Introduction

Long-term rapid economic development and urban expansion in China have caused severe environmental degradation, particularly urban air quality. Air pollution annually causes approximately 1.6 million deaths in China, which accounts for 17% of total deaths in the country (Rohde and Muller, 2015; Yue et al. 2020). Fine particulate matter (PM2.5) pollution is one of the worst types of air pollution (Zhang et al. 2020). An excessive PM2.5 concentration has always been a challenging issue in the governance of urban air pollution (Liu et al., 2024). Under the guidance of macro policies such as the “Air Pollution Prevention and Control Action Plan (APPCAP)” and the “Three-year Action Plan to Fight Air Pollution”, China has achieved remarkable PM2.5 pollution control results, which are specifically reflected in the steady decline in the national annual average PM2.5 concentration. However, in regions such as Beijing–Tianjin–Hebei and the Fenhe-Wei River Plain, there has been a bottleneck in the reduction of heavy PM2.5 pollution. Since 2013, PM2.5 pollution has consistently been the primary pollutant contributing to air quality exceedances, and it is also the single air pollutant with the highest proportion in cities. Although the formation of PM2.5 pollution often involves meteorological conditions, in the long run, the generation of PM2.5 originates from structural and institutional factors such as the local industrial layout, energy structure, population, and economic agglomeration (Zhang et al. 2022). The driving factors stemming from socioeconomic development are fundamental issues that policymakers need to pay attention to.

Given the above, researchers have conducted extensive analyses of spatiotemporal evolution characteristics and comparative studies and examined types of variations in PM2.5 from an empirical perspective (Ji et al. 2018; Zhao et al. 2019; Geng et al. 2021). These research efforts have explored the spatiotemporal evolution of PM2.5 in various cities and regions across the Americas, Europe, Africa, Asia, and even globally (Thongthammachart et al. 2021; Liu et al. 2023). Global analyses have revealed that over the past two decades, PM2.5 concentrations have increased in developing countries, whereas a contrasting trend has been observed in developed nations (Lu et al. 2021a). In particular, areas with high PM2.5 pollution are predominantly concentrated in India and China (Lim et al. 2020). Specifically, within China, PM2.5 reveals significant spatial heterogeneity and aggregation (Bai et al. 2023). Urban dispersion decreases from southeast to northwest (Liu et al. 2021), whereas cities north of the Yangtze River and east of the Hu Line experience a high degree of breakdown (Yan et al. 2021). Simultaneously, the emergence of new urbanization theories, including “smart growth” and “new urbanism,” has led to the utilization of socioeconomic factors such as population density, transportation, and the economy when analyzing the impact of urban development on atmospheric pollutants (Gren et al. 2019; Bibri, 2020). These factors serve as crucial indicators for quantifying urban growth and development. As a result, urban planners must determine whether urban compactness variables can improve environmental quality and increase resource utilization efficiency (Chen et al. 2023). A compact city is a spatial planning paradigm that effectively drives sustainable urban development (Dantzing and Saaty, 1973). Despite the absence of a unified definition in academic circles, a compact city is generally recognized as embodying characteristics such as high density, intense development, mixed land use, and efficient transportation (Harari, 2020). Furthermore, this type of plan has been established as a resource-efficient model for large-scale urban development that is capable of mitigating environmental pollution and addressing energy scarcity (Yao et al. 2022).

In the context of the United Nations Sustainable Development Goals (UN SDGs), the intricate interplay between urban spatial compactness and air pollution has emerged as a pivotal research avenue (Kutty et al. 2020). Current research shows that while the high-density development paradigm of compact cities effectively mitigates urban sprawl and enhances operational efficiency, it also poses challenges such as traffic congestion and air pollution (Yin and Yao, 2024). To gain a more comprehensive understanding of the nexus between compact cities and air quality, scholars have gradually shifted from relying solely on a singular metric to assess compactness to embracing a comprehensive, multifaceted measurement framework (Angel et al. 2020). Initially, research primarily zeroed in on a singular indicator, focusing either on urban morphology or functionality. However, as the understanding of a compact city has deepened, multiple dimensions, such as population density, land utilization, economic activity, and urban function, have been progressively integrated into the compactness evaluation system (Yu et al. 2023; Xing et al. 2024). In terms of methodological approaches, regression analysis, mediation effect models, the geographically weighted regression (GWR) model, and the stochastic impacts by regression on population, affluence, and technology (STIRPAT) model have been extensively utilized to delve into the intricate relationships between compact city development and environmental indicators, including PM2.5 concentrations, carbon emission scales, and energy efficiency (Hu et al. 2022; Van der Borght and Barbera, 2023). Although scholars have investigated the relationship between urban compactness and air pollution, there remains ample scope for further research. First, the indicators chosen are solely based on spatial morphological compactness, but functional compactness is insufficiently considered. In addition, most researchers have not explored the spatial heterogeneity effect of urban compactness factors on air pollution. Finally, existing multi-indicator calculation methods cannot depict the relationship between each factor and PM2.5, and these relationships should be investigated from a multidimensional perspective.

