Beyond Linearity: Uncovering the Complex Spatiotemporal Drivers of New-Type Urbanization and Eco-Environmental Resilience Coupling in China’s Chengdu–Chongqing Economic Circle with Machine Learning

Abstract

Rapid urbanization worldwide has led to ecological challenges, undermining eco-environmental resilience (EER). Understanding the coupling coordination between new-type urbanization (NTU) and EER is critical for achieving sustainable urban development. This study investigates the Chengdu–Chongqing Economic Circle using the coupling coordination degree (CCD) model to evaluate NTU-EER coordination levels and their spatiotemporal evolution. A random forest (RF) model, interpreted with Shapley Additive exPlanations (SHAP) and Partial Dependence Plot (PDP) algorithms, explores nonlinear driving mechanisms, while Geographically and Temporally Weighted Regression (GTWR) assesses drivers’ spatiotemporal heterogeneity. The results reveal the following: (1) NTU and EER levels steadily improved from 2004 to 2022, although coordination between cities still requires enhancement; (2) CCD exhibited a temporal pattern of “progressive escalation and continuous optimization,” and a spatial pattern of “dual-core leadership and regional diffusion,” with most cities shifting from NTU-lagged to synchronized development; (3) environmental regulations (MAR) and fixed asset investment (FIX) emerged as the most influential CCD drivers, and significant nonlinear interactions were observed, particularly those involving population size (HUM); (4) CCD drivers exhibited complex spatiotemporal heterogeneity, characterized by “stage dominance—marginal variation—spatial mismatch.” These findings enrich existing research and offer policy insights to enhance coordinated development in the Chengdu–Chongqing Economic Circle.

1. Introduction

Urbanization has become one of the most significant global trends in the 21st century, with the global urban population expected to increase by 2.2 billion by 2050 [1]. The unprecedented scale and speed of urbanization have significantly enhanced economic prosperity and improved human well-being. However, urbanization has also triggered serious ecological problems [2]. These include population concentration and construction land expansion, leading to losses in arable land, wetlands, and forests, fragmentation of ecosystems, and declining regional ecological resilience. These issues have triggered soil and water erosion, biodiversity loss, and degradation of ecological service functions [3] These hazards affect not only the natural environment [4] but also pose serious challenges to human quality of life and sustainable development. As the world’s second-largest economy, China’s urbanization rate reached 67.00% by the end of 2024, ranking among the highest globally [5]. Over the past 75 years, the urbanization rate of the resident population has increased by 56.36%, with an average annual growth of 0.75% [6]. However, rapid and extensive urban expansion has resulted in resource depletion and environmental pollution, placing further pressure on the eco-environment [7]. In response, the international community has emphasized the need to promote urbanization while focusing on ecological protection, social equity, and sustainable development to achieve high-quality urban growth [8]. Therefore, exploring the complex relationship between urbanization and the eco-environment at both global and national levels can offer valuable insights for coordinating urban construction and ecological conservation.

“New-type urbanization” (NTU), centered on a people-oriented approach, differs from traditional urbanization that mainly emphasizes urban construction. NTU emphasizes harmony between humans and nature, promoting integrated urban–rural development. It seeks to enhance urbanization quality and efficiency, while ensuring balanced resource use and ecological protection. Recognizing its significance for sustainable development, the Chinese government introduced the “14th Five-Year Plan” Implementation Plan for NTU in 2022, which highlights the need to strengthen eco-environmental protection and build urban ecological resilience [9]. In this context, China has prioritized eco-environmental resilience (EER) in the NTU process, aiming to enhance the adaptability, recovery, and sustainable development capacity of urban ecosystems through a comprehensive ecological policy system. This includes improving environmental protection mechanisms, developing green and eco-friendly cities, restoring ecological systems, promoting resource conservation and recycling, strengthening environmental monitoring and evaluation, and integrating urban and rural ecological governance. Additionally, green and smart urban construction, together with enhanced environmental legislation, further supports this effort [10,11]. Coordinating NTU and EER is thus essential not only for China’s high-quality development strategy but also for advancing global research on human–environment relationships and sustainability.

Numerous studies highlight complex interactions between NTU and EER. Initially defined in 1973, eco-environmental resilience refers to an ecosystem’s ability to absorb disturbances and maintain its essential functions [12]. Subsequent studies expanded this definition through systems ecology by including adaptive capacity, learning, and sustainable development potential [13,14,15]. With urbanization accelerating, EER has become critical to understanding urban social–ecological systems [16], specifically denoting a city’s ability to respond, adapt, and recover from ecological disruptions to reduce risks and restore ecosystem functions. The Pressure-State-Response (PSR) model, proposed by Rapport and Friend in 1979 [17], is now widely applied in EER and environmental health assessments [18]. In this model, “Pressure” describes the environmental stress caused by socioeconomic activities; “State” captures the condition and trends of ecosystems and human well-being; and “Response” addresses the actions taken to mitigate, restore, or prevent environmental damage. The model effectively answers the following questions: what happened, why it happened, and what we should do [19]. Since the 21st century, to support sustainable urban development and respond to evolving needs such as low-carbon cities [20], smart cities [21], sponge cities [22], EER has become a widely acknowledged research focus. It encompasses natural background conditions (e.g., resource endowment and topographic–climatic features), ecological quality (e.g., land use types) [23], and environmental governance (e.g., resource carrying capacity and regulatory oversight) [24,25], which together determine an ecosystem’s resistance, recovery ability, and learning capacity [26,27]. Urbanization places pressure on EER, while EER in turn provides feedback to urbanization by regulating ecological space, optimizing resource allocation, and enhancing ecosystem services. Without coordination of these complex interactions and feedback loops, sustainability issues may emerge. Therefore, clarifying this relationship not only enriches theoretical research on human–environment interactions but also supports informed policymaking for the sustainable advancement of new-type urbanization [28,29].

The interaction between urbanization and the eco-environment is a complex, multi-stage system rather than a simple linear process [30,31]. The Environmental Kuznets Curve (EKC) [32] initially provided an intuitive framework to explain the evolution of urban ecological conditions and guide phased policy interventions [33]. However, EKC has limitations, especially regarding non-economic factors such as institutional contexts and spatial structures. Ecological Modernization Theory (EMT) thus emerged, advocating continuous ecological improvements throughout the development process. Subsequently, the concept of “coupling,” initially derived from physics, was introduced to better understand urbanization–eco-environment interactions. In geography and environmental science, coupling describes mutual influences and feedbacks between systems moving toward harmony and integration [34,35]. With the advancement of research on human–land interactions under the Coupled Human and Natural Systems (CHANS) framework, numerous theoretical models have emerged, including the Social–Ecological System (SES) [36], the service–resilience feedback framework [37], the human–nature coupling system [38], and the remote coupling framework [39]. In terms of methodological tools, studies have employed classical coupling indices [40], dynamic coupling models [41], coupling formulas [42], coupling circles, and coupling devices [43]. Research findings suggest that the dynamic interaction between urbanization and the eco-environment generally aligns with the EKC curve, indicating a strong coupling coordination relationship [44]. These approaches go beyond traditional linear or “cascade” models, offering more complex representations of system behavior [45]. In recent years, increasing attention has been given to the coupling and coordination between NTU and EER at the urban agglomeration and regional scales. Studies reveal that their interaction is not strictly linear but features fluctuations and variability, with both NTU and EER, as well as their coupling coordination, exhibiting significant spatial heterogeneity [46,47,48]. Thus, the coupling and coordination between NTU and EER can be understood as a complex, nonlinear process driven by multiple interactive mechanisms [49].

