Evaluating the Performance of Sustainable Urbanization and Its Impacts on Carbon Emissions: A Case Study of Nine Pearl River Delta Cities

Abstract

China is presently placing significant emphasis on sustainable urbanization as a means to facilitate the shift towards high-quality economic development. The concept of sustainable urbanization has gained added intricacy and depth within the framework of ‘carbon peak and carbon neutrality’. A primary concern among practitioners is that while carbon emissions adhere to specific standards, the term sustainable urbanization remains vaguely defined, encompassing multifaceted objectives. The absence of a well-defined evaluative standard for sustainable urbanization creates challenges in driving comprehensive progress. Additionally, owing to the spatially heterogeneous nature of urbanization, its influence on carbon emissions can vary across different cities. This study delves into the shared factors influencing the performance of sustainable urbanization, introducing a Principal Component Analysis (PCA)-based evaluation system to gauge sustainable urbanization performance. Furthermore, we employ spatial regression analyses to explore the spatial differences in the impact of these factors on carbon emissions. Our investigation centers on data from nine cities in the Pearl River Delta, allowing for the ranking of these cities based on their sustainability performance. The outcomes reveal that the key factors influencing carbon emissions differ among cities due to variations in sustainable urbanization characteristics. Notably, our research integrates sustainable urbanization with the parameters of a low-carbon economy. In the realm of policymaking, we offer a quantifiable approach for assessing sustainable urbanization. Furthermore, we assert that cities at distinct stages of sustainable urbanization should prioritize different factors to attain carbon neutrality.

1. Introduction

China’s remarkable economic progress over the past few decades has been predominantly attributed to its swift urbanization. However, this rapid urbanization phase has brought about pressing issues such as resource scarcity, environmental pollution, and population density, which have been exacerbated, leading to an unsustainable developmental trajectory [1]. Consequently, the Chinese government is actively working to devise and enact several policies aimed at transitioning from the outdated model of economic growth to a high-quality paradigm, placing significant emphasis on achieving sustainable urbanization. Simultaneously, the global shift towards a low-carbon economy and China’s resolute commitments to “carbon peak and carbon neutrality” have delineated a clear path for sustainable urbanization [2]. Merging the objectives of sustainable urban development with those of a low-carbon economy is of paramount importance, necessitating a meticulous examination of the influence of sustainable urbanization on carbon emissions.

Measuring sustainable urbanization, however, poses a challenge that is far from straightforward. Past researchers have endeavored to establish a multidimensional framework for evaluating sustainable urbanization. For instance, Tan et al. [3] devised a framework encompassing six distinct facets: policy, housing, population growth, energy supply, land development, and environmental protection. Zhong et al. [1], on the other hand, elaborately constructed a five-dimensional assessment of urbanization quality, encompassing population urbanization, economic development, ecological environment, urban–rural heterogeneity, and basic public service quality. Despite the comprehensive nature of these systems, they lack a streamlined and elegant scoring mechanism for gauging sustainability at an aggregate level. In practical application, the usability of these evaluation standards has diminished due to their complexity.

Further complexity comes from the interplay between sustainable urban development and carbon emissions. In previous studies, the low-carbon scenario has been included in sustainable urbanization, and the role of urbanization is highlighted in the study of carbon emissions ([3,4,5]). However, most studies mainly focus on the national and provincial setting and have said little about how factors affect carbon emissions at the city level in the context of a more specific spatial variation in sustainable urbanization. Given the vast divergence of development among cities in China, sustainable urbanization factors affect carbon emissions in cities with different performances.

In light of this, the core objective of this research is to explore the factors that affect carbon emissions in cities with different performances in sustainable urbanization by taking nine Pearl River Delta cities as case studies.

This research is advanced by two sub-objectives:

(1) To propose an evaluation system to measure the sustainable urbanization performance at the city level. This system is based on Principal Component Analysis, which incorporates all the current measure dimensions.

(2) To examine the spatial heterogeneity of factors impacting city-level carbon emissions based on the varied sustainable urbanization performances using regression and correlation analyses.

Our proposed system is illustrated using data from nine cities from China, but our framework is extendable to incorporate any dimensions. The proposed framework can be generalized to analysis of similar problems in other cities and provide guidance for policymaking.

2. Literature Review

2.1. Sustainable Urbanization

Urbanization refers to the expansion of urban areas and the exploitation of land used for physical space construction [6]. Meanwhile, the urbanization process is accompanied by population migration flows in which people of a certain generation are leaving their homes in rural areas and moving to live and work in urban areas [3]. With its powers to concentrate economic activity and people, as well as the concurrent growth in public services, social welfare, infrastructures, and institutions, urbanization has been regarded as the primary engine of economic growth and social function enhancement [6]. Urbanization makes urban areas prosperous, diversified, and lively. However, sometimes they are overbuilt or overexpanded, resulting in two typical scenarios: a high-density city [7] (Yuan et al., 2017) or a ghost city [8]. The imbalance between economic development, social functioning efficiency, and environmental protection is demonstrated by these iconic situations during rapid urbanization growth.

