Integration of the Digital–Real Economy and Energy-Embedded Green Utilization Efficiency of Urban Land: Causal Evidence from Double Machine Learning

Abstract

Enhancing Energy-Embedded Green Utilization Efficiency of Urban Land (E-GUEUL) is crucial for reconciling economic growth with carbon neutrality targets, with the Integration of the Digital–Real Economy (IDRE) emerging as a key driver. This study measures city-level E-GUEUL using the super-efficiency SBM–Malmquist index model. To rigorously identify the causal effect of IDRE on E-GUEUL and address potential model misspecification and high-dimensional confounding factors, a Double Machine Learning (DML) framework is employed. Findings reveal a robust and significant positive effect of IDRE on E-GUEUL, a conclusion that holds across a series of robustness checks and endogeneity controls. Heterogeneity analysis indicates that the efficiency enhancement is more pronounced in non-resource-based, digitally developed, and eastern or central cities. Mechanism analysis reveals that optimizing Energy Consumption Intensity acts as a short-term driver, while Green Technology Innovation and Environmental Regulation serve as long-term sustainers. Furthermore, moderating effects reveal that Marketization exerts a positive moderating influence. This study provides empirical evidence and policy insights for leveraging IDRE to advance green growth through tailored approaches.

1. Introduction

China continues to confront severe and persistent challenges, including elevated aggregate energy consumption and carbon emissions, alongside an energy consumption structure that remains dominated by conventional fossil fuels and has yet to undergo a fundamental transformation. According to BP’s Statistical Review of World Energy (2023), China ranked first globally in both energy consumption and carbon emission indicators in 2022. Specifically, China’s primary energy consumption reached 5.44 billion tons of standard coal equivalent, accompanied by 11.877 billion tons of carbon dioxide emissions. Further analysis indicates that energy use constitutes the predominant source of national carbon emissions, accounting for approximately 88.83%. Enhancing Energy-Embedded Green Utilization Efficiency of Urban Land (E-GUEUL) is a primary task for accelerating the construction of a modernized energy industry system and for advancing coordinated improvements in carbon mitigation, pollution control, ecological enhancement, and economic growth. Energy-embedded efficiency treats energy use as a binding component of the urban land-use production system. Urban land generates desirable economic output while simultaneously requiring energy inputs and producing environmentally undesirable by-products. E-GUEUL therefore evaluates land-use performance under joint energy and environmental constraints, rather than under a purely desirable output framework. It also represents a necessary avenue for addressing the intertwined constraints of resource scarcity and environmental pressures while advancing the dual carbon goals and high-quality development [1]. As a critical driving force, the deep integration of the digital economy and the real economy, namely the Integration of the Digital–Real Economy (IDRE), can effectively improve E-GUEUL and, by leveraging regional endowments, foster the targeted development of new, high-quality productive forces in the energy sector [2]. Accordingly, IDRE has become a key pathway for accelerating supply-side structural reform in the energy sector, promoting the transformation of energy consumption patterns, enhancing E-GUEUL, and advancing the development of a modernized energy industry system [3].

Figure 1 shows that, over the period of 2011 to 2020, the national mean of E-GUEUL and the level of IDRE exhibited a concurrent upward trajectory. This pattern is consistent with a pathway through which IDRE facilitates the realization of green value and supports economic momentum in the context of the ongoing energy transition. This study is motivated by the need to identify whether deeper IDRE improves city-level E-GUEUL and to clarify the channels and institutional conditions through which such effects may arise. While the growing literature associates digitalization or the digital economy with greener performance, existing evidence is largely correlational and provides limited causal identification for the IDRE–E-GUEUL relationship at the city level. Reverse causality, policy selection, and unobserved city characteristics may jointly influence both IDRE and E-GUEUL, leaving a clear identification gap. From the perspective of thematic relatedness, the literature most closely aligned with this study can be grouped into two categories. In contrast to single-indicator measures, total factor energy efficiency is not confined to energy inputs alone; it evaluates the overall effectiveness with which all production inputs, including energy, capital, and labor, are utilized to achieve a given level of desirable output. A key advantage of this metric is that it incorporates the complex interdependencies among multiple inputs within a unified analytical framework, thereby providing a more complete characterization of actual energy-use performance. As an extension of conventional total factor energy efficiency, E-GUEUL introduces energy consumption and environmentally undesirable outputs as binding constraints in the accounting system, thereby establishing a more comprehensive framework for assessing the sustainability of economic activity. This framework captures not only improvements in production performance but also the contributions of technological progress and environmental management to green growth [4]. From the perspective of thematic relatedness, the literature most closely aligned with this study can be grouped into two categories. The first concerns the measurement of E-GUEUL [5]. Among these, DEA-based models [6] and the SBM framework, which explicitly incorporate undesirable outputs, have increasingly become mainstream [7]. The second category examines the determinants of E-GUEUL. Early work suggests that environmental regulation of different types and environmental information disclosure constitute effective channels for improving total factor energy efficiency [8]. Policy instruments, such as emissions trading schemes, dual-pilot programs for low-carbon and smart cities, and pilot schemes for energy-use rights trading, also contribute to improvements in total factor energy efficiency [9]. With the iterative upgrading of digital technologies, an expanding body of research emphasizes that the utilization of data factors, digital technology innovation, and digital economy development provides enabling support for improvements in E-GUEUL [10].

IDRE emphasizes the complementarity between digital capabilities and real-economy production processes, and thus provides a structural perspective beyond digitalization in isolation [11]. The literature most relevant to this study can be traced to three main strands. First, a broad set of studies focuses on conceptualizing IDRE and developing its theoretical framework. In this view, digital technologies and data as a production factor jointly form a dual driving force behind deeper integration, and the theoretical content of the transformation and upgrading of the real economy exhibits a substantial affinity with the practical features of industrial digitalization [12]. Other contributions emphasize that the expansion and upgrading of the real economy can, in turn, create extensive market space and diversified application scenarios for the digital economy, thereby supporting rapid growth in the latter. Accordingly, IDRE is more appropriately characterized as an innovation-centered form of integrative economic activity, in which the digital economy and the real economy are connected and interact through digital technologies and data factors, exhibiting bidirectional reinforcement and synergistic co-development [13]. Second, with respect to measurement approaches and evolutionary patterns, constructing an indicator system and applying methods such as the entropy-weight method and the coupling coordination degree model have become the mainstream strategies for quantifying the level of IDRE [14]. Although the overall level of IDRE remains at a relatively early stage and interregional disparities in integration have shown a widening tendency, the depth of integration has strengthened over time, with a spatial evolution characterized by multi-nodal expansion and a progression from points to corridors and then to broader regional surfaces. Third, empirical research on the green effects of IDRE has focused on outcomes such as green total factor productivity, the green transformation of development patterns, and coordinated improvements in industrial pollution control and carbon mitigation, providing evidence on the effects of IDRE and the underlying mechanisms [15].

In summary, the existing literature on IDRE and E-GUEUL has become increasingly substantial, providing a solid theoretical foundation for further inquiry, yet meaningful scope for additional research remains. Although some studies have recognized that deeper IDRE can generate green effects, the influence of IDRE on E-GUEUL has not received commensurate attention. Prior work has largely emphasized the enabling role of digital technologies and the digital economy in improving E-GUEUL. However, as the concept of IDRE has been invoked with growing frequency, relatively few studies have focused on the extent to which deeper IDRE serves as a driver of improvements in E-GUEUL and systematically respond to this practical concern. Against this background, to provide comprehensive and reliable empirical evidence regarding the relationship between the two, this study employs a balanced city-level panel of 273 cities over the period 2011 to 2020. From a multidimensional perspective, it constructs an indicator system to measure the level of IDRE. Building on this measurement, the study employs a double machine learning framework to estimate the effect of IDRE on E-GUEUL, document its multidimensional heterogeneity, identify three core pathways through which IDRE operates, and extend the analysis by examining moderating effects. This design directly targets the causal identification gap by flexibly controlling for high-dimensional confounders and reducing specification-driven biases that commonly arise in conventional parametric regressions.