Oriented by ecological civilization and high-quality development, research on the coupling of NTU and EER within urban areas has become urgently needed. It is crucial to analyze its spatiotemporal characteristics and identify scientifically based regional coordination paths for sustainable human–land relationships. In China, many studies have begun to try to construct assessment models applicable to the Chinese context. Among them, the “Coupling Coordination Degree” (CCD) model has become one of the more mainstream analytical methods. By establishing an “urbanization-ecosystem” coupling indicator system, this approach analyzes complex systems composed of multiple interacting indicators and dimensions [50]. It effectively quantifies the matching degree and nonlinear correlations between factors such as urban population density, construction land expansion, economic growth, and ecosystem services [51], thereby determining the developmental stages and trends of their coordinated progression. This kind of research has to some extent broken through the limitations of the earlier “univariate-single-indicator” analysis, and started to view the system as a whole of interactions and assess whether these interactions are harmonious or conflicting, which is crucial for reaching a harmony between the process of urban development and ecosystem protection [44].

However, there are significant limitations in the existing coupling analysis methods. Traditional CCD models rely on exponential weighting and linear calculation, which make it difficult to reveal the dynamic feedback and nonlinear mechanisms within the system. For example, when the expansion of urban land causes the loss of green space, does it inevitably lead to ecological degradation, and is there a certain threshold or window of adaptation? These questions have not been effectively modeled in existing studies. In terms of the identification of influencing factors, some of the studies still remain at the macro-related level, and lack in-depth analysis of the intrinsic mechanism of “what factors affect the coupling process” [52], which leads to a certain degree of ambiguity in the policy recommendations for ecosystem management [53,54]. From theoretical constructs to modeling methods, although existing studies have made progress in the study of coupling mechanisms, there are still many gaps, such as how to understand the dynamic tension between NTU and EER at the regional scale, and how to realize zoning classification and fine regulation in policy operation. These questions need to be supported by more explanatory models and theoretical frameworks that are more in line with local practices. The GTWR model, an advanced form of geographically weighted regression, effectively captures spatial and temporal variability, providing clearer insights into variable dynamics [55,56]. This model has been widely used in several research fields, such as ecosystems, urbanization levels, etc., and has become an effective tool for analyzing phenomena with spatiotemporal heterogeneity at coupled scales [57]. In addition, although some researchers have explored various drivers affecting the dependent variable, the driving mechanisms among them are not clear. For this reason, some more flexible data-driven methods have been introduced, such as the random forest [58] model in machine learning, which can identify the key factors affecting the coordinated evolution of urbanization and ecology [59], breaking through the reliance of the traditional methods on linear assumptions [60], which can reflect the dominant mechanisms behind the complex interactions more clearly, elucidating the contributions of the various driving factors, and providing a robust method for solving the limitations of the traditional analytical methods [61].

The Chengdu–Chongqing Economic Circle, as a key strategic region in western China, has been formally incorporated into the national NTU strategy. It has become an important growth pole for promoting high-quality national development and has made significant contributions to national competitiveness [62]. Its economic growth and urbanization rates have significantly exceeded the national average, reflecting strong development momentum. However, rapid economic, social, and urban development has brought severe challenges, including intense competition for environmental resources, disordered land use, and ecological degradation. These issues have seriously constrained the progress of NTU and the sustainable development of the eco-environment. Therefore, examining the coupling and coordinated development of NTU and eco-environmental resilience in the Chengdu–Chongqing Economic Circle holds important practical value for policymaking and implementation. There is an urgent need to investigate the current state of coupling coordination, identify underlying driving mechanisms, and explore strategies and pathways that can promote their mutual reinforcement [63].

Building on the above context and rationale, this study sets out the following research objectives: focusing on the Chengdu–Chongqing Economic Circle, it examines the coupling coordination relationship and driving mechanisms between NTU and EER, with particular emphasis on identifying potential nonlinear interactions and spatiotemporal heterogeneity between CCD and its influencing factors. To achieve these aims, evaluation indicator systems for NTU and EER are constructed for 16 cities in the Chengdu–Chongqing Economic Circle, covering the period from 2004 to 2022. The CCD model is applied to assess the degree of coupling coordination and to analyze its spatiotemporal evolution. Building upon this, a random forest (RF) machine learning algorithm is used, along with interpretability methods such as SHAP (Shapley Additive exPlanations) and PDP (Partial Dependence Plot), to quantify the relative importance and directional effects (positive/negative) of CCD driving factors, and to identify nonlinear influence patterns. Additionally, the GTWR model is employed to explore the dynamic spatiotemporal variation of these influencing factors.

The structure of this paper is as follows: Section 2 outlines the study area, data sources, research indicator system, and specific analytical methods and models. Section 3 presents the empirical findings, including CCD levels, classifications, spatiotemporal evolution patterns, and the relative importance, mechanisms, and spatiotemporal heterogeneity of CCD driving factors. Section 4 discusses the results, emphasizing policy recommendations for enhancing the coupling coordination of NTU and EER, while also highlighting the study’s innovations, limitations, and directions for future research. Section 5 concludes with a summary of the theoretical and practical contributions of this work. Ultimately, this study seeks to address gaps in existing literature through a clearly defined research focus and an innovative framework, providing scientific and practical insights for promoting coordinated urban–ecological development and sustainable urbanization within the Chengdu–Chongqing Economic Circle and similar urban agglomerations.

2. Materials and Methods

2.1. Research Framework

Building upon existing research, we propose a more innovative research framework (Figure 1). To investigate the coupling coordination relationship between NTU and EER and their driving mechanisms, we established a comprehensive indicator system for evaluating both NTU and EER systems. Using the CCD model, we quantified the CCD between NTU and EER and analyzed the spatiotemporal and typological evolution patterns of CCD. On this basis, the RF model is used to reveal the driving mechanism of CCD, and the model results are interpreted by combining the SHAP and PDP methods, a method that can more scientifically and accurately identify the relative importance of the drivers and the nonlinear interactions. Subsequently, the GTWR model is used to clarify the spatial and temporal variability of CCD drivers. This model, compared with the traditional GWR model, introduces the time dimension, which can more comprehensively reflect the dynamic change process of the influence magnitude and direction of the drivers in time and space.