China has made substantial progress in controlling PM2.5 pollution over the past decade. As the economic development model has shifted from rapid growth to high-quality development, there is an urgent need to update the thinking and policy framework for urban air pollution control. Most scholars have explored the causes of PM2.5 pollution from the perspectives of natural mechanisms, meteorological conditions, and environmental management, but there is a lack of empirical research exploring urban PM2.5 pollution from the perspective of urban space or functional morphology. The marginal contributions of this study are primarily twofold. First, this study constructs a theoretical analysis framework for the effect of urban compactness on environmental quality and elaborates on how population compactness, economic compactness, land-use compactness, and transportation compactness affect environmental quality. On this basis, four hypotheses are proposed, which enrich the theoretical application of urban economics and environmental economics. Second, the geographical and temporal weighted regression (GTWR) model was employed to reveal the heterogeneity effects of population compactness, economic compactness, land-use compactness, and transportation compactness on PM2.5 pollution in 285 prefecture-level cities in China, expanding the research on factors that influence PM2.5 pollution. Finally, the empirical results illuminate the influencing mechanism of urban compactness on air pollution, which can offer a new perspective for meeting the needs of urban air pollution prevention and control in the new era and provide tailored decision-making bases for cities with different functional orientations.

The research roadmap of this study is shown in Fig. 1. The first part outlines the background and reviews the relevant literature. Section “Theoretical analysis framework” constructs a theoretical framework and proposes the main hypotheses. The data sources and methods are introduced in the section “Methods”. Section “Results and discussion” presents the results and discussion. Section “Conclusions and policy implications” summarizes the major conclusions and provides policy implications, and the future research plan is presented in the section “Future research directions”.

Fig. 1
figure 1

Research roadmap.

Full size image

Theoretical analysis framework

Grounding from the perspective of population compactness

Population compactness manifests spatially through the concentration of residents, and this results in an increase in population density. Notably, this spatial distribution of residents often yields reduced levels of carbon emissions (Wang and Li, 2021). Given that pollutants stemming from the combustion of fossil fuels used in residents’ daily lives constitute a primary source of haze pollution, the development of urban spaces characterized by population compactness facilitates a centralized energy supply. This, in turn, enables economies of scale in residents’ energy consumption and consequently mitigates environmental pollution (Wang, 2022). Vande and Kennedy (2007) reported that high-density development promotes a more compacted residential and occupational lifestyle, resulting in adverse effects on urban energy consumption and carbon emissions. Furthermore, the pollution mitigation effect of the compact urban development model varies greatly depending on the scale of the city. Comparatively speaking, the overconcentration of population in first-tier cities has led to a crowding effect, diluting the scale effect of centralized pollution treatment. On the basis of the aforementioned discussion, Hypothesis 1 is proposed as follows.

Hypothesis 1: Higher population compactness contributes to improved urban environmental quality.

Grounding from the perspective of economic compactness

Economic compactness is manifested primarily as the spatial agglomeration of economic activity, and this agglomeration of factors is conducive to improving environmental quality. Some scholars believe that cities with high economic agglomeration are more energy efficient and environmentally friendly than rural areas with low economic agglomeration (Wang et al. 2022). Furthermore, under certain conditions, economic agglomeration can exert a “home market effect,” which can alleviate the “pollution haven” phenomenon (Zeng and Zhao, 2009).

In terms of an influence mechanism, industrial agglomeration exerts a positive influence on environmental governance through three externalities: the common labor market, intermediate inputs, and knowledge spillover (Wang and Han, 2023). Enterprises can precisely match skilled laborers in the labor pool according to changes in product market demand. To meet environmental regulatory requirements, enterprises can acquire laborers with higher pollution treatment capabilities. From the perspective of intermediate inputs, the production activities of enterprises in agglomeration areas generate similar pollutants, and industrial agglomeration helps to achieve scale effects in pollution control and reduce control costs (Hong et al. 2020). Many enterprises with “forward linkages” and “backward linkages” can achieve pollutant reuse by constructing a circular economy development model, thereby achieving energy conservation and emission reduction (Wang et al. 2021). From the perspective of knowledge spillover, industrial agglomeration serves as a platform for various enterprises to engage in technological exchange and information sharing. Furthermore, the exchange of energy-saving and emission-reduction experiences among enterprises is facilitated, accelerating technological innovation in pollution control. Therefore, we propose the following hypothesis.

Hypothesis 2: Economic compactness is positively correlated with urban environmental quality.

Grounding from the perspective of compact land use

Against the backdrop of rapid urbanization, city boundaries are constantly being broken and expanding toward suburbs, which leads to the phenomenon of urban sprawl and impacts environmental quality (Chettry, 2023). First, urban sprawl can lead to spatial decentralization of the population and job‒housing separation, which makes residents increasingly dependent on motor vehicles for travel and results in increased energy consumption and atmospheric pollution emissions in the region (Lu et al. 2021b). Second, during the urban sprawl process, large swaths of agricultural land are converted into urban development areas, leading to a continuous decline in the proportion of agricultural land. This reduction in urban green spaces weakens the purification capacity of the environment and exacerbates haze pollution (Arshad et al. 2022). Third, pollution emissions from the secondary and tertiary sectors are significantly higher than those from agriculture (Lyu et al. 2016). Urban sprawl reduces the agricultural share in the industrial structure, thereby leading to a decline in urban environmental quality. Additionally, urban sprawl reduces population density and economic density, increasing the demand for buildings. This, in turn, leads to the emission of dust pollutants during building construction, which increases haze pollution. Therefore, this research proposes that a compact land-use pattern can effectively mitigate the impacts on environmental quality caused by urban sprawl, such as vehicle emissions, building pollution, and greenery destruction. Hypothesis 3 is proposed as follows.

Hypothesis 3: Increasing land use compactness contributes to improved urban environmental quality.