The marginal contributions of this study are threefold. First, building on prior research, the measurement of IDRE is implemented at the prefecture-level city scale. This focus helps move beyond the limitations inherent in province-level measurement and facilitates a more comprehensive assessment of the current state of IDRE across Chinese cities [16]. Second, the study develops a unified analytical framework that jointly examines IDRE and E-GUEUL, providing a systematic discussion of their underlying logic and operating mechanisms [17]. Using a double machine learning approach, the analysis estimates the effect of IDRE on E-GUEUL, thereby mitigating estimation biases that may arise from complex nonlinearities and model misspecification in conventional econometric specifications. These efforts establish a more systematic and reliable empirical basis for subsequent work that seeks further to uncover the internal linkages between IDRE and E-GUEUL. Third, the study proposes and empirically validates the mechanisms of green technological innovation, energy consumption intensity, and environmental regulation [18]. It also examines the moderating role of marketization in the relationship through which IDRE contributes to improvements in E-GUEUL. Together, these analyses clarify the internal mechanisms connecting IDRE to E-GUEUL, enrich and extend the literature on IDRE in the domain of green development, and provide theoretical and practice-relevant insights for advancing deeper IDRE.

The primary objective of this study is to identify and quantify, at the city level, the aggregate effect of IDRE on E-GUEUL and to provide robust empirical evidence regarding their relationship. As secondary objectives, the study systematically examines heterogeneity in the IDRE–E-GUEUL relationship with respect to resource endowments, digital foundations, and regional location; it also offers an in-depth and up-to-date analysis of the principal transmission pathways involving green technological innovation, energy consumption intensity, and environmental regulation. In addition, the extension analysis evaluates whether marketization strengthens the baseline effect estimated in the benchmark regressions. The remainder of the paper is organized as follows. Section 2 develops the theoretical framework and research hypotheses. Section 3 describes the research design, including the empirical specification, data sources, and variable construction. Section 4 presents the empirical results, including descriptive statistics, baseline estimates, robustness checks, endogeneity considerations, heterogeneity analysis, and mechanism tests. Section 5 reports the extension analysis on moderating effects. Section 6 concludes with the main findings and policy implications.

2. Theoretical Analysis and Research Hypotheses

2.1. Direct Effect of IDRE on E-GUEUL

IDRE exerts a dual influence on production factors. It improves the performance of conventional inputs while also establishing data as a production factor. As a new factor characterized by non-rivalry, externalities, and multiplier effects, data support the real economy in reshaping the internal composition of productive capacity through intelligent transformation.

To examine the effect of IDRE on E-GUEUL and the associated mechanisms, this study incorporates data as an input factor into a partial equilibrium framework [19]. The setup is as follows. Consider a firm whose production relies on an input bundle comprising land (N), labor (L), capital (K), and data (D). The production process is accompanied by energy use and carbon emissions (E), which give rise to social environmental costs (S). When the firm organizes production using data, the deep embedding of digital technologies and data factors into the real economy’s production process constitutes a concrete manifestation of IDRE. In this setting, the production function can be written as follows:

$$Y={N^{\alpha }L}^{\beta }{K}^{\gamma }{D}^{\delta }$$

(1)

In the production function, α and β denote the output elasticities of labor and capital, respectively. Their values lie in the interval (0, 1), and they satisfy the returns-to-scale restriction 0 < α + β + γ + δ ≤ 1. Under deeper IDRE, firms in the real sector can leverage digital technologies and data factors to shift their production mode from an extensive pattern characterized by high resource use and high emissions toward a more intensive pattern, thereby improving energy-use performance and reducing total pollutant emissions. Accordingly, carbon emissions and total output can be represented by the following functional relationship:

$$E=\pi \left(\theta , T, C, R\right)Y$$

(2)

E denotes total carbon emissions. $\pi (\theta , T, C, R)$ denotes the emissions coefficient, which depends on abatement effort $\theta$, green technological innovation T, the level of emissions E, and the intensity of environmental regulation R. Carbon emissions E are modeled as a function of energy consumption C and regulatory intensity R. Carbon emitted by the real economy imposes environmental degradation and pollution. The damage function $S\left(E\right)$ is therefore determined by the level of carbon emissions. Accordingly, green output ${ Y}^G$ is defined as:

$${ Y}^G=\left(1-S\right)Y$$

(3)

The marginal contribution of the data factor to green output can be characterized by the following partial derivative:

$${\displaystyle \frac{\partial Y^G}{\partial D}}={\displaystyle \frac{\partial \left(1-S\right)}{\partial D}}Y+\left(1-S\right){\displaystyle \frac{\partial Y}{\partial D}}$$

(4)

China is currently advancing the market-based allocation of data as an economic factor while continuously refining data rights confirmation mechanisms. Digital technologies are being widely applied across various sectors beyond environmental protection, with data collection primarily sourced from smart manufacturing and smart city development rather than solely from energy and environmental monitoring. Furthermore, the strategic initiatives such as “Broadband China” and “East Data, West Computing” prioritize the rapid growth of the digital economy over green governance. Overall, the supply and development of data as an economic factor are primarily driven by the digitalization process, rather than green technological innovation, energy consumption, or environmental regulations.

Therefore, it can be assumed that the data factor is exogenous to green technological innovation, energy consumption and environmental regulation namely, when ${\displaystyle \frac{\partial (1-S)}{\partial D}}$ = 0, data constitutes an essential input in total output. Since ${\displaystyle \frac{\partial Y}{\partial D}}$ =${{ \delta N}^{\alpha }L}^{\beta }{K}^{\gamma }{D}^{\delta -1}$ > 0, it follows that:

$${\displaystyle \frac{\partial Y^G}{\partial D}}=\left(1-S\right){\displaystyle \frac{\partial Y}{\partial D}}>0$$

(5)

Based on the derivation in Equation (5), data input has a significant positive effect on green output. This result suggests that as data resources are integrated into the real economy, they can facilitate knowledge spillovers and information-enabled upgrading, thereby promoting more efficient resource allocation and restructuring the production system. These adjustments strengthen green production capacity and, in turn, generate a favorable marginal effect on E-GUEUL improvements. Accordingly, this study proposes the following hypothesis:

H1: 

IDRE significantly improves E-GUEUL.

2.2. Indirect Effects of IDRE on E-GUEUL

The indirect impact of IDRE on E-GUEUL is a dynamic process characterized by temporal heterogeneity. Specifically, this study posits that IDRE exerts an immediate influence through the optimization of energy consumption intensity in the short term, while providing sustained momentum through green technological innovation and environmental regulatory reinforcement in the long term.

