Regional Integration, Technological Innovation, and Digital Economy: Evolution and Evidence of the Decarbonization Driving Forces in the Yangtze River Delta

Abstract

Economic decarbonization has gained attention as a viable way to achieve carbon neutrality without sacrificing high-quality economic development. Regional integration (RI), technological innovation (TI), and the digital economy (DIE) are important decarbonization driving forces. To analyze how they impact carbon reduction and the evolutionary patterns observed in urban development, this study focuses on the urban agglomeration of the Yangtze River Delta (YRD) and uses the spatial Durbin model (SDM) to decompose the carbon reduction effects into spatial spillover effects and short-term and long-term effects. The research findings include the following: (1) The decarbonization driving forces have promoted economic decarbonization in the YRD and exhibit distinct evolutionary patterns. (2) The decarbonization effects exhibit spatiotemporal heterogeneity depending on the economic situation, period, and energy status of the cities. Over time, TI and DIE demonstrate better decarbonization effects. Economically developed cities are more likely to expand their carbon reduction advantages, while cities with dense energy supplies struggle to achieve their full carbon reduction potential due to traditional energy structures. (3) Industrial structure upgrading (UIS) is a critical pathway through which TI and DIE reduce carbon emissions. These findings enrich existing research on decarbonization drivers and offer valuable insights for governments to formulate further policies for economic decarbonization.

1. Introduction

Sea level rise, extreme weather events, the destruction of ecosystems, and the increasing scarcity of resources owing to global warming pose a great threat to both the survival and development of humankind [1,2]. Carbon emissions contribute to over 60% of the greenhouse effect and have become a top priority for controlling climate change [3,4]. As a major global emitter of carbon and pollutants, China has relied heavily on traditional fossil fuels for energy supply, driving rapid economic growth over the past three decades at the cost of surging carbon emissions [5,6]. China’s total carbon emissions crept to 12,730.64 million tons in 2020 from 6827.46 million tons in 2005 [7], contributing approximately 30% of global emissions [8], while non-fossil fuels represented only 15.9% of its energy mix [9]. Despite setting “dual carbon” goals (carbon peak by 2030 and carbon neutrality by 2060), China’s energy structure remains reliant on fossil fuels in the short term, making economic decarbonization a central proposition for achieving green development. Therefore, exploring effective decarbonization drivers from multiple perspectives has become crucial for researchers seeking sustainable development solutions.

The existing literature has extensively explored decarbonization drivers. Drivers such as industrial structure [10,11], population density [12,13,14], economic growth [15], openness to foreign trade [16,17], China’s market-oriented reforms [18,19], government policies [20], energy intensity, and energy mix [11] have been found to either promote or inhibit carbon emissions. Among them, RI, TI, and DIE have demonstrated remarkable carbon emission reduction potential under the promotion of the “dual carbon” goals and national policies. RI has been endowed with new connotations (regional economic and ecological integration) by national strategies such as the construction of a “Resource-conserving and Environment-friendly Society” and the “Beautiful China” initiative, enhancing resource efficiency through policy coordination [21,22]. As the core driver of the “Innovation-Driven Development” strategy, green TI promotes emission reduction by directly transforming green patents to upgrade industrial structures [23,24]. China’s “Digital China” strategy clearly states that “Digitization Drives Green Production Modes,” regarding DIE as a key vehicle to achieve the “dual carbon” goals, which amplifies the emission reduction achievements of technologies through a multiplier effect [25,26,27]. However, most studies focus on single drivers, and a consensus on their relationship with carbon emissions remains elusive. Moreover, due to China’s reform approach of implementing pilot programs before nationwide adoption, cities often exhibit diverse development characteristics. Consequently, the specific impacts of these three drivers on carbon emissions and their evolutionary patterns across different urban development stages remain underexplored. Addressing this knowledge gap holds significant implications for enriching socioeconomic decarbonization theories and informing scientifically grounded development planning.

This study focuses on the YRD urban agglomeration and employs spatial econometric models to systematically investigate the spatial dynamics and operational mechanisms of decarbonization drivers. The main contributions are as follows: (1) The respective carbon emission performance of RI, TI, and DIE was analyzed from the perspective of the spatial spillover effect. (2) The decarbonization effectiveness and evolution patterns of RI, TI, and DIE at the city level are examined in terms of the period, economic situation, and energy intensity heterogeneity. (3) The possible mechanisms of the decarbonization driving forces on carbon emissions by industrial structure upgrading are explored.

The remainder of the article is arranged as follows: Section 2 introduces the study area, variables, data, and methodologies. Section 3 reports the empirical results, including a heterogeneity analysis and mechanism analysis. Section 4 discusses the economic implications of the findings, their relationships with relevant research, and proposes policy recommendations alongside research limitations. Section 5 summarizes the key findings of this study. Figure 1 shows the main framework of the study.

2. Materials and Methods

2.1. Study Area

The YRD urban agglomeration, located in southeastern China, encompasses 26 cities across Shanghai Municipality and the provinces of Anhui, Jiangsu, and Zhejiang [28]. First, despite occupying less than 4% of China’s land area, the YRD generates nearly 25% of the nation’s GDP and one-third of its total import–export volume, making it the most economically robust, globally integrated, and strategically vital region in China [29]. The region’s population has grown steadily, comprising approximately one-eighth of China’s total population by 2021 [30]. However, like many economically developed and densely populated areas, the YRD faces profound conflicts between ecological protection and development imperatives [31]. Second, as a globally renowned manufacturing and trade hub, the YRD has a substantial share of labor-intensive and energy-intensive secondary industries [32]. Rapid urbanization and industrialization have driven a continuous rise in regional energy consumption, with fossil fuels dominating the energy structure [33]. Finally, the YRD has progressively shifted away from its traditional high-energy-consumption growth model in recent years. Concurrently, the Chinese government aims to establish the YRD as a demonstration zone for RI, emphasizing industrial synergy and innovation [34]. Further, the size of DIE in the YRD accounts for about 44% of the regional GDP [35]. Therefore, it is an ideal location to systematically study the decarbonization effectiveness of RI, TI, and DIE (as in Figure 2).

2.2. Variables

2.2.1. Dependent Variable

According to the research of [36,37,38,39], carbon emissions at the city level were calculated based on the IPCC (Intergovernmental Panel on Climate Change) carbon emission inventory and energy balance table. The carbon emissions (CE) of most cities showed a positive growth trend in the YRD (as in Figure 3). Then, the logarithmic value of carbon emissions was used as the dependent variable while eliminating heteroscedasticity.

2.2.2. Key Explanatory Variable

In this paper, it is necessary to find or construct proxy variables as key explanatory variables to indicate the development level of RI, TI, and DIE, respectively. Since patent authorization better reflects the actual improvement of technology [40], this paper selects green invention patent authorizations to reflect the technological innovation capacity to capture the dynamics of green innovation in society. The technological innovation capacity of the cities in the YRD cluster is growing overall, with a few “peak” cities experiencing rapid growth and other “slow” cities experiencing small increases (as in Figure 4). Figure 5 depicts the spatial and temporal changes in technological innovation in the cities of the YRD, as mapped by ArcGIS 10.8. It can be seen that over time, the spatial agglomeration effect of high-level regions has become more obvious, forming a technological innovation plateau centered on the “peak” cities (such as Shanghai, Hangzhou, and Nanjing), while the gap within the region has also widened.

The level of RI and the development of DIE are measured by a composite system of evaluation indicators.

Table 1. The measurement indicator system of the regional integration.