Grounding from the perspective of compact transportation

Compact transportation is characterized by diverse transportation infrastructures and strong accessibility, which has a significant impact on three main aspects of environmental pollution. First, transportation infrastructure plays a pivotal role in promoting local economic development. This, in turn, increases the demand for environmental quality among the populace, compelling governments to strengthen environmental regulations. Consequently, a ‘cleansing effect’ has been observed (Yang et al. 2019). Notably, upgrades in transportation infrastructure reduce the cost of factor mobility and significantly restrict the entry of polluting enterprises into developed regions. Second, traffic congestion increases the operating time of motor vehicles, leading to increased fuel consumption and pollutant emissions (Wang et al. 2023a). Improvements in public transportation density, such as rail transit, encourage travelers to shift away from private cars and nonpublic transportation modes, resulting in a ‘transportation shift effect’ or ‘Mohring effect’ (Fielbaum et al. 2020). This shift leads to reduced tailpipe emissions from road traffic vehicles due to decreased usage, ultimately lowering urban air pollution emissions. Finally, compact transportation reduces pollution emissions through economies of scale, particularly in the case of public transportation. While road traffic often experiences congestion due to excessive demand, which leads to air pollution, public transportation benefits from economies of scale in terms of passenger capacity. This reduction in traffic congestion offers distinct advantages in improving urban air quality (Singh et al. 2023). In summary, we propose Hypothesis 4 as follows.

Hypothesis 4: Higher levels of transportation compactness contribute to improved urban environmental quality.

Urban compactness manifests in four distinct dimensions, and on the basis of each compactness factor, we proposed four research hypotheses. The theoretical research framework is illustrated in Fig. 2.

Fig. 2
figure 2

Theoretical analysis framework.

Full size image

Methods

Data sources

This research utilizes panel data from 285 prefecture-level cities in China spanning from 2010 to 2021. However, owing to a severe lack of data and inconsistencies in statistical standards, Taiwan, Hong Kong, Macao, Xinjiang, and Tibet are excluded from the research scope. The PM2.5 pollution concentration data are sourced from the Atmospheric Composition Analysis Group at Washington University in St. Louis. This dataset, which has been validated for global use due to its comprehensive coverage, extensive duration, and high accuracy, is estimated via information from satellites, simulations, and monitoring sources (Lee, 2020). The remaining variable data related to socioeconomic situation and urban development are sourced from the China City Statistical Yearbook, China Urban Construction Statistical Yearbook.

Urban compactness indicators

The compact city theory underpins urban compactness measurement. Breheny (1997) defines a compact city as one that prioritizes urban density, land use mix, public transportation, and resident quality of life. Based on the preexisting research, this study selects urban compactness indicators from four dimensions: population, economic, land use and transportation compactness. Specifically, indicators of urban compactness factors specifically cover population density, Economic density, Land development equilibrium degree and road network density.

Population compactness is dominated by population density (PD). According to the theory of compact cities, increasing population density in city centers creates agglomeration economy effects, reduces land waste, and improves urban space usage, but it also raises rigid demand for urban housing, infrastructure, and transportation, which may increase urban haze. Economic density (ED) measures economic compactness. High economic density cities will attract a large labor force and firms, affecting energy consumption and industrial structure. It’s subject to regional development levels, industry dispersion, and spatial disparities.

Land use diversity is another hallmark of compact cities. Land development equilibrium degree (LED) reflects land function blending, which is influenced by three factors: more construction land, less forest and arable land, and a higher land usage rate. The equilibrium of the urban land use can be easily defined based on the information entropy formula:

$${{\rm {LED}}}=\frac{H}{{H}_{{\rm {m}}}}=\frac{-\sum _{i}{P}_{i}\log {P}_{i}}{\log N},$$
(1)

where \({{\rm {LED}}}\) denotes land development equilibrium degree, \(H\) denotes the information entropy of the land use, \({H}_{{\rm {m}}}\) denotes the maximum entropy value, \({P}_{i}\) corresponds to the probability of the event and \(N\) denotes the number of urban land use types.

Road network density (RN) measures transportation compactness. Per capita road area represents the city’s transportation intensity. High road density causes traffic congestion, lowers urban transportation efficiency, and may harm the environment. Road network density indicates city accessibility. Higher road network density shortens driving distances and reduces exhaust emissions.

Control variables

Regarding control variables, per capita GDP, population density, and foreign direct investment were considered by Yu et al. (2022). Cheng et al. (2020) demonstrated that foreign direct investment can significantly exacerbate PM2.5 pollution in Chinese cities. In the study by Wang et al. (2023b), the control variables included average electricity consumption by GDP per land, construction land, and total land. On the basis of the above studies and the availability of data at the city level, the urban greening level (GC), energy consumption (EC), industrial structure (SGDP), and openness (FGDP) are considered control variables. The variables and their definitions are displayed in Table 1 and a statistical description of all the variables can be found in Appendix Table A1.