2.2. Research Area

The Chengdu–Chongqing Economic Circle is located in Southwest China (103°36′ E–108°53′ E, 28°10′ N–31°26′ N), with Chengdu and Chongqing as its core cities. It encompasses 15 prefecture-level cities across Sichuan Province and the Chongqing Municipality (a municipality directly under the central government), covering a total area of 185,000 square kilometers (Figure 2). The region features complex and diverse topography, including mountainous, hilly, and basin landscapes. In 2004, Sichuan Province and Chongqing Municipality signed the “Framework Agreement on Strengthening Economic and Social Cooperation between Sichuan and Chongqing and Jointly Promoting the Development of the Upper Yangtze River Economic Zone,” marking the beginning of a more in-depth phase of regional collaborative development between the two areas. As a key area within China’s western development strategy, the economic output of the Chengdu–Chongqing Economic Circle grew from 6.3 trillion yuan in 2019 to 8.2 trillion yuan in 2023, reflecting a growth rate of 6.1%. In terms of spatial layout, the region has gradually developed a dual-core structure centered on Chengdu and Chongqing. According to statistics, the urbanization rates of Chengdu and Chongqing reached 80.8% and 72.14%, respectively, in 2024 [64,65], significantly exceeding the national average [66]. From 2004 to 2022, the regional coordinated development of the Chengdu–Chongqing Economic Circle underwent a new phase of spatial restructuring and institutional innovation, transitioning from strategic deepening to the construction of a national strategic growth pole.

Rapid urbanization has exerted dual pressure on land resources and the eco-environment. Over the past decade, the average annual expansion rate of urban construction land in the Chengdu–Chongqing region exceeded 5%, particularly in the Chongqing metropolitan area and around the Chengdu Plain, where land development intensity is especially high. On one hand, this growth has fueled regional economic development and industrial agglomeration; on the other hand, excessive development has intensified land resource stress and led to ecological degradation in certain areas. Moreover, with ongoing industrialization and urbanization, environmental issues such as air pollution and water scarcity have become increasingly prominent, posing significant challenges to the region’s sustainable development. Therefore, the Chengdu–Chongqing Economic Circle is selected as the study area. Investigating the coupling coordination relationship and driving mechanisms between its NTU and EER can support the advancement of green, innovative, and sustainable development strategies.

2.3. Data Sources and Pre-Processing

The period from 2004 to 2022 encompasses a significant cycle in the Chengdu–Chongqing region, from the initial establishment of regional collaboration mechanisms to the elevation to a national strategic level. This period also satisfies the requirements for data integrity and continuity. Selecting this period as the research timeframe aids in accurately reflecting the coordinated evolution mechanism of NTU and EER in the Chengdu–Chongqing Economic Circle under differentiated policy contexts. Therefore, this study has collected statistical and remote sensing data on socioeconomic and ecological environments from 16 cities in the Chengdu–Chongqing Economic Circle during this period. Data related to NTU indicators and CCD driving factors were sourced from provincial and municipal statistical yearbooks and bulletins. NDVI remote sensing data were obtained from the MOD13A3 dataset, regularly released by NASA (https://www.earthdata.nasa.gov/, accessed on 22 January 2025). Meteorological data were sourced from the National Centers for Environmental Information (https://www.ncei.noaa.gov/, accessed on 22 January 2025).

Given that some of the selected research indicators vary in scale and exhibit both positive and negative directional effects, we applied a dimensionless transformation using polar normalization. Separate normalization formulas were used for positive and negative indicators. The formulas are as follows:

$$\mathrm{Positive} \mathrm{indicator}:{ \mathrm{r}}_{\mathrm{ij}}={\displaystyle \frac{\left({\mathrm{X}}_{\mathrm{ij}}-{ \min \{\mathrm{X}}_{\mathrm{j}}\}\right)}{\left({\max \{\mathrm{X}}_{\mathrm{j}}\} -{ \min \{\mathrm{X}}_{\mathrm{j}}\}\right)}}$$

(1)

$$\mathrm{Negative} \mathrm{indicator}:{ \mathrm{r}}_{\mathrm{ij}}={\displaystyle \frac{\left({\max \{\mathrm{X}}_{\mathrm{j}}\} -{ \mathrm{X}}_{\mathrm{ij}}\right)}{\left({\max \{\mathrm{X}}_{\mathrm{j}}\} -{ \min \{\mathrm{X}}_{\mathrm{j}}\}\right)}}$$

(2)

where i is the year, j is the indicator, ${\mathrm{r}}_{\mathrm{ij}}$ is the standardized value, ${\mathrm{X}}_{\mathrm{ij}}$ is the original value, and max${\{\mathrm{X}}_{\mathrm{j}}\}$ and min${\{\mathrm{X}}_{\mathrm{j}}\}$ are the maximum and minimum values of indicator j, respectively. All standardized values are in the range [0, 1].

2.4. Evaluation Index System of NTU and EER

In order to explore the coupling coordination relationship between NTU and EER in the Chengdu–Chongqing Economic Circle, we developed a comprehensive, multi-dimensional indicator system based on existing literature and guided by the principles of scientific objectivity, representativeness, and validity. The NTU system is constructed across four dimensions: demographic urbanization, economic urbanization, social urbanization, and land urbanization [47,67]. Meanwhile, the EER system adopts the PSR model as its foundation, encompassing three dimensions: eco-environmental pressure, state, and response [49,68,69]. Based on considerations of multicollinearity and data availability, the NTU system ultimately includes 9 indicators across 4 dimensions (

Table 1. Chengdu–Chongqing Economic Circle NTU evaluation index system.

System Layer	Criterion Layer	Indicator Layer	Indicator Description	Attribute
NTU (new-type urbanization)	Population urbanization	Proportion of urban population (%)	Reflecting the proportion of urban population to the total population, it is an important indicator of the level of urbanization.	+
		Population density (people/km2)	Indicates the number of people per unit area, reflecting the degree of urban population concentration.	+
	Economic urbanization	GDP per capita (CNY)	Reflects the level of regional economic development and the economic productivity of residents and is the core indicator of economic urbanization.	+
		Urban residents’ disposable income (CNY)	Representing the level of residents’ income, it is a key indicator of the quality of life and consumption capacity.	+
	Social urbanization	Urban registered unemployment rate (%)	Describes the efficiency of urban labor market operation and employment stability; the lower the unemployment rate, the better.	−
		Public library collections per capita (volumes)	Reflects the level of cultural and educational resources enjoyed by urban residents.	+
		Share of education spending in fiscal spending (%)	Measures the strength of government investment in education and is an important indicator of social equity and education quality.	+
	Land urbanization	Proportion of built-up area to urban area (%)	Indicates the intensity of urban development; a key indicator of urban space expansion and land use efficiency.	+
		Urban road area per capita (m2)	Measures the level of urban transportation infrastructure, reflecting the convenience of residents’ travel.	+

), while the EER system comprises 9 indicators within 3 dimensions (

Table 2. Chengdu–Chongqing Economic Circle EER evaluation index system.