First, in the short term, IDRE promotes improvements in E-GUEUL by improving energy consumption intensity. Compared to the long cycles required for R&D and institutional changes, the optimization of energy use can be realized rapidly through the application of existing digital infrastructure. Improvements in energy-use performance depend primarily on adjustments in both the industrial structure and the energy consumption structure [20]. By promoting industrial digital transformation, IDRE enables production processes in traditionally energy-intensive sectors to become more intelligent and more finely managed, thereby improving energy-use performance per unit of output. On the one hand, the adoption of digital tools such as the Internet of Things, big data, and intelligent manufacturing allows real-time monitoring and dynamic control of production processes, which improves the allocation of energy and materials, enhances overall operational efficiency, and makes energy use along the industrial chain more precise and efficient [21]. On the other hand, IDRE supports the expansion of services and high-technology manufacturing, guiding industrial upgrading and limiting an increase in the share of energy-intensive industries. In turn, energy-intensive activities are more likely to achieve efficiency improvements or shift toward less energy-intensive sectors, which contributes to improvements in aggregate energy consumption intensity at the macro level. In addition, IDRE can guide capital flows toward green and low-carbon industries, thereby providing financing support for economic activities related to environmental improvement, climate change mitigation, and resource recycling [22]. At the same time, deeper IDRE can reshape preferences and demand in energy consumption by reducing reliance on fossil fuels and natural resources, strengthening the propensity for clean energy use and circular economy development, and increasing the share of clean energy in total consumption, thereby promoting improvements in the energy consumption structure [23]. Accordingly, this study proposes the following hypothesis:

H2a: 

IDRE promotes improvements in E-GUEUL by improving energy consumption intensity.

Second, from a long-term perspective, IDRE provides a core driving force for E-GUEUL by accelerating green technological innovation. Unlike the immediate operational improvements seen in energy consumption, technological innovation is a cumulative process characterized by long R&D cycles, high uncertainty, and a gestation period. The Porter hypothesis emphasizes that technological innovation and efficiency improvements constitute fundamental channels through which firms strengthen competitiveness [24]. To begin with, the deep integration of the digital economy and the real economy supports industrial upgrading toward digitalization. The adoption of digital technologies can reduce firms’ costs of acquiring external information and, in turn, improve the efficiency of R&D resource use. Moreover, by deploying techniques such as artificial intelligence and machine learning, firms can enhance the efficiency and accuracy of information processing. This enables firms to identify market demand in a timely manner, reduce the trial-and-error costs associated with R&D activities, shorten the time required for the development and application of green technological outcomes, and accelerate the commercialization and diffusion of green innovations [25].

In addition, IDRE can foster green development by promoting collaborative innovation and industrial upgrading. Specifically, IDRE advances the green transition through multiple channels. It facilitates the formation of efficient collaborative networks linking firms, universities, and research institutions, thereby reducing informational frictions across actors and creating favorable conditions for the rapid diffusion and industrial application of green technologies. Furthermore, supported by digital platforms and data-sharing arrangements, upstream and downstream firms along supply chains can achieve deeper coordination and joint innovation [26]. Beyond the supply chain, IDRE can blur traditional industrial boundaries and promote technological spillovers and resource complementarities across sectors, giving rise to emerging green business forms and models, including the circular economy and the energy internet. At the operational level, firms’ digital transformation enhances refined cost management, enabling a more targeted reallocation of resources toward R&D and innovation activities. This contributes to the formation of production systems that are both environmentally compatible and economically sustainable, thereby providing persistent momentum for growth in E-GUEUL. Overall, IDRE operates through technology-enabled upgrading and managerial transformation to improve allocative efficiency and reduce avoidable losses, while embedding green governance into production and organizational practices in a more systematic manner, thus laying a solid foundation for high-quality development. In this context, reductions in unit production costs and improvements in output value and profitability can further reinforce firms’ innovation incentives, generating a virtuous cycle that supports sustained improvements in E-GUEUL. Accordingly, this study proposes the following hypothesis:

H2b: 

IDRE promotes improvements in E-GUEUL by raising the level of green technological innovation.

Third, also operating as a long-term mechanism, IDRE can strengthen environmental regulatory intensity, thereby providing an institutional basis for improvements in E-GUEUL. From a theoretical perspective, environmental regulation frameworks suggest that stringent environmental policies can induce firms to upgrade their technologies and managerial practices, promoting greener production systems and thereby supporting improvements in E-GUEUL [27]. However, traditional regulatory instruments are often constrained by inherent information barriers and high implementation costs, which limit both enforcement efficiency and targeting precision. Over time, by embedding information technologies into governance processes, IDRE can enhance the implementation capacity and monitoring effectiveness of environmental regulations, thereby providing institutional support for improvements in energy-use performance [28].

On the one hand, digital monitoring and data platforms enable real-time collection of data on pollutant emissions and energy use, thereby reducing information asymmetries and lowering regulatory costs for government agencies. Within an internet-enabled supervision framework, such tools raise the cost for firms of concealing excessive energy use and strengthen the constraining force of environmental policies on firms’ energy-related behavior. On the other hand, IDRE facilitates the operation of market-based mechanisms, such as emissions trading, thereby improving the timeliness and specificity of both governmental and market oversight and enabling market mechanisms to allocate resources more effectively toward energy conservation and emission reduction. From the perspective of social oversight, IDRE also strengthens public participation and media scrutiny. Digital platforms can make environmental information disclosure more transparent and timely, increasing the social costs associated with noncompliance and thereby reinforcing behavioral constraints that encourage firms to engage in energy conservation and emission mitigation [29]. Accordingly, this study proposes the following hypothesis:

H2c: 

IDRE promotes improvements in E-GUEUL by strengthening environmental regulatory intensity.

2.3. Moderating Effect of IDRE on E-GUEUL

Marketization is an essential indicator of the maturity of resource allocation mechanisms and the openness of the institutional environment, and it plays a positive moderating role in the process through which IDRE contributes to improvements in E-GUEUL. First, in highly marketized settings, the mobility of production factors, such as data, technology, capital, and talent, is stronger, allowing digital resources to be allocated more efficiently toward firms or sectors with greater innovative capacity and energy-saving potential [30]. This dynamic optimization of factor allocation raises the marginal returns to IDRE and enables it to exert a more substantial influence on improvements in E-GUEUL. Second, insights from endogenous growth theory suggest that the intensity of innovative activity depends on the degree of market competition and expected returns [31]. Higher levels of marketization are typically associated with stronger property rights protection and more effective competitive mechanisms, which strengthen firms’ incentives to undertake digital transformation and invest in green innovation, thereby increasing the effectiveness of IDRE in promoting improvements in E-GUEUL [32]. Third, in more marketized regions, government intervention is relatively limited, while information disclosure tends to be more transparent and policy implementation more effective. Under these conditions, digital governance platforms and energy-use supervision systems are more likely to function effectively. This reduces institutional frictions in the implementation of IDRE and, through market-signal mechanisms, guides firms toward improved energy-use performance [33]. Accordingly, this study proposes the following hypothesis:

H3: 

Marketization positively moderates the effect of IDRE on improvements in E-GUEUL.

In summary, the theoretical framework of this study is presented in Figure 2.

