“Urban Sprawl” or “Urban Compactness”? Differentiated Impacts of Urban Growth Patterns on the Coupling Coordination Between Pollution and Carbon Emissions

Abstract

Rapid urbanization in China has reshaped the coupling coordination between pollution and carbon emissions. However, existing studies largely rely on linear approaches and lack multidimensional and nonlinear assessments of urban growth patterns. Using panel data for 289 prefecture-level cities from 2010 to 2023, including built-up land, nighttime lights, CO₂ emissions, and PM_2.5 concentrations, this study develops three indicators: Urban Expansion Intensity (UEI), Urban Sprawl Index (USI), and Urban Compactness (UC). By integrating a coupling coordination model, K-means clustering, Geographically and Temporally Weighted Regression (GTWR), and interpretable XGBoost-SHAP analysis, four urban growth patterns are identified: High-Speed Low-Efficiency Expansion (HLE), Low-Speed Low-Efficiency Expansion (LLE), High-Speed High-Efficiency Compact (HHC), and Low-Speed High-Efficiency Compact (LHC). Results indicate that: (1) USI and UC exhibit significant nonlinear threshold effects on CCD; moderate expansion and higher compactness enhance synergy, whereas excessive dispersion or over-compactness weakens coordination. (2) UEI plays a relatively indirect and spatially heterogeneous role. (3) HHC and LHC cities achieve the highest CCD levels, while HLE cities perform the lowest. (4) Urban expansion shows an overall contraction trend, yet substantial regional disparities persist. These findings highlight nonlinear and spatially heterogeneous mechanisms linking urban growth patterns and pollution–carbon coupling coordination, providing implications for differentiated spatial governance.

1. Introduction

Since the reform and opening up, China has undergone a rapid process of urbanization, accompanied by substantial increases in urban construction, infrastructure development, and population agglomeration [1,2]. Urban areas have gradually become the primary spaces for economic growth, industrial upgrading, and resource concentration, playing a decisive role in the restructuring of regional development patterns [3,4,5]. At the same time, however, this process has also been associated with disorderly land expansion, intensive resource consumption, and mounting environmental pressures [6,7]. Among these challenges, the persistent increase in air pollution and carbon emissions has become particularly prominent [6,8,9,10]. Although China’s development strategy has gradually shifted from scale-oriented expansion to quality-oriented improvement [11], and national policies have increasingly emphasized the coordinated advancement of pollution reduction and carbon mitigation [12], how to steer urban expansion toward a more sustainable trajectory remains a critical issue in contemporary urban governance.

Urban expansion is not merely a process of increasing construction land; rather, it constitutes a multidimensional transformation involving expansion intensity, sprawl tendency, and compactness characteristics [13,14]. Existing studies have shown that appropriately managed expansion can foster industrial agglomeration, improve infrastructure provision, and enhance land-use efficiency, whereas excessive outward growth often results in land fragmentation, longer commuting distances, inefficient infrastructure allocation, and rising energy demand [15,16,17]. This suggests that the effects of urban expansion are inherently multidimensional, encompassing not only changes in urban land use, transportation, and ecosystems, but also differentiated impacts on environmental quality and carbon emissions. In other words, urban expansion should not be simplistically understood as an increase in built-up area. Its environmental consequences depend not only on the scale of expansion itself, but also on how urban land is spatially organized, how different land units are connected, and the extent to which urban functions are integrated within the broader urban system [18,19,20].

From an ecological perspective, urban expansion may undermine habitat integrity, reduce ecosystem connectivity, and erode regional carbon sink capacity [10,21,22]. From the perspective of the urban environment, it may also intensify air pollution, increase carbon emissions, exacerbate the urban heat island effect, and lead to the loss of ecological land [9,23,24,25]. Importantly, these environmental consequences are shaped not only by the scale of expansion, but also by the spatial organization of urban land. More dispersed forms of development are often associated with land fragmentation, rising transport demand, and increased energy consumption [26,27,28], whereas more compact development may improve infrastructure efficiency and resource allocation. However, when spatial agglomeration exceeds a reasonable threshold, it may also give rise to congestion and pollutant accumulation [29,30]. Therefore, urban expansion should not be interpreted solely in terms of quantitative growth in built-up area, but also examined through the lens of spatial organization and the development quality reflected therein. This issue is particularly important in China, where many cities are currently at a critical stage of transition from incremental expansion to a model that places equal emphasis on stock optimization. On the one hand, there is a pressing need to curb the ecological and environmental pressures induced by disorderly outward expansion; on the other hand, cities are increasingly required to improve the efficiency and quality of existing urban space through urban renewal, the redevelopment of inefficient land, and the optimization of spatial structure [31,32,33].

Beyond the eco-environmental effects associated with urban expansion, growing attention has also been paid to its relationship with the coordinated development of pollution reduction and carbon mitigation [12,34,35,36]. Against the backdrop of China’s “dual carbon” goals and high-quality development strategy, clarifying whether urban expansion promotes or constrains the coordinated improvement of pollution reduction and carbon mitigation has become an increasingly important research question [37,38]. Because air pollutants and carbon emissions often share common sources, transmission pathways, and governance mechanisms, they are increasingly being examined within a unified analytical framework [36,39,40,41]. Relevant studies have investigated the spatial co-agglomeration of pollution and carbon emissions, as well as key driving factors such as industrial structure, population density, technological innovation, policy intervention, and environmental regulation, and have explored measurement approaches based on coupling coordination methods and related analytical frameworks [34,36,42,43]. In particular, the coupling coordination degree model has been widely adopted to evaluate the coordination status among multiple subsystems and has provided an important methodological foundation for studies of environmental synergy [35,44].

At present, approaches to measuring urban expansion have become increasingly sophisticated, and a substantial body of research has examined this issue from the perspectives of land-use change and efficiency, expansion intensity, spatial morphology, and spatial regulation [13,14,45,46]. In quantitative studies, urban expansion is commonly characterized using urban form indicators, urban expansion intensity, landscape pattern indices, urban sprawl indices, and remote sensing-based spatial analysis [13,22,47]. In recent years, some socioeconomic indicators related to land development efficiency, such as the ratio of land consumption rate to population growth rate and urban land-use efficiency, have also been incorporated into the assessment framework of urban expansion [48]. Nevertheless, existing studies still tend to focus on a single dimension of urban expansion, making it difficult to systematically identify and compare the relative roles of expansion speed, sprawl degree, and compactness within a unified analytical framework [49,50]. These studies collectively suggest that urban expansion should not be treated as a one-dimensional change in built-up area, but rather as a multidimensional process.

Despite the growing body of research on urban expansion and its eco-environmental effects, several limitations remain. First, many studies still rely on single-dimensional indicators to characterize urban expansion, making it difficult to fully capture the multidimensional nature of urban growth [14,47,48]. Second, although urban expansion is widely recognized as an important driver of environmental change, its effects on pollution–carbon coordination are still often discussed in linear or average terms, with insufficient attention paid to nonlinear thresholds and spatial heterogeneity [30,36]. Third, although the synergy between pollution reduction and carbon mitigation has become an important policy objective in Chinese urban development, research remains limited on how different urban growth patterns shape such synergy across different city types and regions [11,34]. Consequently, the existing literature still provides limited support for differentiated urban planning and place-based spatial governance.