System Layer	Criterion Layer	Indicator Layer	Indicator Description	Attribute
EER (eco-environmental resilience)	Pressure resilience	Industrial wastewater emissions per capita (t)	Reflects the pressure of human activities on the environment of water bodies and is one of the important indicators of environmental pollution load.	−
		Industrial SO2 emissions per capita (t)	Indicates the intensity of sulfur dioxide pollution from industrial sources and is a key parameter for assessing air quality risk.	−
		Industrial soot and dust emissions per capita (t)	Measures the contribution of industrial emissions to atmospheric particulate matter pollution, reflecting the pressure on the urban air environment.	−
	State resilience	Green coverage in built-up areas (%)	Reflects visually the urban air quality and the health risk of the residents and is the core indicator of the state of the atmospheric environment.	+
		Park green area per capita (m2)	Indicates the level of urban ecological space construction. Reflects the function of improving microclimate and ecological regulation.	+
		Usable volume of water resources per capita (m3)	Measures the equity of urban green space and the degree of access to ecological welfare of residents.	+
	Response resilience	Hazard-free treatment rate of household garbage (%)	Reflects the freshwater resources available per capita, a key indicator of water ecological security.	+
		Urban sewage treatment rate (%)	Measures the capacity of urban solid waste management and the level of environmental management.	+
		Comprehensive utilization rate of industrial solid waste (%)	Reflects the ability for urban sewage collection and treatment, which is an important manifestation of water environment management.	+

). Additionally, indicators are classified as either positive or negative according to their effect on system development. A positive indicator contributes to the system’s improvement when its value increases, whereas a negative indicator hinders development as its value rises.

After constructing the evaluation index systems, we applied the entropy weighting method to calculate the weights of the indicators for both the NTU and EER systems within the Chengdu–Chongqing Economic Circle. This objective weighting technique helps eliminate subjective biases and enhances the scientific rigor of indicator weight assignment [70]. On this basis, the comprehensive development index of NTU and EER is calculated. The formula is as follows:

$${\mathrm{U}}_1={\displaystyle \sum }_{\mathrm{j}=1}^{\mathrm{m}}{\mathrm{X}}_{\mathrm{ij}}^{‘}{\mathsf{\omega }}_{\mathrm{j}}$$

(3)

$${\mathrm{U}}_2={\displaystyle \sum }_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{Y}}_{\mathrm{ij}}^{‘}{\mathsf{\omega }}_{\mathrm{j}}$$

(4)

where U₁ and U₂ are the combined development indices of NTU and EER, respectively. ${\mathrm{X}}_{\mathrm{ij}}^{‘}$ and ${\mathrm{Y}}_{\mathrm{ij}}^{‘}$ are the normalized values of NTU and EER, respectively. ${\mathsf{\omega }}_{\mathrm{j}}$ is the weights of the indicators.

2.5. Methodology

2.5.1. Coupling Coordination Degree Modeling

We employ the coupling coordination degree (CCD) model to analyze the coupling coordination and relative development levels of NTU and EER. The specific formula is as follows:

$$\mathrm{C}=2 \times {\left[\left({\mathrm{U}}_1{ \times \mathrm{U}}_2\right){/\mathrm{U}}_1{+\mathrm{U}}_2\right]}^{{\displaystyle \frac{1}{2}}}$$

(5)

$$\mathrm{T}=\mathsf{\alpha }{\mathrm{U}}_1+\mathsf{\beta }{\mathrm{U}}_2$$

(6)

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

(7)

$${\mathrm{E}=\mathrm{U}}_1{/\mathrm{U}}_2$$

(8)

where C is the coupling degree of NTU and EER. U₁ and U₂ are the composite indices of NTU and EER, respectively. T reflects the overall level of NTU and EER. $\mathsf{\alpha }$ and $\mathsf{\beta }$ represent the contributions of NTU and EER, respectively. Referring to the existing study [71,72], NTU and EER are equally important, so $\mathsf{\alpha }$ and $\mathsf{\beta }$ are set to 0.5. D is the CCD of NTU and EER. E reflects the relative development level of NTU and EER. Based on the existing studies [46,73], we categorize the level and type of CCD of NTU vs. EER into 5 levels and 15 types (

Table 3. Classification of CCD in NTU and EER.

CCD	CCD Level	Relative Development Level	Type
0 ≤ D ≤ 0.2	Severe imbalance	0 < E ≤ 0.8	NTU lag
	Severe imbalance	0.8 < E < 1.2	Synchronous development
		E ≥ 1.2	EER lag
0.2 < c ≤ 0.4	Moderate imbalance	0 < E ≤ 0.8	NTU lag
		0.8 < E < 1.2	Synchronous development
		E ≥ 1.2	EER lag
0.4 < c ≤ 0.6	Antagonism	0 < E ≤ 0.8	NTU lag
		0.8 < E < 1.2	Synchronous development
		E ≥ 1.2	EER lag
0.6 < c ≤ 0.8	Moderate coordination	0 < E ≤ 0.8	NTU lag
		0.8 < E < 1.2	Synchronous development
		E ≥ 1.2	EER lag
0.8 < c ≤ 1	High coordination	0 < E ≤ 0.8	NTU lag
		0.8 < E < 1.2	Synchronous development
		E ≥ 1.2	EER lag

).

2.5.2. Selection of Driving Factors for CCD

In order to study the influencing factors driving the coupling coordination of NTU and EER in the Chengdu–Chongqing Economic Circle and their spatial and temporal heterogeneity, we combine the existing studies and select two types of factors, i.e., socioeconomic and natural factors, with a total of nine indicators [74], as shown in

Table 4. Drivers of CCD between NTU and EER in the Chengdu–Chongqing Economic Circle.

Dimension	Indicators	Symbol	Indicator Description
Socioeconomic factors	Government intervention level	GOV	Government expenditure as a percentage of GDP
	Environmental regulations	MAR	Total retail sales of consumer goods as a percentage of GDP
	Fixed asset investment	FIX	Fixed asset investment as a percentage of GDP
	Degree of openness to the outside world	OPE	Foreign direct investment as a percentage of GDP
	Population size	HUM	Annual resident population at the end of the year
	Employment structure	EMP	Employment in the tertiary sector as a percentage of total employment
Natural factors	Precipitation amount	PRE	Average annual precipitation
	Normalized vegetation index	NDVI	Vegetation growth and coverage
	Average annual temperature	TEM	Average annual temperature

.

2.5.3. Random Forest (RF) and Interpretable Algorithms

The random forest (RF) model is one of the most widely used and powerful ensemble learning algorithms. It enhances model accuracy and robustness by constructing multiple decision trees and aggregating their predictions through voting or averaging methods [75,76]. In this study, the RF model is implemented using the sklearn toolkit in Python 3.8. To further analyze the complex driving mechanisms behind CCD, we incorporate two machine learning interpretability techniques: SHAP and PDP.