In this context, sustainable urbanization has emerged as a concept for coping with the unsustainable issues of rapid urbanization growth [3]. The concept of sustainable urbanization can be comprehended through a holistic view. First, sustainable urbanization is a dynamic process where urban socioeconomic issues faced or caused by cities are being solved. New solutions and approaches are created to ensure the sustainable usage of economic, social, and environmental resources [6]. Second, sustainable urbanization is viewed as a kind of urbanization strategy to govern the urbanization process under the participation of multiple actors with a long-term goal of a ‘harmonious society’ [9]. Third, sustainable urbanization is fundamentally a multidimensional endeavor, encompassing various theoretical inspirations and empirical practices in monitoring and assessing the development conditions and levels of society. For example, de Jong et al. [10] identified new categories of ‘city’ as the city development strategies towards the goal of sustainable urbanization based on the review of scholarly work and policy discourse, such as Smart City, Green City, Knowledge City, Resilient City, and Livable cities. Building on de Jong et al.’s work [10], we recognize that sustainability transcends singular dimensions like low-carbon energy production and consumption; it demands a holistic integration of economic efficiency, social equity, ecological resilience, and institutional innovation, requiring careful balancing of trade-offs between these interconnected priorities. Tan et al. [3] established a framework for evaluating sustainable urbanization in China that considers six different aspects: policy, housing, population growth, energy supply, land development, and environmental protection. Bian et al. [11] combined the perspectives of origin and modernization in examining sustainable urbanization performance, with the elements of nature and culture falling under the perspective of origin and the elements of the economy, society, and intelligence falling under the perspective of modernization. Zhong et al. [1] comprehensively developed five aspects to assess urbanization quality, and they are population urbanization, economic development, ecological environment, urban–rural heterogeneity, and basic public service quality.

The understanding of sustainable urbanization is multifaceted. To better understand and achieve sustainable urban development, there is a need to position sustainable urbanization in a broader and more open context in which developing challenges from the global and regional dimensions will reshape and affect the city’s sustainable urbanization rather than focusing just on sustainable problems encountered or caused by the city itself.

2.2. The Integration Between Sustainable Urbanization and Low-Carbon Society

The global trend of low-carbon economic growth has pointed out a clear direction for sustainable urban development [2]. The transition of sustainable urbanization [12] is consequently taking place toward a low-carbon society, serving as a pathway to optimize the conditions of energy utilization and emissions in different fields of society, such as transportation and infrastructure construction. In the research on sustainable urbanization, the studies related to the objective of a low-carbon society have attracted more and more attention. For instance, Tan et al. [3] highlighted energy efficiency and renewable energy as essential aspects of the practice of sustainable urbanization. Li and Liu [4] emphasized energy consumption as the key criterion for assessing the level of sustainable urbanization.

Meanwhile, China is the world’s second-largest economy, and its influence on global energy transition and low-carbon economic growth is enormous. China’s target for carbon neutrality by 2060 has also aroused scholars’ interest in analyzing issues of carbon emissions in the study of sustainable urbanization. For instance, Chang and Chen [5] examined the spatio-temporal patterns of carbon emissions during the process of sustainable urbanization in eastern China, revealing a significant negative correlation between carbon emissions and land economic efficiency. Sun et al. [13] analyzed the influencing factors of carbon emissions under China’s commitments to carbon neutrality, suggesting that sustainable development capacity including real GDP and industrial structure would positively impact carbon emissions.

According to the existing studies of sustainable urbanization and low-carbon society, we can reach a consensus on the interplay between sustainable urban development and carbon emissions. In these studies, the low-carbon scenario has been included in sustainable urbanization, and the role of urbanization is highlighted in the study of carbon emissions. However, most studies from the above fields mainly focus on the national and provincial setting and have said little about how factors affect carbon emissions at the city level in the context of a more specific spatial variation in sustainable urbanization. Therefore, this research seeks to explore the factors that affect carbon emissions in cities with different performances in sustainable urbanization. This core research objective will be decomposed into two sub-objectives, including (1) evaluating the sustainable urbanization performance at the city level and (2) examining the spatial heterogeneity of factors influencing city-level carbon emissions based on the different performances of sustainable urbanization.

2.3. An Overview of Geographical Statistical Method and Its Application in Solving Issues of Sustainable Urbanization and Carbon Emissions

2.3.1. An Overview of the Geographical Statistical Method

The application of statistical methods in analyzing spatial problems or phenomena has been an essential part of quantitative geography [14]. As there is a wide range of statistical methods involved in quantitative geography research, the application of statistical methods can be summarized into the following two aspects: the domains in which statistical methods can improve the geographical study, and the geographical data or spatial information that can enrich statistical methods.

First, the domains in which statistical methods can improve the geographical study encompass two sub-areas, including extrapolating the distribution of a set of spatial features and measuring the spatial relationship between features. On the one hand, the distribution of spatial features involves many types, such as clustered, dispersed, or random distribution of observed points. There are four main statistical approaches that can provide support to extrapolate the patterns of spatial feature distribution, including nearest-neighbor analysis, quadrat analysis, kernel density analysis, and Ripley’s K-function [14,15]. On the other hand, the measurement of the spatial relationship between observed features is based on the First Law of Geography [16], trying to unfold the potential correlation between one observed spatial object and another. Containing the global and local measure tests, spatial autocorrelation is recognized as the effective approach to assessing the spatial relationship of the observed spatial object. Furthermore, apart from the spatial correlation that is in line with the First Law of Geography, the notion of spatial relationship can be extended to another important character of spatial features which is spatial heterogeneity. Developed from the multivariate regression approaches, Geographically Weighted Regression is widely used to deal with the spatial heterogeneity of observed objects in different sub-areas of the research area by considering the spatial conditions of objects.