Primary Dimension	Secondary Indicator	Unit	Indicator Attribute	References
Social life stratification plane	Urban basic pension insurance participants	104	+	[42]
	Urban basic medical insurance participants	104	+	[42]
	Number of medical institutions	-	+	[42]
	Beds in medical institutions	-	+	[42]
	Average wage of full-time-employed workers	CNY 104	+	[42,43]
	Domestic waste disposal rate	%	+	[42]
	Urban green coverage area	104 m2	+	[40,41]
	Total collection of public libraries	104	+	[44]
	College students per 10,000 people	-	+	[20,45,46]
	Proportion of education expenditure in GDP	%	+	[42,47]
Economic life stratification plane	Balance of financial institution loan	CNY 108	+	[20,25]
	Balance of financial institution deposit	CNY 108	+	[25]
	Total retail sales of consumer goods	CNY 108	+	[42]
	Highway passenger transport quantity	104	+	[47]
	Total fixed assets investment	CNY 108	+	[48]
	Total profit of industrial enterprises above designated size	CNY 108	+	[42]
	Industrial enterprises above designated size	-	+	[42]

displays the measurement indicator system of the regional integration, including two levels of social and economic life. This paper incorporates 17 indicators in the areas of public services, human capital, and ecological construction, making the evaluation of regional integration more reasonable. Public services are related to the quality of life and social welfare of residents, and this indicator can measure the level and balance of various public services provided by the government for residents in the region. Human capital is one of the core elements of regional economic development, and this indicator is selected to reflect the quality and ability of the labor force in the region. In order to measure the degree of attention paid to the ecological environment and the effectiveness of its protection in the process of economic development, the indicator on ecological construction is used to assess the state of ecological environmental protection and sustainable development in the region. The urban green coverage area reflects the government’s efforts to increase carbon sinks. The domestic waste disposal rate reflects the sanitary conditions of the city. In addition, the literature has pointed out that reducing pollution has similarities with carbon emissions in addressing climate change and sustainable economic development [41,42]. Given data availability, future research could supplement other pollutant emission or treatment indicators to reflect the efforts of local governments in ecological governance more accurately. This comprehensive indicator is calculated using the entropy weight (EW) method.

Figure 6 represents the spatiotemporal evolution of RI in the YRD city cluster in 2011, 2016, and 2021. During the study period, the level of RI in the urban agglomeration increases significantly and shows regional synergies. The high-value area gradually expands and focuses on the core cities. In terms of the spatial pattern, the high-value areas are mainly concentrated in Southern Jiangsu and Shanghai. The change reflects the increasing capacity for regional cooperation and harmonious development of the cities in the YRD driven by the integration strategy and also reveals that there are still some development gaps between the regions.

As shown in

Table 2. The measurement indicator system of the digital economy.

Primary Dimension	Secondary Indicator	Unit	Indicator Attribute	References
Fundamentals of industrial development	Total postal business	CNY 108	+	[49,50,51]
	Total telecom business	CNY 108	+	[49,52,53]
	Proportion of ICT employees in tertiary industry	-	+	[45]
The scale of digital applications	Internet broadband access users	104 households	+	[40,43,49]
	Digital financial inclusion index	-	+	[40,52]
	Mobile phone users	104 households	+	[40,45]
	Total express delivery businesses	108	+	-

, DIE has spawned potential growth drivers in the logistics industry over the past decade. Therefore, adding the express delivery business indicator to the traditional internet and communication indicators can more scientifically reflect the scale of digital applications. A principal component analysis (PCA) is used for evaluation.

The overall level of DIE in the cities of the YRD cluster increased yearly over the study period, with regional differences gradually narrowing (as shown in Figure 7). In view of the spatial pattern, the high-value areas are mainly concentrated in the central cities of Hangzhou, Nanjing, and Shanghai and spread to the periphery of the surrounding cities, with these core cities as the center. Over time, areas such as Wuxi, Hefei, and Jinhua also witnessed significant DIE growth, leading to a leveling of development gradients across regions. This spatial diffusion and equalization signify a transition from single-core to multi-core DIE development within the YRD, propelling regional integration and coordinated growth.

2.2.3. Other Variables

Studying the drivers of decarbonization in such a systematic framework requires a consideration of other important factors in carbon emissions. The population density (PD), GDP per capita (PGDP), and actual utilized foreign investment (FAI) are selected to represent population, economic prosperity, and openness in the rapid urbanization of China [20,45,54]. The efficiency of resource allocation is frequently discussed with market-oriented reform in China [18,19]. However, it remains unclear how the marketization level (ML) impacts carbon emissions. Therefore, referring to existing studies, the share of private and individual employees in the urban workforce was adopted to measure the ML [19]. The general public financial budget expenditures (FE) and R&D investments (RD) are utilized to indicate the behavior of the government in maintaining social operations and promoting reforms [20]. In addition, while the structure of the energy supply may not undergo rapid transformation, energy supply efficiency (ESE) is an indispensable factor in the study of carbon emissions. ESE is evaluated using the ratio of GDP to the total social electricity, LPG, and total gas supply after conversion to standard coal [5]. Finally, industrial structural upgrading (UIS) is used as a transmission variable to explore how decarbonization drivers affect carbon emissions through industrial reshaping. UIS was measured as the labor productivity ratio between the tertiary and secondary industries, referring to Wu et al.’s study [55].

2.3. Data Sources and Processing

This paper draws upon relevant professional research databases, primarily the China Stock Market & Accounting Research Database [56], China Emission Accounts and Datasets [57], and Chinese Research Data Services [58]. Other statistical information comes from the Provinces Statistics Yearbook, the China City Statistical Yearbook, the Cities Statistical Yearbook, etc. The GDP data are converted into constant prices in 2011. Any missing data are provided using interpolation and other fitting methods. This study used Stata 18.0 statistical analysis software.

Table 3. Descriptive statistics of variables.

Variable	Obs	Mean	Std. Dev.	Min	Max
CE	286	59.301	54.422	7.214	236.487
RI	286	0.377	0.144	0.092	0.823
TI	286	375.972	587.767	1.000	3190.000
DIE	286	0.200	0.165	0.004	0.944
PD	286	674.736	390.267	158.471	2354.392
ESE	286	199,723.26	89,498.938	74,716.941	503,335.65
PGDP	286	104,654.263	58,653.690	18,779.008	288,118.612
ML	286	0.578	0.125	0.157	0.856
FAI	286	327,134.907	570,688.151	7792.000	5,271,624.000
RD	286	1,724,412.333	2,486,769.414	18,158.000	18,197,700.000
FE	286	8,365,626.916	12,730,015.964	706,635.000	84,308,600.000

presents the descriptive statistics of each variable, and the correlation coefficients between variables are shown in

Table 4. Correlation coefficient between variables.

	LnCE	RI	TI	DIE	PD	ESE	PGDP	ML	FAI	RD	FE
LnCE	1.000
RI	−0.004	1.000
TI	0.612	0.252	1.000
DIE	0.597	0.328	0.884	1.000
PD	0.682	0.109	0.658	0.593	1.000
ESE	−0.311	0.043	0.324	0.183	0.257	1.000
PGDP	0.757	0.134	0.640	0.706	0.602	0.339	1.000
ML	−0.052	0.234	0.043	0.044	0.137	0.067	0.135	1.000
FAI	0.426	0.182	0.702	0.708	0.493	0.265	0.481	−0.071	1.000
RD	0.659	0.167	0.870	0.866	0.805	0.280	0.704	−0.049	0.681	1.000
FE	0.572	0.179	0.804	0.807	0.822	0.232	0.561	−0.086	0.656	0.951	1.000

. To stabilize the data and mitigate heteroscedasticity, all variables were subjected to logarithmic transformation in subsequent experiments.

2.4. Methodology

First, the following fixed effects model was developed to investigate the “city–year” effect of the drivers of decarbonization.

$$LnC{E}_{it}={\alpha }_0+{\alpha }_1{K}_{it}+{\alpha }_2{C}_{it}+{\mu }_i+{\upsilon }_t+{\epsilon }_{it}$$

(1)

In the above equation, the subscripts $i$ and $t$ refer to the city and year, respectively, so $LnC{E}_{it}$ (i.e., the dependent variable in this paper) represents the natural logarithm of carbon emissions for city $i$ in year $t$. $K_{it}$ refers to the key explanatory variables (i.e., the decarbonization driving forces in this paper). Depending on the driver of decarbonization, $K_{it}$ is represented in Equation (1) as the variable for regional integration, technological innovation, and the digital economy, respectively. $C_{it}$ denotes all control variables in the study. ${\mu }_i$ and ${\upsilon }_t$ control for city and year fixed effects, respectively. ${\epsilon }_{it}$ is a random disturbance term.

