investigating air pollution across entire cities and urban agglomerations. However, recent work has demonstrated that neighborhood-scale factors (i.e., land use and building patterns) also impact pollutant concentrations (Chen, M. and Dai, 2022; Lee, 2021; Yuan, Song et al., 2019). Few microscale studies are available, and the mechanisms driving the effects of interactions between land use and building patterns have not been described. Second, most existing studies treat construction land as a single land use variable, neglecting how differences between various components (e.g., public service facilities, municipal facilities, and storage lots) impact air pollution. Finally, scale differences have resulted in inconsistent findings. However, the influence of scale is rarely considered, and the effects of scale and zoning on pollutant concentrations need to be explored for each built environment parameter.
The scale effect is one type of MAUP, in which different spatial divisions of the same geographic space result in different analytical conclusions, yielding results that do not reflect the real situation of the original spatial information. This problem results in artificial spatial patterns and arises from 1) dividing a continuous geographic area into artificial spatial units and 2) collating and collecting spatial information according to artificially defined spatial units (Wu, J., 2000). Related MAUP problems can be divided into scale and zonal effects. The scale effect refers to the impact of using different scales of spatial division in the same geographic space, and the zoning effect refers to the impact of using different methods of spatial attribute division in the same geographic space (Openshaw and Taylor, 1979).
Though first confined to the field of statistics, MAUP research has expanded to multiple disciplines, such as geography, ecology, and urban planning, where researchers have begun to consider the influence of MAUPs in their approaches to urban development and transportation planning (Budde and Neumann, 2019; Li, P., Zhao et al., 2020). However, MAUP research still focuses primarily on questions within landscape ecology and demography, with few studies investigating their influence on urban planning and the urban built environment, despite growing evidence suggesting that MAUPs contribute to uncertainty in spatial analyses (Liu, 2016).
Here, we explore the effects of various built environment parameters, including land use and building morphology indicators, on air pollutants at the block scale. We also investigate scale and zoning effects in the association between air pollutant levels and the built environment by using different scale and zoning approaches.
This study was conducted in Gulou District, Nanjing, China. This neighborhood is located in the northwest portion of the city on the south bank of the lower reaches of the Yangtze River. It is one of Nanjing’s central urban areas and has an administrative area of 54.18 km2. With a large population and a developed economy, Gulou District is an important political, cultural, and educational center in Jiangsu Province, and is representative of high-density, built-up areas.
The base unit of most multi-scale studies is 500 m. In Chinese cities, 500 m is also a common block unit scale, so we selected 500m*500m as the base raster in addition to five larger scales: 1000m*1000m, 1500m*15000m, 2000m*2000m, 2500m*2500m, 3000m*3000m. We divided each of these six scales using two different raster division methods, allowing us to maximize the difference in raster divisions at the same scale between the two methods. PM2.5 and PM10 concentrations and built environment parameters were first measured in the basic raster. These values were then used for calculations at other scales. The study area and the raster division of each scale are shown in Figure 1.
PM2.5 and PM10 are important components of air pollution and refer to particulate matter measuring 2.5 and 10 microns, respectively. Similar studies use pollution data obtained from national air quality testing stations, but insufficient data was available for our 50 km2 study area, which is monitored by only two stations. To improve accuracy, we manually measured air pollution.
We used a Xinsit HT9600 handheld particulate matter detector to collect PM2.5 and PM10 concentration readings every 50 seconds at a height of 1.2 m. Pollutant concentrations were measured between 8:00 a.m. and 11:00 a.m. on November 29, 2020, in clear weather with no wind. Nine devices were used to collect readings in 500m grid cells at the same time, with each investigator recording air pollutant concentrations twice in each of 30 points. Pollutant concentrations were calculated as the average of measurements collected in each grid cell. Our measurements of maximum, minimum, and average air pollution were similar to those reported by national meteorological monitoring stations (Table 1); our pollutant measurements were within 20% of official measurements, corroborating the robustness and credibility of our approach.