To address these gaps, this study examines 289 prefecture-level and above cities in China over the period 2010–2023 and investigates the differentiated effects of different urban growth patterns on the coupling coordination degree (CCD) of pollution reduction and carbon mitigation. Specifically, this study develops a multidimensional indicator system consisting of urban expansion intensity (UEI), urban sprawl index (USI), and urban compactness (UC), and organizes the analytical methods within a stepwise framework. The CCD model is first used to measure the coordinated status of pollution control and carbon mitigation. K-means clustering is then applied to identify urban growth patterns based on the three expansion dimensions. On this basis, GTWR is employed to examine the spatiotemporal heterogeneity in the relationships between urban expansion and CCD, while XGBoost-SHAP is used to further identify nonlinear effects, threshold characteristics, and relative variable contributions. By combining multidimensional measurement, pattern identification, local regression, and interpretable machine learning, this study offers a more nuanced understanding of the complex interactions between urban expansion and environmental coordination processes.

The contributions of this study are threefold. First, by jointly considering expansion intensity, sprawl tendency, and compactness characteristics, this study develops a multidimensional analytical framework for identifying urban growth patterns, thereby moving beyond the limitations of single-indicator analysis. Second, it reveals the nonlinear and threshold effects of different dimensions of urban expansion on the coordination of pollution reduction and carbon mitigation, demonstrating that the environmental effects of urban expansion are not monotonic. Third, it shows that the relationship between urban expansion and the synergy of pollution reduction and carbon mitigation varies significantly across city types and regions, thereby providing more targeted evidence for differentiated spatial governance and urban planning. Overall, these contributions not only enrich the literature on urban expansion and environmental effects, but also provide robust empirical evidence for advancing green, low-carbon, and high-quality urban development in China.

2. Theoretical Analysis

Urban growth should not be understood merely as the increase in built-up land, but rather as a multidimensional process involving expansion intensity, spatial sprawl, and compactness [13,15]. Different urban growth patterns can reshape the spatial organization of population, industry, transport, and infrastructure, thereby influencing the coupling coordination of pollution reduction and carbon mitigation [34,51]. More specifically, urban expansion may affect the coupling coordination degree (CCD) through multiple pathways, including land-use efficiency, commuting distance, infrastructure allocation, agglomeration economies, and congestion effects [51,52,53]. Hence, its environmental consequences depend not only on the speed of urban growth, but also on how urban space is organized [13,15].

Urban expansion intensity mainly reflects the speed and scale of land development [14]. Moderate expansion may help optimize urban functions and generate scale benefits, whereas excessively rapid expansion may lead to ecological land loss, infrastructure redundancy, and rising environmental pressure [16,17,46]. Therefore, the effect of urban expansion intensity on CCD is unlikely to be purely linear, but may instead exhibit nonlinear or threshold characteristics [54,55].

Urban sprawl captures the dispersed and discontinuous dimension of urban growth patterns [47]. Excessive sprawl often leads to spatial fragmentation, longer commuting distances, greater transport dependence, and lower infrastructure efficiency, thereby weakening pollution–carbon coordination [25,45]. By contrast, moderate spatial expansion may help optimize spatial functions and relieve excessive concentration [26,30]. Accordingly, the effect of sprawl on CCD is also unlikely to be purely linear, but may vary across different value ranges [54,55].

Urban compactness reflects the concentration and organizational efficiency of urban spatial structure [26,56]. A relatively compact form may improve land-use intensity, accessibility, and infrastructure efficiency, thereby promoting pollution–carbon coordination [30,57,58]. However, excessive compactness may also lead to congestion, crowding, and pollutant accumulation [23,59]. Therefore, the relationship between compactness and CCD may involve an optimal interval rather than a simple linear effect [29,30].

Compared with expansion intensity, sprawl and compactness more directly capture the spatial structure and quality of urban growth [26,60]. Since transport efficiency, land-use intensity, and infrastructure allocation are more directly shaped by spatial configuration than by growth speed alone, spatial structure variables are more likely to show stronger explanatory relevance for CCD [26,61]. Meanwhile, differences in development stage, industrial structure, and governance conditions may cause these effects to vary across urban growth patterns [27,28]. Therefore, the effects of UEI, USI, and UC on CCD should be understood as context-dependent rather than spatially uniform.

3. Materials and Methods

3.1. Study Area

This study selects 289 prefecture-level and above cities in China as the analytical units (Figure 1). China provides a unique context for examining the environmental implications of urban expansion due to its rapid urbanization, large spatial scale, and pronounced regional disparities. Over the past decades, Chinese cities have undergone substantial spatial restructuring, accompanied by rising pressures from air pollution and carbon emissions. Such dynamic transformation offers an appropriate setting for identifying nonlinear and spatially heterogeneous effects of expansion patterns on pollution–carbon coordination (CCD). After excluding cities with substantial missing data, a balanced panel for 2010–2023 was constructed. The sample covers eastern, central, western, and northeastern regions, representing diverse development stages, industrial structures, and spatial morphologies. This regional heterogeneity provides sufficient spatial contrast for testing threshold effects and mode-specific environmental responses, thereby enhancing the robustness and generalizability of the empirical findings.

3.2. Methods

3.2.1. Coupling Coordination Degree Model

The coupling coordination model effectively captures the synergistic development between subsystems and evaluates overall system performance. Accordingly, this study employs the Coupling Coordination Degree (CCD) model [34] to assess the coordinated status of pollution control and carbon reduction from 2010 to 2023, operationalized using CO₂ emissions and PM_2.5 concentration after negative normalization. The model is defined as follows:

$$D = \sqrt{C \times T}$$

(1)

$$C=2\sqrt{U_1 \times { U}_2/{(U_1+{ U}_2)}^2}$$

(2)

$$T = a{U}_1+ b{U}_2$$

(3)

where $D$ represents the coupling coordination degree, reflecting the interactive and coordinated relationship between the two subsystems. A higher $D$ value indicates stronger coupling coordination between pollution and carbon emissions ($D\in [0,1]$). $U_1$ denotes total ${\mathrm{CO}}_2$ emissions, and $U_2$ represents the annual average ${\mathrm{PM}}_{2.5}$ concentration. C is the coupling degree, measuring the strength of interaction between the subsystems, while $T$ reflects the overall development level. In this study, a higher CCD value indicates that the pollution and carbon-emission subsystems achieve a more coordinated state, characterized by relatively lower pollution and carbon-emission pressures after normalization. By contrast, a lower CCD value reflects a less coordinated state, suggesting weaker synergy between pollution control and carbon reduction.

Given the dominant role of ${\mathrm{CO}}_2$ in greenhouse gas emissions and the significant environmental relevance of ${\mathrm{PM}}_{2.5}$, equal weights were assigned to the two subsystems ($a=b=0.5$). To ensure consistent directional interpretation of environmental performance, ${\mathrm{CO}}_2$ emissions and ${\mathrm{PM}}_{2.5}$ concentrations were negatively normalized prior to calculation. After negative normalization of ${\mathrm{CO}}_2$ emissions and ${\mathrm{PM}}_{2.5}$ concentration, the CCD is interpreted as an index of the coordinated improvement of pollution control and carbon reduction, rather than a direct coupling of two negative concepts. In practical terms, lower CCD values indicate weaker coordination between pollution control and carbon mitigation, whereas higher values indicate a more synchronized improvement process with relatively lower environmental pressures after normalization. Therefore, the CCD in this study should be understood not merely as a mathematical coupling coefficient, but as a synthetic indicator of the practical coordination status of pollution–carbon synergy. It should also be noted that CCD reflects a relative composite status of coordination rather than a direct measure of causal policy effectiveness. Following previous studies [35,44,62], the classification criteria for CCD are shown in

Table 1. Classification criteria for the coupling coordination degree.