The core algorithm of SHAP quantifies the contribution of each individual feature and aggregates these contributions using an additive model to generate a comprehensive interpretation. It provides insights into model outcomes across both global and local dimensions [77]. PDP visualizes how specific features influence model predictions, clearly illustrating the marginal effect each feature has on the predicted outcomes. For complex black-box models (e.g., random forests, neural networks, etc.), PDP can provide meaningful and intuitive explanations [78].

2.5.4. Geographically and Temporally Weighted Regression (GTWR) Model

GTWR introduces the time dimension into the traditional geographically weighted regression (GWR) model, which well solves the limitation of the GWR model in observing the changes in the time series, so that it can more accurately reveal the heterogeneity of the variables in time and space [56,79]. As the socioeconomic resources and natural environment in the Chengdu–Chongqing Economic Circle region are dynamically changing in time and space [80]. Therefore, the GTWR model can be utilized to better identify the magnitude and direction of the influence of the drivers of CCD in different time and space. The specific model expression is as follows:

$${\mathrm{Y}}_{\mathrm{i}}={\mathsf{\alpha }}_0\left({\mathrm{u}}_{\mathrm{i}}{,\mathrm{v}}_{\mathrm{i}}{,\mathrm{t}}_{\mathrm{i}}\right)+{\displaystyle \sum }_{\mathrm{k}=1}^{\mathrm{n}}{\mathsf{\alpha }}_{\mathrm{k}}\left({\mathrm{u}}_{\mathrm{i}}{,\mathrm{v}}_{\mathrm{i}}{,\mathrm{t}}_{\mathrm{i}}\right){\mathrm{x}}_{\mathrm{ik}}+{\mathsf{\theta }}_{\mathrm{i}}$$

(9)

where Y_i denotes the CCD of NTU and EER of city i. (u_i, v_i, t_i) denotes the spatiotemporal coordinates of city i. α₀(u_i, v_i, t_i) denotes a constant, i.e., the intercept term, for city i. α_k (u_i, v_i, t_i) is the regression coefficient of xk. X_ik represents the CCD of city i driver x_k. θ_i is the random error term. n represents the number of drivers.

In constructing the GTWR model, to enhance model performance and result stability, this paper adopts the corrected Akaike Information Criterion (AICc) as the optimal bandwidth selection criterion. The bandwidth optimization process mainly includes the following three steps: First, a complete GTWR data framework is constructed based on the geographical coordinates and temporal variables of sample points in the study area. Second, under both fixed and adaptive bandwidth settings, the corresponding AICc metrics are calculated by progressively traversing different bandwidth values, and the model performance is compared. Finally, the adaptive bandwidth form with the smallest AICc value is selected as the optimal model setting. This selection takes into account the characteristics of uneven urban distribution and significant spatial density variations of samples in the study area, enabling the model to better adapt to the local variation characteristics of both spatial edge zones and dense zones.

3. Results

3.1. Comprehensive Development Level of NTU and EER in the Chengdu–Chongqing Economic Circle

3.1.1. Time Evolution Trend

Between 2004 and 2022, the NTU and EER levels in the Chengdu–Chongqing Economic Zone demonstrated steady improvement, albeit with slight variations in growth rates across different phases (Figure 3). The core cities, represented by Chengdu and Chongqing, have achieved continuous improvement in both dimensions, especially after 2015. Most of the remaining cities, such as Mianyang, Luzhou, and Yibin, on the other hand, showed an accelerated pace of development. Overall, the magnitude and pace of the two indexes still differ significantly among different cities, reflecting the existence of stage misalignment and structural imbalance in the development of the Chengdu–Chongqing Twin Cities Economic Circle, and the regional synergy still faces certain challenges.

3.1.2. Spatial Evolution Trend

From the perspective of spatial distribution, the development of NTU and EER presents the spatial hierarchy of “core cities taking the lead in making breakthroughs, with neighboring regions gradually following” (Figure 4). Between 2004 and 2022, the high-value NTU areas expanded outward from Chengdu to surrounding regions such as southern Sichuan and northeastern Chongqing, while the high-value EER areas clustered from Ya’an and Chengdu toward Chongqing, Zigong, and Luzhou. This indicates that during the rapid advancement of NTU, its growth primarily spread along transportation and economic nodes, whereas EER regressed toward mountainous regions with better ecological foundations, resulting in spatial dislocation between the two. Evidently, in the course of development, the Chengdu–Chongqing Economic Circle still needs to further coordinate its spatial layout to ensure synchronized expansion of NTU and enhancement of EER.

3.2. CCD Analysis of NTU and EER in the Chengdu–Chongqing Economic Circle

3.2.1. CCD Spatiotemporal Characteristics

The overall CCD level of the Chengdu–Chongqing Economic Circle has shown a continuous upward trend since 2004, reflecting the gradual enhancement of the synergistic relationship between NTU and EER (Figure 5). In 2004, most of the area was at a lower coupling level (D < 0.5), while by 2022, the high CCD area is gradually concentrating towards Chengdu, Chongqing, and its neighboring cities, forming a “twin core” with the coordinated development belt centered on the “double core”. Luzhou, Neijiang, Yibin, etc., showed a significant jump after 2016, indicating that the regional coordination mechanism is gradually taking effect under the impetus of policies, and the Chengdu–Chongqing region as a whole shows a coordinated evolution trend of “expanding from the core to the periphery”.

3.2.2. Evolution of Coupling Coordination Type

Figure 6a demonstrates that the CCD level exhibited an overall evolutionary trajectory of “progressive escalation and continuous optimization” between 2004 and 2022. In the early stages, most cities were in a state of “moderate imbalance” or “antagonism” displaying weak coordination. However, with the advancement of development strategies emphasizing both new urbanization quality and ecological construction, the number of regions achieving “antagonism” and “high coordination” significantly increased after 2016. Notably, Chengdu and Chongqing gradually approached the optimal level, with steadily improving intra-regional collaborative capabilities.

Further analysis of relative development types reveals that the equilibrium–lag relationship among cities in the Chengdu–Chongqing Economic Circle underwent a marked evolution during 2004–2022 (Figure 6b). From 2004 to 2015, most cities fell under the “NTU lag” type, while Neijiang distinctly differed from others as an “EER lag” type, indicating its urbanization outpaced ecological construction. By 2022, only Deyang, Guang’an, Nanchong, Neijiang, Suining, and Ziyang had transitioned to a “synchronous development” type in the region. This shift reflects enhanced synergy between urbanization quality and ecological construction under policy guidance prioritizing both. However, localized asymmetric development persists among many cities, with some still requiring further improvements in EER or NTU levels, leaving room for more balanced bidirectional quality development across the region.