Table 1 Definition of all variables.
Full size table

Spatial correlation test

The spatial autocorrelation measures the degree of correlation between physically adjacent, reflecting the spatial distribution of research items in the study area. The global Moran’s I is used to depict the spatial dependence of variables within the study area as a whole, which is calculated as follows:

$${{\rm {Mora}{n}}}^{{\prime} }{\rm {s}}\,I=\frac{n\mathop{\sum}\nolimits_{i=1}^{n}\mathop{\sum}\nolimits_{j=1}^{n}{w}_{{ij}}\left({x}_{i}-\bar{x}\right)({x}_{j}-\bar{x})}{{\mathop{\sum }\nolimits_{i=1}^{n}\mathop{\sum }\nolimits_{j=1}^{n}{w}_{{ij}}\left({x}_{i}-\bar{x}\right)}^{2}},\forall j\,\ne\, i,$$
(2)

where \({x}_{i}\) present the PM2.5 concentrations in spatial regions i, and \(\bar{x}\) is the average PM2.5 concentration in spatial regions. \({w}_{{ij}}\) is the spatial weight matrix, and n is the sample size. The significance level of Moran’s I is judged by the z-value, which is calculated as follows:

$$Z=\frac{1-E\left(I\right)}{\sqrt{{{\rm {Var}}}\left(I\right)}}.$$
(3)

As Appendix Fig. B1 illustrates, all of the twelve years passed the 1% significance test, showing the geographical distribution of PM2.5 in prefecture-level cities in China has a significant positive association and spatial aggregation.

Local Moran’s I is the specific analysis method for local spatial autocorrelation. The specific formula regarding the local Moran’s I is

$${{\rm {Local}}\; {\rm {Mora}{n}}}^{{\prime} }{\rm {s}}\,I=\frac{n\left({x}_{i}-\bar{x}\right)\mathop{\sum }\nolimits_{j=1}^{m}{w}_{{ij}}({x}_{j}-\bar{x})}{\mathop{\sum }\nolimits_{i=1}^{n}{\left({x}_{i}-\bar{x}\right)}^{2}},\forall j\,\ne \,i,$$
(4)

where m represents the number of cities adjacent to city \(i\). The local indicators of spatial association (LISA) cluster maps depict the spatial agglomeration of Local Moran’s I.

The four clustering states for PM2.5 pollution in Chinese cities, namely, high–high (H–H), low–low (L–L), high–low (H–L), and low–high (L–H), are presented in Fig. 3. From 2010 to 2021, high PM2.5 aggregation zones were located in Beijing–Tianjin–Hebei and north of the Yangtze River. L–L clustering reveals that cities located mainly in Inner Mongolia, northeastern Heilongjiang, southwestern cities, and southeastern coastal cities, are spatially clustered at low PM2.5 values. By utilizing the calculated results of the global Moran’s I index, GeoDa software was employed to generate Moran scatter plots depicting the PM2.5 concentrations across the years 2010, 2013, 2017, and 2021 (see Appendix, Fig. B2). Specifically, the first and third quadrants represent the H–H and L–L clustering types, respectively, whereas the second and fourth quadrants signify the L–H and H–L clustering types, respectively.

Fig. 3: LISA cluster of the PM2.5 concentrations in 2010, 2013, 2017 and 2021.
figure 3

The figure shows the local spatial autocorrelation characteristics of PM2.5 concentrations across 285 cities during four time periods: a 2010, b 2013, c 2017, and d 2021.

Full size image

Through a comparative analysis of the number of points distributed across these quadrants, Liaocheng, Cangzhou, Shijiazhuang, Hengshui, and Puyang clearly stand out as the most influential cities within the first quadrant. These cities not only present high levels of pollution but also contribute significantly to the elevation of PM2.5 concentrations in neighboring cities, which are primarily concentrated in the regions of Shandong, Hebei, and Henan. Consequently, urban planners in these cities should prioritize strategies to mitigate the spatial spillover effects of pollutants. Conversely, Sanya, Hulunbuir, and Lijiang emerge as cities with the strongest L–L aggregation influence, primarily situated in provinces that have superior ecological resources, such as Hainan, Inner Mongolia, and Yunnan. The findings revealed that the PM2.5 concentration had considerable spatial autocorrelation among the 285 cities. For analysis, an acceptable spatial econometric model should be used.

Geographically and temporally weighted regression model

Fixed spatial variable interactions are commonly assumed in spatial econometrics, which violates geospatial heterogeneity and non-stationarity. The geographically weighted regression model (GWR) is a crucial tool for analyzing spatial data spatial variability. GWR, on the other hand, examines cross-sectional data without a time dimension, resulting in a bias towards panel data analysis. The geographically and temporally weighted regression model (GTWR) incorporates the time dimension and may explain the non-stationary spatiotemporal fluctuation of regression coefficients, thus overcoming the GWR model’s shortcomings (Cao et al. 2023). In this study, we first constructed a fundamental model:

$$\begin{array}{ll}{\mathrm{ln}}{{\rm {PM}}}_{{it}}={\beta }_{0}+{\beta }_{1}({\mathrm{ln}{{\rm {PD}}}}_{i})+{\beta }_{2}({\mathrm{ln}{{\rm {ED}}}}_{i})+{\beta }_{3}({\mathrm{ln}{{\rm {LED}}}}_{i})\\ \qquad\qquad\,\,+\,{\beta }_{4}({\mathrm{ln}{{\rm {RN}}}}_{i})+{\beta }_{5}({{{\rm {GC}}}}_{i})+{\beta }_{6}({\mathrm{ln}{{\rm {EC}}}}_{i})\\\qquad\qquad\,\,+\,{\beta }_{7}({{{\rm {SGDP}}}}_{i})+{\beta }_{8}({{{\rm {FGDP}}}}_{i})+{\varepsilon }_{{it}},\end{array}$$
(5)

where \({{{\rm {PM}}}}_{i}\) refers to the annual average PM2.5 concentration of the ith city in year \(t\); \({\beta}_{0}\) represents the coefficient of intercept term; \({\beta}_{1} - {\beta}_{7}\) are the coefficients of each independent variable; \({\varepsilon}_{{it}}\) is the independent random error term. The 10 independent variables are PD, ED, LED, RN, GC, EC, SGDP, and FGDP. To remove magnitudes from the data and reduce geographical heteroskedasticity, the numerical data were logarithmically processed while the percentage data were untreated. The superiority of GTWR lies in the introduction of temporal and spatial variables in the model. The mathematical expression is as follows:

$${y}_{i}={\beta }_{0}\left({u}_{i},{v}_{i},{t}_{i}\right)+\mathop{\sum }\limits_{k=1}^{p}{\beta }_{k}\left({u}_{i},{v}_{i},{t}_{i}\right){x}_{{ik}}+{\varepsilon }_{i},\,i=1,2,\ldots ,n.$$
(6)

where \(\left({u}_{i},{v}_{i},{t}_{i}\right)\) denotes the spatiotemporal coordinates at sample point i. \({\beta }_{k}\left({u}_{i},{v}_{i},{t}_{i}\right)\) is the kth regression coefficient at the ith sample point, \({\varepsilon }_{i}\) is the random error, and \({\varepsilon }_{i} \sim N\left(0,\,{\sigma }^{2}\right)\). The regression coefficients can be estimated using the local weighted least-squares method, calculated as

$$\hat{{\boldsymbol{\beta }}}({u}_{i},{v}_{i},{t}_{i})={[{{\boldsymbol{X}}}^{{\rm{T}}}{{\boldsymbol{W}}}_{{\boldsymbol{i}}}({u}_{i},{v}_{i},{t}_{i}){\boldsymbol{X}}]}^{-1}{{\boldsymbol{X}}}^{{\rm{T}}}{{\boldsymbol{W}}}_{{\boldsymbol{i}}}({u}_{i},{v}_{i},{t}_{i}){\boldsymbol{Y}},$$
(7)

where \({W}_{i}\left({u}_{i},{v}_{i},{t}_{i}\right)\) is the spatiotemporal kernel function. The expression for the elements of row i and column of the spatiotemporal kernel function matrix is

$${{w}_{{ij}}}^{{{\rm {ST}}}}=\exp \left(-{\left(\frac{{{d}_{{ij}}}^{{{\rm {ST}}}}}{{h}^{{{\rm {ST}}}}}\right)}^{2}\right),$$
(8)

where \({{d}_{{ij}}}^{{{\rm {ST}}}}\) is the spatiotemporal distance calculated based on the spatiotemporal ellipsoidal coordinate system, and \({h}^{{{\rm {ST}}}}\) is the spatiotemporal window width parameter. This study applied the Akaike information criterion to calculate the optimal adaptive bandwidth.

Results and discussion

Spatial and temporal variations of PM2.5 concentrations

Temporal variation characteristics

Since national PM2.5 concentration data are fragmented, this study splits national interannual PM2.5 values into six intervals according to Chinese ambient air quality standards (GB3095-2012). The classification of PM2.5 levels in 285 Chinese cities from 2010 to 2021 is presented in Fig. 4, highlighting the progressive improvement in air quality over this period. Beginning in 2013, a significant upward shift in air quality was noticeable, as the percentage of cities experiencing excellent or good air quality increased steadily to 76.49% by 2021. Moreover, the percentage of cities experiencing slight haze pollution decreased from 46.67% in 2010 to 23.51% in 2021, reflecting a marked improvement in the majority of urban areas.

Fig. 4
figure 4

Proportion of 285 cities with different PM2.5 concentrations.

Full size image

After 2014, the proportion of cities with moderate pollution decreased from 23.86% to 0% in 2021. Notably, the number of heavily polluted cities has remained zero since 2017. Although cities experiencing severe pollution were first recorded in 2011, 2013, and 2014, they accounted for a minimal proportion, peaking at 1.75% in 2013 before subsequently diminishing. Following the implementation of the APPCAP by the Chinese central government in 2013, national and regional air quality improvement trends are obvious. However, 25.31% of cities remain slightly polluted. Haze pollution levels remain far from the WHO threshold of 10 μg/m3, indicating that China should implement more air pollution management efforts.

Spatial variation characteristics

Accurate identification of the spatial pattern evolution features and distribution patterns of PM2.5 pollution is critical for the development of cross-regional linkage pollution management strategies in China. The spatial dispersion of the PM2.5 concentration in China from 2010 to 2021 is shown in Fig. 5. Over the past 12 years, there has been a marked increase in the number of cities with annual average PM2.5 concentrations below 35 μg/m3, indicating a gradual expansion of the trend toward improved air quality from regions, such as East, South, Southwest, and North China, to nationwide coverage. Notably, this trend became particularly pronounced in 2017, with cities exhibiting good and excellent air quality primarily clustered in Inner Mongolia, Guangdong, and Fujian. By 2021, this positive trend had extended south of the Yangtze River.

Fig. 5: Spatial distributions of the PM2.5 concentration in 2010, 2013, 2017 and 2021.
figure 5

The figure depicts the spatial distribution characteristics of PM2.5 concentrations across 285 cities during four time periods: a 2010, b 2013, c 2017, and d 2021.

Full size image

In contrast, the number of polluted cities with annual average PM2.5 concentrations exceeding 35 μg/m3 significantly decreased. Prior to 2013, cities with severe pollution were predominantly concentrated in regions such as Beijing–Tianjin–Hebei, Shandong, and Henan, with a tendency to spread to neighboring cities. However, in recent years, particularly by 2021, pollution in these regions has significantly decreased.