Then, this paper focuses on spatial econometric models based on the spatial spillover perspective. The spatial Durbin model (SDM) has significant advantages over the spatial error model (SEM) and spatial autoregressive model (SAR) in understanding spatial correlations and spillover effects and addressing omitted variables [59]. As a result, the SDM is a pleasant starting point in the framework of spatial economic analysis. The spatial panel Durbin model that controlled the “city–year” effect is as follows.

$$LnC{E}_{it}={\beta }_0+\rho WLnC{E}_{it}+{\beta }_1{K}_{it}+{\lambda }_{1}W{K}_{it}+{\beta }_2{C}_{it}+{\lambda }_{2}W{C}_{it}+{\mu }_i+{\upsilon }_t+{\epsilon }_{it}$$

(2)

It is necessary to consider the specific impact of the spatial lag terms of the variables on the explanatory variables [60]. In order to avoid endogeneity issues and demonstrate the dynamic characteristics in carbon emission research, the dynamic SDM is developed to further improve the explanatory power.

$$LnC{E}_{it}={\beta }_0+\rho WLnC{E}_{it}+\eta WLnC{E}_{it-1}+{\beta }_1{K}_{it}+{\lambda }_{1}W{K}_{it}+{\beta }_2{C}_{it}+{\lambda }_{2}W{C}_{it}+{\mu }_i+{\upsilon }_t+{\epsilon }_{it}$$

(3)

In Equations (2) and (3), $\rho$ is the spatial autoregressive coefficient, $W$ is the spatial weight matrix, and $WLnC{E}_{it}$, $W{K}_{it}$, and $W{C}_{it}$ represent the spatial lag terms of the corresponding variables. $WLnC{E}_{it-1}$ refers to the spatiotemporal associative lag term of carbon emissions. Other parameters are set as in Equation (1).

Finally, in order to understand the role played by UIS on carbon emissions, Equations (4)–(6) are combined, together with Equation (2), to explore possible transmission mechanisms [61].

$$LnUI{S}_{it}={\gamma }_0+\rho WUI{S}_{it}+{\gamma }_1{K}_{it}+{\delta }_{1}W{K}_{it}+{\gamma }_2{C}_{it}+{\delta }_{2}W{C}_{it}+{\mu }_i+{\upsilon }_t+{\epsilon }_{it}$$

(4)

$$LnC{E}_{it}={\gamma }_0+\rho WLnC{E}_{it}+{\gamma }_{3}UI{S}_{it}+{\delta }_{3}WUI{S}_{it}+{\gamma }_2{C}_{it}+{\delta }_{2}W{C}_{it}+{\mu }_i+{\upsilon }_t+{\epsilon }_{it}$$

(5)

$$LnC{E}_{it}={\gamma }_0+\rho WLnC{E}_{it}+{\gamma }_{3}UI{S}_{it}+{\delta }_{3}WUI{S}_{it}+{\gamma }_1{K}_{it}+{\delta }_{1}W{K}_{it}+{\gamma }_2{C}_{it}+{\delta }_{2}W{C}_{it}+{\mu }_i+{\upsilon }_t+{\epsilon }_{it}$$

(6)

Equations (4) and (5) test whether the key explanatory variables can significantly affect UIS and whether UIS significantly influences carbon emissions. Next, Equation (2) is combined with Equation (6) to test whether UIS is a transmission channel.

In addition, considering that the adjacency matrix does not easily express the spatial relationships of cities that are geographically close but not adjacent (e.g., Jiaxing and Shaoxing), the baseline regression to analyze spatial effects is based on the geographical distance matrix ($W_1$). In the robustness test, the asymmetric economic geographic distance matrix ($W_2$) is included to strengthen the role of city economic ties in the spatial effects.

$$W_1=\left\{\begin{array}{l}1/d_{ij},i\ne j\\ 0,i=j\end{array}\right.$$

(7)

$$W_2=W_1\times diag\left({\displaystyle \frac{P_1}{\overline{P}}},{\displaystyle \frac{P_2}{\overline{P}}},\dots ,{\displaystyle \frac{P_i}{\overline{P}}}\right)$$

(8)

where $d_{ij}$ is the distance between the city $i$ and $j$ calculated based on latitude and longitude. $\overline{P}$ and $P_i$ denote the per capita GDP of the YRD and each city, respectively.

3. Results

3.1. Spatial Correlation Test

The spatial correlation test acts as a preliminary spatial data exploration. Before establishing a spatial econometric model, this study constructs a standardized geographic distance matrix ($W_1$). Conducting Moran’s I index test and reporting the significance level in

Table 5. Moran’s I index test of carbon emissions under $W_1$ (2011–2021).

Year	Geographic Distance Matrix (${\mathit{W}}_1$)
	Moran’s I	z-Values
2011	0.020 **	2.012
2012	0.016 **	1.869
2013	0.016 **	1.848
2014	0.020 **	1.983
2015	0.021 **	2.038
2016	0.021 **	2.023
2017	0.016 **	1.856
2018	0.031 ***	2.346
2019	0.031 ***	2.355
2020	0.031 ***	2.382
2021	0.031 ***	2.354

Note: ***, ** means at a significance level of 1%, 5%, respectively.

can effectively show whether spatial autocorrelation exists. The Moran index value indicates that the carbon emissions in the YRD were significantly and spatially positively correlated from 2011 to 2021. Therefore, there was positive urban spatial dependency in carbon emissions in the last decade. Moran’s scatter plots in Figure 8 reflect the spatial heterogeneity of carbon emissions, indicating that spatial spillover is influenced by some social and economic drivers.

3.2. Benchmark Regression Results

The regression results of the ordinary least squares (OLS) method might not reflect the true relationship between the variables because of the spatial dependence of regional carbon emissions. In practical terms, carbon emissions from different geographic locations are related not only to the local area but also to other cities in space. To incorporate the city differences and period factors into account in the model, according to Equation (1), this study first uses the non-spatial fixed effects model (FE) in

Table 6. Regression results of RI, TI, and DIE based on the FE model.

Dependent Variable	LnCE
Key Explanatory Variable	Model 1: Non-Spatial Fixed Effects Model
	RI	TI	DIE
RI	−0.1140 **
	(0.0505)
TI		0.1329 ***
		(0.0244)
DIE			0.1282 ***
			(0.0384)
PD	1.2170 ***	0.8925 ***	0.7302 **
	(0.3243)	(0.3006)	(0.3234)
ESE	0.0163	0.0399	0.0084
	(0.0433)	(0.0412)	(0.0429)
PGDP	1.4261 ***	1.1434 ***	1.0503 ***
	(0.2996)	(0.2819)	(0.2999)
ML	−0.0503	−0.0582	−0.0602
	(0.0527)	(0.0502)	(0.0522)
FAI	−0.0057	−0.0357 *	−0.0171
	(0.0191)	(0.0185)	(0.0187)
RD	0.0986 ***	0.0775 **	0.0723 **
	(0.0350)	(0.0333)	(0.0348)
FE	−0.1013	−0.0811	−0.0724
	(0.0987)	(0.0941)	(0.0981)
City effect	Yes	Yes	Yes
Year effect	Yes	Yes	Yes
Hausman test	19.55	37.47	37.13
Prob > chi2	0.0122	0.0000	0.0000
R2	0.978	0.980	0.979
N	286	286	286

Note: The standard errors are values in parentheses; *, **, and *** indicate 10%, 5%, and 1% significance levels, respectively.

to estimate the decarbonization driving forces.