3. Research Design

3.1. Model Specification

IDRE may involve network effects, threshold effects, and complex factor reallocation. These intricate interactions are difficult to capture precisely within traditional linear interaction models (OLS). The partially linear double machine learning (DML) framework, proposed by Chernozhukov et al. (2018) [34], is adopted as the primary approach for causal inference in this study to accurately identify the effect of IDRE on E-GUEUL. Specifically, the analysis employs cross-fitted estimators based on Neyman-orthogonal moment conditions. By combining orthogonal moment construction with sample-splitting and cross-fitting, this framework jointly addresses three key econometric challenges. First, it effectively mitigates estimation bias arising from nonlinear relationships among covariates and alleviates the curse of dimensionality in high-dimensional settings. Second, it overcomes limitations common to conventional machine learning applications in empirical work, including overfitting tendencies and difficulties in constructing valid confidence intervals. Third, under a two-way fixed effects specification, it improves the precision of parameter estimation and enhances the reliability of statistical inference. Therefore, compared to conventional econometric models, double machine learning models (DML) represent a superior choice.

$${E\text{-}\mathit{GUEUL}}_{it}=\theta {\mathit{IDRE}}_{it}+f\left(X_{it}\right)+U_{it}$$

(6)

$$E\left(U_{it}\mid {\mathit{IDRE}}_{it},X_{it}\right)=0$$

(7)

$${\mathit{IDRE}}_{it}=g\left(X_{it}\right)+V_{it}$$

(8)

$$E\left(V_{it}\mid X_{it}\right)=0$$

(9)

Here, $i$ indexes cities and $t$ indexes years. ${\mathrm{IDRE}}_{\mathrm{i}\mathrm{t}}$ is the explanatory variable capturing the level of IDRE, and ${E\text{-}\mathit{GUEUL}}_{it}$ is the outcome variable capturing E-GUEUL. $\theta$ is the coefficient of interest used to assess whether IDRE is associated with improvements in E-GUEUL. A significantly positive $\theta$ indicates that IDRE is conducive to improvements in E-GUEUL, whereas a negative $\theta$ suggests an adverse association with the evolution of E-GUEUL. $X_{it}$ denotes a vector of control variables that may affect the outcome. $f\left(X_{it}\right)$ and $g\left(X_{it}\right)$ are unknown functions to be approximated using machine learning methods. $U_{it}$ and ${ V}_{it}$ are idiosyncratic disturbance terms that are mutually independent and have zero mean.

Building on the existing methodology, this study further employs the double machine learning framework to construct a partially linear instrumental variables model. The model is specified as follows:

$${E\text{-}\mathit{GUEUL}}_{it}=\theta {\mathit{IDRE}}_{it}+f\left(X_{it}\right)+U_{it}$$

(10)

$${\mathit{IV}}_{it}=g\left(X_{it}\right)+V_{it}$$

(11)

Here, ${\mathit{IV}}_{it}$ denotes the instrumental variable, and the meanings of the remaining variables are the same as above.

3.2. Variable Definition and Explanation

3.2.1. Explanatory Variable

The key explanatory variable is the city-level measure of IDRE. IDRE characterizes a systematic and dynamic process through which the digital economy and the real economy, via technological penetration, business-model innovation, and value co-creation, ultimately form a development paradigm and economic ecosystem featuring deep synergy and co-evolution. The level of IDRE can be reflected in the degree of interactive integration and coordinated development between these two systems. Accordingly, following standard practice in the related literature, this study refers to existing research and constructs an index system to measure the development level of the digital economy (DE) based on five indicators, including Digital penetration, Digital workforce, Digital service output, Mobile digital access and Digital finance [35]. And it builds an index system for the development level of the real economy (RE) based on four dimensions: output, investment, consumption, and credit [36]. The specific explanation of each indicator is shown in

Table 1. Indicator System for Measuring IDRE.

Overall Construct	First-Level Indicator	Second-Level Indicator	Third-Level Indicator	Attribute
Digital–real economy integration	Digital development supporting urban land-system functions	Digital penetration in urban space	Internet users per 100 persons (urban connectivity intensity)	+
		Digital workforce in urban land-system economy	Employment share in computer services and software (digital employment structure)	+
		Digital service output in urban space	Per capita mobile telecommunications services (digital service intensity)	+
		Mobile digital access in urban space	Mobile Internet users per 100 persons (mobile access intensity)	+
		Digital finance enabling urban land-system activities	Digital Inclusive Finance Index (digital financial inclusion)	+
	Land-based real-economy development	Land-based industrial output in urban production space	Share of secondary-sector value added (industrial structure)	+
			Profits of above-scale industrial enterprises per 10,000 persons (industrial profitability)	+
			Number of above-scale industrial enterprises (industrial agglomeration)	+
		Investment in urban production space	Per capita fixed-asset investment (capital deepening)	+
		Consumption capacity supporting urban land-system demand	Per capita retail sales of consumer goods (consumption intensity)	+
		Credit support to urban land-system economy	Share of outstanding RMB loans of financial institutions (credit availability)	+

. In terms of methodology, the entropy method is first employed to conduct a comprehensive evaluation and quantitative measurement of the digital economy development subsystem and the real economy growth subsystem. On this basis, a coupling coordination degree model is introduced to rigorously assess the level of synergistic development and the overall coordination state between the two systems. The specific measurement procedure follows the steps below. In the first step, to eliminate potential interference arising from differences in units and magnitudes across indicators, the constructed indicator system is standardized. De-pending on whether an indicator contributes positively or negatively to the system, the following standardization formulas are applied to benefit-type and cost-type indica-tors, respectively.

For benefit-type indicators, the standardized value is computed as

$$A_{ij}=\left(X_{ij}-min\left(X_{ij}\right)\right)/\left(max\left(X_{ij}\right)-min\left(X_{ij}\right)\right)$$

(12)

whereas for cost-type indicators, the standardization is given by

$$A_{ij}=\left(max\left(X_{ij}\right)-X_{ij}\right)/\left(max\left(X_{ij}\right)-min\left(X_{ij}\right)\right)$$

(13)

Here, $X_{ij}$ denotes the raw data. $min\left(X_{ij}\right)$ and $max\left(X_{ij}\right)$ are, respectively, the minimum and maximum values of indicator $j$ in the sample. $A_{ij}$ is the standardized value of indicator $j$.

In the second step, the share of province $i$ in indicator $j$ is calculated as

$$P_{ij}=X_{ij }/\sum _{i=1}^n{X}_{ij}\left(\mathrm{i}=1,2,\dots ,\mathrm{n}; \mathrm{j}=1,2,\dots ,\mathrm{m}\right)$$

(14)

The information entropy of indicator $j$ is then computed as

$${ e}_j=-\left(1/\mathrm{l}\mathrm{n}\left(n\right)\right)\sum _{i=1}^n{P}_{ij} ln\left(P_{ij}\right)$$

(15)

The information utility (redundancy) is defined as $g_{ij}=1-e_j$. A larger information utility value indicates that the corresponding indicator plays a greater relative role in the evaluation system.

The entropy weight for indicator $j$ is defined as

$$W_j={ g}_j/\sum _{j=1}^m{g}_j\left(1\le \mathrm{j}\le \mathrm{m}\right)$$

(16)

In the third step, the composite evaluation score is calculated as

$$\mathrm{F}=\sum _{\mathrm{i}=1}^{\mathrm{m}}{\mathrm{W}}_{\mathrm{j}} ·{ \mathrm{A}}_{\mathrm{ij}}\left(\mathrm{i}=1,2,\dots ,\mathrm{n}\right)$$

(17)

In the fourth step, based on the composite development scores of the digital-economy subsystem and the real-economy subsystem obtained from the entropy method, the coupling coordination degree model is introduced to quantify the level of synergistic development between the two systems. The model is specified as follows:

$$C_i=2\sqrt{D{E}_i\times R{E}_i}/\left(D{E}_i+R{E}_i\right)$$

(18)

$$T_i=\alpha \times D{E}_i+\beta \times R{E}_i$$

(19)

$$\mathit{IDRE}=\sqrt{C_i\times T_i}$$

(20)

Here, $C_i$ denotes the coupling degree between digital-economy development and real-economy development, and $T_i$ represents their coordination level. $\alpha$ and $\beta$ are the respective weights assigned to the digital economy and the real economy, with $\alpha$ + $\beta$ = 1. Given that the two subsystems play equally important roles in supporting high-quality economic development, the weights are set symmetrically at 0.5. When IDRE approaches 1, this indicates a high level of integration between the digital economy and the real economy, reflected in a strong degree of interconnection between the two subsystems.