Second, for the geographical data or spatial information that can enrich statistical methods, we should focus on the manipulation and management of geospatial data (or geospatial big data) in promoting an understanding of spatial phenomena in geographical studies and other social science studies [17]. The classic sources of data collection involve statistical books, questionnaires, and on-site surveys, which might take an abundant amount of time to process. Due to the advancement of technologies, many techniques such as mobile devices (GPS tracker devices), satellites, and environment monitoring devices emerged and can be applied to track and collect data in multiple forms (such as figures, point data, line data, and areal data) in a more efficient, dynamic, and continuous way, significantly facilitating the power of location and space of data in the statistical methods [18]. However, the situation of data deluge may easily occur when the data-poor environment is transforming into a data-rich environment [19]. Therefore, we need to manage the geospatial data before manipulating it. There are three main approaches to realizing the management of geospatial data. Principal Component Analysis and factor analysis are two ways to conduct the dimensionality reduction of data, trying to resource the dimensions of data while avoiding the risk of losing the important data. Cluster analysis is another way to reduce the volume of data by detecting meaningful and useful groups (clusters) of data with similar features and reducing the row space [20]. To sum up, the geographical statistical method is of significance in scientific geographical studies and practice, contributing to quantitative knowledge with a positivist epistemology and avoiding ‘just the facts and experiences’ in the analytical process of geographical issues [21].

2.3.2. Application of Geographical Statistical Methods in Analyzing Sustainable Urbanization and Impacts of Urbanization on Carbon Emissions

Existing studies on sustainable urban development share one point in common regarding the two-step assessment of sustainable urbanization performance, which consists of designing evaluation indicators of sustainable urbanization as the first step and computing an index indicating the level of sustainable urbanization performance based on the indicators as the second step. The indicators are the elements that comprise the evaluation framework of sustainable urbanization performance. Nevertheless, methods used to determine the numbers, dimensions and weights of indicators and build a model to calculate the evaluation index are various. For example, Shen et al. [22] designed an indicator system covering social, economic, and environmental layers and applied the Entropy method to calculate the development and coordination index indicating the speed and quality of urbanization in Jinan, China. Then, the index results were displayed through a hybrid model that integrated the Entropy method and McKinsey Matrix. Xu et al. [23] stated that the Entropy method has limitations in combining the evaluation index or scores of cities into a synthetic value that represents the comprehensive performance of sustainable urbanization of an urban agglomeration. As a result, after developing an indicator system, they used a Full Permutation Polygon Synthetic Indicator approach to calculate the index of sustainable urbanization performance of 20 Chinese urban agglomerations.

In addition, urbanization plays a vital role in carbon emissions. Many studies have considered numerous urbanization indicators in their research on the factors that affect carbon emissions. Different geographical statistical methods have been applied to elaborate on the dynamic relationships between urbanization and its impacts on carbon emissions. For instance, Martínez-Zarzoso and Maruotti [24] used the Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model to analyze the effects of urbanization on carbon emissions in 88 developing countries from 1975 to 2003 and their findings suggested an inverted U-shaped relationship between the level of urbanization and carbon emissions. Yang et al. [25] conducted Global Moran’s I analysis to measure the spatial autocorrelation of carbon emissions in provinces with high-energy-intensive (HEI) industries in China, demonstrating a significant positive spatial autocorrelation of carbon emissions in HEI industries. Then, the local indicators of spatial association (LISA) were used to identify the high–high, low–low, and high–low spots (spatial clusters) of carbon emissions in HEI industries. The high–high and low–low spots are provinces in HEI industries that have positive local-spatial autocorrelation, while the high–low spots are those that have negative local–spatial autocorrelation. After that, the Geographically Weighted Regression model was employed to reveal the spatial variation in influencing factors on carbon emissions, which showed that urbanization is an important driving force to increase carbon emissions with its impacts on population changes, transportation, and infrastructure construction. Li et al. [26] used the Theil index to decompose China’s regional variation in carbon emissions and then performed Pearson correlation analysis to show the correlation between carbon emissions and other urbanization-related factors. According to the findings of Li’s research, there is a negative correlation between carbon emissions per capita (CCE) and the proportion of employees in the tertiary sector. In contrast, there is a positive correlation between CCE and the proportion of employees in the secondary sector.

The literature reviewed above indicates the importance of geographical statistical methods in addressing issues with evaluating sustainable urbanization performance and the impacts of urbanization on carbon emissions. Based on the research gap identified through the literature review about the integration between sustainable urbanization and low-carbon society (see Section 2.2), this research seeks to apply Principal Component Analysis to evaluate sustainable urbanization performance. Then, this research will employ Geographically Weighted Regression to identify factors that affect carbon emissions at the city level under a more specific spatial variation in sustainable urbanization. This research is anticipated to have policymaking implications for promoting China’s sustainable urbanization that is low-carbon-focused.

3. Methodology and Data Source

3.1. The Indicator System of Sustainable Urbanization

According to the literature, the five most common dimensions to evaluate the performance of sustainable urbanization include economy, society, people, living environment [22,23,27,28] and knowledge (including technology and innovation) [10]. To advance systematic assessment, this study refines the conceptual scope of the economy and people dimensions while operationalizing precise indicators across all dimensions. The resulting framework, summarized in

Table 1. Indicator system for sustainable urbanization.

