A study on synergistic development of urban digitalization and greening under the diffusion of low-carbon pilot policies

Abstract

Low-carbon pilot policies, as comprehensive environmental strategies, provide a valuable perspective for examining the integrated development of urban digitalization and environmental sustainability. However, from the perspective of an integrated “pilot-diffusion” approach, the varying effects of pilot policies on the synergistic development of urban digitalization and greening, under different diffusion modes, are still not well clear. This study uses panel data from 278 Chinese cities between 2005 and 2022 and employs a double machine learning model to effectively control for high-dimensional confounders, reducing biases commonly associated with traditional methodologies. The results indicate that low-carbon city pilot policies effectively promote the synergistic development of urban digitalization and greening, with hierarchical diffusion demonstrating stronger policy effects compared to horizontal diffusion. Additionally, the upgrading of industrial structures and green technological innovation were found to regulate the effects of the policies, and the promotional effect of low-carbon pilot policies on the urban digitalization and greening development development of cities exhibits heterogeneity across different regions and city types. Through the analysis of policy diffusion patterns, this paper clarifies how different diffusion mechanisms influence policy outcomes, providing targeted policy recommendations and theoretical insights to support the coordinated progress of urban digitalization and greening. To enhance future low-carbon pilot policies, this paper emphasizes the need to adapt policy designs to specific diffusion modes and regional contexts, encouraging targeted and differentiated strategies to foster sustainable urban development.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-17783-8.

Introduction

In the comprehensive pursuit of socialist modernization, the《Outline of the Fourteenth Five-Year Plan for the National Economic and Social Development of the People’s Republic of China and the Vision for 2035》emphasizes the necessity of “accelerating digital development” and “promoting green development”. As the “Dual-Carbon” strategy is progressively implemented and digital technologies continue to advance, digitalization and greening have emerged as twin drivers of high-quality development. The synergy between digitalization and greening is steadily intensifying, gradually emerging as a pivotal trend in driving economic and social transformation. While digitalization fosters advancements in greening, the reciprocal role of greening in supporting digitalization has also gained notable traction¹. The report of the 20th Party Congress further highlights a strategic vision for building a digital China, advocating for a modernized framework that harmonizes human-nature coexistence. This vision provides essential theoretical underpinnings and practical guidance for advancing the synergy between digitalization and greening². In addition, projections from the white paper《Digital Carbon Neutrality (2021)》, published by the China Academy of Information and Communications Technology, estimate that digital technologies could reduce China’s total carbon emissions by approximately 12–22% by 2030 and contribute 10–40% of the potential carbon reduction across all sectors. As the greening process imposes growing demands on digitalization, the pace of innovation in digital technology has accelerated, establishing a robust foundation for socio-economic development characterized by “integration and symbiosis” as well as “mutual reinforcement and advancement.“³Consequently, systematically evaluating the synergy between urban digitalization and greening, while exploring strategies for its optimization, has become a key element in digitally enabling green and high-quality development.

To effectively address environmental pollution and promote green transformation, the National Development and Reform Commission issued the Notice on Promoting Low-Carbon Provinces and Cities Pilot Programs in 2010, marking the launch of the first batch of low-carbon city pilot projects. The second batch of pilots was officially approved in November 2012, followed by the third batch in January 2017 (see Table 1 for details on the pilot scope).Compared with other environmental policies, the low-carbon pilot policy has achieved remarkable results in reducing emissions, with its key advantage being the “pilot-diffusion” model⁴. Unlike the first batch, which was designated through a“top-down”approach, the second and third batches were initiated by local governments through voluntary applications. The central government then reviewed and finalized the list, reflecting a “bottom-up” approach characterized by autonomy and some degree of randomness.As a result, the low-carbon pilot programs developed along two paths: “top-down” hierarchical diffusion and “bottom-up” horizontal diffusion. This step-by-step expansion greatly improved the policy’s feasibility and effectively advanced green urban development. The diffusion of low-carbon pilot policies is not only a vital step in implementing the national low-carbon transition strategy but also a driving force for the coordinated development of digitalization and green transformation in cities.The policy innovations and practical experiences generated by the low-carbon pilots provide clear guidance for integrating digital transformation with green goals. They enable cities to pursue both environmental sustainability and economic efficiency in a balanced way.