Global regression results

Exploring the spatiotemporal heterogeneity of the effects of compactness factors on PM2.5 in each city is important. Before the GTWR model was employed, global ordinary least squares (OLS) were used to verify that the selected variables globally affect the dependent variable. According to Table 2, five explanatory variables passed the 1% significance level test, with FGDP being the exception at a significance level of 10%. The estimated coefficients of the effects of ED, RN, GC, and FGDP on the PM2.5 concentration were negative, and the coefficients of the effects of PD, LED, EC, and SGDP were all positive. The variance inflation factor (VIF) values of all independent variables were less than 3, indicating that there was no multicollinearity among the variables. The summary of the GTWR regression coefficients clearly changed from negative to positive. The above result suggests that the effect of each variable on PM2.5 is spatially heterogeneous and requires further geographic analysis.

Table 2 Coefficient results from the global OLS and GTWR methods.
Full size table

For comparison, in this study, the TWR, GWR, and GTWR models were also calculated, and the findings can be found in Appendix Table A2. Overall, R2 increased from 32% for the global OLS model to 46% for the TWR model, 79% for the GWR model, and 87% for the GTWR model, whereas AICc decreased from −4822.65 to −5542.94 for the TWR model, −8757.69 for the GWR model, and −10,108.2 for the GTWR model. The model with the lower AICc and higher R2 values achieved more robust applicability. The GTWR model could match the data well since this model could manage both geographical and temporal variations. Compared to the TWR model, the GWR model achieved a better fit, revealing that temporal nonstationarity is lower than spatial nonstationarity.

Heterogeneity analysis of the urban compactness factors

Temporal variation analysis

On the basis of the GTWR regression results, the characteristics of each core explanatory variable were aggregated and presented at the 95% significance level. Figure 6 displays the temporal change trend of the coefficients of each compactness factor. The effect of each compactness factor on PM2.5 is temporally nonstationary. PD, ED, LED, and RN all contribute to PM2.5 pollution.

Fig. 6: Temporal effects of the four compactness factors on PM2.5 pollution.
figure 6

The figure shows the temporal variation of the regression coefficients from 2010 to 2021 for four compactness factors: a population density, b economic density, c land development equilibrium degree, and d road network density.

Full size image

PD increased air pollution slowly but significantly over 12 years. From 2010 to 2013, the impact of PD on air pollution remained consistently high, whereas from 2013 to 2021, this influence began to decline, leading to a narrowing of the impact gap during this period. In contrast, LED had a relatively stable positive effect on air pollution during this timeframe, particularly after 2014, when this trend became increasingly evident. However, compared to PD and LED, ED, and RN had contrasting effects, showing significant alleviation of urban PM2.5 pollution over the past 12 years. Notably, the relationship between PM2.5 pollution and ED follows an inverted U-shaped curve (see Appendix Fig. B3), indicating the existence of the environmental Kuznets curve (EKC), where the suppressive effect of ED on air pollution gradually diminishes after a certain threshold is reached.

To further verify the robustness of our findings, we utilized GDP per capita (PGDP), construction land development intensity (CI), and urban road area per capita (PR) as alternative variables to ED, LED, and RN, respectively, with the aim of better capturing economic compactness, land use compactness, and transportation compactness. After applying the GTWR model, we observed an AICc value of −10,020.5 and an R2 of 0.86 (see Appendix, Table A3). Further insights into the temporal variations in these coefficients after variable substitution indicate that each influencing factor continues to contribute to PM2.5 pollution in a manner that is consistent with the patterns observed for PD, ED, LED, and RN prior to substitution (see Appendix, Fig. B4).

Geographically variation analysis

The regression coefficients that passed the 95% significance level among the 285 cities were mapped to assess the directional influence and magnitude of the urban compactness factors on the PM2.5 concentration. To capture the potential variations in trends, we focused on years where discontinuities or significant changes were anticipated. Therefore, the regression analyses for 2010, 2013, 2017, and 2021 were selected as representative points to reflect the nonsmooth patterns observed over time.

  1. (1)

    Population compactness: PD represents population compactness, and according to the estimation results of the GTWR model, from 2010 to 2021, the proportion of cities where PD contributed to an increase in PM2.5 pollution rose from 88.77% to 91.22%. As shown in Fig. 7, in terms of the spatial distribution of the regression coefficients, PD has a positive promoting effect on PM2.5 pollution in most cities, indicating that the higher the degree of population agglomeration is, the more severe the urban PM2.5 pollution becomes. This is inconsistent with Hypothesis 1. A possible reason is that the energy consumption caused by large-scale population agglomeration exceeds the scale effect of energy-intensive use.

    Fig. 7: Regression coefficients of PD in 2010, 2013, 2017, and 2021.
    figure 7

    The figure demonstrates the spatiotemporal evolution characteristics of the population compactness regression coefficients across 285 cities during four time periods: a 2010, b 2013, c 2017, and d 2021.

    Full size image

    More specifically, cities where PD has a mitigating effect on PM2.5 pollution are located primarily in South China, such as Guangdong and Fujian Provinces. These cities have inherently low PM2.5 pollution levels and have not undergone excessive urban spatial expansion. The compact population agglomeration effect has improved urban transportation operations and resource utilization efficiency, thereby facilitating the improvement of urban air quality (Yan et al. 2022). Cities where PD exacerbates PM2.5 pollution are located mainly in North China and Northwest China, such as Beijing–Tianjin–Hebei, Shanxi, and Inner Mongolia. In these cities, the large-scale agglomeration of the population has led to an increase in demand for transportation, housing, and energy consumption, which has promoted the emission of air pollutants and resulted in a greater destructive effect of compact urban development patterns on the environment (Li et al. 2020). The above results indicate that although Hypothesis 1 suggests that an increase in population compactness helps reduce urban haze pollution, this effect varies greatly depending on the size of the city.