Model 1 accounted for the city–year fixed effects. As Table 6 shows, the regression results of RI, TI, and DIE passed the Hausman test with a significance level of at least 5%, suggesting that the FE model was more appropriate than the random effects model (RE). The problem of heteroscedasticity among variables has also been avoided. The direct effect coefficient of RI is significantly negative at the 5% level, implying that a 1% increase in RI leads to a decrease of 0.1140 logarithmic percentage points in carbon emissions. Conversely, the direct effect coefficients of TI and DIE are significantly positive at the 1% level, implying that a 1% increase in their respective levels results in respective increases of 0.1329 and 0.1282 logarithmic percentage points in carbon emissions. Among the control variables, both PD and PGDP show significantly positive coefficients, suggesting their stimulating effects on carbon emissions. This demonstrates that urbanization and economic growth serve as drivers of carbon emissions. The significantly positive coefficient of ESE further exacerbates carbon emissions, reflecting that reliance on fossil energy remains challenging to reverse in the short term. However, the FE model renders it difficult to explain the spatial correction between explanatory variables and the dependent variable of neighboring units, so the estimation results are also biased. For example, some local governments have strongly emphasized economic development, and fiscal expenditures have increased year after year. The development climax of the traditional infrastructure triggered by fixed assets investment has resulted in huge amounts of carbon emissions. Because of the competition among local governments, the adverse effects of carbon emissions may be boosted. Therefore, while controlling for city-level differences, there is an urgent need for spatial econometric modeling to explore similar spatial relationships to those mentioned above.

The SDM provides a comprehensive understanding of the spatial spillover effect on carbon emissions and has a more accurate statistical performance. It reported more complex spatial linkages in the decarbonization driving forces of the YRD, and the static estimation results are given in

Table 7. Regression results of static spatial Durbin model under $W_1$.

Variable	Dependent Variable: LnCE
	Model 2: Static SDM
${\mathit{W}}_1$	Direct Effect			Spillover Effect
	SDM for RI	SDM for TI	SDM for DIE	SDM for RI	SDM for TI	SDM for DIE
RI	−0.1162 **			−0.0018
	(0.0454)			(0.1359)
TI		0.1295 ***			−0.1193 *
		(0.0217)			(0.0676)
DIE			0.1883 ***			−0.4347 ***
			(0.0442)			(0.1018)
PD	1.4590 ***	1.2348 ***	1.2428 ***	0.5504	0.6762	1.1119
	(0.3614)	(0.3299)	(0.3330)	(0.8840)	(0.8358)	(0.9054)
ESE	0.0887 **	0.1196 ***	0.1262 ***	0.4267 **	0.3909 **	0.5929 ***
	(0.0368)	(0.0351)	(0.0363)	(0.1792)	(0.1774)	(0.2076)
PGDP	1.6662 ***	1.4638 ***	1.5304 ***	0.7471	0.8454	1.3469
	(0.3359)	(0.3106)	(0.3132)	(0.8171)	(0.7828)	(0.8528)
ML	−0.0711	−0.0747 *	−0.0675	0.3618 ***	0.3308 **	0.4478 ***
	(0.0476)	(0.0446)	(0.0452)	(0.1395)	(0.1420)	(0.1582)
FAI	−0.0281	−0.0579 ***	−0.0394 *	0.0369	0.0717	0.0533
	(0.0223)	(0.0213)	(0.0213)	(0.0535)	(0.0532)	(0.0554)
RD	0.0807 **	0.0611 **	0.0675 **	−0.0709	−0.1220	−0.0067
	(0.0323)	(0.0307)	(0.0316)	(0.1488)	(0.1469)	(0.1644)
FE	−0.1244	−0.1140	−0.0810	1.1431 ***	1.1648 ***	1.2056 ***
	(0.0954)	(0.0911)	(0.0921)	(0.2250)	(0.2209)	(0.2399)
ρ	−2.0668 ***	−1.9602 ***	−1.8110 ***
R2	0.523	0.529	0.508
N	286	286	286

Note: The standard errors are values in parentheses; *, **, and *** indicate 10%, 5%, and 1% significance levels, respectively.

. In general, the carbon emissions indicated negative spatial spillover effects due to significantly negative spatial autoregressive coefficients. This is consistent with the “low–high” value aggregation characteristic of carbon emissions reflected in the Moran scatter plot, further supporting the need to establish spatial joint prevention and control emission reduction strategies to avoid “carbon leakage” in some cities.

The Chi values of the Hausman test, LR test, and Wald test shown in

Table 8. Statistical test of Model 2.

Dependent Variable	LnCE
Key Explanatory Variable	Model 2: Static SDM
	RI	TI	DIE
Hausman test	100.69	101.02	82.91
Prob > chi2	0.0000	0.0000	0.0000
LR test for SAR	40.48	39.88	47.54
Prob > chi2	0.0000	0.0000	0.0000
LR test for SEM	39.11	40.59	53.18
Prob > chi2	0.0000	0.0000	0.0000
Wald test for SAR	46.69	45.34	53.75
Prob > chi2	0.0000	0.0000	0.0000
Wald test for SEM	43.67	45.88	60.28
Prob > chi2	0.0000	0.0000	0.0000

are all significant at the 1% level, demonstrating that the selected spatiotemporal dual fixed SDM can reasonably represent the impact of key explanatory variables on the dependent variable.

The direct effect reflects the role of decarbonization drivers in local carbon emissions. As shown in column (2) of Table 7, the direct effect coefficient of RI is 0.1162 and statistically significant at the 5% level, indicating that RI suppresses carbon emissions. This demonstrates that the YRD region is committed to building an ecologically friendly integrated development and has shown certain emission reduction effects. However, columns (3) and (4) reveal that the direct effect coefficients of TI and DIE are 0.1295 and 0.1883, respectively, both significant at the 1% level. This implies that their development demands substantial energy inputs, thereby exacerbating local carbon emissions. Moreover, DIE exhibits a stronger carbon-enhancing effect than TI, as evidenced by its higher coefficient magnitude. From the perspective of the control variables, both PD and PGDP exhibit positive coefficients significant at the 1% level. This is a testament to the scale effect triggered by rapid urbanization and the pursuit of wealth at the expense of the environment. Furthermore, with a significance level of at least 5%, ESE implies that cities consume large amounts of cheap fossil fuels through high rates of economic development. In other words, urban development has lowered the threshold of fossil fuel utilization, leveraging fossil fuels to promote the high-speed operation of the city. Additionally, R&D investment promotes carbon emissions, while marketization and government financial expenditures have some but not significant effects on reducing emissions. This indicates a lack of local government investment and support for green industries. FAI significantly abates carbon emissions when TI and DIE are key explanatory variables, suggesting that multinational enterprises are more inclined to disseminate green production technologies in regions that emphasize ecological construction.

Spatial spillover effects reveal how the decarbonization drivers of neighboring cities influence local carbon emissions. As shown in column (5) of Table 7, the spatial spillover coefficient of RI is −0.0018, exhibiting a negligible and statistically insignificant spatial spillover effect, which indicates that local regional integration does not significantly affect neighboring cities’ carbon emissions. Columns (6) and (7) reveal that the spatial spillover coefficients of TI and DIE are −0.1193 and −0.4347, respectively, significant at the 10% and 1% levels. Both exhibit substantially stronger negative spatial spillovers, thereby promoting carbon emission reductions in adjacent cities. Furthermore, a comparison of the two impact coefficients on carbon emissions reveals that DIE facilitates the “learning effect” and therefore shows a stronger decarbonization capacity than TI. With respect to the control variables, there is no obvious spillover effect of PD and PGDP among the cities in the YRD, reflecting the similarity of the population size and economic development level. The energy supply structure remains unchanged in the short term. The coefficients for ML, FAI, and FE are statistically significant and positive. This implies that increases in ML, FAI, and FE in neighboring cities contribute to higher carbon emissions in the local area, reflecting competitive behaviors among local governments.

3.3. Robustness Check

Three main methods are used to examine the robustness of the regression results in this paper. The findings of the robustness check not only add confidence in the authenticity of the study but also deepen the understanding of the decarbonization patterns of RI, TI, and DIE.

3.3.1. Change the Spatial Matrix

The matrix changes to the asymmetric economic geographic distance matrix ($W_2$) and the results are shown in

Table 9. Regression results of static spatial Durbin under $W_2$.