Table 1.

Scope of Low-Carbon City Pilots.

batches	first batch	second batch	third batch
Implementation time	July 2010	November 2012	January 2017
Scope of the pilot	Guangdong, Shaanxi, Liaoning, Hubei, Yunnan 5 provinces; Tianjin, Chongqing, Hangzhou, Shenzhen, Xiamen, Nanchang, Guiyang, Baoding 8 cities	Hainan Province; Beijing, Shanghai, Shijiazhuang, Qinhuangdao, Jincheng, Hulunbeier, Suzhou, Qingdao, Zunyi, Kunming, Yan’an and other 28 cities	Wuhai, Shenyang, Dalian, Chaoyang, Nanjing, Changzhou, Jiaxing, Jinhua, Quzhou, Hefei, Huaibei, Huangshan, Lu’an, Xuancheng, Sanming, Ji’an, Fuzhou, Jinan, Yantai, Weifang and other 45 cities (districts and counties)

Therefore, exploring the varying impacts of low-carbon pilot policies on the synergistic development of urban digitalization and greening under different diffusion models can help us understand how to optimize policy design based on diffusion pathways in the complex, multi-level policy implementation process. This will better facilitate the integration of low-carbon and digitalization goals.Does the low-carbon pilot policy have different effects on the synergistic development of urban digitalization and greening under various policy diffusion models? Focusing on this question, the contribution of this paper lies in several key areas: First, it enriches the theoretical framework for the synergistic development of low-carbon policies and urban digitalization and greening. Existing literature mainly examines the roles of digitalization and green development in their individual domains, without systematically exploring their synergistic effects. This study uses low-carbon pilot policies as a starting point and analyzes their impact on the combined development of digitalization and green growth, focusing on the bidirectional interaction between the two. This provides a new theoretical basis for how low-carbon policies can drive sustainable urban transformation.Second, it uncovers the differential effects of policy diffusion models on the synergistic development of digitalization and green growth. The paper systematically analyzes two policy pathways: “top-down” hierarchical diffusion and “bottom-up” horizontal diffusion, exploring how policies differ under each diffusion model. This comparative analysis helps understand the effectiveness of low-carbon policies in multi-level implementation and addresses a gap in research on policy diffusion patterns in the coordinated development of digitalization and greening.Third, it offers a quantifiable evaluation method for assessing coordinated development, providing empirical evidence for policy optimization. The paper establishes a city-level evaluation system for the synergy between digitalization and green development, systematically assessing the real-world effects of low-carbon policies under different diffusion models. This provides quantitative evidence for optimizing and adjusting policies. Furthermore, the evaluation method not only supports the effectiveness of low-carbon policies but also offers a reference for policy synergy research in other areas.

Literature review

Policy diffusion

Policy diffusion is understood as the process through which policies are transferred and adopted from one region or sector to another (Berry, 1994)⁵. Within the framework of policy diffusion’s fundamental principles, scholars hold varying perspectives on its characteristics. These views can be broadly categorized into three theoretical models: top-down diffusion, horizontal diffusion, and bottom-up diffusion. (Wu, et al.,2022)⁶. Specifically, the majority of scholars argue that policy diffusion typically follows an “S-shaped” curve. (Boehmke,2011)⁷.The studies of Wang and Ji based on the diffusion of institutional innovations in development zones, as well as Zhang’s study on the temporary residence permit system, also reveal that the track of the diffusion path of policy diffusion usually conforms to an “S” curve, exhibiting proliferation characteristics of slow in the early stages, rapid in the middle stages and stabilizing in the later stages^8,9. However, Yang et al. (2020) challenged the universality of the “S” curve, suggesting that policy diffusion may also take a non-gradual form, such as explosive diffusion. This pattern, characterized by sudden and widespread adoption, is not constrained by the “S” curve and highlights the role of environmental variables in shaping diffusion patterns¹⁰.