  2. (2)

    Economic compactness: ED represents economic compactness, and from 2010 to 2021, the proportion of cities where ED mitigated PM2.5 pollution increased from 82.46% to 93.68%. As shown in Fig. 8, in terms of the spatial distribution of regression coefficients, ED has a mitigating effect on PM2.5 pollution in most cities. This finding is consistent with Hypothesis 2, which suggests that economic agglomeration deepens the degree of industrial agglomeration within cities and that industrial agglomeration facilitates centralized pollution control, technological innovation, and the exchange and learning of knowledge related to pollution reduction and carbon emission reduction.

    Fig. 8: Regression coefficients of ED in 2010, 2013, 2017, and 2021.
    figure 8

    The figure exhibits the spatiotemporal evolution characteristics of the economic compactness regression coefficients across 285 cities during four time periods: a 2010, b 2013, c 2017, and d 2021.

    Full size image

    More specifically, cities where ED has a mitigating effect on PM2.5 pollution are concentrated in Central, East, and North China. These cities present a stronger economic foundation and have improved production efficiency with the help of industrial structure optimization and technological advancement (Zhou et al. 2021). Although the economic levels of cities in the Southwest region are relatively lagging, they have learned from the experiences and lessons of the development process in the central and eastern regions, focusing on a green economic development model and thus avoiding the old path of pollution first and treatment later. The cities where ED has a positive impact on PM2.5 pollution are primarily concentrated in Northeast China, such as Heilongjiang and Inner Mongolia. Some cities are resource-dependent, with energy-related industries as their pillar industries, and energy processing and production lead to increased air pollution. Simultaneously, the mitigating effect of ED on the PM2.5 concentration is strengthened in cities in Northeast China. This trend underscores the effectiveness of efforts in Northeast China to facilitate the transformation and upgrade of traditional industries, optimization, and adjustment of the energy structure, and reduction in the reliance on resource-intensive economic activity (Zhang et al. 2021).

  3. (3)

    Land use compactness: LED represents land-use compactness as LED, and from 2010 to 2021, the proportion of cities where LED was positively correlated with PM2.5 pollution decreased from 81.05% to 75.08%. This result suggests that the more compact the land development and utilization are, the more severe the PM2.5 pollution in cities is, which is not fully consistent with Hypothesis 3. The possible reason is that the rapid growth in urbanization speed in China has led to a disorderly sprawl of urban space, resulting in a significant increase in land use for transportation and a decrease in land use for farmland, orchards, and grasslands. This has hindered the improvement effect of compact land use on urban environmental quality (He and Zhou, 2024).

    As shown in Fig. 9, in terms of the spatial distribution of the regression coefficients, the cities where LED mitigates PM2.5 pollution are located primarily east of the Hu line, with the most significant mitigation effect observed in cities in Southwest China and South China, including Sichuan, Yunnan, Guizhou, and Guangxi. These cities possess more rational land development planning and efficient land utilization, thereby enhancing the inherent capacity of the land for pollution purification. The cities where LED has an aggravating effect on PM2.5 pollution are located mainly in North China and Northeast China. In these cities, agriculture accounts for a relatively small proportion of the industrial structure while residential land and industrial, mining, and transportation land account for a greater proportion. The disorderly structure of land use and the decrease in land-use efficiency have led to a decline in environmental quality.

    Fig. 9: Regression coefficients of LED in 2010, 2013, 2017 and 2021.
    figure 9

    The figure illustrates the spatiotemporal evolution characteristics of the land use compactness regression coefficients across 285 cities during four time periods: a 2010, b 2013, c 2017, and d 2021.

    Full size image
  4. (4)

    Transportation compactness: RN represents transportation compactness, and from 2010 to 2021, the proportion of cities where RN was negatively correlated with PM2.5 pollution increased from 72.63% to 88.42%. This finding indicates that the more diversified public transportation options are, the better the accessibility and higher the urban air quality is, which basically verifies Hypothesis 4. With the application of new-generation information technologies such as artificial intelligence and the Internet of Things in urban transportation infrastructure management, policymakers are increasingly focusing on building resource-saving and environmentally friendly transportation network patterns and improving transportation operation efficiency (Sun et al. 2021). All of the above efforts play important roles in improving air quality.

    As shown in Fig. 10, in terms of the spatial distribution of the regression coefficients, the cities where RN has the most significant mitigation effect on PM2.5 pollution are located mainly in the northeastern provinces, such as Liaoning, Jilin, and Heilongjiang, followed by the southwestern provinces, such as Sichuan, Yunnan, and Guizhou. The construction of road infrastructure in these cities effectively improves traffic efficiency and contributes to the improvement in air quality (Huang et al. 2021). The cities where RN promotes the aggravation of PM2.5 pollution are located mainly in North China and Northwest China. This may be due to the high intensity of road transportation in these areas, which increases the number of motor vehicles per unit area and leads to excessive emissions of pollutant gases. Notably, the cities located in Northeastern China and North China have shifted from a pollution exacerbation effect in 2010 to a pollution mitigation effect in 2021. This change implies that cities in this region have begun to attach importance to the green development of urban transportation infrastructure, thereby exerting the clean, pollution transfer, and scale effects of compact transportation modes.