| Collection Method | Max | Min | Mean | |
|---|---|---|---|---|
| Sampling instrument | PM2.5 | 87.53 | 43.26 | 65.51 |
| PM10 | 105.01 | 48.20 | 75.31 | |
| Meteorological monitoring station | PM2.5 | 63 | 49 | 55.3 |
| PM10 | 102 | 57 | 82.4 | |
Because the data we collected changes abruptly at the edge of the raster and is not consistent with a continuous distribution of air pollutants, it was necessary to interpolate measurements from each sampling point to generate continuous maps of PM2.5 and PM10 concentrations. Previous research demonstrates that Kriging interpolation optimizes accuracy (Ding, H., Yu et al., 2016), so we employed this method to obtain PM2.5 and PM10 concentrations for each sampling point using ArcGIS. We then calculated concentration data for the base raster by averaging PM2.5 and PM10 concentrations within each 500 m * 500 m raster.
Built environment dataResearch investigating the impact, of the built environment on air quality mainly uses two types of variables: land use and building form indicators. Here, we selected seven categories of land use indicators and four categories of building form indicators.
Land use indicatorsTo calculate the area of road, residential, public service facility, municipal facility, commercial service, and storage land in each grid cell, we used land use data collected from the main urban area of Nanjing in 2009 that had been corroborated using current remote sensing images. Because green space and water system land use data does not include green or blue space in other land use categories (e.g., central green space in residential land, green rooves, public green spaces within land used for roads), we used the GIS maximum likelihood method to decode land use from remote sensing images. The area belonging to each of the seven land use types in each raster cell was divided by the cell’s area to obtain the percentage belonging to each land use type (Table 2).
| Indicator Name | Calculation method | Indicator Meaning | |
|---|---|---|---|
| Percentage of green space and water system |
|
|
|
| Percentage of traffic land | |||
| Percentage of land for public service facilities | |||
| Percentage of land for municipal facilities | |||
| Percentage of residential land use | |||
| Percentage of land for commercial service facilities | |||
| Percentage of storage land |
We obtained the outlines and heights of buildings in the main urban area of Nanjing using the TianDiTu data platform ( https://www.tianditu.gov.cn/) and compiled indicator calculation methods to estimate building density (Wang, W., Li et al., 2019), average building height (Ding, W., Hu et al., 2012), floor:area ratio, and building dispersion (Wang, P., Jiang et al., 2019) within each of the raster cells (Table 3).
| Indicator Name | Calculation method | Indicator Meaning | |
|---|---|---|---|
| Building density |
|
|
|
| Average building height |
|
|
|
| Floor:Area Ratio |
|
|
|
| Building dispersion |
|
|
To investigate the relationship between built environment parameters and pollutant concentrations, we employed multiple linear regression analysis, using PM2.5 and PM10 concentrations as dependent variables and built environment parameters as independent variables. Statistical analysis was performed using SPSS23 software.
Model variable screeningThe number of independent variables examined here was large, so we adopted a "single factor followed by multi-factor" approach to regression analysis. Bivariate correlation analysis was first performed at each scale for PM2.5, PM10 concentrations, after which built environment parameters and Pearson’s correlation coefficient were used to assess correlation. Independent variables that were significantly correlated with dependent variables were screened out (confidence level = 95%).
Multiple linear regressionWe used multiple linear regression to conduct correlation analyses, consistent with approaches used in previous studies (Pan, Zheng et al., 2020). Variables were selected using stepwise regression, and those with significance levels > 0.1 were removed from the model. Inter-variate covariance was determined using the variance inflation factor (VIF), with VIF > 10 indicating severe covariance within a variable (Ziegel, 1992). Appropriate linear regression models were selected for air quality parameters based on the model’s regression statistic (R2, P-value) and whether all linear model assumptions were met.