D Value	Category	D Value	Category
[0, 0.1)	Extreme imbalance	[0.5, 0.6)	Bare coordination
[0.1, 0.2)	Severe imbalance	[0.6, 0.7)	Primary coordination
[0.2, 0.3)	Moderate imbalance	[0.7, 0.8)	Intermediate coordination
[0.3, 0.4)	Mild imbalance	[0.8, 0.9)	Good coordination
[0.4, 0.5)	Near imbalance	[0.9, 1.0)	High-quality coordination

, ranging from extreme imbalance to high-quality coordination.

3.2.2. Urban Expansion Intensity

Urban expansion intensity (UEI) reflects the rate and magnitude of built-up land growth over a given time period and captures the dynamic spatial change in urban development [14]. It is calculated as:

$$\mathrm{UEI} = {\displaystyle \frac{U_b -{ U}_a}{U_a \times T}} \times 100\%$$

(4)

where UEI denotes urban expansion intensity; $U_a$ and $U_b$ represent the built-up land area at the beginning and end of the study period, respectively; and T is the time interval. A higher UEI value corresponds to a stronger intensity of urban expansion, indicating more rapid growth of built-up land and a greater tendency toward extensive land development.

3.2.3. Urban Sprawl Index

The Urban Sprawl Index (USI) is used to capture the dispersed dimension of urban growth patterns and is widely applied to characterize differences in urban spatial development [47]. In this study, USI is quantified using calibrated nighttime light data, following Wu et al. (2022), and is defined as [63]:

$$\mathrm{USI} = 0.5 \times \left(L\% - H\%\right) + 0.5$$

(5)

where $L\%$ denotes the proportion of urban area with nighttime light intensity below the national average, and $H\%$ represents the proportion exceeding the national average. Higher USI values indicate greater spatial dispersion and a stronger sprawl tendency. Calibrated nighttime light data provide a useful proxy for the spatial distribution of human activities and built-up intensity at the city scale. In large-sample intercity comparisons, a higher proportion of low-intensity and spatially scattered light patches generally corresponds to fragmented development and low-density outward growth [13,64]. Nighttime light data also provide relatively strong temporal continuity, broad spatial coverage, and intercity comparability, which supports long-term nationwide comparison across prefecture-level cities. At the same time, it should be acknowledged that such data cannot fully capture fine-grained morphological characteristics of urban sprawl, nor can they adequately distinguish land-use functions, population-density patterns, or urban design features. In addition, the indicator may be affected by brightness saturation in highly developed urban cores and by non-residential lighting sources such as industrial facilities. Therefore, the USI is interpreted here as a proxy for the dispersed and fragmented dimension of urban growth patterns rather than a direct one-to-one morphological measure of urban sprawl.

3.2.4. Urban Compactness

Urban compactness (UC) reflects the concentration and morphological regularity of urban spatial structure and is used as a proxy for land-use efficiency. In this study, UC represents the compactness dimension of urban expansion and is interpreted as an indicator of spatial concentration and land-use efficiency. Higher UC values indicate a more centralized spatial structure and greater resource utilization efficiency.

UC is measured using the Herfindahl–Hirschman Index (HHI) based on calibrated DMSP-OLS and NPP-VIIRS nighttime light data (2010–2023) [65,66]. It is calculated as the sum of squared shares of nighttime light intensity across county-level units within each prefecture-level city:

$$\mathrm{UC} ={ \mathrm{HHI}}_{\mathit{jt} }= \sum _{j=1}^n{\left(l_{\mathit{jt}} / l\right)}^2 = \sum _{j=1}^n{T}_{\mathit{jt}}^2$$

(6)

where $l_{\mathit{jt} }$ represents the average nighttime light intensity of subunit j in year t; l is the total nighttime light intensity of the prefecture-level city; and n denotes the number of subunits. UC ranges from (0, 1], with values approaching 1 indicating monocentric development, while lower values suggest polycentric structures. From the perspective of indicator validity, nighttime light concentration reflects the spatial agglomeration of population, economic activities, and infrastructure. A higher concentration of light within fewer subunits generally indicates a more centralized and contiguous urban form, whereas a more even distribution of light usually implies a looser and more dispersed structure. Therefore, the HHI-based UC index is used here to represent the compactness dimension of urban spatial organization at the prefecture level [67,68]. Nighttime light concentration also permits relatively consistent comparison of spatial concentration across a large number of cities over time and is therefore suitable for nationwide panel analysis. Nevertheless, it should also be noted that this indicator cannot fully capture all dimensions of urban compactness, such as street-network configuration, functional land-use mix, three-dimensional built form, or neighborhood-scale density variation. Accordingly, the UC index is interpreted here as a proxy for relative spatial concentration and compactness at the prefecture level, rather than a complete morphological representation of urban form.

3.2.5. Geographically and Temporally Weighted Regression

Geographically and Temporally Weighted Regression (GTWR) is a local linear regression model that simultaneously accounts for spatial and temporal non-stationarity, enabling more accurate estimation of the dynamic effects of urban expansion on pollution–carbon coupling coordination [69]. Compared with traditional global regression models, GTWR can identify variations in the effects of variables across space and time and thus has advantages in capturing spatiotemporal heterogeneity. In this study, it helps reveal the spatial differentiation and temporal evolution of the effects of urban expansion variables on CCD. The GTWR model is specified as:

$$y_i ={ \beta }_0\left(u_i,v_i,t_i\right) +\sum _k{\beta }_k\left(u_i,v_i,t_i\right)x_{\mathit{ik}} +{ \epsilon }_i$$

(7)

where $y_i$ denotes the coupling coordination degree (CCD) between pollution and carbon emissions for city i; ${\beta }_0\left(u_i,v_i,t_i\right)$ represents the spatiotemporally varying intercept; ${\beta }_k\left(u_i,v_i,t_i\right)$ is the local regression coefficient of explanatory variable k at location $\left(u_i,v_i\right)$ and time t_i; x_ik denotes the value of explanatory variable k for city i; and ɛ_i is the random error term.

3.2.6. XGBoost-SHAP Analysis

Extreme Gradient Boosting (XGBoost), a tree-based ensemble learning algorithm, demonstrates greater robustness and flexibility than traditional methods such as Random Forest (RF) and Gradient Boosting Decision Tree (GBDT), particularly in handling missing values and large-scale datasets [70]. Moreover, XGBoost effectively alleviates multicollinearity among explanatory variables [71]. Its main advantage lies in its ability to capture complex nonlinear relationships and interaction effects between variables and the target outcome. In this study, XGBoost helps overcome the limitations of traditional linear models and identify the nonlinear effects of urban expansion variables on CCD. The objective function is defined as:

$$L\left(\varnothing \right) = \sum l\left(y_i,{\widehat{y}}_i\right) + \sum \mathsf{\Omega }(f_k)$$

(8)

where $L\left(\varnothing \right)$ denotes the overall objective function; $l\left(y_i,{\widehat{y}}_i\right)$ represents the loss term measuring prediction error; and $\mathsf{\Omega }(f_k)$ is the regularization term that penalizes model complexity to prevent overfitting.