Variable	Dependent Variable: LnCE
	Model 2: Static SDM
${\mathit{W}}_2$	Direct Effect			Spillover Effect
	SDM for RI	SDM for TI	SDM for DIE	SDM for RI	SDM for TI	SDM for DIE
RI	−0.1161 **			0.0049
	(0.0469)			(0.2570)
TI		0.1135 ***			−0.1030
		(0.0228)			(0.1437)
DIE			0.1676 ***			−0.6978 ***
			(0.0443)			(0.2217)
PD	1.4092 ***	1.1215 ***	1.1298 ***	−0.3461	−0.3959	0.8574
	(0.3514)	(0.3253)	(0.3314)	(1.5575)	(1.4981)	(1.6097)
ESE	0.0811 **	0.0988 ***	0.1119 ***	0.7289 *	0.5738	0.9132 **
	(0.0400)	(0.0382)	(0.0401)	(0.4222)	(0.3867)	(0.4329)
PGDP	1.6314 ***	1.3767 ***	1.4245 ***	0.3807	0.1598	1.5964
	(0.3223)	(0.3023)	(0.3073)	(1.3927)	(1.3912)	(1.5082)
ML	−0.0736	−0.0688	−0.0688	0.7878 **	0.5419 *	0.8194 **
	(0.0479)	(0.0452)	(0.0464)	(0.3372)	(0.3203)	(0.3579)
FAI	−0.0353	−0.0581 ***	−0.0419 **	0.0134	0.0264	0.0265
	(0.0216)	(0.0209)	(0.0209)	(0.0871)	(0.0914)	(0.0883)
RD	0.0930 ***	0.0777 **	0.0835 **	0.2412	0.2530	0.3141
	(0.0345)	(0.0335)	(0.0343)	(0.2931)	(0.2923)	(0.3002)
FE	−0.1271	−0.1104	−0.1005	1.4226 ***	1.3347 ***	1.4642 ***
	(0.0916)	(0.0880)	(0.0895)	(0.4651)	(0.4580)	(0.4793)
ρ	−0.7241 ***	−0.6658 ***	−0.6665 ***
R2	0.491	0.508	0.462
N	286	286	286

Note: The standard errors are values in parentheses; *, **, and *** indicate 10%, 5%, and 1% significance levels, respectively.

. The signs and significance levels of the core explanatory variables remain consistent, demonstrating the robustness of the baseline regression results. The implication of $W_2$ is described in Section 2.4 of the paper.

3.3.2. Change the Dependent Variable

The per capita carbon emissions (PCE) can also reflect the direct impact of carbon emissions on human activities because of the regional population differences. In addition, this study calculated the carbon emission intensity (CEI) by considering the share of carbon emissions in the urban GDP, reflecting the dynamic change of carbon emissions with the economy. After replacing the dependent variable,

Table 10. Robustness check results (change the dependent variable).

Variable	Dependent Variable: LnPCE			Dependent Variable: LnCEI
	Model 2: Static SDM			Model 2: Static SDM
${\mathit{W}}_1$	SDM for RI	SDM for TI	SDM for DIE	SDM for RI	SDM for TI	SDM for DIE
Direct effect
RI	−0.1165 **			−0.1160 **
	(0.0461)			(0.0452)
TI		0.1299 ***			0.1294 ***
		(0.0220)			(0.0217)
DIE			0.1888 ***			0.1883 ***
			(0.0449)			(0.0448)
Spillover effect
RI	−0.0009			−0.0026
	(0.1351)			(0.1392)
TI		−0.1180 *			−0.1195 *
		(0.0663)			(0.0678)
DIE			−0.4242 ***			−0.4239 ***
			(0.0990)			(0.0989)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
ρ	−2.1132 ***	−2.0283 ***	−1.9049 ***	−1.9964 ***	−1.9561 ***	−1.9177 ***
R2	0.298	0.270	0.252	0.129	0.166	0.159
N	286	286	286	286	286	286

Note: The standard errors are values in parentheses; *, **, and *** indicate 10%, 5%, and 1% significant levels, respectively.

presents the estimation results that are basically in line with the baseline regression, further confirming the robustness of the findings.

3.3.3. Replace the Estimate Model

Considering the need to mitigate the endogeneity problem and the lagged or persistent impact of decarbonization drivers in reality, the dynamic spatial Durbin model (i.e., Model 3) is established based on Equation (3). The dynamic model in

Table 11. Robustness check results (Model 3: dynamic SDM).

Variable	Dependent Variable: LnCE
	Model 3: Dynamic SDM
${\mathit{W}}_1$	SDM for RI		SDM for TI		SDM for DIE
	Short-Term	Long-Term	Short-Term	Long-Term	Short-Term	Long-Term
Direct effect
RI	−0.0884 **	−0.1062
	(0.0447)	(0.7656)
TI			0.0949 ***	0.1691 **
			(0.0215)	(0.0722)
DIE					0.1964 **	0.4006 **
					(0.0832)	(0.1825)
PD	0.8565 **	−0.5580	0.7806 **	−0.0899	0.7784 **	0.0320
	(0.3466)	(21.3900)	(0.3201)	(1.0388)	(0.3236)	(0.6952)
ESE	0.1360 ***	0.0170	0.1531 ***	0.1007	0.1587 ***	0.0941
	(0.0434)	(1.5577)	(0.0418)	(0.1358)	(0.0435)	(0.0893)
PGDP	1.0587 ***	−0.3119	1.0110 ***	0.2247	1.0258 ***	0.3429
	(0.3300)	(22.0630)	(0.3117)	(0.9914)	(0.3151)	(0.6756)
ML	−0.0521	−0.3119	−0.0634	−0.2229 *	−0.0494	−0.1841 *
	(0.0464)	(2.0557)	(0.0452)	(0.1324)	(0.0458)	(0.1003)
FAI	−0.0144	−0.0086	−0.0360 *	−0.0680	−0.0134	−0.0229
	(0.0222)	(0.3392)	(0.0216)	(0.0643)	(0.0222)	(0.0456)
RD	0.0442	0.0113	0.0360	0.0459	0.0460	0.0579
	(0.0300)	(0.5990)	(0.0284)	(0.0685)	(0.0294)	(0.0610)
FE	0.0275	−0.5106	0.0301	−0.3547	0.0464	−0.2495
	(0.0950)	(6.5219)	(0.0922)	(0.2880)	(0.0942)	(0.2012)
Spillover effect
RI	−0.1241	0.0141
	(0.1384)	(0.7738)
TI			−0.0177	−0.1337
			(0.0688)	(0.0865)
DIE					−0.3838	−0.4834 **
					(0.2480)	(0.2416)
PD	5.4397 ***	3.2835	5.0107 ***	2.7345 **	5.3871 ***	2.7150 ***
	(1.1948)	(21.3908)	(1.0809)	(1.1903)	(1.2199)	(0.9265)
ESE	0.6089 ***	0.3050	0.5801 ***	0.2336	0.7155 ***	0.2948 **
	(0.2035)	(1.5652)	(0.1913)	(0.1759)	(0.2242)	(0.1466)
PGDP	5.7033 ***	3.2390	5.3052 ***	2.6596 **	5.7662 ***	2.6830 ***
	(1.2124)	(22.0715)	(1.1068)	(1.1506)	(1.2564)	(0.9123)
ML	0.4473 ***	0.4832	0.4044 ***	0.3787 **	0.4673 ***	0.3703 ***
	(0.1422)	(2.0600)	(0.1399)	(0.1591)	(0.1579)	(0.1364)
FAI	0.0121	0.0076	0.0156	0.0586	0.0050	0.0191
	(0.0512)	(0.3408)	(0.0499)	(0.0735)	(0.0530)	(0.0568)
RD	0.0891	0.0462	0.0623	−0.0011	0.0722	−0.0054
	(0.1470)	(0.6051)	(0.1385)	(0.1119)	(0.1523)	(0.1088)
FE	1.3878 ***	1.1234	1.4323 ***	1.0226 ***	1.4326 ***	0.9089 ***
	(0.2657)	(6.5221)	(0.2538)	(0.3053)	(0.2936)	(0.2437)
ρ	1.9547 ***		2.0169 ***		1.8394 ***
R2	0.340		0.349		0.352
N	260		260		260

Note: The standard errors are values in parentheses; *, **, and *** indicate 10%, 5%, and 1% significance levels, respectively.

not only verified the robustness of the static model but also decomposed into short-term and long-term impacts of RI, TI, and DIE on carbon emissions.