3.2.2. Explained Variable

The key dependent variable in this study is E-GUEUL. This study conceptualizes each prefecture-level city as an integrated urban land system and spatial production unit, in which production activities are embedded in land-based socioeconomic space and generate environmental pressures borne by local land–water systems. Compared to the traditional GUEUL indicator, the E-GUEUL assessment focuses not only on input factors such as land, labor, and capital when evaluating the green utilization of urban land systems, but also emphasizes the energy–environment-embedded production process with undesirable emissions. At present, there is no unified standard in the literature for measuring city-level E-GUEUL. Among the available approaches, methods developed within the data envelopment analysis (DEA) framework have been widely adopted because they can simultaneously accommodate multiple inputs, desirable outputs, and undesirable outputs—features that are well suited to evaluating the green utilization performance of urban land systems under environmental pressures. The super-efficiency SBM–Malmquist index model combines the strengths of the super-efficiency SBM model in handling slacks with the Malmquist index for dynamic analysis, making it particularly suitable for the precise measurement and intertemporal comparison required in this study. By incorporating undesirable outputs into the efficiency evaluation system, the model yields a more comprehensive and informative assessment. Moreover, the Malmquist index enables a dynamic decomposition of efficiency changes over time, clarifying both the direction and potential sources of variation. In this way, the approach addresses limitations of conventional DEA models, including the omission of undesirable outputs and the lack of intertemporal comparability. It provides a more systematic and objective basis for measuring and evaluating efficiency.

Accordingly, this study sets 2011 as the base year and conducts a dynamic measurement of E-GUEUL using the super-efficiency SBM–Malmquist index model. In order to more comprehensively measure the efficiency of land use as a core factors, land, labor, capital and energy are selected as inputs: Land is measured by the area of urban built-up zones (sq.km); labor is proxied by the number of employed persons in municipal districts; capital is measured by the capital stock constructed via the perpetual inventory method [37], with the depreciation rate set at 9.6%; and the energy input is proxied by total energy consumption for each prefecture-level city. Desirable output is measured by real regional GDP. Additionally, industrial sulfur dioxide emissions, industrial smoke and dust emissions, and industrial wastewater discharge are incorporated as undesirable outputs, representing pollution loads associated with land-based industrial activities in urban production space.

3.2.3. Mechanistic Variables

To systematically examine potential indirect channels through which IDRE may affect city-level E-GUEUL, this study follows the theoretical framework and hypotheses developed above. Consistent with the related literature, it selects green technological innovation, energy consumption intensity, and environmental regulatory intensity as mechanism variables [38].

(1) Green technological innovation (lnGTI). Using targeted searches of the China National Intellectual Property Administration’s patent database and compiling the results by category, this study collects the number of green invention patents and green utility model patents. The two counts are summed to obtain the total number of green patent applications, and the natural logarithm of (total green patent applications + 1) is taken to measure the level of green technological innovation. (2) Energy consumption intensity (ECI). Following the approach of Pang G et al. (2025) [39], total energy consumption is first derived from DMSP/OLS nighttime light data. Energy consumption intensity is then computed as the ratio of total energy consumption to regional GDP. (3) Environmental regulation intensity (ER). This study processes the text of government work reports using Python (version 3.12)-based tokenization. It then constructs a quantitative baseline indicator of environmental regulation intensity as the ratio of the frequency of environment-related regulatory keywords to the total number of words in the text [39].

3.2.4. Control Variables

To effectively control for and minimize estimation bias arising from omitted variables, thereby enabling a more accurate identification of the effect of IDRE on city-level E-GUEUL and the associated mechanisms [40], this study includes the following control variables: (1) educational attainment (lnHuman), measured as the natural logarithm of the ratio of the number of enrolled college students to the year-end permanent resident population. (2) government intervention (Gov), proxied by the share of local fiscal budget expenditures in regional GDP; (3) infrastructure development (lnRoad), measured as the natural logarithm of urban road area per capita; (4) built-up area greening (Gre), captured by the ratio of green space area within built-up areas to the total built-up area; and (5) technological innovation capacity (TI), proxied by the share of science and technology expenditures in total fiscal budget expenditures.

3.3. Data Sources and Descriptive Statistics

Considering both sample adequacy and data availability, this study excludes cities in regions with substantial missing information, such as Tibet and Xinjiang. The final sample comprises 273 prefecture-level-and-above cities in China. The study period spans from 2011 to 2020, and a balanced panel is constructed for the empirical analysis [41]. The data are primarily drawn from the official publications of the China City Statistical Yearbook, the China Urban Construction Statistical Yearbook, and the China Energy Statistical Yearbook, supplemented with relevant information from individual city statistical yearbooks and national economic and social development statistical bulletins.

The digital inclusive finance data are obtained from the Digital Inclusive Finance Index jointly compiled by the Institute of Digital Finance at Peking University and Ant Group. Green patent data are collected through systematic searches and category-based tabulation of the China National Intellectual Property Administration’s patent database. Total energy consumption is estimated from DMSP/OLS nighttime light data using remote sensing inversion techniques. For variables with partially missing observations, the study further supplements the dataset using city-level annual statistical reports and applies linear interpolation to complete the imputation. Descriptive statistics for all variables are reported in

Table 2. Descriptive Statistics.

Variable	Obs	Mean	Std. Dev.	Min	Max
E-GUEUL	2730	1.0689	0.165	0.43	2.85
IDRE	2730	0.3155	0.075	0.16	0.66
lnHuman	2730	4.7356	1.540	−3.24	8.67
Gov	2730	0.1947	0.086	0.04	0.68
lnRoad	2730	2.7875	0.428	0.31	4.10
Gre	2730	0.4031	0.048	0.15	0.65
TI	2730	0.0170	0.017	0.00	0.21
lnGTI	2730	5.0245	1.642	0.00	10.27
ECI	2730	0.7634	0.411	0.08	3.89
ER	2730	0.0058	0.002	0.00	0.02

.

Because the construction of the E-GUEUL index requires the initial year to serve as the base period and to be normalized to 1, this study selects 2012 and 2020 as benchmark time points and plots the spatiotemporal evolution of IDRE and E-GUEUL across Chinese cities in Figure 3 and Figure 4. From a spatial perspective, both IDRE and E-GUEUL exhibit a multi-nodal distribution; however, the overall pattern follows a west-to-east gradient, with higher levels observed in the eastern regions. This spatial differentiation remains relatively stable over the eight-year observation window and does not display pronounced structural shifts. From a temporal perspective, in 2012, both IDRE and E-GUEUL were broadly at an early stage, with integration depth and efficiency performance still having substantial scope for improvement. With the sustained development of the digital economy and the accelerated convergence of related factors, both indicators are expected to show marked improvement by 2020.

4. Empirical Results

4.1. Baseline Regression

This study employs the double machine learning (DML) framework for causal identification to estimate the effect of IDRE on city-level E-GUEUL and to examine the associated mechanisms. In the baseline implementation, the sample-splitting ratio is set to 1:4, and cross-fitting is performed with a random forest learner. The estimation results are reported in

Table 3. Baseline regression results.

	(1)	(2)	(3)	(4)	(5)
Variable	E-GUEUL	E-GUEUL	E-GUEUL	E-GUEUL	E-GUEUL
IDRE	0.549 ***	1.399 ***	0.427 ***	0.427 ***	0.427 ***
	(0.061)	(0.075)	(0.072)	(0.088)	(0.131)
Controls	YES	NO	YES	YES	YES
City FE	NO	YES	YES	YES	YES
Year FE	NO	YES	YES	YES	YES
N	2730	2730	2730	2730	2730

Note: * indicates statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level; the same applies hereinafter. Standard errors are reported in parentheses for columns (1)–(3), robust standard errors are reported in parentheses for column (4), and city-level clustered standard errors are reported in parentheses for column (5).