Concentration data were processed using Kriging interpolation (Li, Yuan, He et al., 2013) to evaluate PM2.5 and PM10 concentrations in the study area (Figure 2), after which measurements were averaged in each grid cell to provide pollutant concentrations in the 500 m base grid (Figure 3). Concentration distributions of PM2.5 and PM10 were highly correlated, with low and high concentrations generally occurring in the north and south, respectively. Using the natural breakpoint method, we divided concentrations into five levels. High concentration areas were mainly located in the southwestern part along the river and the southeastern part of the Xinjiekou Commercial Center. Areas with low concentrations were mainly located in the northeastern part of the district, near the scenic area of Xuanwu Lake. Medium concentrations were generally distributed across the northwestern part of the study area.
| Air pollutant category | Scale | Sample size | Maximum | Minimum | Average | Standard deviation |
|---|---|---|---|---|---|---|
| PM2.5 | 500m | 274 | 87.53 | 43.26 | 65.51 | 11.44 |
| 1000m | 77 | 87.43 | 45.37 | 65.84 | 11.60 | |
| 1500m | 39 | 84.70 | 45.73 | 65.73 | 11.63 | |
| 2000m | 27 | 87.43 | 46.14 | 65.55 | 12.14 | |
| 2500m | 17 | 84.85 | 45.98 | 65.24 | 11.83 | |
| 3000m | 13 | 84.22 | 48.36 | 65.87 | 11.47 | |
| PM10 | 500m | 274 | 105.01 | 48.20 | 75.31 | 13.96 |
| 1000m | 77 | 103.58 | 51.40 | 75.78 | 13.94 | |
| 1500m | 39 | 100.91 | 52.44 | 75.61 | 13.79 | |
| 2000m | 27 | 103.58 | 52.36 | 75.47 | 14.43 | |
| 2500m | 17 | 100.58 | 53.60 | 74.99 | 14.11 | |
| 3000m | 13 | 100.91 | 55.22 | 76.04 | 13.65 |
Multiple linear regression results of the relationship between built environment parameters and PM2.5 and PM10 concentrations for six scales ranging from 500-3000 m are shown in Table 5 and Table 6.
The two multiple linear regressions demonstrate that raster classification method did not affect the number of variables with significant correlations and that eight built environment parameters were significantly correlated with PM2.5 and PM10 concentrations. Regression coefficients indicate that green space, water system, and public service facility land area was significantly correlated with PM concentrations at the 2500 m scale and that building dispersion was significantly correlated at the 1000 m and 15000 m scales. Variable coefficients fluctuated slightly but the signs did not change.
At different scales, 8 of the 11 built environment parameters were significantly correlated with PM2.5 and PM10 concentrations (p<0.1). The number of building parameters significantly correlated with pollutant concentrations decreased gradually with increasing spatial scale. The 2500 m scale had the highest fitting degree (adjusted R2) for both zoning approaches (Figure 4) where both models have four significant built environment parameters. This indicates that 2500 m is the best scale to use for exploring the influence of built environment parameters on PM2.5 and PM10 concentrations.
Five land use type indicators and three building form indicators were significantly correlated with the dependent variable, with land use indicators more strongly correlated than building form indicators. Indicators can be divided into three categories according to how they affect the dependent variable. First, average height was significantly correlated with the dependent variable at all six scales, and the regression coefficient increased as the scale increased. Second, some built environment parameters were correlated at partial scales. Building density was significantly negatively correlated at scales between 500-2000m, with a large regression coefficient. Green space and water system land use was significantly and negatively correlated at the 500-1500 m scales. Road land use was significantly positively correlated at scales between 500-1500m. Public service facility land use showed a significantly positive correlation at scales between 500-2500 m, and municipal facility land use was significantly positively correlated at all five scales except 1500 m. Finally, the proportion of residential land use and building dispersion changed the direction of correlation at some scales. Residential land use had a weak, negative correlation at the 500 m scale and a significant positive correlation at the larger scales of 2000 and 25000 m. Building dispersion was significantly negatively correlated at the 1500 m scale in the second division, and the rest of the scales were positively correlated. However, both of these correlation coefficients were too small and the significance level too low, and the weak correlation had little effect on the dependent variable.