The dynamic model provides three important findings: (1) RI has a smaller long-term carbon reduction effect than it does in the short term. (2) TI and DIE show stronger spatial spillover effects. (3) Population, energy supply, and economic structure will have no significant impact in the short term but will continue to put pressure on carbon reduction in the long term.

3.4. Heterogeneity Analysis

In recent years, the Chinese government has prioritized high-quality development. This is a long-term initiative aimed at bringing about new changes in the urban economy, energy structure, and other aspects. To this end, the heterogeneous performance of RI, TI, and DIE in terms of decarbonization effectiveness in the YRD region is further examined from the perspectives of temporal heterogeneity, economic disparity, and energy supply level. Briefly, we need to construct dummy variables for each of the above three aspects. Taking 2017 as the cut-off point, the sample is separated into two time periods: 2011–2016 and 2017–2021. Cities in the 2017–2021 time period are assigned a value of 1, while those in the previous time are assigned a value of 0. By calculating the median per capita GDP of each city in each year, the cities above the median are set as 1, and vice versa, as 0. The same method is used to determine the median energy supply efficiency of each city each year. Cities that are higher than the median of that year are energy-intensive cities, assigned a value of 1, and vice versa, assigned a value of 0. The interaction terms are added between the key explanatory variables and their respective dummy variables, and benchmark regression estimation is performed again.

Table 12. Heterogeneity analysis from the perspectives of temporal heterogeneity, economic disparity, and energy supply level.

Variable	Temporal Heterogeneity		Economic Disparity		Energy Supply Level
	Direct Effect	Spillover Effect	Direct Effect	Spillover Effect	Direct Effect	Spillover Effect
	(1)	(2)	(3)	(4)	(5)	(6)
RI * Dhetero	0.1209 *	−0.0831 *	0.0599 **	−0.0420 **	0.2653 ***	−0.1846 ***
	(0.0729)	(0.0503)	(0.0279)	(0.0199)	(0.0655)	(0.0476)
TI * Dhetero	−0.0276 ***	0.0188 **	−0.0218 **	0.0146 **	0.0130 **	−0.0090 **
	(0.0106)	(0.0073)	(0.0085)	(0.0057)	(0.0058)	(0.0041)
DIE * Dhetero	−0.0593 **	0.0405 **	−0.0498 ***	0.0335 ***	0.0537 **	−0.0365 **
	(0.0261)	(0.0181)	(0.0118)	(0.0083)	(0.0233)	(0.0162)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes	Yes	Yes
N	286	286	286	286	286	286

Note: D_hetero = 1 or 0. The value will be determined based on the construction of the above dummy variable for the heterogeneity analysis. The standard errors are values in parentheses. *, **, and *** indicate 10%, 5%, and 1% significance levels, respectively.

displays the heterogeneity analysis results.

Columns (1) and (2) reflect the temporal heterogeneity. In 2017, the Chinese government clearly proposed the promotion of coordinated regional development and the “Digital China” strategy to enhance the independent innovation capacity. In terms of direct effects, the coefficient of RI becomes statistically significant and positive, suggesting that factor-driven regional integration inhibits urban carbon reduction efforts. Instead, the coefficients of TI and DIE turn statistically significant and negative, indicating their direct contributions to urban decarbonization. Driven by these policies, both factors have evolved to exert measurable carbon mitigation effects. Regarding the spatial spillover effects, the signs of coefficients for all three decarbonization drivers (RI, TI, and DIE) are reversed. Locally, they exhibit contrasting impacts on neighboring cities’ emissions: regional integration facilitates inter-city emission reductions, whereas technological innovation and digital economy development inhibit neighboring areas’ decarbonization progress. Columns (3) and (4) demonstrate economic development heterogeneity, yielding conclusions largely consistent with the temporal heterogeneity analysis. In economically advanced cities, the direct effects of TI and DIE remain statistically significant and negative, demonstrating their enhanced efficacy in leveraging innovation and digitalization for emission abatement. Columns (5) and (6) reveal that in energy-intensive cities, all decarbonization drivers show statistically significant positive direct effects, failing to deliver local emission reductions. However, their negative spatial spillover coefficients (statistically significant) suggest that neighboring cities’ progress in these drivers effectively promotes local carbon mitigation.

3.5. Mechanism Analysis

The majority of the questions about the actual effects and evolution patterns of decarbonization driving forces in the YRD have been answered in the above analysis. However, there is a lack of necessary transmission mechanism research on how to promote the decarbonization capabilities of TI and DIE. A prominent foothold of China’s economic reform is the upgrading of industrial structure. Hence, this subsection first adopts the “three-step method” to explore whether UIS can serve as a transmission channel for TI and DIE to impact carbon emissions.

Table 13. Upgrading industrial structure in the mechanism analysis (three-step method).

Step	Variable Relationship	Significance Test	Analysis Results
Step 1	LnTI→LnCE (Equation (2))	0.1132 ***	The transmission channel can be tested.
		(0.0203)
	LnDIE→LnCE (Equation (2))	0.1364 ***
		(0.0370)
Step 2	LnTI→LnUIS (Equation (4))	0.1490 *	The transmission channel exists.
		(0.0774)
	LnDIE→LnUIS (Equation (4))	0.4329 ***
		(0.1326)
	LnUIS→LnCE (Equation (5))	−0.0422 ***
		(0.0163)
Step 3	LnTI→LnCE (Equation (6))	0.1229 ***	LnUIS is not the only transmission channel for
		(0.0201)	LnTI to reduce LnCE.
	LnUIS→LnCE (Equation (6))	−0.0559 ***
		(0.0154)
	LnDIE→LnCE (Equation (6))	0.1575 ***	LnUIS is not the only transmission channel for
		(0.0370)	LnDIE to reduce LnCE.
	LnUIS→LnCE (Equation (6))	−0.0534 ***
		(0.0160)

Note: The standard errors are values in parentheses; * and *** indicate 10% and 1% significance levels, respectively.

lists these mechanism analysis results.

In the first step, according to Equation (2), LnTI and LnDIE are significantly and positively associated with carbon emissions, respectively (${{\beta }_1}^{\prime }=0.1132$, ${{\beta }_1}^{″}=0.1364$, p < 0.01). According to Equation (4), the effectiveness of LnTI and LnDIE on LnUIS is tested separately in step 2. Afterward, we concluded that LnUIS significantly inhibits the carbon emissions (${\gamma }_3=-0.0422$, p < 0.01), according to the regression results of Equation (5). This performance indicates that UIS is a transmission channel to reduce carbon emissions. In the last step, when LnUIS is combined with LnTI or LnDIE, which is included in Equation (6) to study carbon emissions, the coefficient is also significant. According to the estimation results, UIS is not the only transmission channel. Moreover, the estimated coefficients of LnTI and LnDIE have slightly increased, indicating that industrial structure upgrading shows a better capability to reduce emissions. This suggests that cities in the YRD region have neglected the further integration of TI and DIE with UIS in their economic decarbonization development. As a result, the positive efforts of UIS on emission reduction have been obscured. As shown in

Table 14. The mechanism analysis of industrial structure upgrading (Bootstrap test).