3.3. Driving Mechanism of CCD

In order to study the mechanisms driving the coupling coordination of NTU and EER in the Chengdu–Chongqing Economic Circle, we used random forests and combined SHAP and PDP to identify the nonlinear interactions between each driver and CCD.

3.3.1. Model Training and Evaluation

Before constructing the random forest model, we tested the covariance between all independent variables by Pearson’s correlation coefficient, and all correlation coefficients were less than 0.8, which indicated that there was no multicollinearity between the indicators (Figure 7). We compared the random forest model with three models, decision tree, linear regression, and support vector machine, setting the same parameters. In total, 80% of the dataset was selected as the training set and 20% as the test set, and 5-fold cross-validation was used to ensure the stability of the model. The results show that random forest has the best performance in R², MAE, MSE, and RMSE (Figure 8). Therefore, random forest was chosen as our model and SHAP and PDP analysis were performed based on it.

3.3.2. Relative Importance of Drivers

Figure 9a,b show the relative importance and SHAP value distribution of the drivers, respectively.

Table 5. Mean SHAP values and relative importance results of drivers.

Dimension	Indicator	Average SHAP Value	Relative Importance (%)	Ranking
Socioeconomic factors	MAR	0.048	36.60	1
	FIX	0.047	36.19	2
	OPE	0.011	8.78	3
	HUM	0.007	6.55	4
	GOV	0.007	4.98	5
	EMP	0.004	2.75	6
Natural factors	TEM	0.003	2.20	7
	NDVI	0.002	1.24	8
	PRE	0.001	0.71	9

shows the results of the average SHAP values and the relative importance measured by the average SHAP values. The results show that among all the drivers, two socioeconomic indicators, MAR (36.60%) and FIX (36.19%), are the two most influential factors among the top five indicators, followed by OPE (8.78%), HUM (6.55%), and GOV (4.98%). TEM (2.20%) is the most influential factor among the natural factors.

From Figure 9b, it can be observed that both MAR and FIX show a significant positive effect on CCD. This indicates that the larger the proportion of total retail sales of consumer goods and fixed asset investment to total GDP, the more positively it contributes to the coupled harmonization of NTU and EER. However, when the proportion is small, it suppresses the coupling coordination degree of NTU and EER. The high values of SHAP (red points) of TEM, a natural factor, are mainly located on the left side of the y-axis, while its low values (blue points) are mainly located on the right side. This pattern suggests that there is a negative effect of TEM on CCD, i.e., the coupling coordination between NTU and EER will be suppressed as the annual mean temperature increases. Although the relative importance of NDVI is lower compared to other factors, it still plays a positive and active role in the coupling coordination degree of NTU and EER.

3.3.3. Nonlinear Relationships Between CCDs and Their Important Drivers

To further explore how these drivers are associated with the CCD, we selected the top five drivers affecting the CCD in order of relative importance (MAR, FIX, HUM, OPE, GOV) and used the PDP to analyze the nonlinear relationship between these factors and the CCD. The PD is particularly useful for understanding the relationship between the predicted results and the features, especially when the relationship exhibits a nonlinearity. Figure 10 shows how one feature interacts with another to affect the predicted value of the CCD. The results demonstrated noteworthy nonlinear interaction patterns of multiple significant driving factors on CCD.

We found that in the low-MAR region (0.5), the influence of FIX on CCD gradually increased. This reflects that as the consumer market continues to expand, the demand for green services has also gradually increased, driving green investment. The interaction effect between the two promoted the synergistic enhancement of NTU and EER, and this nonlinear pattern confirmed the core proposition of the Ecological Modernization Theory. Additionally, the nonlinear effect of HUM on CCD aligns with the “inverted U-shaped” pattern of the Environmental Kuznets Curve. From the perspective of HUM and GOV, when GOV is constant, as HUM gradually increases, the predicted CCD value gradually decreases, from 0.595 to 0.545. Moreover, the predicted CCD values in the low HUM region are generally greater than those in the high HUM region. FIX also exhibits a similar pattern. This indicates that when the population size is small, the scale effect of NTU can enhance resource utilization efficiency and promote CCD. However, when the population size becomes too large, population aggregation intensifies resource consumption, exceeding the environmental carrying capacity, triggering negative feedback that leads to a decline in EER and suppresses CCD.

3.4. Spatial and Temporal Variability of CCD Drivers

3.4.1. GTWR Model Construction

To thoroughly reveal the spatiotemporal differentiation characteristics of CCD driving factors, the extended GTWR model of GWR was selected for modeling analysis. In the model construction, we used nine driving factors (including six socioeconomic factors—GOV, MAR, FIX, OPE, HUM, EMP—and three natural factors—PRE, NDVI, TEM) as independent variables to build the GTWR model, optimizing the bandwidth parameter using the AICc criterion. On this basis, we further obtained the local regression coefficients of the variables across temporal and spatial dimensions for subsequent heterogeneity analysis.

To validate the applicability of the GTWR model, this study compares its performance with traditional OLS and GWR models (

Table 6. Comparison of OLS, GWR, and GTWR model performance.

Model	R2	Adjustment R2	AICc
OLS	0.617	0.606	−708.75
GWR	0.899	0.869	0.869–945.50
GTWR	0.982	0.974	−1367.01

). The model validation results show that the GTWR model has an AICc value of −1367.01, which is significantly better than those of OLS (−708.75) and GWR (−945.50). The results indicate that the GTWR model achieves a high R² of 0.982, with an adjusted R² of 0.974, significantly outperforming OLS (0.617) and GWR (0.899). Its AICc value is the lowest (−1367.01), demonstrating its significant advantage in revealing the spatial heterogeneity and non-stationary characteristics of CCD driving factors, and it can provide more precise spatiotemporal support for subsequent policy analysis and regional governance.

3.4.2. Temporal Heterogeneity

Figure 11 shows that, in terms of socioeconomic factors, GOV initially shows obvious negative inhibition, and then the negative effect gradually diminishes and maintains a weak negative direction in the late stage; MAR and FIX play a positive facilitating role in the advancement stage of NTU, but as economic growth enters into the middle and late stages, their pulling effect gradually declines and tends to converge; OPE imposes constraints on the coupling coordination in the lack of openness, but its role quickly turns into a positive–positive push as openness deepens; EMP impacts show multiple positive and negative alternations; and EMP impacts show multiple positive and negative alternations. With deepening openness, its role quickly turns into a positive–positive push; EMP’s impact is characterized by multiple positive–negative alternations: it plays a positive pulling role during economic expansion, and turns into a negative constraint during the stage of industrial adjustment or market fluctuation, and then returns to a positive direction; HUM maintains a positive effect on coupling coordination as a whole, but shows a negative inhibitory effect as the level of NTU rises and the size of the population increases, and overall shows a trend from enhancement to smoothness and then a slight decline. To summarize, GOV, MAR, FIX, OPE, and other factors are mainly reinforcing effects in the early stage, and generally attenuate or level off in the middle and late stages, while the effects of EMP and HUM are characterized by both stage fluctuations and stability.