Although XGBoost provides global feature importance rankings, its ability to interpret complex feature–target relationships may diminish as model complexity increases. To enhance interpretability and quantify the marginal contribution of urban expansion variables (UEI, USI, and UC) to CCD, this study integrates SHAP (Shapley Additive Explanations) as a post hoc interpretation framework [72]. SHAP has the advantage of explaining feature importance, effect direction, and variation patterns simultaneously. In this study, it helps further uncover the differences in marginal contributions and nonlinear response patterns of different urban expansion dimensions to CCD. The SHAP value for feature i is calculated as:

$${\theta }_i \left(f,x\right) = \sum _{r\in R}{\displaystyle \frac{1}{I}}\left[f_x\left(P_i^r\cup \left\{i\right\}\right) - f_x\left(P_i^r\right)\right]$$

(9)

where ${\theta }_i\left(f,x\right)$ denotes the Shapley value of feature $i$ for instance x; $R$ represents all possible permutations of the feature set; $P_i^r$ is the subset of features preceding feature $i$ in permutation $r$; and $I$ is the total number of input features.

3.2.7. Classification of Urban Growth Patterns Using K-Means

This study employs the K-means clustering algorithm to classify urban growth patterns among 289 standardized city samples. The algorithm partitions observations into homogeneous groups by minimizing within-cluster variance while maximizing between-cluster differences [73]. Clustering is based on three indicators—UEI, USI, and UC—and iteratively updates cluster centroids to minimize the sum of squared errors (SSE):

$$\mathrm{SSE} ={ {\sum }_{i=1}^k{\sum }_{x\in C_i}(x_i - C_i)}^2$$

(10)

where $C_i$ denotes the centroid of cluster $i$, and $x$ represents observations within cluster $i$. To ensure interpretability and theoretical consistency with urban expansion typologies, the number of clusters was predefined as $k = 4$. Clustering performance was evaluated using the silhouette coefficient, ranging from −1 to 1, with higher values indicating greater intra-cluster similarity and inter-cluster separation [74].

Based on the relative means of UEI, USI, and UC compared with overall medians, and grounded in urban spatial development theory, four expansion patterns were identified: HLE, characterized by rapid expansion, high sprawl, and low compactness; LLE, marked by slow expansion, high sprawl, and low compactness; HHC, featuring rapid expansion, low sprawl, and high compactness; and LHC, defined by slow expansion, low sprawl, and high compactness. This indicator-based clustering framework has been validated in prior urban expansion studies [13]. Each city was assigned to a corresponding category, providing the analytical basis for subsequent spatiotemporal, nonlinear, and visualization analyses. The identified types represent the overall classification results derived from integrated multi-year urban expansion characteristics over the study period, rather than year-specific categories.

3.3. Data Sources and Processing

Built-up land area data were obtained from the China Urban Construction Statistical Yearbook for the period 2010–2023. Nighttime light data used to calculate UC were derived from the NPP/VIIRS and DMSP/OLS datasets provided by the National Oceanic and Atmospheric Administration (NOAA). CO₂ emission data were sourced from the Emissions Database for Global Atmospheric Research (EDGAR) (https://edgar.jrc.ec.europa.eu/ (accessed on 6 November 2025)). EDGAR provides global anthropogenic CO₂ emission gridmaps at a spatial resolution of 0.1° × 0.1°. In this study, the gridded data were matched to the administrative boundaries of 289 prefecture-level and above cities in China to derive city-level estimates. Its standardized methodology and nationwide comparability make it suitable for prefecture-level intercity analysis [75,76,77]. ${\mathrm{PM}}_{2.5}$ concentration data were obtained from the Atmospheric Composition Analysis Group at Washington University in St. Louis, calibrated using a geographically weighted regression approach (https://sites.wustl.edu/acag/surface-pm2-5/ (accessed on 18 December 2025)), as well as from the China National Environmental Monitoring Centre (https://air.cnemc.cn:18007/ (accessed on 12 January 2026)).

The base map used in this study was obtained from the Standard Map Service of the Ministry of Natural Resources of China (Map Approval No. GS(2023)2767), and no modifications were made to the original content. Data processing, modeling, and visualization were conducted using MATLAB R2023a, Python 3.12.2, and ArcGIS 10.8.2, with GTWR analysis implemented via the GTWR plugin.

Based on the above data and methodological procedures, the overall research framework is presented in Figure 2.

4. Results

4.1. Identification of Urban Growth Patterns

Using three standardized indicators (UEI, USI, and UC), K-means clustering classified the 289 prefecture-level and above cities into four urban growth patterns (

Table 2. Identification Results and Characteristics of Four Urban Growth Patterns.

Urban Growth Patterns	Number of Cities	Share (%)	Mean (SD)			Characteristics
			UEI	USI	UC
High-Speed Low-Efficiency Expansion (HLE)	70	24.22	0.0468 (0.0426)	0.6782 (0.0791)	0.1721 (0.0662)	Rapid growth, high sprawl, low compactness, and low land-use efficiency
Low-Speed Low-Efficiency Expansion (LLE)	95	32.87	0.0299 (0.0346)	0.8718 (0.0523)	0.2098 (0.0563)	Slow growth, high sprawl, low compactness, and low land-use efficiency
High-Speed High-Efficiency Compact (HHC)	56	19.38	0.1765 (0.0375)	0.8397 (0.0833)	0.2382 (0.0842)	Rapid growth, low sprawl, high compactness, and high land-use efficiency
Low-Speed High-Efficiency Compact (LHC)	68	23.53	0.0285 (0.0366)	0.8612 (0.0778)	0.4234 (0.0821)	Slow growth, low sprawl, high compactness, and high land-use efficiency

). LLE cities form the largest group (95 cities, 32.87%) and are characterized by a relatively high USI mean (SD) of 0.8718 (0.0523) and low UC of 0.2098 (0.0563), suggesting that low-speed and low-efficiency expansion remains the dominant trajectory in China’s urbanization process. HLE (70 cities, 24.22%) and LHC (68 cities, 23.53%) follow; among them, HLE has the lowest UC 0.1721 (0.0662), whereas LHC shows the highest UC 0.4234 (0.0821). HHC cities account for the smallest share (56 cities, 19.38%) but exhibit the highest UEI 0.1765 (0.0375), indicating that the ideal pattern of rapid yet efficient growth has not been widely achieved.

Overall, sprawl-oriented expansion predominates, encompassing 165 cities (57.1%), which is 1.4 times the number of compact-development cities (124). At the macro level, this confirms that despite long-standing policy advocacy for intensive and compact development, extensive and dispersed growth remains prevalent in practice.