In terms of policy diffusion drivers, most of the studies expand and extend the drivers affecting policy diffusion into two categories: internal factors and external factors. Specifically, Mitchell (2018) points out that policy attributes are internal factors affecting policy diffusion¹¹. Lindsey et al. (2018) and Yi et al. (2018) further suggested that the competence and characteristics of actors are internal factors that drive behavior¹². −¹³ DeMora et al. (2019) and Zhang et al. (2019) argued that external pressures, such as those from parties and interest groups, serve as external drivers of policy diffusion¹⁴. −¹⁵These findings broaden the understanding of policy diffusion, showing that it is not a simple process but a complex phenomenon influenced by time, location, and context.

Effectiveness evaluation of low-carbon pilot policies

Since the introduction of low-carbon pilot policies, their effects on the economy, environment, and social welfare have drawn considerable academic interest. Previous studies have examined the diverse impacts of this policy from various theoretical and empirical perspectives, offering valuable insights into their mechanisms and improvements. Firstly, in terms of economic effects, scholars have examined how low-carbon pilot policies influence corporate innovation and the upgrading of industrial structures. Shi et al. (2024) explored the incentive mechanisms of this policy for green technological innovation and found that they effectively encourage enterprise R&D activities and strengthen green innovation capacity, highlighting their potential to drive innovation¹⁶. Zheng et al. (2022) showed that low-carbon pilot policies are essential to accelerate the transformation and upgrading of industrial structures, facilitating the shift from high-carbon to low-carbon industries¹⁷. Zhou et al. (2023) confirmed the beneficial influences of low-carbon pilot policies on enterprise growth and competitiveness, highlighting their contribution to high-quality development and their role in improving economic quality¹⁸. In addition, the low-carbon city pilot policies has positively influenced the digital transformation of businesses. Research indicates that the policy has notably enhanced the digital focus of businesses, with an average increase of 8.5% in the digital engagement of companies located in low-carbon pilot cities(Wang&Chen,2024)¹⁹. Secondly, in terms of environmental impacts, scholars have conducted empirical studies to evaluate how low-carbon pilot policies affect emission reduction and environmental management. He et al. (2024), using a DID model, assessed the effectiveness of this policy in reducing urban air pollution. Their results indicated that low-carbon pilot policies led to a significant decrease in air pollution levels, confirming their direct role in environmental improvement²⁰. Yan et al. (2021) highlighted that the development of low-carbon cities was key to reducing haze pollution. These effects were especially notable in eastern regions, larger urban areas, and non-resource-dependent cities, where the impact was more significant²¹. Chen and Wang (2021), using a multidimensional fixed-effects model, examined the combined effects of integrating low-carbon pilot policies with industrial policies. Their analysis revealed that this approach significantly lowered PM2.5 concentrations, confirming that aligning low-carbon policies with other strategies can enhance environmental outcomes²². The goal of establishing low-carbon cities is not only to conserve energy, reduce emissions, and enhance the ecological environment, but also, and more importantly, to foster the transition to a green economy. Research by scholars has demonstrated that low-carbon city pilot policies can effectively stimulate green economic growth in pilot cities, with the impact being particularly pronounced in cities located in the eastern region(Cheng et al.,2019; Wang and She, 2020)²³. −²⁴Finally, in terms of welfare effects, research has explored how low-carbon policies impact public health and social welfare. Zhang et al. (2022) showed that low-carbon pilot policies play a significant role in improving national health literacy, emphasizing their positive effect on health-related welfare²⁵. Wang et al. (2024) also showed that implementing low-carbon pilot policies has a beneficial effect on public health²⁶. Furthermore, Wang et al. (2023) found that the policy promotes employment growth, highlighting its combined benefits for both social welfare and economic sustainability²⁷. Regarding the optimization of policy effects, existing studies consistently reveal that foreign direct investment (FDI) inflows (Liu et al., 2020), increased economic density (Zhu and Lee, 2022), and advancements in technological innovation (Chen et al., 2021) are significant strategies for enhancing the effectiveness of low-carbon pilot policies^28–30. These strategies enhance the policy’s internal dynamics and improve its adaptability to changing environmental conditions, providing a strong foundation for its long-term and effective implementation. To further enhance the scientificity and accuracy of policy assessment, scholars have begun to adopt double machine learning methodology to assess policy effects in recent years. Using a double machine learning model, Han et al. (2024) evaluate the synergistic effects of the low-carbon city pilot policy and the Broadband China pilot policy. Their findings suggest that the green digital policy has a significant co-promotional effect on the development of a digital ecological civilization³¹.