    Fig. 10: Regression coefficients of RN in 2010, 2013, 2017 and 2021.
    figure 10

    The figure demonstrates the spatiotemporal evolution characteristics of the transportation compactness regression coefficients across 285 cities during four time periods: a 2010, b 2013, c 2017, and d 2021.

    Full size image

Conclusions and policy implications

Conclusion

This study first constructs a theoretical analysis framework for urban compactness development with respect to environmental quality and then clarifies the pathways through which population, economic, land-use, and transportation compactness affect environmental quality. Second, on the basis of panel data from 285 prefecture-level cities in China from 2010 to 2021, the spatial correlation of PM2.5 pollution is verified. Finally, the GTWR model is employed to elucidate the spatial heterogeneous impacts of the four urban compactness factors on PM2.5 pollution. The main findings are summarized as follows.

  1. (1)

    From 2010 to 2020, the proportion of cities in China with good air quality increased significantly, whereas the proportion of cities with moderate and severe pollution notably decreased. These findings suggest that the series of air pollution control policies or measures implemented in China have achieved good results. In terms of spatial distribution characteristics, the PM2.5 concentrations in most cities exhibit high-high or low-low clustering patterns. The high-high clustering cities are located in Beijing, Tianjin, Hebei, Shandong, and Henan Provinces. Among them, the cities of Xingtai, Shijiazhuang, and Hengshui exhibit significant spatial spillover effects.

  2. (2)

    The influence of the compactness factor on China’s urban PM2.5 pollution exhibits spatial heterogeneity. First, for most cities, the higher the degree of population compactness is, the more severe the urban PM2.5 pollution becomes. This is because large-scale population agglomeration exceeds the scale effect of energy-intensive use. Second, economic compactness has a mitigating effect on urban PM2.5 pollution in Central, East, and North China, while it has an intensifying effect in other regions such as Heilongjiang and Inner Mongolia. Notably, some cities in Northeast China are actively undergoing the transformation and upgrade of traditional industries, which enhances the pollution mitigation effect in these cities.

  3. (3)

    The higher the degree of land-use compactness is, the more severe the urban PM2.5 pollution is in most cities. A possible reason is the severe loss of arable land due to the disorderly expansion of urban space in recent years. Cities where land-use compactness mitigates PM2.5 pollution are located primarily east of the Hu line. The better the accessibility of urban public transportation facilities is, the higher the urban air quality. The cities where transportation compactness has the most significant mitigation effect on PM2.5 pollution are located primarily in the provinces of Northeast China, followed by the provinces in Southwest China.

Policy implications

Based on the experimental results and findings, the following recommendations were proposed in this study.

  1. (1)

    Regional integration should be promoted as an entry point to generate inhibitory effects of urban compactness factors on air pollution. The regional ecological environment is an organic whole. Hence, comprehensive management of haze pollution should not be separately implemented in the control of local pollutant emission sources and should simultaneously overcome administrative boundaries, thus achieving collaborative prevention and management of regional haze pollution. From the spatial and temporal evolution patterns of PM2.5 pollution in China, the hardest-hit areas in terms of PM2.5 pollution were mainly concentrated in Shandong, Henan, Shanxi, and the Beijing–Tianjin–Hebei urban agglomeration. Therefore, haze prevention and control in China should first focus on this part of the high-pollution cycle, strengthen the accountability of emission regulation, and enhance environmental planning and environmental legislation in regional planning.

  2. (2)

    The population structure should be optimized, and regional economic development should be developed to maximize the positive effects of economic agglomeration. This research reveals that the beneficial impact of population compactness on urban environmental pollution is only significant in cities with relatively small populations, at the same time, economic compactness has significantly decreased urban environmental pollution. Therefore, in the urban planning process, decisions should be made based on city conditions, and by guiding and establishing the optimal urban population density, the positive economic impact of agglomeration should be maximized to promote the rapid development of the city and regional economy. Cities in North and Northwest China, attention should be given to improving energy consumption efficiency, optimizing the housing and transportation infrastructure configuration, promoting the reasonable concentration of population, and reducing the negative impact of population concentration on haze control.

  3. (3)

    Mixed urban land development patterns should be reasonably planned, and the accessibility to urban transportation should be enhanced. In the process of urbanization, attention should be directed toward rationally structuring land use, particularly in balancing construction land expansion with sustainable spatial organization. Urban development will shift from incremental expansion to the optimization of existing spaces. During this optimization phase, especially in cities in South, North, and Northeast China, focus should be directed toward restructuring urban spaces. Additionally, this study reveals that improved road network density can significantly improve environmental quality. Therefore, urban road networks should be rationally planned to enhance urban traffic accessibility, traffic efficiency, and compact development. In the inland northwest regions, in particular, high-density network traffic patterns within urban areas should be accelerated to reduce the road surface density, thus improving the superiority of road compactness in regard to haze control.

Future research directions

The core conclusions of this research provide specific practical value and scientific basis for urban planning decision-makers, yet there are still some limitations. These are primarily manifested in the following aspects. Firstly, the systematic construction of an urban compactness level indicator system that aligns with China’s actual situation has not been achieved, and only single indicators have been selected from the four dimensions of population, economy, land, and transportation. Future research could incorporate more indicators related to urban spatial morphology. Secondly, there is still room for improvement in the division of urban samples. More urban grouping methods could be considered, such as grouping by urban agglomerations. Thirdly, time variables could be considered as threshold variables to further analyze the impact of compactness factors on urban air pollution at different development stages and its changing patterns.