Figure 3 presents the overall classification results of urban growth patterns for the 289 prefecture-level and above cities, based on the integrated clustering results derived from multi-year data over the study period using the K-means algorithm, rather than data from a specific year. The figure shows marked spatial heterogeneity and regional clustering in urban growth patterns, as characterized by UEI, USI, and UC. HHC cities exhibit rapid expansion while maintaining high compactness and land-use efficiency, indicating an intensive growth trajectory. In contrast, HLE cities show strong outward expansion, elevated sprawl intensity, and low compactness. LHC cities expand at a slower pace but retain compact spatial structures, whereas LLE cities are characterized by low-density expansion and persistently low land-use efficiency.

Spatially, HHC cities are mainly concentrated in the eastern coastal region, HLE cities are more common in parts of western China and some inland transitional areas, LHC cities are primarily distributed in some relatively mature cities in the east and south, and LLE cities are widely located in central China and parts of the northeast. This distribution suggests that compact-oriented cities are more likely to occur in regions with stronger economic agglomeration, more mature infrastructure systems, and relatively higher land-use efficiency, which helps sustain compact spatial organization under both rapid and slow expansion [26,46,63]. By contrast, sprawl-oriented cities are more common in regions where land-dependent growth, weaker agglomeration capacity, or uneven infrastructure provision make outward and dispersed expansion more prevalent [14,28,78]. In this sense, the concentration of HHC and LHC cities in the eastern coastal and southern regions indicates that urban growth in these areas is more easily translated into compact and efficient spatial development, whereas the broader distribution of HLE and LLE cities in western, central, and northeastern China suggests that urban expansion in these regions is more often accompanied by lower compactness and weaker land-use efficiency [26,27,60].

4.2. Spatiotemporal Evolution

4.2.1. Spatiotemporal Evolution of UC, USI, and UEI

Overall, urban expansion across the study area exhibited a clear contraction trend. Urban Expansion Intensity (UEI) declined from 0.087 in 2010 to 0.007 in 2023 (−92%), while inter-city disparities gradually narrowed over time. By 2023, 96.5% of cities had UEI values within the 0–0.1 range, indicating that most cities had shifted from relatively rapid outward growth to a low-expansion or near-stable stage.

The Urban Sprawl Index (USI) also decreased during the study period, indicating that the degree of spatial dispersion was alleviated to some extent. However, sprawl levels remained relatively high in most cities, suggesting that the inertia of low-density and fragmented expansion has not been fundamentally removed. Pronounced declines were observed in the Beijing–Tianjin–Hebei region, the Shandong Peninsula, and southeastern cities, while east–west differentiation became more evident. This pattern implies that cities differ substantially in their spatial governance capacity and development stage: some regions have begun to move away from extensive outward expansion, whereas others continue to face persistent sprawl pressure.

Urban Compactness (UC) remained generally low, with only about 5% of cities reaching an effective expansion threshold. This indicates that, although expansion intensity has slowed in most cities, a more efficient and concentrated spatial structure has not yet been widely achieved. UC values were higher in some cities of the Hexi Corridor and South China than in cities of Central China and the coastal regions, indicating significant regional differences in spatial organization and land-use efficiency. As shown in Figure 4, the average UC in 2020 remained approximately 27% below the estimated optimal level, suggesting that most cities were still far from the compact development state most conducive to stronger pollution–carbon coordination. In other words, the key challenge of current urban development lies less in whether cities continue to expand and more in how to improve spatial organization and development quality under a low-expansion regime.

4.2.2. Spatiotemporal Evolution of CCD

Using model-estimated CCD values for prefecture-level cities from 2010 to 2023, the spatial patterns were mapped according to Table 2 (Figure 5a). Most cities were classified as being in Good coordination or High-quality coordination, and by 2023, 88% of cities had reached these levels. This indicates that the coordinated advancement of pollution control and carbon reduction has become the dominant trend across Chinese cities. However, low-value clusters remained concentrated in the Beijing–Tianjin–Hebei region, the Central Plains, and the Shandong Peninsula, suggesting that these areas still face relatively strong constraints in achieving environmental synergy, likely due to heavier industrial pressure, denser emissions, or more complex transition burdens [12,34]. In contrast, Northeast China and the southeastern coast maintained relatively high CCD levels and showed further spatial expansion, implying that the benefits of coordinated pollution–carbon governance were becoming more stable and geographically widespread in these regions.

The kernel density curves (Figure 5b) show a clear rightward shift in national CCD, with the peak increasing from 0.88 to 0.91, indicating an overall movement toward a higher coordination stage. This shift suggests that improvements in pollution–carbon synergy were not limited to a few leading cities, but were becoming more common across the national urban system. The boxplots (Figure 5c) further reveal marked differences among urban growth patterns. HHC and LHC cities had the highest and most concentrated CCD distributions, indicating that compact and relatively efficient expansion patterns were generally associated with stronger and more stable coordination outcomes. By contrast, HLE cities had the lowest CCD and the largest dispersion, suggesting that rapid but inefficient expansion may weaken synergy and produce greater uncertainty in environmental performance. LLE cities were at an intermediate level but remained highly variable, implying that slow expansion alone does not necessarily improve coordination if structural inefficiency persists.

4.3. GTWR Results

To further reveal the spatiotemporal heterogeneity in the effects of the four urban growth patterns on CCD, this study introduces the GTWR model for analysis. Compared with conventional OLS and GWR models, GTWR incorporates both temporal and spatial weights, enabling a more refined characterization of the spatiotemporal nonstationarity in variable relationships. A comparison of model fitting results is presented in

Table 3. Comparison of Model Fitting Results.

Model	R2	Adjusted R2	AICc
OLS	0.598	0.597	−924.173
GWR	0.612	0.612	−1300.421
GTWR	0.717	0.714	−734.768

. In terms of goodness of fit, the GTWR model shows both ${\mathrm{R}}^2$ and adjusted ${\mathrm{R}}^2$ values above 0.71, indicating satisfactory explanatory power and model fit.

The GTWR results indicate that the effect of UEI on CCD exhibits significant spatial heterogeneity, and that this heterogeneity intensified over time, evolving from regional differentiation to a more pronounced pattern of spatial restructuring (Figure 6a). During 2010–2015, negative UEI effects expanded across parts of northeastern and western China, whereas positive effects increased but remained largely concentrated along the eastern coast. This suggests that, in the earlier stage, rapid expansion in many inland and old industrial regions was more likely to generate environmental pressure than coordination gains, while eastern coastal cities were better able to convert expansion into improvements in pollution–carbon coordination. By 2023, the effect of UEI had become predominantly positive nationwide, although spatial contrasts remained strong. This implies that expansion intensity itself may no longer be the primary constraint on CCD in many regions; instead, differences in spatial organization and development quality appear to have become more important.

When differences across urban growth patterns are considered, UEI exerted a significantly positive effect in HLE cities, while LHC cities showed a strengthening negative effect relative to the national average (Figure 6b). This indicates that, in HLE cities, a certain degree of expansion may still help release scale effects and support coordination improvement, even under relatively low efficiency. In contrast, the increasing negative effect in LHC cities suggests that, under already compact and relatively stable expansion conditions, further increases in expansion intensity may weaken coordination performance rather than enhance it. Overall, the effect of UEI differed markedly across urban growth patterns in both direction and magnitude, highlighting the differentiated role of expansion intensity under distinct spatial development conditions.