Digitalisation and greening of cities

Current research on the relationship between digitalization and green development mainly concentrates on two key areas. On the one hand, scholars have examined the theoretical mechanisms and pathways through which digitalization and greening can be integrated. These studies generally emphasize how digitalization facilitates urban greening, highlighting the central role of digital technology in advancing urban green transformation³². Ma and Zhu (2022) argued that digital development can directly promote sustainable urban green growth by upgrading industrial structures, leading to a balanced benefit for both economic growth and environmental sustainability³³. Han and Chen (2022) proposed that the inherent green attributes of digital technology effectively address information asymmetry caused by time and spatial dislocation in traditional economic systems. They emphasized that digital technology provides robust support for green development by optimizing industrial structures³⁴. In addition, Guo et al. (2022) highlighted that, with the goal of promoting ecological civilization, China faces several challenges, including the transformation of economic development models and industrial structures, inefficient circulation of ecological information, and difficulties in digital governance³⁵. On the other hand, Scholars have discussed the specific effects of urban digital development on the green transition from both macroeconomic and microeconomic perspectives. At the macroeconomic level, Lin(2024) highlighted the effective carbon emission reduction effects driven by digital development³⁶. Yang and Hui (2024) emphasized the effectiveness of digital finance in enhancing the efficiency of green innovation³⁷. Wang and Shao (2023) further showed that, within the process of enhancing urban energy efficiency, the effect of the digital economy can’t be ignored³⁸. At the microeconomic level, research has mainly concentrated on the impact of digital transformation in enterprises on green development. For instance, Cao et al. (2023), using a longitudinal single-case study approach, analyzed the stages of Sany Group’s digital development and its intrinsic mechanisms and impacts on the green transformation of manufacturing corporations³⁹. Additionally, other studies, using data from listed companies, have examined the effects of digitalization on the green development of manufacturing and high-pollution industries⁴⁰. −⁴¹These studies reveal that digitalization plays a dual role of “incremental improvement” and “qualitative optimization” in driving green technological innovation in businesses.

In summary, existing research on low-carbon pilot policies has predominantly concentrated on evaluating policy outcomes, while studies on urban dualization have explored the impact of digital development on greening and the pathways for integrating the two. However, there has been limited focus on whether the different diffusion modes of low-carbon pilot policies can facilitate the synergistic development of digitalization and greening in urban areas. This raises several critical questions: Can low-carbon pilot policies foster the synergistic development of urban digitalization and greening? Do differences exist in this synergistic development under various diffusion modes? What mechanisms underlie the impact of low-carbon pilot policies, implemented through different diffusion modes, on the synergistic development of urban digitalization and greening? This paper hopes to explore these questions in depth, offering new research perspectives and empirical evidence to support the advancement of the synergistic development of urban digitalization and greening.

Theoretical analysis and research hypotheses

As a core strategy for advancing green and low-carbon transformation in China, the low-carbon city pilot policy provides essential institutional support and policy backing for the deep integration of urban digitalization and environmental sustainability. The policy’s implementation not only prioritizes carbon emission reduction but also accelerates digital transformation across various regions through clear goal orientation. This approach closely integrates environmental governance with digital technologies, fostering a virtuous cycle where green development and digital transformation mutually reinforce each other. On one hand, the policy has expedited the application of digital technologies in environmental governance, improving energy management efficiency and optimizing resource allocation. Technologies such as big data and the Internet of Things are widely employed in pollution monitoring, energy consumption management, and other areas, effectively advancing the achievement of green development goals. On the other hand, the policy’s technological incentives have facilitated the green industry’s transition to digitalization, further enhancing the digitalization of energy systems and resource management. By promoting the integration of digital technology with green development, the low-carbon city pilot policy has generated a dual benefit, reducing pollution and carbon emissions while driving digital economic growth. This has contributed to the synergistic development of urban digitalization and greening sustainability.