Concerning natural factors, PRE has stage fluctuations on coupling coordination: initially, it is mainly a negative constraint, and then it turns into a significant positive promotion, and eventually its positive effect also gradually declines and tends to stabilize; the vegetation index NDVI continues to provide positive support in the early stage of ecological restoration, but with the improvement of the coverage and the ecological use efficiency, its positive effect will experience attenuation or even a brief negative fluctuation, and finally return to a stable positive effect; TEM and HUM have both stage fluctuations and stability. The influence of TEM on coupling coordination alternates several times, first showing weak negative constraints, followed by deepening of the negative effect, then turning to positive promotion, and finally tending to be negative again. Overall, PRE, NDVI, and TEM are able to bring obstacles to coupling coordination at different development stages, as well as provide key support when EER improves or green development advances. The nonlinear time-varying characteristics of natural factors on the coordination process of NTU and EER are highlighted.

3.4.3. Spatial Heterogeneity

Figure 12 shows that the regression coefficients of GOV, MAR, FIX, OPE, EMP, and HUM exhibit obvious spatial heterogeneity in terms of socioeconomic factors. Among them, GOV shows the strongest negative effect in Chongqing, gradually weakening to near zero toward the surrounding areas; the positive coefficient of MAR spreads to the periphery with Yibin as the center, and gradually decreases toward Chengdu, Meishan, and Chongqing; the positive effect of FIX is the most prominent in Chengdu and the western cities (e.g., Ya’an, Leshan, and Meishan areas), and significantly decreases to the east Sichuan riverine areas such as Luzhou and Chongqing; the positive effect of OPE is dominated by Nanchong–Dazhou–Guangan, and gradually decreases southwestward to Luzhou, Yibin, and toward Chongqing; HUM has the highest positive coefficients in the mountainous areas of southern Sichuan, such as Ya’an, Leshan, and Yibin, and gradually decreases toward Chongqing, Dazhou, Mianyang, and other cities; and EMP exhibits positive impacts in the cities of Ya’an, Meishan, Leshan, and Chengdu, and to Chongqing, Luzhou, it turns to a weak negative influence.

Regarding natural environmental factors, the effects of PRE, NDVI, and TEM also have significant spatial heterogeneity; PRE has the highest positive coefficients in Ya’an, Leshan and other western regions, and turns to nearly zero in the Chongqing Plain and the Northeast Sichuan Basin; NDVI has the most significant positive effect in Chongqing, and is negative or weakly positive in the northwestern cities of the study area; TEM has weakly negative coefficients in the cities in Northeast Sichuan, such as Nanchong, Dazhou, and Guangan, and weakly negative coefficients in the cities in Northeast Sichuan, and weakly negative in the cities of Chongqing, Luzhou, and Chengdu. The TEM has a weak negative coefficient in Nanchong, Dazhou, Guang’an, and other cities in northeast Sichuan, with an increasing negative effect toward the central Sichuan region (Chongqing and Suining), and the strongest negative effect in Yibin and Luzhou in the south.

4. Discussion

4.1. Spatial and Temporal Evolution Characteristics of CCD

The study indicates that during the period of 2004–2022, the CCD of the Chengdu–Chongqing Economic Circle presents a spatial diffusion pattern of “dual-core leadership and regional diffusion” with Chengdu and Chongqing as the core, which reflects the evolutionary trend of NTU and EER progressively and synergistically advancing within the region [28]. Similar spatial differentiation characteristics are also found in other urban agglomerations. For example, in the middle and lower reaches of the Yangtze River [81], although there is a significant spatial mismatch between the level of NTU and EER, the overall CCD level still increased during 2005–2020. The spatial distribution is a “block-like agglomeration”, the type of coordination is dominated by “basic coordination”, and there are differences in the dominant driving dimensions among cities, which reflects the complex interaction between the level of urban development and the ecological infrastructure conditions. In the case of Shaanxi Province, there are significant spatial differences in the types of CCD coordination within the region, and the development status is characterized by the co-existence of “NTU lag, synchronous development and EER lag”, reflecting the incomplete synergy between urbanization and ecological construction [82]. However, although the spatial and temporal evolution of CCD in the Chengdu–Chongqing area to some extent reflects the common trend of the development of urban agglomerations in China [48], such as the overall increase in the level of coordination and the co-existence of spatial heterogeneity, whether the diffusion pattern of “dual-core leadership and regional diffusion” is widely applicable still needs to be further verified through systematic comparisons and empirical studies in more regions. Further verification is needed through systematic comparison and empirical research in more regions.

4.2. Important Drivers and Nonlinear Interaction Mechanisms of CCDs

Clarifying the important drivers affecting CCD and determining the nonlinear relationship between them are crucial to promote the coordinated development of NTU and EER [83]. Therefore, we used RF combined with SHAP and PDP to analyze nine socioeconomic factors and natural factors in the Chengdu–Chongqing Economic Circle. The two socioeconomic factors, MAR and FIX, are the drivers that have the greatest impact on CCD. This means that the more active and larger the consumer market is, and the stronger the domestic demand power is, the more effective it is in promoting the coupling coordination development of NTU and EER. The development of the market economy can promote the process of NTU, while the upgrading of consumption promotes the high quality demand for EER, thus prompting urban construction and management to pay more attention to eco-friendliness [84]. In addition, the increase in the proportion of fixed asset investment, especially in the fields of green infrastructure and smart cities, has a positive effect on the environmental carrying capacity and EER of cities [71,85]. The improvement of infrastructure is also exactly the key to NTU.

TEM, as the most influential driver of CCD among the natural factors, shows a negative inhibitory effect on CCD. This may be attributed to the proposition that as the temperature rises, it usually has an inhibitory effect on vegetation [86], creating a heat island effect for the city, which in turn attenuates the EER. In addition, the NDVI reflects the growth of vegetation, and the degree of vegetation cover is critical to the eco-environment [74]. The abundance of PRE will affect the potential growth capacity of vegetation. These natural factors affect CCD by directly influencing EER; however, TEM, NDVI, and PRE showed relatively weak and stable effects in this study, which is in line with the expectations. This is because the importance of these factors lies in maintaining the basic functions and services of the ecosystem [87,88]. Only drastic changes (e.g., severe drought, large-scale destruction of vegetation) may bring significant impacts. However, the magnitude of changes in these drivers in the Chengdu–Chongqing region during 2004–2022 is small enough to have a large impact effect on the CCD of NTU and EER.