The GTWR results show that the effect of USI on CCD exhibits clear spatial heterogeneity and temporal evolution (Figure 7a). In 2010, positive USI effects were mainly concentrated in northeastern, northern, and parts of central China, while effects were weaker in southern and southeastern coastal cities, indicating a pronounced north–south contrast. This pattern suggests that, at the early stage, a certain degree of spatial expansion was more likely to be associated with coordination improvement in northern and inland cities, whereas its contribution was more limited in southern and coastal regions. By 2015, the positive influence expanded into central and eastern China, forming a more continuous positive belt, which indicates that the beneficial effect of USI on CCD became more widespread across the national urban system. In 2023, positive effects became more consolidated nationwide, with stronger effects in northeastern and parts of southwestern China, whereas effects in eastern coastal core urban agglomerations were relatively moderate. This implies that the relationship between sprawl and coordination was not uniform across space: in some regions, moderate spatial expansion may have helped optimize urban functions and release coordination gains, while in more mature coastal agglomerations its marginal contribution was comparatively weaker.

A comparison among different urban growth patterns further shows that USI had the strongest positive effect in HLE cities and remained above the national average, whereas its effect was weaker in LHC cities (Figure 7b). This indicates that, in HLE cities, CCD was more sensitive to changes in USI, suggesting that under rapid but inefficient expansion, spatial expansion form played a more prominent role in shaping coordination performance. By contrast, the weaker effect observed in LHC cities suggests that, under relatively compact and stable development conditions, CCD was less sensitive to changes in sprawl. Overall, the effect of USI varied substantially across urban growth patterns, indicating that the role of spatial dispersion in shaping pollution–carbon coordination depended strongly on the underlying development mode.

The GTWR results indicate a stable positive effect of UC on CCD, with relatively smooth temporal evolution and limited spatial redistribution (Figure 8a). From 2010 to 2023, high-value UC coefficients remained concentrated in central-eastern China and parts of Northeast China, while western regions generally showed weaker effects. This pattern suggests that the contribution of compact spatial structure to pollution–carbon coordination was consistently stronger in regions with relatively better agglomeration conditions and land-use organization, whereas its effect remained more limited in areas with weaker spatial concentration. The spatial pattern became slightly more continuous by 2015 and remained broadly stable by 2023, indicating that the effect of UC on CCD was not characterized by major directional shifts over time, but rather by a gradual consolidation of its positive role.

The pattern-specific results shown in Figure 8b further indicate that HLE cities exhibited the strongest positive UC effect. This indicates that, under rapid but relatively inefficient expansion, improvements in compactness were more strongly associated with higher CCD, suggesting that spatial concentration played a more important role in alleviating the environmental pressure generated during fast outward growth. HHC cities also exhibited positive effects, but at slightly lower levels, implying that once a relatively efficient and compact structure had already been formed, the additional contribution of further compactness became less pronounced. By contrast, LLE and LHC cities showed comparatively weaker UC effects close to the national average, indicating that under slow-expansion conditions, CCD was less sensitive to changes in compactness. Overall, the results suggest that the effect of UC on CCD was generally positive, but its strength varied considerably across different urban growth patterns.

4.4. Nonlinear Relationship Analysis Based on an Interpretable XGBoost-SHAP Model

While GTWR captures spatiotemporal heterogeneity in local coefficients, it is less suited to identifying nonlinear effect-transition thresholds. Therefore, XGBoost-SHAP is further employed to quantify nonlinear marginal effects and threshold behaviors.

4.4.1. Model Performance and Factor Contributions

An XGBoost regression model was used to examine the nonlinear effects of urban expansion variables on CCD. Hyperparameters were optimized using grid search with 3-fold cross-validation. The optimal settings were learning rate = 0.05, n_estimators = 100, and max_depth = 5. To reduce overfitting and improve generalization, random subsampling was applied (subsample = 0.7; colsample_bytree = 0.7). All predictors were standardized, and the dataset was randomly split into training and test sets (7:3). The model was implemented in Python 3.12.2 and achieved an MSE of 0.0039 on the test set, indicating satisfactory predictive performance.

SHAP values were further calculated to assess the relative importance of each predictor and to interpret their nonlinear associations with CCD. As shown in Figure 9a, USI had the largest mean absolute SHAP value, accounting for 52.2% of the total contribution, indicating that variation in CCD was more strongly associated with the sprawl dimension than with the other two expansion dimensions within the model framework. UC and UEI ranked second and third, contributing 38.2% and 9.6%, respectively. These results suggest that spatial structure variables, especially USI and UC, were substantially more important than expansion intensity alone in explaining model-predicted differences in CCD. In particular, the much smaller contribution of UEI implies that expansion speed by itself was less informative than spatial organization in distinguishing pollution–carbon coordination outcomes across cities. Percentage contributions were obtained by normalizing the mean absolute SHAP values across predictors. It should be noted that the thresholds identified from the PDPs in the following analysis represent model-implied effect-transition points in the XGBoost-SHAP framework, rather than structural threshold estimates in a causal or econometric sense.

4.4.2. Nonlinear Relationships

Based on the XGBoost-SHAP framework, the nonlinear response patterns of urban expansion variables to CCD were further identified, and effect-transition points were characterized using partial dependence plots (PDPs) (Figure 9). Overall, the effects of USI, UC, and UEI on CCD were not monotonic, but changed across different value intervals. The most analytically meaningful finding is that the associations between urban expansion dimensions and CCD varied markedly by range: spatial structure variables corresponded to more distinct shifts in coordination outcomes, whereas the effect of expansion intensity was comparatively weaker.

As shown in Figure 9b, USI exhibited a pronounced nonlinear relationship with CCD. When USI was below 0.58, its SHAP contribution remained negative, indicating that relatively low levels of spatial expansion were generally associated with lower pollution–carbon coordination; when USI exceeded 0.58, the contribution turned positive, suggesting that a moderate degree of spatial expansion was more likely to coincide with higher CCD. This pattern implies that very limited spatial expansion was not necessarily associated with stronger coordination, whereas moderate expansion was more often linked to better functional organization. UC showed a similarly clear threshold pattern. When UC was below 0.16, its contribution to CCD was negative; once UC exceeded 0.16, the contribution became positive, indicating that higher compactness was generally associated with more favorable coordination outcomes after a certain level had been reached. This suggests that the advantages of compact development do not emerge at low levels of concentration, but become more evident once compactness reaches a sufficient threshold. By contrast, UEI displayed a relatively weaker nonlinear pattern. Its contribution was positive when UEI was below 0.00, but became negative when UEI exceeded 0.00, indicating that slower expansion was generally associated with higher CCD, whereas excessively rapid expansion was more often associated with weaker coordination. Overall, these results suggest that, compared with expansion intensity, spatial structure variables showed more pronounced nonlinear relationships with CCD, and that the threshold characteristics of USI and UC were of greater analytical significance.

4.4.3. Heterogeneity of Nonlinear Effects Across Urban Growth Patterns

Given differences in development priorities and resource endowments across urban growth patterns, it is necessary to examine the heterogeneity and nonlinear mechanisms underlying pollution–carbon synergy. Based on UEI, USI, and UC, this study classified 289 cities into four urban growth patterns (HLE, LLE, HHC, and LHC) using cluster analysis. An XGBoost-SHAP model was then constructed with USI, UEI, and UC as predictors and CCD as the response variable to identify the major contributors to pollution–carbon synergy and their heterogeneous effects across urban growth patterns.