As a distinctive approach to policy innovation in China, the low-carbon pilot policies follow a “pilot-diffusion” model, in which the central government plays a pivotal guiding role. In China’s unitary system, the government structure is highly hierarchical and centralized, characterized by a command-and-obey relationship between higher and lower levels of government. From the viewpoint of intergovernmental relations, the “top-down” diffusion model (hierarchical diffusion) functions through administrative directives and policy guidelines, with the central government exercising authority over local governments. The driving force in this model primarily stems from the authority and directive power of the central government. Conversely, in the “bottom-up” diffusion model (horizontal diffusion), local governments adopt policies through learning, imitation, and competition. Here, the diffusion is driven by the autonomy and initiative of local governments. In the process of policy diffusion, both central and local governments play distinct functions within two diffusion models, which form the key distinction between hierarchical and horizontal diffusion, and influence the synergistic effectiveness of policy. In selecting low-carbon pilot cities in China, the first batch of pilot cities was directly designated by the central government in a top-down manner, reflecting the centralized and mandatory nature of the hierarchical diffusion model. In contrast, the second and third batches were initiated by local governments and selected by the central government, demonstrating the flexibility inherent in the horizontal diffusion model. Institutional change theory argues that the effects of policies depend not only on how the policy is designed but also on how formal and informal rules interact within the institutional environment (North, 1990)⁴². And the adoption and implementation of policies are influenced by a range of factors, including political, economic, social, and institutional elements. Different diffusion models activate various transmission mechanisms in the institutional environment, resulting in varying outcomes in policy implementation (Berry&Berry,2018)⁴³. For cities experiencing hierarchical diffusion, the central government usually controls and allocates critical fiscal, technical, and informational resources. Through funding disbursements, project approvals, and technical guidance, the central government drives local investment in digital infrastructure and promotes the intelligent upgrading of green industries, thereby lowering the barriers for local governments to implement policies. This supply relationship of external key resources makes local governments dependent on the central government when resources are scarce, and this dependence, in turn, enhances the central government’s policy control power(Pfeffer & Salancik, 1978)⁴⁴. At the same time, the unified deployment of policies has brought about a high degree of policy replicability and standardization. The vertically structured policy network also facilitates a more concentrated and rapid flow of information (Rhodes, 1997)⁴⁵. Together with mechanisms such as pilot evaluations and supervisory assessments, it has strengthened the execution constraints and promotion impetus of local, enabling the policy effects to be effectively released. In pilot cities for horizontal diffusion, the process relies more heavily on the proactive initiatives of local governments and their strategic grasp of opportunities for central resource allocation. Local governments, driven by the expectation of enhancing regional competitiveness and securing policy benefits, voluntarily apply to become pilot cities. Through mobilizing local fiscal and social capital to guide corporate R&D and market promotion, and build green infrastructure and digital governance platforms. However, due to the lack of stable central resource support, this pathway is easily constrained by insufficient funds and technological reserves. The capacity bottlenecks faced by local governments in policy implementation are a typical manifestation of external supply shortages under resource dependency relationships (Pfeffer & Salancik, 1978)⁴⁴, thereby affecting the depth and stability of policy implementation. While such horizontal policy networks exhibit high flexibility, their decentralized characteristics result in slower information transmission speeds and narrower coverage compared to vertical networks (Rhodes, 1997)⁴⁵. Additionally, the absence of unified performance evaluation and mandatory incentives may lead some local governments to underperform in the sustainability of policy implementation, thereby weakening the effectiveness of low-carbon pilot policies.Based on this, this paper proposes Hypothesis H1 and Hypothesis H2.

H1: The implementation of low-carbon pilot policies, whether through the hierarchical diffusion model or the horizontal diffusion model, facilitates the synergistic development of urban digitalization and greening.

H2: In the diffusion process of low-carbon pilot policies, cities under the hierarchical diffusion model demonstrate greater effectiveness than those under the horizontal diffusion model in promoting the synergistic development of digitalization and greening.