| Air quality parameter | Regression coefficient | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Building Form Indicator | Land Use Indicator | Constants | Adjusted R2 | ||||||||
| Average building height | Building density | Building dispersion | Percentage of green space and water system | Percentage of road land | Percentage of land for public service facilities | Percentage of land for municipal facilities | Percentage of residential land use | ||||
|
500 m |
PM2.5 | 0.347** | -66.682** | 0.001 | -11.373* | 52.664** | 12.504** | 20.568* | -6.140 | 63.919 | 0.223 |
| PM10 | 0.430** | -65.689** | 0.001 | -12.024* | 67.661** | 13.577** | 23.448* | -7.256 | 73.115 | 0.205 | |
| 1000m | PM2.5 | 1.065** | -115.178** | -18.648* | 84.435** | 27.640** | 30.908 | 60.644 | 0.493 | ||
| PM10 | 1.175** | -147.847** | -22.459* | 113.544** | 33.641** | 71.507 | 0.462 | ||||
| 1500m | PM2.5 | 0.785** | -178.135** | -42.030** | 139.186** | 29.589* | 55.260 | 0.527 | |||
| PM10 | 0.964** | -197.149** | -40.045** | 163.625** | 35.833* | 79.025 | 0.469 | ||||
| 2000m | PM2.5 | 2.116** | -95.083** | 48.379** | 145.109* | 44.586 | 0.695 | ||||
| PM10 | 2.518** | -114.942** | 59.378** | 144.261* | 50.802 | 0.673 | |||||
| 2500m | PM2.5 | 1.992** | 99.016* | 42.049** | 50.353 | 0.853 | |||||
| PM10 | 2.268** | -71.163 | 116.090* | 49.558** | 56.679 | 0.819 | |||||
| 3000m | PM2.5 | 2.249** | 136.787 | 35.896 | 0.720 | ||||||
| PM10 | 2.697** | 41.618 | 0.654 | ||||||||
Note: Correlation coefficients are shown only for p<0.1
*. Significant at the 0.05 level
**. Significant at the 0.01 level
Table 6. Multiple linear regression models of the effects of built environment parameters on PM2.5 and PM10 concentrations at different scales (Zonal approach Ⅱ)
| Air quality parameter | Regression coefficient | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Building Form Indicator | Land Use Indicator | Constants | Adjusted R2 | ||||||||
| Average building height | Building density | Building dispersion | Percentage of green space and water system | Percentage of road land | Percentage of land for public service facilities | Percentage of land for municipal facilities | Percentage of residential land use | ||||
|
500 m |
PM2.5 | 0.347** | -66.682** | 0.001 | -11.373* | 52.664** | 12.504** | 20.568* | -6.140 | 63.919 | 0.223 |
| PM10 | 0.430** | -65.689** | 0.001 | -12.024* | 67.661** | 13.577** | 23.448* | -7.256* | 73.115 | 0.205 | |
| 1000m | PM2.5 | 0.426** | -135.993** | -29.587** | 119.739** | 13.838 | 72.497 | 0.424 | |||
| PM10 | 0.525** | -146.283** | 0.003 | -27.735** | 143.266** | 18.979* | 78.916 | 0.408 | |||
| 1500m | PM2.5 | 1.557** | -188.125** | -0.002* | -28.436** | 107.823** | 37.601** | 63.974 | 0.752 | ||
| PM10 | 1.806** | -227.369** | -0.002 | -30.891** | 129.008** | 48.901** | 73.362 | 0.720 | |||
| 2000m | PM2.5 | 2.098** | -140.402** | 66.457** | 57.906* | 55.245* | 51.274 | 0.605 | |||
| PM10 | 2.445** | -169.851** | 79.753** | 63.154* | 49.456** | 58.823 | 0.594 | ||||
| 2500m | PM2.5 | 2.548** | -26.573* | 56.086* | 56.125** | 57.262 | 0.835 | ||||
| PM10 | 1.977** | -43.237** | 71.041** | 47.442** | 83.065 | 0.893 | |||||
| 3000m | PM2.5 | 1.751** | 142.881* | 42.474 | 0.5656 | ||||||
| PM10 | 1.984** | 138.978 | 50.013 | 0.484 | |||||||
Note: Correlation coefficients are shown only for p<0.1
*. Significant at the 0.05 level
**. Significant at the 0.01 level
Our findings demonstrate that the two types of raster division have only a small effect on the final results, indicating that the zoning effect is weak. They reveal that the built environment factors examined here do influence pollutant concentrations significantly, and that their influence is relatively stable and does not change with the analytic unit division.