For HLE cities, the most noteworthy result is that CCD was more strongly associated with compactness than with expansion intensity or sprawl (Figure 10a). Among the three variables, UC showed the strongest contribution, and its effect became clearly more favorable when UC exceeded 0.11. This suggests that, within HLE cities, insufficient compactness is a more critical constraint than expansion speed itself. In practical terms, rapid expansion does not necessarily correspond to weaker coordination; rather, the more analytically meaningful distinction lies in whether such expansion is accompanied by a sufficiently compact spatial structure. By contrast, UEI contributed relatively little and displayed only weak variation, indicating that expansion intensity alone had limited explanatory value in distinguishing CCD outcomes within this type.

Another notable finding is that USI exhibited a type-specific threshold pattern that differed from the pooled-sample result. When USI remained below 0.50, its association with CCD was more favorable, whereas values above 0.50 corresponded to weaker coordination performance. This indicates that, for HLE cities, additional spatial dispersion beyond a certain level was more often associated with reduced pollution–carbon coordination, rather than improvement. Spatially, positive SHAP values for UC and USI were more frequently observed in eastern coastal cities, whereas negative values were concentrated in central-western and northeastern areas (Figure 10b). This pattern suggests that even within the same HLE category, structurally similar expansion does not correspond to the same coordination outcome across regions; instead, the implications of rapid and low-efficiency expansion vary markedly with regional development conditions.

For LLE cities, CCD was more strongly associated with USI and UC than with expansion intensity (Figure 11a). Among the three variables, USI showed the clearest threshold effect. When USI was below 0.79, its contribution to CCD remained negative, whereas values above this threshold were associated with more favorable coordination outcomes. This indicates that, in LLE cities, excessively weak spatial expansion was more often associated with lower CCD, while a moderate increase in spatial expansion corresponded to better coordination performance. UC also displayed a relatively stable positive pattern. Once UC exceeded 0.15, its contribution became more favorable, indicating that improvements in compactness were consistently associated with stronger pollution–carbon coordination in this city type. By contrast, UEI contributed relatively little and showed only limited variation, suggesting that expansion speed itself was not the main factor distinguishing CCD differences within LLE cities.

Figure 11b shows clear regional differences in the SHAP values of the three variables. Positive SHAP values for USI were more concentrated in parts of eastern and southern China, whereas negative values appeared more frequently in central-western and northeastern regions. For UC, positive effects were more evident in eastern China, especially in the Pearl River Delta and the Yangtze River Delta, while weaker or negative effects were more common in areas with less developed spatial organization. This pattern suggests that even within the same LLE category, cities with relatively better compactness and functional organization were more likely to exhibit higher CCD. Overall, the LLE pattern can be characterized as a low-speed, low-efficiency, and weak-agglomeration structure, in which differences in USI and UC were more closely associated with CCD than differences in UEI.

For HHC cities, CCD was more strongly associated with USI and UC than with expansion intensity (Figure 12a). Among the three variables, USI showed the clearest threshold pattern. When USI was below 0.85, its contribution to CCD remained relatively weak or negative, whereas values above this threshold were more often associated with favorable coordination outcomes. This indicates that, within HHC cities, a sufficiently high level of spatial expansion was more likely to coincide with stronger coordination performance. UC also showed a relatively stable positive association. Once UC exceeded 0.18, its contribution became more favorable, suggesting that higher compactness was generally associated with better CCD in this type. By contrast, UEI contributed relatively little and showed only limited variation, indicating that expansion intensity itself was not the main factor distinguishing coordination differences within HHC cities.

Figure 12b further shows pronounced regional variation in SHAP values. Positive SHAP values for USI were more concentrated in central-western and northeastern China, whereas the positive effect of UC was more evident in central-western and southwestern regions. In contrast, both effects were relatively limited in the eastern coast and in the Pearl River Delta and Yangtze River Delta. This pattern suggests that even among HHC cities, stronger coordination was not uniformly associated with the same spatial dimension across regions: in some areas, coordination outcomes were more closely linked to the expansion dimension represented by USI, whereas in others they were more closely associated with compactness. Overall, the HHC pattern can be characterized as a high-speed, high-efficiency, and high-compactness structure, in which the differences in USI and UC were more closely associated with CCD than the differences in UEI.

For LHC cities, CCD was more strongly associated with USI and UC than with expansion intensity (Figure 13a). Among the three variables, USI showed the largest contribution and the clearest threshold pattern. When USI remained below 0.93, its contribution to CCD was relatively weak, whereas values above this threshold were more often associated with favorable coordination outcomes. This indicates that, within LHC cities, only when spatial expansion reached a relatively high and orderly level did USI correspond more clearly to higher CCD. UC ranked second and showed a relatively stable positive pattern. Once UC exceeded 0.29, its contribution became more favorable, indicating that higher compactness was generally associated with stronger pollution–carbon coordination in this type. By contrast, UEI contributed the least and displayed only limited variation, suggesting that expansion intensity itself had little ability to distinguish coordination differences within LHC cities.

Figure 13b reveals that the positive SHAP values of both USI and UC were mainly concentrated in central-western and southwestern China, while weaker or negative effects were more common in the eastern coast and parts of northeastern China. This spatial pattern suggests that in central-western and southwestern LHC cities, differences in orderly expansion and compactness were still closely linked to CCD variation, meaning that spatial structure remained an important dimension of differentiation. In contrast, in eastern coastal areas, where compact development conditions were already relatively mature, the marginal variation associated with USI and UC was more limited. Overall, the LHC pattern can be understood as a low-speed but already compact development structure, in which compactness provides the basic condition for coordination, while further differences in CCD depend more on whether expansion remains orderly rather than on how fast cities grow.

5. Discussion

5.1. Nonlinear Relationships Between Urban Expansion and CCD

One of the central findings of this study is that CCD is more strongly associated with urban spatial structure than with expansion speed, and that these associations are clearly nonlinear. Because CCD jointly reflects the coordinated status of pollution and carbon emissions, mechanisms that shape carbon-emission outcomes through urban form (e.g., transport demand, land-use efficiency, and agglomeration) may also influence CCD through shared pathways [79,80,81]. This study finds significant nonlinear and threshold effects of USI, UC, and UEI on CCD, which is consistent with prior empirical evidence on the relationship between urban spatial form and carbon emissions [82]. Yin and Yao (2024), based on 281 Chinese cities, reported an inverted U-shaped relationship between urban compactness and carbon emissions, suggesting that a moderate level of compactness is most conducive to emission reduction, whereas excessive compactness or dispersion may increase emissions; this effect is further moderated by city size and regional characteristics [30]. Xing et al. (2024) further noted that compact spatial forms tend to reduce carbon emissions in small and medium-sized cities, but may increase energy use and emissions in large and megacities when compactness becomes excessive. In addition, factors such as industrial diversity and jobs-housing balance may significantly regulate this relationship [83]. Consistent with these findings, this study also identifies a positive effect on CCD under moderately compact urban spatial conditions, while further showing that expansion speed exhibits a threshold effect: in cities with high compactness and relatively rapid expansion, faster expansion may be associated with traffic congestion and energy pressure, thereby corresponding to less favorable pollution–carbon coordination. Using a panel threshold model for 295 Chinese prefecture-level cities, Ding et al. (2022) found a double-threshold effect of urban compactness on CO₂ emissions, with the mitigation effect weakening significantly beyond a certain compactness level; this is consistent with the threshold characteristics identified in this study [29]. From a broader land-use perspective, Searchinger et al. (2017) emphasized the importance of land carbon opportunity cost and land-use patterns in determining overall mitigation potential, which provides theoretical support for understanding the interaction between spatial configuration and carbon emissions during urban expansion [58].