In the diffusion of low-carbon pilot policies, the digital and green development of cities is influenced by multiple factors, making it challenging to achieve synergistic development goals through policy incentives and constraints alone. To address this, industrial structure upgrading and green technology innovation are proposed as regulatory mechanisms. Together, these mechanisms facilitate the comprehensive implementation of low-carbon policies from both economic and technical levels, promoting sustainable urban development. Firstly, the upgrading of the industrial structure provides a robust economic foundation and structural support, contributing to enhancing the efficiency of policy transmission and the capacity for implementation. A key objective of the low-carbon pilot policies is to reduce carbon emissions and drive the green transformation of industries, with industrial structure upgrading serving a critical regulatory role in this process. Building on a high-quality industrial structure, cities are better positioned to harness the endogenous dynamics of green transformation and respond positively to low-carbon city pilot policies. This adjustment in economic structure not only contributes to carbon emission reduction and environmental improvement but also creates substantial market opportunities for promoting advanced low-carbon technologies and the application of digital tools. By fostering demand for green industries, this shift makes the development of digitalization and greening mutually complementary and reinforcing, thereby enhancing the effectiveness of policy outcomes(Ding et al.,2024)⁴⁶. Secondly, green technological innovation, as a key driver of technological progress, provides the technical foundation necessary to achieve the objectives of low-carbon pilot policies. Green technology innovations not only improve energy efficiency and reduce carbon emissions but also drive the digital transformation of cities (Li et al., 2024). The application of technologies such as smart grids, intelligent transport systems, and green buildings has enabled cities to become progressively smarter and more efficient in the areas of energy, transport, and buildings, driven by low-carbon pilot policies^47–49. According to Rogers’ (2003) theory of the diffusion of innovations⁵⁰, when technology diffusion proceeds smoothly, the widespread adoption and application of green technologies can enhance the implementation of low-carbon city pilot policies, strengthen the effectiveness of the policies, and help achieve a broader set of objectives. Thus, the continuous innovation of green technologies serves as a key driver for both digitalization and greening, enabling their organic integration and deep synergy under the guidance of low-carbon policies. Low-carbon pilot policies complement the upgrading of industrial structures and green technology innovation, fostering not only the low-carbon transformation and green development of cities but also enhancing their resilience and capacity to address global climate change and sustainable development challenges. Based on this, this paper proposes hypothesis H3.

H3: Under different diffusion modes, both industrial structure upgrading and green technology innovation act as moderating factors in fostering the synergistic development of digitalization and greening driven by low-carbon pilot policies.

Research design

Methodology

Baseline regression model

Currently, most studies evaluating policy effects rely on classical econometric approaches, such as the difference-in-differences (DID) method and the synthetic control method (SCM). However, the application of DID models is often constrained by stringent data requirements, particularly the assumption of parallel trends. This assumption dictates that the experimental and control groups must exhibit similar developmental trajectories before policy implementation. In practice, identifying a control group that satisfies this condition can be challenging. To address this issue, the synthetic control method improves upon DID by constructing weighted control groups, thereby relaxing the need for identical pre-policy trends between groups. Despite its benefits, SCM also has limitations, especially in “many-to-many” scenarios involving multiple units and long time series.

To address the limitations of traditional causal inference models, scholars have increasingly turned to machine learning techniques for causal analysis. Chernozhukov et al. (2018) introduced Double Machine Learning (DML), which combines a doubly robust estimation method with cross-fitting based on Neyman orthogonal estimating equations. This approach effectively reduces the strict assumptions needed by traditional causal inference models⁵¹. Firstly, this model can effectively address confounding factors in high-dimensional data environments. Compared to traditional regression models, double machine learning (DML) automatically identifies and selects variables that have a significant predictive effect on causal relationships, using regularization algorithms such as Lasso regression. This approach helps avoid the dimensionality problem caused by an excessive number of control variables. During the model training phase, DML typically employs K-fold cross-validation (e.g., 5-fold or 10-fold) to select optimal tuning parameters, balancing model bias and variance while ensuring the robustness of variable selection. The results of cross-validation are not only used for variable screening but also play a role in parameter tuning, thereby enhancing the accuracy of causal effect estimates. Second, the effects of low-carbon city pilot policies are often nonlinear, and traditional linear regression models struggle to capture such complex relationships. Double machine learning, with its flexible algorithmic structure, can accommodate nonlinear effects and reduce the influence of confounding factors through sample partitioning techniques, leading to more robust and unbiased causal effect estimates. In contrast, methods such as difference-in-difference and synthetic control may not handle these issues with the same precision. The strength of double machine learning lies in its ability to adapt to complex data structures, nonlinear relationships, and high-dimensional control variables, making it highly flexible and robust for evaluating low-carbon pilot policies. Despite these advantages, double machine learning remains sensitive to data quality and the selection of variables. Although DML can reduce the impact of model setting errors through the Neyman orthogonality condition, the issue of omitted variable bias still requires attention. If key variables are excluded from the model, particularly when important control variables are omitted, double machine learning may fail to remove bias, resulting in inaccurate estimates. Therefore, caution is needed when using double machine learning to identify causal relationships, and its findings should be validated through robustness tests. Based on this, this study refers to the approach of Yang et al.(2020)⁵², using Double Machine Learning to construct the model.