. Columns (1)–(3) present, in turn, the estimates obtained without controls, with controls only, and with both controls and city-by-year two-way fixed effects. Across all specifications, the estimated coefficient on the key explanatory variable, IDRE, is positive and statistically significant at the 1% level. This pattern indicates a stable positive association between IDRE and improvements in city-level E-GUEUL, providing initial empirical support for Hypothesis H1.

Furthermore, Columns (4) and (5) report results using robust standard errors and city-level clustered standard errors, respectively. The estimates remain unchanged, with the IDRE coefficient consistently at 0.427 and significant at the 1% level. These results provide statistically robust evidence that IDRE is an important channel supporting improvements in E-GUEUL, thereby offering stronger validation for Hypothesis H1.

4.2. Robustness Checks and Endogeneity Considerations

4.2.1. Alternative Machine Learning Learners

To minimize the possibility that the learner’s choice drives the baseline DML results, this study conducts robustness checks by replacing the machine learning algorithm. Specifically, gradient boosting and support vector machines are used for re-estimation to assess whether changes in the learner affect the baseline conclusion [42]. Columns (1) and (2) of

Table 4. Robustness check results (1).

	(1)	(2)	(3)	(4)
Variable	E-GUEUL	E-GUEUL	E-GUEUL	E-GUEUL
IDRE	0.415 ***	0.164 ***	0.445 ***	0.445 ***
	(0.064)	(0.049)	(0.071)	(0.073)
Controls (linear terms)	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
N	2730	2730	2730	2730

report estimates based on Lasso gradient boosting and support vector machines, respectively. The results show that IDRE continues to have a statistically significant positive effect on E-GUEUL at the 1% level, which is highly consistent with the baseline findings and further supports the robustness of the main conclusion.

4.2.2. Alternative Sample-Splitting Ratios

Another common hyperparameter robustness strategy in machine learning applications is to vary the split between the training and estimation samples in the partially linear model. This study replaces the baseline 1:4 split with 1:2 and 1:6 to examine the sensitivity of the results to alternative sample-splitting choices [43]. The corresponding estimates are reported in Columns (3) and (4) of Table 4. Across both splits, the estimated coefficient on IDRE remains positive and statistically significant at the 1% level, providing consistent empirical support for Hypothesis H1.

4.2.3. Two-Sided Winsorization

City- and firm-level data may be affected by measurement error, definitional inconsistencies, or transitory shocks. A small number of extreme observations can inflate variance and destabilize estimated coefficients, thereby weakening the reliability of inference. Accordingly, this study applies two-sided winsorization at the 1% level to all continuous variables to limit the influence of extreme values on estimation results [44]. As shown in Column (1) of

Table 5. Robustness check results (2).

	(1)	(2)	(3)	(4)	(5)
Variable	E-GUEUL	E-GUEUL	E-GUEUL	E-GUEUL	E-GUEUL
IDRE	0.419 ***	0.416 ***	0.282 ***	0.318 ***	3.289 ***
	(0.074)	(0.074)	(0.053)	(0.121)	(0.337)
Controls (linear terms)	YES	YES	YES	YES	YES
Controls (quadratic terms)	NO	YES	NO	NO	NO
City FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
N	2730	2730	2380	2730	2400

, the main findings remain intact after this treatment.

4.2.4. Nonlinear Terms for Controls

Although Chernozhukov et al. showed that DML can mitigate the curse of dimensionality associated with high-dimensional controls [34], estimation performance may still depend on the empirical setting, including sample construction and the underlying data-generating process. Expanding the set of covariates can also amplify multicollinearity, potentially affecting estimation precision [45]. To assess sensitivity to richer functional forms, this study augments the baseline specification with nonlinear terms for the control variables. Column (2) of Table 5 shows that the key estimate remains stable, supporting the robustness of the conclusion.

4.2.5. Excluding Core Cities

To reduce the risk that resource agglomeration associated with administrative status confounds the estimates, this study excludes municipalities directly under the central government, provincial capitals, and sub-provincial cities, retaining only ordinary prefecture-level cities for a robustness check [46]. This approach helps reduce systematic differences arising from policy advantages, location-related benefits, and unequal digital infrastructure. As reported in Column (3) of Table 5, IDRE continues to show a statistically significant positive association with improvements in E-GUEUL after excluding core cities.

4.2.6. Replace the Calculation Method of the Explanatory Variable

The explanatory variable (IDRE) holds a core position in this study. To mitigate potential measurement errors and enhance the credibility of the findings, a sensitivity analysis was conducted by altering the calculation method of the explanatory variable. The weights α and β for the digital economy and real economy were respectively set to 0.4 and 0.6, and regression estimates were recalculated. As shown in Column (4) of Table 5, the conclusions remain robust and credible after altering the calculation method for the explanatory variables.

4.2.7. Instrumental Variables Strategy

Omitted variables and reverse causality may bias the estimates. Following the related literature, this study applies an instrumental variables approach to mitigate potential endogeneity. Specifically, it constructs a panel-suitable instrument by interacting the number of fixed-line telephones per 10,000 persons in each prefecture-level city in 1984 with the national number of internet users in the previous year, and uses this interaction as an instrument for IDRE [47]. The identification strategy relies on three considerations. First, fixed-line telephony reflects historical communication infrastructure and path dependence in regional communications technology, which is plausibly correlated with contemporary IDRE, satisfying relevance. Second, historical fixed-line provision has no direct contemporary channel affecting E-GUEUL in a modern sense. It is orthogonal to the control variables and the idiosyncratic error term, supporting the exclusion restriction. Third, interacting the cross-sectional historical measure with the evolving national internet penetration converts the instrument into a time-varying series suitable for panel analysis, addressing the limitation that the original historical measure is time-invariant.

Using the partially linear IV-DML specification in Equations (5) and (6), the results in Column (5) of Table 5 show that the IDRE coefficient remains positive and statistically significant at the 1% level. This indicates that the baseline conclusion is preserved after addressing endogeneity with instrumental variables.

4.3. Heterogeneity Analysis

4.3.1. Resource Endowment Heterogeneity

Natural resources form a material foundation for urban energy systems, and differences in resource endowments may generate pronounced regional heterogeneity in the effect of IDRE on E-GUEUL. Accordingly, this study examines heterogeneity along the resource endowment dimension by dividing the 273 prefecture-level cities into two groups: 108 resource-based cities and 165 non-resource-based cities. It then estimates the differential effects of IDRE on E-GUEUL across the two groups [48]. The results, reported in Columns (1) and (2) of

Table 6. Heterogeneity analysis results.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variable	Resource-Based	Non-Resource-Based	Digitally Advanced	Digitally Less Advanced	Eastern	Central	Western
IDRE	0.231 ***	0.564 ***	0.499 ***	0.379 ***	0.816 ***	0.309 ***	0.156 **
	(0.060)	(0.103)	(0.118)	(0.071)	(0.162)	(0.075)	(0.061)
Controls	YES	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES	YES
N	1080	1650	1360	1370	980	1000	750

, show that the estimated coefficient on IDRE is 0.231 for resource-based cities and 0.564 for non-resource-based cities, both positive and statistically significant at the 1% level. Thus, IDRE is associated with improvements in E-GUEUL in both groups, with a substantially larger effect in non-resource-based cities. A plausible explanation is that non-resource-based cities face less severe path dependence and have more flexible, diversified industrial structures. With an industrial base centered on high-value-added services and technological innovation, these cities can more rapidly move into strategic positions in digital and green technologies. This first-mover advantage may facilitate stronger absorption of innovation spillovers, thereby yielding a more pronounced contribution to improvements in E-GUEUL.