Explanatory Variable	Dependent Variable	Transmission Variable	Effect Type	Observed Coefficient	Bootstrap Std. Err.	[95% Conf. Interval]
LnTI	LnCE	LnUIS	Indirect effect	−0.0112	0.0064	[−0.0253 −0.0008]
			Direct effect	0.1441	0.0255	[0.0919 0.1937]
LnDIE	LnCE	LnUIS	Indirect effect	−0.0278	0.0119	[−0.0544 −0.0077]
			Direct effect	0.1561	0.0428	[0.0787 0.2479]

, the Bootstrap method is also used to revalidate the existence of UIS as a transmission channel, thereby confirming the credibility of the results. The 95% confidence intervals (CIs) are derived from 5000 Bootstrap resampling. The confidence intervals do not include zero in either the direct or indirect effects analysis. Thus, UIS is the transmission mechanism that promotes the decarbonization capabilities of TI and DIE.

4. Discussion

4.1. Spatiotemporal Evolution Patterns of Decarbonization Drivers

Since the 21st century, carbon emission pressures have emerged as a critical challenge for developing nations, with China confronting analogous issues. While the country achieved rapid economic growth following its reform and opening-up policy, the extensive development model led to technological stagnation in production, the overexploitation of resources, and environmental degradation. Since 2007, the central government has promoted regional economic and ecological integration, implementing a series of policies to build a “Resource-saving and Environment-friendly Society”, to alleviate emission reduction pressures. This study reveals that the carbon abatement effect of RI exhibits “short-term effectiveness but long-term limitations”, partially aligning with the inverted-U-shaped hypothesis identified by Zhang et al. [62]. The baseline regression results confirm its direct emission reduction effects, demonstrating that policy-driven resource integration facilitates decarbonization. RI provides a favorable environment for economically active and densely populated areas, facilitating the formation of economies of scale, enhancing enterprises’ capabilities to develop low-carbon technologies [63,64], and eliminating regional barriers to accelerate the diffusion of green technologies and management experiences [65]. It encourages the utilization of comparative advantages under market competition and influences carbon emissions by fostering innovation networks [66]. Notably, the YRD region has dedicated itself to ecologically friendly integrated development. By improving the convenience and sharing of public services, the government has enhanced public welfare and guided the adoption of green lifestyles. Environmental greening and governance have promoted green investment and consumption, while prioritizing public education has nurtured talent with sustainable development mindsets. Moreover, the spatial analysis reveals negligible and statistically insignificant spatial spillover effects from RI. This may stem from persistent administrative barriers and market fragmentation, coupled with divergent understandings and implementation approaches toward urban sustainability among local governments. Temporally, its long-term carbon reduction effect is weaker than its short-term impact. As integration progresses, an industrial structure lock-in effect may emerge, where energy-intensive industries form stable regional supply chains, weakening long-term emission reduction outcomes. In conclusion, the long-term development of RI is influenced by multiple factors [67], and its sustained effectiveness depends on incorporating more ecological elements.

TI plays a pivotal role in accelerating green economic transitions. This study reveals that the spatiotemporal evolution of TI follows a “short-term cost versus long-term dividend” pattern. Although green innovation technologies are widely regarded as core decarbonization drivers [68] capable of reducing emissions through production mode transformation and energy structure optimization [69,70,71], our empirical results show that TI initially increases local carbon emissions through direct effects. Initial technological investment is often accompanied by the “inertial expansion” of traditional industries. This paradoxical outcome may stem from productivity expansion in traditional industries prioritizing economic output over ecological efficiency [72], inadequate infrastructure for green technology adoption, and high costs of green innovation implementation triggering carbon rebound effects [73], suggesting a time-lagged decarbonization impact of technological innovation. However, further analysis demonstrates TI’s substantially stronger long-term decarbonization potential and enhanced spatial spillover effects. Over extended periods, green technological innovation facilitates the growth of digital finance industries to achieve emission reduction–economic growth dual objectives [54], while generating knowledge externalities that reduce cross-regional technology adoption costs through R&D spillovers. Inter-regional knowledge diffusion via labor mobility and corporate partnerships, particularly from high-R&D regions, enables broader technological dissemination. This evidence underscores the urgency to establish regional green technology trading platforms by leveraging patent sharing and joint R&D programs to dismantle innovation silos.

In China, DIE has emerged as a dynamic economic paradigm, flourishing alongside deepening RI and waves of TI. Unlike traditional systems reliant on massive fossil fuel consumption, DIE operates as a green, sustainable decarbonization driver [45], potentially catalyzing the decoupling of economic growth from carbon emissions [74,75]. This study confirms that DIE’s emission reduction capacity strengthens with enhanced spatial coordination. While it initially increases local carbon emissions through short-term direct effects, DIE demonstrates a significant long-term decarbonization potential and suppresses neighboring cities’ emissions via negative spatial spillovers. Furthermore, there is a certain synergy between DIE and TI. DIE is highly dependent on TI, but the adoption and scaling of technology require a maturation period. In the early stage of its integration with traditional industries, it may trigger a “scale paradox”: on the one hand, the digital transformation of enterprises requires the deployment of a large number of hardware facilities such as servers and communication equipment, whose manufacturing and operation generate direct carbon emissions; on the other hand, the empowerment of traditional industries by digital technology may stimulate capacity expansion in the short term, such as e-commerce driving the growth of logistics and freight volume, indirectly pushing up energy demand. In the absence of renewed energy from intensive R&D activities [76] and energy efficiency improvements from green technological innovations, DIE will lead to an undesirable increase in the output of traditional industrial products, instead promoting carbon emissions. The fact that the coefficient of DIE on carbon emissions is higher than that of TI in the direct effect results also confirms this point. This result is consistent with the “spatial imbalance” observation of Gu et al. [77], and further points out that the digital economy can strengthen the spatial spillover of technological innovation through the “learning effect”, effectively promoting carbon emission reduction. Similarly, its “learning effect” depends on infrastructure and policy support. Without green innovation support, it is likely to become a tool for the expansion of traditional industries. Therefore, policies should promote the “bundled” investment in digital infrastructure and green technologies.

These findings provide us with more insights. TI and DIE are more advanced decarbonization driving forces, and their improvement in urban energy and economic structure needs constant attention. Carbon reduction requires the market to provide a new low-carbon economic transition, rather than relying solely on the government’s master plan for the region. For example, due to the constraints of urban spatial planning, there is a saturation value for public services and urban greening, which weakens public contribution to low-carbon actions, and the physiological activities of biological sources increase carbon emissions [78].

4.2. Heterogeneity of Decarbonization Effects

The findings of this study exhibit significant heterogeneity based on the characteristics of urban development time, economic differences, and energy structures. Firstly, the temporal heterogeneity analysis reveals that RI initially increases urban carbon emissions. After the 19th National Congress of the Communist Party of China in 2017, cities in the YRD were influenced by policies promoting regional coordinated development. The enhanced regional cooperation led to increased energy demand, which, to some extent, resulted in higher urban carbon emissions. Additionally, following the implementation of relevant policies, high-pollution industries were transferred. Environmental pollution in the YRD’s central cities decreased, while that in the peripheral cities increased, potentially leading to an overall rise in the environmental pollution level [79]. This further demonstrates the complexity of RI’s impact on carbon emissions: depending on government policies, pollutants, and the selection of research areas, the decarbonization effect of RI may yield opposite results [67]. Guided by the “Digital China” policy, the emission-reducing effects of TI and DIE gradually emerged, and the three decarbonization drivers showed opposite impact characteristics. After 2017, cities in the YRD paid more attention to achieving their ecological construction goals through the development of DIE and a new round of technological revolution. However, competition among local governments was evident. The actions of neighboring cities to promote RI can improve the regional integration framework of the local city and lead to a greater focus on ecological benefits. Nevertheless, it is difficult to find a more appropriate synergy between TI and DIE in mitigating climate change and reduce carbon emissions.

Secondly, regarding the decarbonization performance of different city types, economically advanced cities demonstrate a greater capacity to achieve “dual carbon” goals, while energy-intensive cities are held back by their traditional energy structures. Economically developed cities (such as Shanghai and Hangzhou) have sufficient capital investment and strong technology absorption capabilities, making it easier for them to leverage their capital and technology advantages to unleash their decarbonization potential. However, there are few ways to further expand this effect. The decarbonization effects of energy-intensive cities (such as Xuzhou and Ma’anshan) are severely restricted by their traditional energy structures. Such cities need to prioritize the transformation of their energy structures rather than blindly pursue the expansion of DIE. This finding confirms the impact of economic development levels and energy structures on carbon emissions [80] and further refines the analysis of differences among different city types in the decarbonization process. Policy formulation needs to be tailored to local conditions, and differentiated emission-reducing strategies should be formulated according to the characteristics of different cities.