First, a partial linear model is constructed:

1

2

where i is the city, t is the year, and denotes the dependent variable, which is the degree of synergistic development of digitalization and greening of the city; is the core explanatory variable, which is the policy variable of the low-carbon pilot city, taking the value of 1 after the implementation of the pilot policy, and 0 when the pilot is not implemented;is the disposition coefficient, which is the estimated coefficient focused on in this study;is the set of control variables, which affects the covariates of the explanatory variables through the function g. g is a nonlinear function whose specific form is unknown. Its estimated value is obtained through machine learning models such as random forests or gradient boosting algorithms; is the error term with a conditional mean of 0.

Secondly, auxiliary equations for the partial linear model are constructed:

3

4

The estimation steps are as follows: machine learning model is employed to obtain the estimate of m in Eq. (3), which leads to the estimate of the residual term .Secondly,is used as an instrumental variable for, and a machine learning model is applied to estimate , the value of the function g, which in turn provides the estimate of . The details are as follows:

5

where n represents the total sample size.

Moderating mechanism testing model

For the moderation mechanism test model, this paper adopts the Double Machine Learning model, following the approach of Ahrens et al. (2025)⁵³. It introduces the moderating variable M_i and constructs the moderation mechanism test model as follows:

6

7

8

The estimation steps are identical to those of the baseline regression model and are outlined as follows:

9

Variables and measurements

Dependent variable

The dependent variable in this paper is the synergistic development of urban digitalization and greening (DGSP). A comprehensive index system for evaluating the development levels of digitalization and greening is developed, building on relevant research⁵⁴ and following the principles of data availability and scientific rigor. In evaluating the level of digital development, consider the entire “foundation-use-innovation” process, encompassing the path from infrastructure development to user engagement and technological investment. Six specific indicators are selected across three dimensions: the scale of digital business, the foundation of digital users, and the environment for digital innovation. The scale of digital business focuses on the degree of industrialization and marketization of the digital economy. To assess this, the proportion of telecommunications business revenue and postal business revenue is selected, as they reflect the importance of the information technology and service industries in the regional economy, and, by extension, the development level of the city’s digital economy. The digital user base examines the acceptance and frequency of digital technology usage among city residents and businesses. Measured by mobile phone and internet penetration rates, this dimension reflects the city’s investment in digital infrastructure and its impact on widespread adoption, indicating the extent of digital participation among residents and enterprises. The innovation environment assesses the city’s capacity for technological innovation and R&D. Measured by the ratio of fiscal expenditure on science and technology and the intensity of R&D investment, this dimension aims to evaluate the city’s policy support and resource allocation for digital technology innovation. A robust digital innovation support system provides sustained momentum for the city’s digital development and serves as a key driver of digital transformation. In evaluating the level of greening development, eight specific indicators are selected across three dimensions: greening production, greening life, and greening ecology. Greening production primarily assesses environmental pollution control and the efficiency of resource conservation and reuse in industrial sectors. Key indicators, such as industrial wastewater emissions per unit of GDP, industrial SO₂ emissions per unit of GDP, and the comprehensive utilization rate of industrial solid waste, provide insight into industrial pollution and resource use efficiency. These indicators help measure the environmental sustainability of a city’s industrialization process and assess its effectiveness in promoting green production. Greening life focuses on the transformation of residents’ daily activities, particularly in terms of energy conservation, emission reduction, and green consumption. Key indicators, such as the rate of harmless treatment of domestic waste, household water usage, and the number of public transport vehicles, reflect green lifestyles. These indicators help assess the effectiveness of cities in promoting green consumption, energy savings, emissions reductions, and sustainable transportation. Greening ecology primarily evaluates the health of the urban ecosystem and the resilience of the natural environment. Two key indicators, the green area and the green coverage rate of built-up areas, provide a comprehensive measure of urban greening and the quality of ecological spaces, reflecting the city’s progress in developing a sustainable green environment. The comprehensive evaluation index system for the level of digitalization and greening is presented in Table 2.