4.3.2. Digital Economy Heterogeneity

The digital economy is an economic form driven by new-generation digital technologies, with data resources serving as a key production factor and modern information networks as a central carrier. It exerts broad and deep influences on the transformation of growth patterns and the pursuit of high-quality development. Because the level of digital-economy development directly shapes the degree of IDRE, it may also induce heterogeneity in E-GUEUL outcomes. This study first measures the level of digital economy development for each prefecture-level city and then divides the sample at the median into two groups: a more developed group and a less developed group for subgroup estimation [49]. Columns (3) and (4) of Table 6 report the results. The estimated IDRE coefficients are 0.499 for the more developed group and 0.379 for the less developed group, both of which are positive and significant at the 1% level. Thus, irrespective of the level of digital economy development, IDRE is associated with improvements in E-GUEUL, with the effect being more pronounced in cities with more advanced digital economy development. One underlying reason is that, relative to less-developed cities, more-developed cities possess more comprehensive digital infrastructure, more advanced digital technologies, and more abundant data, which together strengthen the capacity of IDRE to contribute to improvements in E-GUEUL.

4.3.3. Spatial-Location Heterogeneity

Differences in spatial location constitute a basis for the spatial differentiation of IDRE, and such heterogeneity can be transmitted through specific channels to generate corresponding regional differentiation in E-GUEUL. Based on geographic location, this study divides the 273 cities into three subsamples—98 eastern cities, 100 central cities, and 75 western cities—and estimates the model separately for each group [50]. The results in Columns (5)–(7) of Table 6 show that the estimated coefficient on IDRE is 0.816 for eastern cities and 0.309 for central cities, both significant at the 1% level, while the estimate coefficient for western cities is 0.156 and significant at the 5% level. This pattern indicates a gradient in the association between IDRE and improvements in E-GUEUL that weakens from east to west. Although the coefficient for the western region passed the significance test, its marginal effect remains significantly lower than that of the eastern region. This statistical finding suggests that pure digital investment in western cities may face constraints of diminishing marginal returns. This significant spatial disparity suggests the existence of deep-seated structural barriers, which will be discussed in detail in Section 5.

4.4. Mechanism Tests

Building on the theoretical framework developed above, this study conducts mechanism tests across three dimensions, namely green technological innovation, energy consumption intensity, and environmental regulation intensity. The results are reported in

Table 7. Mechanism test results.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	ECI	lnGti	lnGti1	ER	ER1	E-GUEUL
IDRE	−1.266 ***	8.246 ***	8.467 ***	0.007 ***	0.008 ***
	(0.159)	(0.361)	(0.361)	(0.001)	(0.002)
IDRE_Market						0.356 ***
						(0.056)
Controls (linear terms)	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
N	YES	YES	YES	YES	YES	YES

.

4.4.1. Energy Consumption Intensity

To evaluate whether energy consumption intensity constitutes a mechanism through which IDRE contributes to improvements in E-GUEUL, energy consumption intensity is defined as the ratio of total energy consumption to regional GDP [17]. The results are reported in Column (1) of Table 7. The mechanism test reveals that the estimated coefficient of IDRE on energy consumption intensity is statistically significant and negative at the 1% level, indicating that IDRE is associated with a reduction in energy consumption intensity, namely, lower energy use per unit of output. This provides evidence for an important transmission pathway through which improvements in energy consumption intensity contribute to improvements in E-GUEUL. A substantial empirical literature has documented that improvements in energy consumption intensity are typically accompanied by adjustments in the energy structure, gains in energy conversion efficiency, and the development of cleaner energy, and have therefore been treated as a central lever for improving E-GUEUL. Consistent with this reasoning, the results indicate that IDRE contributes to improvements in E-GUEUL through the energy consumption intensity channel, providing support for H2a.

4.4.2. Green Technological Innovation

Green technological innovation is measured as the logarithm of the number of green patent applications and grants, with a +1 adjustment applied before taking the logarithm. This measure is used to assess the mediating role of green technological innovation in the relationship between IDRE and E-GUEUL. Columns (2) and (3) of Table 7 present the estimation results. The mechanism tests indicate that the coefficient associated with green technological innovation as a key mediator is positive and statistically significant at the 1% level. This finding supports the channel through which IDRE strengthens green technological innovation, thereby contributing to improvements in E-GUEUL. Related research suggests that neutral green technological progress can raise the productivity of multiple inputs proportionally without increasing the use of other production factors. In contrast, biased green technological progress primarily operates through the adoption of energy-saving technologies and equipment, thereby strengthening the output elasticity of energy as a production factor [51]. Overall, the evidence suggests that IDRE provides crucial enabling support for green technological innovation, serving as an effective transmission mechanism for improvements in E-GUEUL, consistent with the theoretical prediction of H2b.

4.4.3. Environmental Regulation Intensity

To assess whether environmental regulation intensity serves as a mechanism through which IDRE contributes to improvements in E-GUEUL, this study operationalizes environmental regulation intensity using text-mining methods. Specifically, the measure is constructed as the relative frequency of environment-related terms in government work reports [52]. Columns (4) and (5) of Table 7 report results based on two text specifications, namely the full-mode total word-frequency measure and the precise-mode total word-frequency measure. The estimated coefficients for environmental regulation intensity are positive and statistically significant at the 1% level, indicating that IDRE is associated with stronger environmental regulation intensity. Existing studies suggest that, although formal and informal environmental regulation may differ in their short-run effects, both tend to support improvements in E-GUEUL over longer horizons. Taken together, these findings suggest that IDRE enhances environmental regulation intensity, thereby contributing to improvements in E-GUEUL, and provide empirical support for H2c.

4.5. Moderating Effect

Building on the theoretical analysis above, this study further evaluates whether marketization moderates the relationship through which IDRE contributes to improvements in E-GUEUL. Specifically, it employs a prefecture-level marketization index Market and constructs an interaction term between marketization and IDRE, denoted IDRE_Market. The baseline explanatory variable is then replaced by this interaction term for estimation [53]. The corresponding results are reported in Column (6) of Table 7. The estimated coefficient on IDRE_Market is positive and statistically significant at the 1% level. This evidence suggests that higher marketization, by enhancing institutional arrangements and enhancing allocative efficiency, significantly reinforces the positive contribution of IDRE to improvements in E-GUEUL, thereby providing robust support for Hypothesis H3.

5. Discussion

5.1. Interpretation of Regional Heterogeneity

Our findings in Section 4.3.3 reveal a gradient from East to West. This aligns with the structural barriers described by Morán et al., regarding the Heihe–Tengchong Line [54], and Glaeser and Xiong, regarding structural imbalances in urbanization and population distribution [55]. A plausible explanation is the existence of a “critical scale” in the empowerment of digitalization on the real economy. Although national strategies such as “East Data and West Computing” have accelerated the digitalization process in the western region, the phenomenon of “computing power produced in the west, but the application is in the east” may lead to the dislocation of value spillover and energy consumption costs, such that the digital infrastructure investment (such as data centers) in the western region has not been converted into local energy efficiency improvement and therefore has not yet triggered the critical point of E-GUEUL’s leap, with high input cost and benefits that have not yet appeared. In addition, despite decades of policies aimed at achieving regional equilibrium, population density and urban development patterns have been consistent with the Heihe–Tengchong line, suggesting that regional differences between the eastern, central and western regions are not only due to inadequate technological or digital infrastructure, but also to structural mismatches between population and industrial agglomeration. Digitalization empowers the real economy, and the population density needs to reach a certain threshold, while many cities in the west are limited by the Heihe–Tengchong line and have not reached this critical point.