On the other hand, the number of built environment factors that have significant effects on PM2.5 and PM10 concentrations decreased as scale increased, with only two factors having a significant effect at the 3000 m scale. This may be because, as the study scale increases, built environment parameters become less differentiated between groups, with fewer significant differences between them. Thus, fewer built environment parameters are significantly correlated with PM2.5 and PM10 concentrations. Some built environment parameters gradually became insignificant (p > 0.1) with increasing scale, which may also have resulted from decreasing differentiation between groups as study scale increased. The 2500 m scale may represent the threshold at which the effects of differences in the urban built environment become indistinguishable from one another.
Land use indicatorsBecause of anthropogenic disturbance and land use heterogeneity, land use categories are inevitably correlated with air pollution. Our work shows that the proportion of land allocated for green spaces and water systems, roads, public service facilities, municipal facilities, and residences is an important factor affecting changes in PM2.5 and PM10 concentrations.
Consistent with the results of other research, green space and water systems play a crucial role in reducing PM2.5 and PM10 concentrations. Our analysis across six scales ranging from 500 to 3000 m reveals that the effect of this land use type gradually increases with increasing spatial scale. This suggests that the more connected and aggregated green space and water systems are at the macroscopic scale, the greater the absorption and retention of PM2.5 and PM10. This reflects the influence of scale in analyses of urban ecosystem service provision. Road land use percentage was significantly and positively correlated with PM2.5 and PM10 concentrations. As road area increased, so did PM2.5 and PM10 concentrations, reflecting the contributions of urban motor vehicle traffic emissions to air pollutant concentrations. In addition, the percentages of land used for public services and municipal facilities was significantly and positively correlated with PM2.5 and PM10 concentrations, but the drivers of this relationship are unclear. However, field surveys demonstrate that, in contrast to public service and municipal facility land, residential land in the study area is generally scattered and has a scale smaller than 500 m. Residential space is more homogenous at smaller scales, while PM2.5 and PM10 concentrations are more homogenous at larger scales. In contrast, the distribution of residential space is more homogeneous at smaller scales, with higher variability between samples at larger scales. This results in a weak positive correlation with PM2.5 and PM10 concentrations, followed by a significant negative correlation.
Building form indicatorsWe found that building height, density, and dispersion significantly influenced PM2.5 and PM10 concentrations. The positive correlation between average building height and pollutant concentrations may be because increasing building height and decreasing air mobility can promote the accumulation of particulate matter, thereby decreasing air quality. At the same time, the correlation coefficient for building height increased with increasing scale. This may be due to the fact that increasing building height does not interfere with air circulation at small scales, whereas an increase in average building height indicates an overall increase in the height of most building at larger scales. This often strongly influences air circulation, thereby increasing the concentration of air pollutants and increasing PM2.5 and PM10 concentrations significantly.