Dimension	Indicator	Unit	Rationale of Indicator Selection
Economic growth	GDP per capita	%	GDP per capita directly indicates the level of economic development
Sustainable industrialization	Advanced degree of industrial structure	N/A	Advanced degree of industrial structure indicates the formation of a sustainable industrial structure, which will support the high-quality development of the economy.
Social vitality	Total raster pixel values of nighttime lights	N/A	Total raster pixel values of nighttime lights refer to the accumulation value of nighttime light intensity of total pixels in an observed region [29] (Yuan et al., 2022). It is the spatial data that reveals the social vitality and the concentration of social activities.
Population development	Population density	people/km2	The population density directly indicates population development which is the essence of urbanization.
Labor resource	Number of employees per unit at the end of the year	million people	The number of employees per unit at the end of the year reveals the level of sustainable urbanization because employment relates to job creation, prosperity, capital accumulation, and poverty reduction during the process of urbanization.
Living environment	Green area of parks	ha	Green space in parks contributes to the improvement of the living environment.
Knowledge	Financial expenditure in science and education	million CNY	Financial expenditure in science and education implies the level of appreciation of knowledge in society.

, comprises seven dimensions: economic growth and sustainable industrialization (economy-related), social vitality (society-related), population development and labor resources (people-related), living environment, and knowledge. Economic growth is measured by GDP per capita, reflecting regional productivity, while sustainable industrialization is captured through the advanced degree of industrial structure, which signals sectoral efficiency and innovation. Social vitality, as the society-related dimension, is quantified via total raster pixel values of nighttime lights, serving as a proxy for socioeconomic activity density. Population development and labor resources—the two people-related dimensions—are assessed via population density and the number of employees per unit, respectively, capturing demographic dynamics and workforce sustainability. The living environment dimension is evaluated through the green area of parks, emphasizing ecological quality, while the knowledge dimension is gauged by financial expenditure in science and education, reflecting institutional support for innovation. Building on this comprehensive indicator framework, we can therefore conceptualize sustainable urbanization as a reconfiguration of economic and social activities into high-quality urban systems, enabled by enhanced livability and knowledge-driven innovation.

3.2. Using PCA for Assessment of Sustainable Urbanization Performance

One criticism of using a multidimensional system to evaluate sustainable development is that it is not clear how much weight should be put on each dimension and it is not a unified scoring system. As a consequence, it is difficult to be used as an effective performance evaluation system. In this research, we explore using Principal Component Analysis (PCA) to accomplish this task. Principal Component Analysis involves using matrix operations to reduce the case-by-variable data into its essential features, called principal components. PCA reduces the data dimensions and maximally explains the variance of all variables at the same time. In other words, PCA enables us to ‘regroup’ the data set into a smaller number of ‘components’ [3,26].

Following the existing studies on using PCA for evaluation purposes [26,30], our PCA-based weighting system is designed as follows (shown in Figure 1).

In the PCA, we keep the component with an eigenvalue greater than one which is adequate to describe the data set as the principal components (PCs). These components are each expressed as a linear combination of original variables (OVs). In these linear combinations, the coefficients of each original variable are calculated using the eigenvalues of retained components and the component loadings (as shown in Step 4 in Figure 1, Coefa). For each OV, their respective coefficients are different across different PVs; we then proceed to combine the coefficients for the same OV into one coefficient for the composite score model, using the formula in Step 5 in Figure 1, where we call it Coefb. After obtaining the Coefb for each OV, we use the formula in Step 6 of Figure 1 to calculate the final weights of each OV. The assessment model will consequently be built based on the weights obtained. The whole procedure grasps the important factors beneath the loosely organized OVs using a mathematically rigorous procedure. And each OV’s contributions are well documented and can be used as a unified scoring system to evaluate the sustainable urbanization performance of each city.

3.3. Research Area

Our research area focuses on the Pearl River Delta (PRD) region, which consists of nine cities: Guangzhou, Shenzhen, Dongguan, Foshan, Zhuhai, Huizhou, Zhongshan, Jiangmen, and Zhaoqing. PRD is one of the most socioeconomically developed urban agglomerations in China [31], with a high urbanization rate of 87.5% in 2022. Due to the rapid industrialization, carbon emission in PRD has witnessed a notable increase in recent years [32].

PRD is pioneering in realizing China’s goal of carbon peak and carbon neutrality. However, the nine cities vary in the level of sustainable urbanization due to the uneven resource allocation and urbanization development. As a result, the nine Pearl River Delta cities represent an ideal case to explore factors that affect carbon emissions in cities with different performances in sustainable urbanization.

4. Results

4.1. Principal Component Analysis of Evaluation Indicators

This research adopts a PCA method in analyzing seven evaluation indicators for sustainable urbanization performance using SPSS version 20 and the relevant data in 2019. To ensure the statistical suitability of PCA, this research first conducted Kaiser–Meyer–Olkin (KMO) and Bartlett’s sphericity statistical tests. The KMO value of 0.720 (exceeding the threshold of 0.6) and Bartlett’s test significance (0.000) confirmed the adequacy of the data set for PCA. The result of the total explained variance of PCA is shown in

Table 2. Total explained variance.

Component	Initial Eigenvalues			Extraction Sums of Squared Loadings
	Total	% of Variance	Cumulative %	Total	% of Variance	Cumulative %
1	5.052	72.165	72.165	5.052	72.165	72.165
2	1.223	17.471	89.636	1.223	17.471	89.636
3	0.355	5.067	94.703
4	0.222	3.175	97.879
5	0.112	1.603	99.482
6	0.022	0.315	99.797
7	0.014	0.203	100.000

Extraction method: Principal Component Analysis.