In addition, we find noteworthy nonlinear interaction mechanisms of the drivers on CCD. These relationship patterns can be explained by existing classical theories such as EKC and EMT. In the case of HUM, for example, the facilitating effect on the coordinated development of NTU and EER is stronger when the population size is lower, and the facilitating effect diminishes or even inhibits as the population size rises. It is not difficult to explain that population growth can promote NTU, but it directly exacerbates the consumption of resources such as water, land, and energy and the emission of environmental pollution, increasing the environmental load. At the same time the demand for urban space expansion, infrastructure, and public services (such as housing, transportation, education, health care) increases, crowding out ecological space, which will all pose serious challenges to the eco-environment [89,90]. When the eco-environment is damaged, it will inhibit NTU [91]. Therefore, appropriate population size is crucial for the coordinated development of NTU and EER. Not only that, but the results also show that the drivers interact with each other and act on CCD [92,93]. In terms of MAR and FIX, when MAR is lower, FIX enhancement has a weaker promotion effect on CCD. However, when MAR is higher, the promotion effect of FIX enhancement on CCD becomes significant. This reflects that the two promote each other and act to harmonize the development of NTU and EER.

4.3. Spatial and Temporal Heterogeneity of CCD Drivers

GTWR results reveal that multiple factors drive urban coupling coordination improvements in the Chengdu–Chongqing region, exhibiting substantial spatiotemporal heterogeneity. First, in the time dimension, socioeconomic factors such as GOV, MAR, and FIX have a significant positive influence in the pre-study period. This is consistent with the results of previous studies [48]. Earlier analyses of the Chengdu–Chongqing economic circle revealed that urban-scale and economic development served as primary driving forces during the early stages of coordinated evolution. However, their driving effects have diminished over time in some cities, indicating a declining marginal effect of external push mechanisms. Meanwhile, the green technology innovation capacity (Gtec) shows a significant spatial polarization distribution, with the driving effect increasing in some cities and decreasing in others, reflecting the complexity and non-equilibrium of the coordinated evolution path [29,94].

In addition, the trend of fluctuating changes in the intensity of the role of natural factors (e.g., NDVI, PRE) at different stages of the interaction between ecosystems and urban systems suggests that there is a characteristic of nonlinear evolution over time between ecological–urban systems [28]. It is further confirmed that there is a typical “inverted U-shaped” coupling relationship between NTU and EER, i.e., at the initial stage, NTU promotes the enhancement of EER, but with the deepening of the development, the pressure on EER increases, and the coupling relationship turns into a negative influence [95,96].

Spatially, the CCD driving mechanism shows significant differentiation among regions. The study shows that Chengdu and Chongqing, as two core cities, exhibit significantly higher regression coefficients across most driving factors compared to surrounding regions, demonstrating the agglomeration advantages brought by concentrated policy resources, industrial structures, and capital. On the other hand, the neighboring regions are constrained by ecological fragility and insufficient infrastructure, and the overall driving strength is low. This can be explained by the fact that when cities have better conditions in terms of location, industry, transportation, etc., they have stronger coupling bases, but the coordination process in regions with relatively poor conditions relies more on structural adjustment and technology accumulation [97]. In addition, different cities show “non-smooth” driving pattern differences at different development stages, which suggests that spatial variability and matching of development stages should be considered in the overall development strategy [98].

In conclusion, the spatiotemporal evolution path of CCD driving factors in the Chengdu–Chongqing Economic Circle exhibits multidimensional characteristics of “stage dominance—marginal variation—spatial mismatch.” It is imperative to strengthen phased regulation strategies, spatially adaptive pathways, and integrated mechanisms for internal and external factors in regional coordination policies, thereby effectively supporting the high-quality integrated development of the urban agglomeration.

4.4. Limitations and Future Research Prospects

This study reveals the spatial and temporal evolution of CCD between NTU and EER in the Chengdu–Chongqing Economic Circle, the driving mechanism of CCD and the spatial and temporal heterogeneity of the driving factors, and the application of RF and its related interpretation algorithm, GTWR model, which enriches the content of the existing research. Despite our innovation in research framework and methodology, there are still some limitations. First, the evaluation index system of NTU and EER still needs to be improved due to the availability and accessibility of data. Urbanization and the eco-environment are two huge and complex systems, so future research should focus on how to select more appropriate and comprehensive indicators to evaluate NTU and EER levels more comprehensively, scientifically, and accurately. Second, this study selected relevant socioeconomic and natural factors and explored the influence of these external factors on CCD, but did not focus on the influence of endogenous factors, so in the future, we can use the barrier degree model to analyze the barriers to CCD, and comprehensively consider the external and endogenous factors in evaluating the CCD. Lastly, we believe that in the future, we can start from different spatial scales, such as the county scale, pixel scale, and so on, to be more detailed and multidimensional. Finally, we believe that in the future, we can start from different spatial scales, such as county scale, pixel scale, etc., to reveal the coupling and coordination between urbanization and the eco-environment in a more detailed and multi-dimensional way.

5. Conclusions

In this study, 16 cities in the Chengdu–Chongqing Economic Circle from 2004 to 2022 are taken as the research objects, and the CCD model is used to quantify the coupling and coordination level of NTU and EER and to analyze their spatial and temporal evolution characteristics. Based on this, nine socioeconomic and natural factors are selected to reveal the driving mechanism of CCD using RF model, SHAP, and PDP interpretation algorithms, and furthermore, combined with the GTWR model, the spatiotemporal divergence of CCD driving factors is analyzed.

The research findings indicate that since 2004, the CCD levels of the cities in the Chengdu–Chongqing Economic Circle have shown an evolutionary pattern of “progressive escalation and continuous optimization”, and the synergistic relationship between NTU and EER has been gradually enhanced, showing a spatial diffusion pattern of “dual-core leadership and regional diffusion” with Chengdu and Chongqing as the core.

As far as the driving mechanisms of CCD are concerned, the two socioeconomic factors, MAR and FIX, are the two most influential factors on the coupling coordination development of NTU and EER and show more significant importance than the other factors. Additionally, notable nonlinear interactions were observed among the drivers and CCD. Particularly, the nonlinear interactions involving HUM were especially pronounced. Moreover, the spatial and temporal evolution patterns of CCD drivers generally show the multidimensional characteristics of “stage dominance—marginal variation—spatial mismatch”.

Building upon this foundation, this study proposes actionable spatial adaptation policies based on the spatial distribution characteristics and mechanisms of driving factors:

In centrally located cities where market forces dominate, a green consumption network and ecologically oriented industrial system should be established. In secondary cities primarily driven by investment, priority should be given to green infrastructure and low-carbon construction projects. In densely populated areas, compact urban development and regular assessments of population and ecological carrying capacities are essential. In ecologically sensitive regions, establishing priority ecological function zones and implementing dual controls on land use intensity and ecological risks are critical. Furthermore, the GTWR model results should guide local governments in formulating urban governance strategies tailored specifically to different zones, groups, and development types.

The study deepens sustainable urban development research, guides the improvement of NTU–EER coordination, and provides insights into promoting coordinated, innovative ecological protection and high-quality urban development.