4.3. Transmission Mechanisms of Industrial Structure Upgrading

The mechanism analysis reveals that UIS serves as a critical mediating mechanism through which TI and DIE influence carbon emissions. These findings confirm the importance of industrial structure adjustment in achieving low-carbon development [81]. However, UIS in YRD cities has not fully harnessed its emission reduction potential. As it only partially explains the decarbonization effects of TI and DIE, underlying mechanisms likely exist. This may stem from current industrial upgrading focusing predominantly on expanding the service-sector scale rather than fostering deep green technology penetration. Energy consumption growth in digital industries such as e-commerce and cloud computing partially offsets the benefits of industrial transformation [51,82]. Green technologies may be preferentially adopted in high-value-added sectors like semiconductor manufacturing, where energy consumption has not significantly decreased in the short term. Digital technologies can directly optimize energy use through smart grids and industrial IoT, operating independently of industrial structure changes. This finding provides a more targeted direction for future research and policy design. TI- and DIE-driven industries should avoid unsustainable development in low-end industrialization to prevent triggering a vicious cycle of environmental and economic decline. Emphasizing the goal of decarbonized development through UIS can foster the healthy development of decarbonization drivers.

4.4. Policy Recommendations and Limitations

It can be concluded that the development direction of economic decarbonization is consistent with the evolution trend of decarbonization driving forces. This means that the integration of traditional elements and policy incentives no longer dominates the process of carbon reduction but is replaced by active low-carbon technology applications and economic forms that can improve green growth rates. In light of these findings, several recommendations may be worth considering for policymakers: (1) To cultivate a conducive decarbonization environment for RI, market segregation and administrative barriers need to be removed from institutional mechanisms. Promoting the convenient sharing of public services in various fields and upgrading infrastructure conditions are also beneficial for the decarbonization effect of RI. For example, establishing a regularized information-sharing platform to disseminate timely information on policy dynamics and industrial development in the region, strengthen cooperation and communication among governments, and promote scientific and synergistic government decision-making. In respect of public services and infrastructure, new energy vehicle sharing services should be promoted, new energy vehicle leasing outlets should be increased, etc. More importantly, cities should bolster cooperative ecological environment preservation and administration so as to establish a green development base for RI. For instance, a cross-regional cooperation mechanism should be established for ecological protection and restoration projects and large-scale ecological restoration projects should be jointly carried out. Pollutant emission standards and treatment measures should be harmonized, and the development of clean energy should be promoted. A comprehensive carbon emissions trading system should be established. Low-carbon industries should be encouraged to become stronger and better. (2) Economic decarbonization cannot rely solely on the integration of traditional production factors and the improvement of resource allocation efficiency, but it needs to find long-term, competitive decarbonization driving forces. It is essential to integrate high-quality development elements like green TI and DIE across regions and sectors to serve the needs of “dual carbon” goals. For instance, this could be achieved by encouraging universities, research institutions, and enterprises to jointly carry out research on green technology innovation and promoting the application of digital technology in energy management, intelligent manufacturing, and other fields. For energy supply-intensive cities, it is recommended to implement targeted subsidies for green patents. For example, special funds should be established to support the transformation of patents related to clean coal utilization and industrial carbon capture technologies. The construction of a green patent sharing alliance in the YRD region should be promoted, allowing energy-intensive cities to obtain free access to the patent rights of core cities. Importantly, the government should establish a “non-zero-sum game” concept to synergistically develop and promote the positive spillover effects of green TI and DIE and avoid the regional “carbon leakage” phenomenon caused by the transfer of polluting industries. Establishing an environmental assessment mechanism for regional industrial transfers should be considered, and rigorous carbon emission assessments of industrial projects to be transferred should be conducted. In addition, UIS is a positive way to cultivate decarbonization potential. (3) Carbon emission reduction strategies in cities should be oriented toward sustainable development, follow the pattern of economic decarbonization, and reasonably formulate emission reduction targets and policy initiatives based on the heterogeneity of decarbonization driving forces in cities. Urban development should focus on synchronizing economic and ecological benefits. Cities at a disadvantage in RI should mobilize green production and consumption initiatives, aim for long-term carbon neutrality goals, introduce green technologies, and accelerate the creation and layout of low-carbon industrial development spaces. Specific measures include establishing citizen green consumption incentive mechanisms and setting up special funds for low-carbon technologies. A higher RI should be followed by a shift toward the exploration of emerging decarbonization technologies and economies. For example, to promote the sharing of advantageous resources such as emission-reducing and energy-saving technologies. Moreover, cities can support the digital and clean development of the energy industry, namely, by ensuring that the efficiency of the energy supply does not decline and by accelerating the low-carbon transition of the energy mix.

Finally, it is acknowledged that this article still faces some limitations. First, although this study made significant efforts to explore the evolution and effects of important drivers of decarbonization separately, according to urban development, it still could not address the specific limitations regarding coupled carbon reduction. Future research could further investigate the meaningful impacts from the coupled coordination theory perspective using more appropriate modeling and extend future research by delving into the interaction mechanisms between RI, TI, and DIE. Second, the time span of this study is only up to 2021, so the sample survey time should be updated and extended into the future to bring more current and relevant results to the study. The research method is mainly quantitative analysis, lacking a qualitative discussion of the details of policy implementation. Additionally, applying the theoretical framework or model to more urban agglomerations is necessary to assess its applicability and generalizability. Third, within the constraints of space and different research focuses, future research could also examine the impact of urban water pollution and solid waste management on carbon emissions. Other drivers from different perspectives are present, such as the direction of TI in the industry, the type of digitization, and the allocation and pricing of land, which can also affect decarbonization. This paper only explores UIS as the core mediating mechanism, without addressing diversified transmission pathways such as green finance, circular economy practices, or public behavior changes. Future research could consider incorporating these factors and conducting in-depth analyses to obtain a more comprehensive understanding of the decarbonization process. Nonetheless, this paper provides robust empirical results as well as some valuable insights and enhances confidence in the study of the driving forces of economic decarbonization.

5. Conclusions

Economic decarbonization is a major strategic arrangement that China is and will firmly be determined to promote. Under the vision of “dual carbon” goals, urban agglomerations are key implementation units for realizing these. RI, TI, and DIE have become key decarbonization drivers. This paper primarily utilizes spatial econometric models to examine their respective relationships with urban carbon emissions while attempting to elucidate the evolution patterns of their decarbonization capabilities. Furthermore, this paper offers more comprehensive evidence from the perspectives of heterogeneity and intrinsic mechanisms. In summary, the following conclusions have been drawn: (1) RI, TI, and DIE are spatially connected in terms of their influence on urban carbon emissions and contribute to the economic decarbonization of the YRD. (2) Population growth and traditional energy supply structures are major obstacles to achieving net-zero emissions. TI and DIE exhibit stronger spatial spillover effects, and both are expected to unlock greater decarbonization potential in the long run. (3) RI, TI, and DIE, as driving forces for decarbonization, follow a certain evolutionary pattern. In the short term, RI is more effective in reducing emissions, while the long-term impact hinges on integrating more ecological elements. DIE tends to expand the decarbonization capacity of RI and TI chronically, but this expansion should be predicated on a commitment to the widespread application of low-carbon technologies in the industrial sectors within the area. (4) The temporal heterogeneity shows that TI and DIE demonstrated significant emission reduction effects after 2017. Cities in the YRD region that are relatively economically developed are more capable of realizing “dual carbon” goals, while energy-intensive cities are being dragged down by traditional energy structures. (5) The transmission mechanism is indispensable in studying the impact of TI and DIE on carbon emissions. Through the mediating pathway of UIS, carbon emissions will ultimately be suppressed.