. According to the extraction rule of ‘eigenvalues greater than 1’, Component 1 and Component 2 can be considered sufficient to describe the data set, with their respective eigenvalues as 5.052 and 1.223. The cumulative percentage of the variance of the retained components is 89.636% of the total variance after retaining Component 1 and Component 2.

Table 3. Rotated component matrix.

Variable	Component
	1	2
Population density	0.987	−0.083
Number of employees per unit at the end of the year	0.941	0.151
Financial expenditure in science and education	0.921	0.016
Advanced degree of industrial structure	0.902	0.075
GDP per capita	0.874	−0.333
Green area of parks	0.869	0.361
Total raster pixel values of nighttime lights	0.027	0.978

Extraction method: Principal Component Analysis.

reports the rotated component matrix of the PCA. As we can see, the population density, number of employees per unit at the end of the year, financial expenditure in science and education, advanced degree of industrial structure, GDP per capita, and green area of parks have high loadings on Component 1. Component 1 can hence be named ‘material bases for sustainable urbanization’ [23]. Component 2 can be named ‘vitality level of sustainable urbanization’ since the total raster pixel values of nighttime lights have high loadings.

4.2. Determination of Weight of Evaluation Indicators

We further express Component 1 and 2 as the linear combination of the original variables (following Step 4 of Figure 1). The final equations are listed below, with coefficients presented in

Table 4. Coefficients of the linear combination of variables in Component 1 and 2.

Variable	Component
	1	2
Population density $(x_1$)	0.439	−0.075
Number of employees per unit at the end of the year $(x_2$)	0.419	0.136
Financial expenditure in science and education $(x_3$)	0.410	0.014
Advanced degree of industrial structure $(x_4$)	0.401	0.068
GDP per capita $(x_5$)	0.389	−0.301
Green area of parks $(x_6$)	0.387	0.327
Total raster pixel values of nighttime lights $(x_7$)	0.012	0.884

.

$$y_1=0.439x_1+0.419x_2+0.410x_3+0.401x_4+0.389x_5+0.387x_6+0.012x_7$$

(1)

$$y_2=-0.075x_1+0.132x_2+0.014x_3+0.068x_4-0.301x_5+0.327x_6+0.884x_7$$

(2)

In Equations (1) and (2) (or in Table 4), each variable (like population density) has its respective coefficients in the two equations. These two sets of coefficients can further be combined using the procedure in Step 5 of Figure 1. The result coefficient of each variable is shown in

Table 5. Coefficients in the composite score model.

Variable	Coefficient
Population density $(x_1$)	0.339
Number of employees per unit at the end of the year $(x_2$)	0.364
Financial expenditure in science and education $(x_3$)	0.333
Advanced degree of industrial structure $(x_4$)	0.336
GDP per capita $(x_5$)	0.254
Green area of parks $(x_6$)	0.375
Total raster pixel values of nighttime lights $(x_7$)	0.182
Sum of coefficients	2.183

.

4.3. Sustainable Urbanization Performance of Nine PRD Cities

In

Table 6. Weight of indicator.

Dimension	Indicator (I)	Weight (W)
Economic growth	GDP per capita	0.116
Sustainable industrialization	Advanced degree of industrial structure	0.154
Social vitality	Total raster pixel values of nighttime lights	0.083
Population development	Population density	0.155
Labor resource	Number of employees per unit at the end of the year	0.167
Living environment	Green area of parks	0.172
Knowledge	Financial expenditure in science and education	0.152

, we normalize the coefficients in Table 5 (using the formula in Step 6 of Figure 1). This produces the weights of each original viable (indicators) to score the performance of the cities.

The evaluation results for the performance of nine PRD cities’ sustainable urbanization can be calculated using Equation (3).

$$Performance Score=\left({\sum }_{i=1}^7{\displaystyle \frac{I_i}{max\left(I_{i,cluster}\right)}}\times W_i\right)\times 100$$

(3)

We summarize the results in

Table 7. Sustainable urbanization performance of nine PRD cities.

Rank	City	Performance Score
1	Shenzhen	87.76
2	Guangzhou	86.76
3	Dongguan	55.57
4	Foshan	45.00
5	Zhuhai	43.24
6	Huizhou	37.94
7	Zhongshan	35.16
8	Jiangmen	33.39
9	Zhaoqing	27.84

.

With a score of 87.76, Shenzhen ranks first on the list of nine cities. The second-placed city is Guangzhou, which has a score of 86.76. However, there is a noticeable gap between the score of Guangzhou and that of Dongguan, and Dongguan ranks third, with a score that is 35.9% lower than Guangzhou’s. Foshan and Zhuhai are in fourth and fifth place, with scores of 45.00 and 43.24, which are lower than the average score of 50.30. Huizhou, Zhongshan, Jiangmen, and Zhaoqing comprise the remaining cities in the evaluation ranking.

These nine cities can be put into three groups/clusters. Shenzhen and Guangzhou are in the first group with satisfactory results for sustainable urbanization. Dongguan, Foshan, and Zhuhai are included in the second group with moderate performance. Huizhou, Zhongshan, Jiangmen, and Zhaoqing comprise the third group with subpar performance.