5.2. Mechanism Pathways

The mechanism analysis indicates distinct temporal characteristics. Energy consumption intensity acts as a short-term driver, while green technology innovation and environmental regulation require longer cycles. Clearly, distinguishing between short-term and long-term mechanisms will enhance the internal logical consistency of the argumentation. It will also help further clarify the prioritization and hierarchical classification of policy proposals, facilitating the subsequent formulation of corresponding short-term, medium-term, and long-term policies.

5.3. Limitations and Future Prospects

This study remains subject to several limitations that merit careful scrutiny, while also identifying clear directions for subsequent inquiry. First, the results may, to a certain extent, be shaped by indicator construction and sample specification. Although E-GUEUL is assessed using a super-efficiency SBM–Malmquist framework to capture green land-use efficiency under energy and environmental constraints, its estimation necessarily depends on the selection of input–output variables, the specification of undesirable outputs, and the harmonization of statistical calibers; accordingly, potential measurement error stemming from index construction and data linkage cannot be entirely excluded. As a composite construct, IDRE is operationalized through a multidimensional indicator system; however, given constraints on data availability, it may not fully reflect deeper elements such as platform ecosystems, market-oriented mechanisms for data-factor allocation, industrial-chain coordination, and organizational restructuring, and thus may provide only an approximate representation of the intensity of digital–real integration. Second, data timeliness and the characteristics of city-level statistics may constrain the generalizability of the conclusions: the sample covers only 2011–2020 and therefore does not incorporate structural shifts since 2021 associated with the accelerated evolution of digital technologies, progress in foundational data institutions, and the further consolidation of the “dual-carbon” policy framework, while cross-city heterogeneity in statistical practices, the handling of missing observations, and potential spatial dependence may also influence the robustness of the empirical estimates. Third, methodological application warrants transparency and prudence. Although double/debiased machine learning offers advantages in alleviating model-specification bias and accommodating high-dimensional confounding, the estimates may be sensitive to the choice of algorithms, hyperparameter configurations, sample-splitting procedures, and cross-fitting strategies, and identification continues to rely on key assumptions, including the sufficiency of controlling for observable confounders. In light of these limitations, future research could extend the analysis to more recent periods and reinforce causal identification by exploiting quasi-natural experiments, such as exogenous policy shocks; it could also integrate firm-level microdata, high-frequency energy consumption and carbon emissions records, or remote sensing information to refine the measurement of IDRE and E-GUEUL and to systematically examine spatial spillovers and transmission pathways. Additionally, cross-regional or cross-national comparative analyses could be undertaken to evaluate the applicability of these findings across institutional contexts, thereby contributing to a more broadly generalizable theoretical framework and more policy-relevant implications.

6. Conclusions and Policy Recommendations

6.1. Research Conclusions

This study employs a balanced panel of 273 prefecture-level cities in China from 2011 to 2020 and proceeds with empirical analysis using the following methodological framework. First, it applies the entropy-weight method in conjunction with the coupling coordination degree model to construct a multidimensional composite evaluation system for IDRE and to quantify its development level. Second, it employs the super-efficiency SBM–Malmquist index model to systematically measure city-level E-GUEUL. Building on these measurements, the study employs a double machine learning framework to estimate the effect of IDRE on E-GUEUL, document heterogeneous patterns, and examine the underlying mechanisms. The main findings are as follows. First, IDRE is associated with statistically significant improvements in E-GUEUL, and this result remains robust after applying seven robustness strategies and treating for endogeneity. Second, heterogeneity analyses along dimensions of resource endowments, digital-economy development, and spatial location indicate that the association between IDRE and improvements in E-GUEUL is more pronounced in non-resource-based cities, cities with more advanced digital-economy development, and cities in the eastern and central regions. Third, mechanism analyses indicate that IDRE contributes to improvements in E-GUEUL through three channels: strengthening green technological innovation, improving energy consumption intensity, and strengthening environmental regulation intensity. Fourth, marketization exhibits a statistically significant positive moderating effect, implying that more complete market mechanisms can amplify the green benefits of IDRE. This provides empirical support for institutional innovation as a foundation for achieving the joint benefits of digital transformation and green development.

6.2. Policy Recommendations

Based on these findings, the study distills the following policy implications.

First, implement differentiated policy packages to reinforce targeted supply. In light of the heterogeneity findings—showing more salient effects in non-resource-based cities and in the eastern region—policy instruments should be calibrated to local factor endowments and governance capacity. In the eastern region and in non-resource-based cities, priority should be given to strengthening data-factor institutions and expanding advanced application scenarios by clarifying the boundaries of data rights, standardizing circulation rules, enhancing cross-department interoperability, and scaling full-chain platforms that integrate R&D, production, supply chains, and energy as well as carbon management. These measures can enhance data operability and business-process coordination, thereby improving allocative efficiency and energy-use performance. In central and western regions, a sequenced pathway—strengthening digital foundations before upgrading capabilities—should be adopted by accelerating 5G and industrial-internet coverage, clustering data centers and computing nodes around key industrial parks, and constructing shared platforms that reduce participation thresholds for SMEs. Building on this foundation, projects and anchor firms should be leveraged to attract and cultivate talent by embedding fiscal subsidies, housing support, and school–enterprise joint training into application pilots and job-creation initiatives, thereby establishing stable mechanisms for talent supply and retention. In parallel, adoption incentives should be reinforced through coordination with green finance instruments and appropriately calibrated environmental regulation.

Second, advance policies through a parallel long–short combination. In the short term, efforts should concentrate on rules and foundational capacity by harmonizing statistical calibers for energy consumption and carbon emissions, issuing interoperable technical standards, and expanding cross-department data-sharing and compliance pilots to reduce institutional transaction costs and enhance policy predictability. In the medium term, emphasis should shift to platform diffusion and coordinated governance by extending full-chain digital platforms to key industries and critical links, linking green credit, tax incentives, and targeted subsidies to verifiable energy-saving and emission-reduction outcomes, and improving talent systems that integrate training, recruitment, and career development to strengthen sustained human-capital supply. In the long term, policy should orient toward innovation and institutional maturity by improving market-based allocation mechanisms for data as a factor of production, developing standardized solutions for AI-enabled energy-efficiency optimization, blockchain-based carbon traceability, and IoT-enabled dispatching, and refining evaluation systems that align digital performance with improvements in E-GUEUL. The central government should concentrate on top-level design, nationwide standards, and cross-regional interconnection rules, whereas local governments should prioritize site selection and implementation, scenario pilots, and performance-based policy delivery supported by public disclosure and regular assessments.

Third, optimize spatial deployment along an urbanization gradient. Urbanization levels and population agglomeration influence infrastructure utilization, scenario density, and the robustness of human-capital supply, thereby shaping the effectiveness with which digital investment translates into energy-saving and emission-reduction performance. In areas with higher population concentration, platform-based upgrading should be accelerated around manufacturing clusters and key industrial parks, and integrated solutions for equipment connectivity, process coordination, and carbon accounting should be promoted to amplify scale economies and diffusion effects. In areas with more dispersed settlement patterns, coordinated deployment between regional hubs and edge nodes should be pursued by concentrating data centers and computing platforms in regional cores while placing edge nodes in counties and industrial parks to support production monitoring and energy management; concurrently, shared access to general-purpose algorithms, industrial software, and carbon-management tools should be provided to reduce usage thresholds. At the same time, cross-regional coordination in data circulation, green finance, and technology transfer should be strengthened, and metropolitan areas should be encouraged to share standards, platforms, and training resources with surrounding cities to realize corridor-based spillovers and jointly enhanced innovation capacity.