In addition, PM2.5 and PM10 concentrations were negatively correlated with building density, suggesting that concentrated and compact development can improve air quality. Chen, L. and Yi (2016) reported similar findings from a statistical analysis of compactness index values and air quality indicators. This may be because compact urban environments are more conducive to “green” modes of transportation, such as walking, bicycling, and public transportation. The availability of public transportation may also reduce the proportion of trips made by small cars, thereby reducing air pollution from traffic emissions. Finally, building dispersion was significantly correlated with pollutant concentrations, but the correlation coefficients were small, likely because building dispersion was not significantly variable at the 500-3000 m scale and thus had a relatively minor effect on the results.
Here, we evaluated the association between PM2.5 and PM10 concentrations and built environment parameters at multiple spatial scales using field measurements collected in the Gulou District in Nanjing, China. We found:
(1) There is a clear scale effect on the impacts of the built environment on PM2.5 and PM10 concentrations. First, as scale increased, the direction of correlation changed for a few parameters, such as residential land use and building dispersion in zonal approach II (Table 6). Second, as scale increased, the number of built environment parameters with a significant effect decreased, and model fitting degree first increased before decreasing. Built environment parameters had the greatest explanatory power at the 2500 m scale, suggesting this may be the best spatial scale to study the influence of the built environment on PM2.5 and PM10 concentrations.
(2) The zoning effect is relatively mild. Different zoning methods influence the correlation between built environment parameters and PM2.5 and PM10 concentrations, but only the magnitude of correlation; the number of significantly correlated variables correlation direction do not change across scales.
(3) The influence of land use on PM2.5 and PM10 concentrations is greater than that of building form; but building form significantly influences air quality at all six scales, while the influence of land use is limited to scales between 500-2000 m. The percentage of land used for green space and water systems, roads, public service facilities, and municipal facilities and building height and density are six important factors affecting PM2.5 and PM10 concentrations. Except for building density and percentage of land used for green space and water systems, all six show significant, positive correlations.
This work reveals the influence of scale and zoning on research investigating the effects of the built environment on PM2.5 and PM10 concentrations. It identifies the main variables affecting pollutant concentrations at different scales as well as the optimal scale for evaluating the influence of built environment parameters on PM2.5 and PM10.
The "Healthy China Initiative" has been upgraded to a national strategy, contributing to an increasing academic focus on the relationship between the built environment and public health in recent years. However, the results of this research are often inconsistent or even contradictory. Though there are many explanations for this, out work suggests that one cause may be the scale effect of spatial analysis units. Thus, it may be necessary to conduct multi-scale comparisons when analyzing the effects of different built environment elements on population health to improve planning and design strategies for environmental interventions.
Finally, at the practical application level, our findings can be used to improve planning design indexes, to provide a reference for the best scale to study the built environment and air quality, and to promote differentiated regulation of planning at different scales.
We also acknowledge this work’s deficiencies in indicator selection and data collection. First, our selection of indicators was limited to those related to land use and building form. These have been explored in previous studies, but there are many other indicators that affect air pollutant concentrations, and we could not include all of them due to model limitations. Additional research is needed to explore the effects of other relevant indicators on air pollutant concentrations more comprehensively. Due to limited personnel and monitoring equipment, we only explored the influence of the built environment on air quality at different spatial scales while ignoring temporal variation in PM2.5 and PM10 concentrations. In addition, we did not consider the influence of external sources of air pollution and important natural ecological elements within the study area on air quality, which may have influenced our interpretation of the model. Future studies can improve the accuracy and reliability of their results by considering how PM2.5 and PM10 concentrations vary across time and in response to the surrounding environment.
Conceptualization, K.S. and Z.H.; methodology, K.S., Y.S., L.W., and Y.S.; software, K.S.; investigation, K.S., Y.S., L.W., and Y.S.; resources, K.S. and Z.H.; data curation, K.S.; writing—original draft preparation, K.S., Y.S., L.W., and Y.S.; writing—review and editing, K.S. and Z.H.; supervision, Z.H. All authors have read and agreed to the published version of the manuscript.
The authors declare that they have no conflicts of interest regarding the publication of the paper.