4.4. Spatial Heterogeneity of Factors Influencing City-Level Carbon Emissions

The above PCA system quantified the performance of each city overall. However, to further investigate the spatial heterogeneity of influencing factors, or how each factor influences each city differently, a spatially aware regression analysis is called for. We adopt the Geographically Weighted Regression (GWR) model for this purpose [25,32]. GWR constructs separate equations by incorporating the dependent and explanatory variables of the features falling within the neighborhood of each target feature. The shape and extent of each neighborhood analyzed is based on the Neighborhood Type and Neighborhood Selection Method parameters.

In this study, city-level carbon emissions serve as the dependent variable, representing a key indicator of sustainability. Of course, the carbon emission is just one of the prominent indicators of sustainability. Our methodology and approach developed here can be generalized when other factors (like green GPD output) are employed.

The main idea behind the GWR model is while every city should be governed by the same physical law, the effect of policy preference causes diffident factors to be singled out. The regression model will disclose the regression of adherent cities.

To ensure methodological robustness given the limited sample size (nine cities), we focus on three crucial indicators. These three indicators are GDP per capita as the representative indicator of economic growth, green area of parks as the representative indicator of living environment and ecological construction, and total raster pixel values of nighttime lights as the representative indicator of social vitality and the concentration of socioeconomic activities.

4.5. Analysis of Influencing Factors on Carbon Emissions Under the Spatial Variation in Sustainable Urbanization

Data Source

The research period is from 2003 to 2019. For the statistical data, GDP per capita and green area of parks are obtained from the China City Statistical Yearbook (2003–2019).

The data on carbon emissions is collected from the Open-Data Inventory for Anthropogenic Carbon Dioxide (ODIAC) data set (https://db.cger.nies.go.jp/dataset/ODIAC/DL_odiac2020b.html (accessed on 9 December 2022).

For geospatial data, the total raster pixel value of nighttime lights is the product of average nighttime lights per pixel and the total pixels of the observed area. The nighttime light data from 2003 to 2013 can be obtained from the DMSP-OLS annual data set (Version 4), while that from 2014 to 2019 can be obtained from the NPP-VIIRS annual data set (Version 2) [33,34]. DMSP and VIIRS annual data sets can be downloaded from the online platform of Earth Observation Group (https://eogdata.mines.edu/products/vnl/ (accessed on 9 December 2022).

4.6. Results of Ordinary Least Squares Model and Global Moran’s I Test

Before conducting the GWR model, we perform the Ordinary Least Squares (OLS) analysis and the Global Moran’s I test to demonstrate that GWR analysis is suitable to our problem [25,35]. The procedure for creating the GWR model is shown in Figure 2.

We used ArcGIS to perform Ordinary Least Squares regression analysis and the Global Moran’s I test. The value of the Koenker (BP) Statistic is not significant, according to the OLS diagnostics report on the three time periods of 2003, 2011, and 2019. Thus, we should instead check the Joint F-statistic. The report shows that the value of Joint F-statistic in the OLS diagnostics report is significant, demonstrating the significance and spatial non-stationarity of the OLS model’s output. As shown in

Table 8. Adjusted R-squared and VIF of the OLS model.

Year	Adjusted R-Squared	Explanatory Variable	VIF
2013	0.896	GDP per capita	1.137
		Total raster pixel values of nighttime lights	1.036
		Green area of parks	1.118
2011	0.821	GDP per capita	1.721
		Total raster pixel values of nighttime lights	1.236
		Green area of parks	2.024
2019	0.664	GDP per capita	2.327
		Total raster pixel values of nighttime lights	1.660
		Green area of parks	2.445

, the values of adjusted R-squared of the model are 0.896, 0.821, and 0.664, which indicates a good fit of the predictive regression model to the data set. The values of variance inflation factor (VIF) of explanatory variables are between 1 and 5, indicating moderate correlation between the explanatory variables. This result confirms that there is no multicollinearity in the model and that the explanatory variables are appropriate for GWR model analysis.

The standardized residual (StdResid) is selected as the input field for the Global Moran’s I test, and the resulting feature of the OLS model is selected as the input feature class. As shown in

Table 9. Results of Global Moran’s I test.

Year	Moran’s Index	Variance	z-Score	p-Value
2003	−0.274033	0.03736	0.77105	0.44068
2011	−0.276982	0.039132	0.76829	0.442313
2019	−0.238834	0.043532	0.54559	0.585347

, the result of Global Moran’s I is insignificant, demonstrating the random distribution of the standardized residual and the spatial non-stationarity of the regression model. Therefore, it is necessary to create the GWR model to analyze the spatial heterogeneity of the impacts of explanatory variables on carbon emissions.

4.7. Results of Geographically Weighted Regression Model

The fixed Kernel type and the AICc bandwidth method are applied in GWR modeling. The output of the GWR model is shown in

Table 10. Results of GWR of nine PRD cities in 2003, 2011, and 2019.