Furthermore, by incorporating USI into the analytical framework of urban expansion dynamics, this study finds that its negative effect becomes particularly pronounced beyond the identified threshold, whereas at lower sprawl levels, moderate spatial expansion is more often associated with more favorable pollution–carbon coordination [26,30]. These nonlinear and threshold patterns may be related to differences in city size, regional economic development level, and industrial structure [28]. For example, eastern coastal cities are typically characterized by relatively high compactness and rapid expansion, and their CCD tends to respond more strongly to spatial agglomeration efficiency and industrial upgrading; by contrast, many central and western cities exhibit slower expansion and looser spatial structures, with comparatively weaker CCD responses. Overall, through a multi-indicator and multi-threshold analysis, this study extends existing perspectives and provides new empirical evidence on the differentiated associations between urban spatial structure, expansion dynamics, and pollution–carbon synergy, highlighting the need for governance strategies that consider the complex interaction between urban form and expansion pace.

5.2. Differential Effects Across Urban Growth Patterns

The four urban growth patterns identified in this study exhibit marked differences in how urban expansion is associated with pollution–carbon coordination, which is broadly consistent with previous evidence on the heterogeneous environmental implications of urban spatial structure and development patterns [11,26,28]. Existing studies have shown that agglomeration level, morphological complexity, and functional organization may shape carbon-emission pathways differently across cities, indicating that the environmental consequences of urban expansion are not spatially uniform [26,29,30]. Related research has also suggested that the effects of urban sprawl on environmental performance tend to vary with infrastructure conditions, development stage, and regional context [27,28]. These findings are generally in line with the type-specific heterogeneity observed in this study (Figures S1 and S2).

A clearer contrast can be observed between relatively compact and efficient growth patterns and those characterized by lower structural efficiency. In particular, the comparison between HHC and HLE more directly illustrates how differences in spatial structure and growth quality correspond to variation in CCD. More specifically, the results indicate that the factors more closely associated with CCD differ substantially across urban growth patterns. In HLE cities, coordination outcomes are more closely associated with compactness and the sprawl dimension than with expansion intensity itself, suggesting that structural efficiency may be more closely associated with CCD differences than expansion speed alone in this type. In LLE cities, both USI and UC show clearer associations with CCD than UEI, indicating that weak spatial organization and insufficient compactness are more closely related to CCD variation than slow expansion alone. In HHC cities, USI and UC remain more strongly associated with CCD than UEI, but their spatial patterns differ across regions, suggesting that the structural basis of coordination varies even within relatively efficient and compact development modes. In LHC cities, compactness appears to be more consistently associated with favorable coordination conditions, while further differences in CCD are more closely associated with whether spatial expansion remains sufficiently orderly. Taken together, these results suggest that the environmental implications of urban expansion depend not only on how fast cities grow, but also on whether growth is accompanied by coordinated improvements in spatial organization and development quality.

5.3. Policy Implications

From an urban planning perspective, the findings suggest that, compared with expansion speed alone, spatial structure and expansion quality may deserve greater attention when considering pathways to improve pollution–carbon coordination. Since USI and UC show stronger and more stable associations with CCD than UEI, the results may provide support for greater planning attention to the spatial organization of urban growth, especially the balance between compactness, orderly expansion, and land-use efficiency. In this sense, urban planning may benefit from placing more emphasis on how built-up land is arranged, connected, and integrated with urban functions.

More specifically, differentiated planning priorities may be considered across urban growth patterns. For HHC cities, where coordination levels are already relatively high, one possible planning consideration is how to maintain compact and efficient development while avoiding the potential weakening of coordination under further expansion. For HLE cities, the main concern lies in the combination of rapid growth with low compactness and relatively high sprawl, suggesting that planning attention in such cities may be more usefully directed toward spatial structure and land-use efficiency than toward expansion speed alone. For LHC cities, compactness already constitutes an important structural advantage, and the more critical issue is whether low-speed expansion can remain sufficiently orderly to support further coordination improvement. For LLE cities, weak compactness and insufficient spatial organization appear to be more important constraints than slow expansion itself, which may imply that greater planning attention could be given to the enhancement of spatial concentration and functional organization. Overall, the results may provide a basis for considering more differentiated planning approaches, rather than relying solely on uniform expansion control. In practice, these priorities may be translated into more targeted actions in transport planning, land-use regulation, and industrial restructuring, but the specific policy design should depend on local development conditions and planning capacity.

5.4. Limitations and Future Research Directions

Although this study uses GTWR and XGBoost-SHAP to capture spatial non-stationarity and nonlinear effects, several limitations remain. First, urban spatial structure and expansion dynamics were primarily measured using nighttime light data and administrative-level statistics. Although these data provide strong temporal continuity and nationwide comparability, their spatial resolution may constrain the ability to capture fine-scale urban morphological variation, functional land-use differences, and neighborhood-level density patterns, especially in smaller cities or highly built-up urban cores. Future research could incorporate finer-grained traffic-flow data, functional-mix indicators, and high-resolution land-use data, while also considering variables such as industrial structure and energy-use behavior to improve mechanism interpretation. Second, this study is based on a static panel framework and does not fully identify the long-term impact pathways through which urban expansion dynamics affect CCD. Future studies may strengthen policy evaluation by applying dynamic panel models and causal inference approaches. Finally, although many international studies have quantified the relationship between urban form and carbon emissions, the spatial–topological effects of urban expansion under different national and cultural contexts remain underexplored [80,84,85]. Cross-country comparative research would help develop more generalizable low-carbon urban expansion strategies.

6. Conclusions

Based on an empirical analysis of 289 prefecture-level and above cities in China, this study systematically examined how urban growth patterns are associated with the coupling coordination degree (CCD) between pollution and carbon emissions. The main findings are as follows:

(1)

USI and UC exhibit significant nonlinear threshold effects on CCD. Moderate spatial expansion and an appropriate level of compactness are associated with higher CCD, whereas excessive dispersion or excessive compactness tends to suppress pollution–carbon synergy.

(2)

The effects of USI and UC are substantially stronger than those of UEI. Their stronger associations with CCD may be related to differences in land-use efficiency, commuting distance, and infrastructure configuration.

(3)

Significant spatial heterogeneity exists across urban growth patterns. HHC cities show the highest CCD levels, whereas LLE and HLE cities are more often associated with structural inefficiency and lower expansion quality, which correspond to comparatively less favorable pollution–carbon coordination outcomes.

(4)

Results from GTWR and XGBoost-SHAP suggest that more organized forms of spatial expansion are more often associated with favorable pollution–carbon coordination outcomes, whereas expansion intensity alone appears to be less informative in distinguishing CCD differences.

Overall, the findings provide empirical support for place-based urban planning and differentiated governance considerations in promoting pollution–carbon coupling coordination and provide both theoretical and policy implications for green and low-carbon urban development.