Rank	Year	City	Local R2	Coef #1	Coef #2	Coef #3
1	2019	Shenzhen	0.8889	−257.961	15.831	3238.071
	2011		0.8880	93.813	306.250	2186.334
	2003		0.9346	288.428	253.703	131.563
2	2019	Guangzhou	0.8458	−4.713	91.651	2126.297
	2011		0.8881	94.416	306.331	2185.141
	2003		0.9347	288.335	253.625	131.560
3	2019	Dongguan	0.8638	−187.672	45.821	2814.770
	2011		0.8880	94.066	306.286	2185.818
	2003		0.9347	288.383	253.671	131.563
4	2019	Foshan	0.8100	137.434	206.658	1427.751
	2011		0.8881	94.606	306.447	2184.044
	2003		0.9347	288.036	253.636	131.617
5	2019	Zhuhai	0.8294	−214.696	132.789	2722.165
	2011		0.8880	94.050	306.406	2184.886
	2003		0.9346	288.024	253.721	131.641
6	2019	Huizhou	0.9448	−246.567	−19.754	3346.333
	2011		0.8879	93.850	306.165	2186.972
	2003		0.9346	288.689	253.663	131.506
7	2019	Zhongshan	0.8293	−157.030	139.500	2401.559
	2011		0.8880	94.170	306.387	2184.921
	2003		0.9347	288.108	253.692	131.618
8	2019	Jiangmen	0.7158	14.286	220.486	1742.582
	2011		0.8881	94.474	306.524	2183.557
	2003		0.9347	287.775	253.691	131.678
9	2019	Zhaoqing	0.9481	485.021	264.336	559.655
	2011		0.8882	95.216	306.556	2182.615
	2003		0.9347	287.860	253.569	131.631

. Almost all local R² values are close to or even above 80%, demonstrating that the GWR model fits the data set better than the OLS model. Generally, the green area of parks (Coef #3) has the most significant positive impact on carbon emissions, followed by the GDP per capita (Coef #1). This result implies a potential issue that green spaces, such as parks, may inadvertently increase emissions in some cases due to poor integration with urban systems—for example, fragmented parks requiring energy-intensive maintenance or displaced industrial activities to peripheral zones. The total raster pixel values of nighttime lights (Coef #2) have a comparatively moderate positive impact on carbon emissions. The GDP per capita had negative impacts on carbon emissions in Shenzhen, Guangzhou, Dongguan, Zhuhai, Huizhou, and Zhongshan in 2019, which means that economic development started to play an essential role in carbon emission reduction. The total raster pixel values of nighttime lights had negative impacts on Huizhou in 2019, which suggests that the concentration of socioeconomic activities has been constructive in carbon emission reduction.

Combined with Figure 3, Figure 4 and Figure 5, detailed information on the spatial heterogeneity of influencing factors on carbon emissions is unfolded. The GDP per capita had a prominently positive impact on carbon emissions in Huizhou in 2003 and Zhaoqing in 2011 and 2019 (Figure 3). This result highlights the need to enforce stricter emissions standards for traditional manufacturing and redirect investments to high-value, low-carbon sectors in these cities. The total raster pixel values of nighttime lights have a prominently positive impact on carbon emissions in Zhuhai in 2003. After that, the carbon emissions in Zhaoqing and Jiangmen received relatively significant positive impacts from the total raster pixel values of nighttime lights in 2011, and Zhaoqing was still significantly influenced by it in 2019 (Figure 4). This finding suggests the importance of nighttime economy governance in these cities to use nighttime light data to regulate energy-intensive commercial zones. The green area of parks had the most significant positive impact on carbon emissions in Jiangmen in 2003. Then, carbon emissions in Huizhou and Zhaoqing received the strongest positive impacts from the green area of parks in 2011 and 2019, respectively (Figure 5).

5. Discussion

This research investigates the factors influencing carbon emissions in cities with various sustainable urbanization performances. The results of the GWR model indicate that the impacts of the three explanatory variables on the carbon emissions of cities in the first tier of sustainable urbanization performance are not the strongest. To further discuss the most crucial factors affecting the carbon emissions of Shenzhen and Guangzhou, this research performed a Pearson correlation analysis. The results in

Table 11. Pearson correlation analysis between indicators and carbon emissions in cities.

	Shenzhen	Guangzhou	Dongguan	Foshan	Zhuhai	Huizhou	Zhongshan	Jiangmen	Zhaoqing
GDP per capita	0.569 *	0.889 **	0.319	0.830 **	0.817 **	0.850 **	0.786 **	0.914 **	0.862 **
Advanced degree of industrial structure	0.949 **	0.916 **	0.759 **	0.581 *	0.850 **	0.955 **	0.938 **	0.643 **	0.830 **
Population density	0.891 **	0.835 **	0.142	0.562 *	0.831 **	−0.590 *	0.711 **	0.866 **	0.869 **
Number of employees per unit at the end of the year	0.820 **	0.931 **	0.610 **	0.725 **	0.981 **	0.968 **	0.800 **	0.944 **	0.787 **
Total raster pixel values of nighttime lights	0.633 **	0.843 **	0.387	0.805 **	0.800 **	0.775 **	0.790 **	0.726 **	0.811 **
Green area of parks	−0.742 **	0.867 **	0.440	−0.692 **	0.705 **	0.160	−0.052	−0.914 **	−0.306
Financial expenditure in science and education	0.691 **	0.797 **	0.721 **	0.821 **	0.795 **	0.804 **	0.914 **	0.874 **	0.845 **

*. Correlation is significant at the 0.05 level (two-tailed); **. Correlation is significant at the 0.01 level (two-tailed).

indicate a strong positive linear relationship between the advanced degree of industrial structure and carbon emissions in Shenzhen, with a correlation coefficient of 0.949. There is a strong positive linear relationship between the number of employees per unit at the end of the year and carbon emissions in Guangzhou, with a correlation coefficient of 0.931. Meanwhile, a strong positive linear relationship between the advanced degree of industrial structure and carbon emissions also exists in Guangzhou, with a correlation coefficient of 0.916. The advanced degree of industrial structure and the number of employees per unit at the end of the year are indicators of sustainable industrialization and labor resource in sustainable urbanization, and they are the drivers of high-quality economic growth. Therefore, it can be considered that in China’s developed and wealthiest cities, the most important factors in carbon emissions are the upgrading of industrial structure and the employment situation of the whole society, rather than the basic factors such as GDP per capita.

Conversely, the results of the GWR model show that GDP per capita has the most significant impact on the carbon emissions of cities in the third tier of sustainable urbanization performance, such as Huizhou and Zhaoqing. Cities in the second tier of sustainable urbanization performance, such as Dongguan and Foshan, also received relatively strong impacts from GDP per capita on carbon emissions. From 2003 to 2019, the largest impacts of total raster pixel values of nighttime lights on carbon emissions shifted from cities in the second tier of sustainable urbanization performance (such as Zhuhai) to cities in the third tier (such as Zhaoqing and Jiangmen). The GDP per capita and the total raster pixel values of nighttime lights are the indicators under the dimensions of economic growth and the concentration of socioeconomic activities. Therefore, it can be considered that for the cities in the stage of rapid development and the pre-stage of sustainable urbanization, the most important factors in carbon emission are macro-economic performance and socioeconomic vitality.

In addition, the results of the GWR model reveal that the green area of parks has the strongest impact on the carbon emissions of cities in the third tier of sustainable urbanization performance (e.g., Jiangmen, Huizhou, and Zhaoqing). These cities are less developed than those in the first and second tiers of sustainable urbanization performance. According to the existing policy discourses and urban strategic plans in Jiangmen, Huizhou, and Zhaoqing, it is noticeable that those three cities have paid more attention to ecological civilization and highlighted the importance of eco-city construction to reinforce their comparative advantages. Therefore, eco-city construction can be an effective way to help such cities to achieve the goal of ‘carbon peak and carbon neutrality’.

6. Conclusions and Policy Implications

The achievement of China’s goal of ‘carbon peak and carbon neutrality’ should be processed in a stepwise manner based on different conditions of economic growth and urbanization. The evaluation of cities’ sustainable urbanization performance and the identification of influencing factors on carbon emissions in cities with different sustainable urbanization performances in this research can contribute to more precise and effective policy implications towards the goal of ‘carbon peak and carbon neutrality’. Below are differentiated policy suggestions for cities at various levels of sustainable urbanization. First, for cities in the first tier, policies should prioritize the transition from economic growth to quality-driven decarbonization. This can be achieved by accelerating industrial upgrades in high-value sectors (e.g., finance, electronics) through the principles of a circular economy. Implementing AI-driven energy management systems can optimize urban efficiency, while establishing green finance mechanisms, such as carbon trading platforms, will facilitate investment in sustainable practices. Second, for cities in the second tier, policies should focus on balancing economic vitality with ecological modernization. This involves retrofitting low-efficiency manufacturing and developing greenbelts between urban cores and peri-urban areas to limit sprawl. Additionally, promoting agro-tourism can help diversify local economies. Governments can leverage nighttime light data to monitor energy-intensive activities and redirect them toward low-carbon alternatives. Third, for cities in the third tier, the focus here should be on avoiding carbon-intensive pathways while utilizing ecological assets. Governments can leverage cities’ comparative advantages in environmental protection to aid in carbon emission reduction. For example, they can capitalize on green spaces by developing carbon-neutral industrial parks powered by distributed solar and biomass energy and promote eco-tourism to reduce reliance on fossil fuel-dependent agriculture, supported by farmer training programs. Finally, cross-tier synergies and regional coordination should be considered in policymaking at the regional level. Governments can link municipal carbon trading schemes across tiers to enable offset mechanisms, such as allowing third-tier cities to sell forest carbon credits to higher-tier cities. Additionally, creating a regional platform for first-tier cities to mentor those in other tiers on green technology deployment can foster collaboration and innovation on governing the sustainable urbanization process.

7. Limitations and Future Work

While this research provides valuable insights into the relationship between sustainable urbanization and carbon emissions in PRD cities, several limitations should be acknowledged to guide future research. First, the analysis focused on nine cities, which may limit the statistical robustness of PCA and GWR. A larger sample size, encompassing more cities across China, would enhance the generalizability of the findings and reduce sensitivity to outliers. Future studies could expand the scope to include secondary cities or other urban agglomerations (e.g., Yangtze River Delta, Beijing–Tianjin–Hebei region) to validate the evaluation framework. Second, due to the small sample size, only three key indicators, namely GDP per capita, green area of parks, and total raster pixel values of nighttime lights, were selected for the GWR model. This simplification may overlook other critical factors influencing carbon emissions, such as industrial energy consumption, transportation infrastructure, or renewable energy adoption. Incorporating additional variables could provide a more comprehensive understanding of spatial heterogeneity. Third, climate-related factors, such as temperature, precipitation, and natural vegetation cover, were not included in the analysis. These variables can significantly affect energy demands (e.g., heating/cooling) and carbon sequestration capacities, particularly in climatically diverse regions like China. Future research should integrate climate data to refine the assessment of urbanization’s environmental impacts.