Table 2.

Comprehensive evaluation index system for the development level of digitalization and Greening.

System	Dimension	Indicator Description	Property	Weights
The development of digitalization	Digital Business Scale	Telecommunication Service Revenue Share	+	0.098
		Postal Revenue Share	+	0.160
	Digital User Base	Mobile Penetration Rate	+	0.099
		Internet Penetration Rate	+	0.156
	Digital Innovation Environment	Share of Financial Science and Technology	+	0.193
		R&D Investment Intensity	+	0.303
The development of greening	Green Production	Industrial Wastewater Discharge per GDP	-	0.004
		Industrial SO2 Emissions per GDP	-	0.001
		Industrial Solid Waste Utilization Rate	+	0.039
	Green Living	Non-hazardous Waste Disposal Rate	+	0.027
		Household Water Consumption	-	0.004
		Year-End Public Buses (Trams) in Operation	+	0.469
	Greening Ecology	Green Space	+	0.450
		Greening Coverage of Urban Areas	+	0.010

Based on the comprehensive evaluation index system for digitalization and greening levels, the overall evaluation indexes for urban digitalization and greening are calculated separately using the entropy value method. The corresponding calculation formulas are as follows:

1. Data Normalization

The data for both positive and negative indicators were normalized using the polar deviation method. For positive indicators, the following formula is applied:

10

For negative indicators, the following formula is applied:

11

Where represents the raw data value of the jth indicator for the ith city, and andrepresent the maximum and minimum values of the jth indicator, respectively. is the standardized value of the indicator.

• (2)Calculate the characteristic weight of the ith city for the jth indicator.

12

• (3)Calculate the entropy of the jth indicator.

13

• (4)Calculate the coefficient of variation for the jth indicator.

14

• (5)Calculate the weight of the jth indicator.

15

• (6)Calculate the composite evaluation index U.

16

The coupling coordination model was then used to measure the level of coupled and coordinated development of urban digitalization and greening, with the following formula⁵⁵−⁵⁶:

17

18

19

Where C is the coupling degree of digitalization and greening, ranging from 0 to 1; U₁ and U₂ present the levels of digital and green development, respectively; T is the comprehensive coordination index of the two systems, considering the equal importance of both systems to the overall system(Qin&Huang, 2023)⁵⁷, and are both taken as 0.5; D is synergetic development between digitalization and greening.

Core explanatory variable

The core explanatory variable in this paper is the low-carbon city pilot policies shock (Policy). Based on notifications issued by the National Development and Reform Commission (NDRC) and the implementation of three batches of low-carbon city pilot projects, the policy’s inception years are identified as 2010, 2012, and 2017, respectively. Considering the overlapping designations among the three batches of pilot cities, this study employs the methodology proposed by Liu et al.(2020)²⁸. For cities that belong to multiple batches simultaneously, the year of policy initiation is determined by the earliest batch designation. In this paper, cities designated as low-carbon pilots function as the experimental sample, whereas those not designated serve as the control sample. To capture policy effects, we introduce a policy dummy variable. During the sample period, low-carbon pilot cities are assigned a value of 1 starting from the year of their selection and for all subsequent years, while non-low-carbon pilot cities are consistently assigned a value of 0 for all years.

Regulatory variable

Industrial structure upgrading (Ins) is measured following the approach of Ren et al. (2013)⁵⁸, using the ratio of the value added of the tertiary industry to that of the secondary industry as an index to quantify the degree of structural upgrading. Green technological innovation (Gpatent) is assessed based on the method proposed by Shen et al. (2021)⁵⁹, where the number of authorized green patent applications is used to represent the level of green technological innovation.

Control variable

①Economic development (GDP): Represented by regional GDP, it reflects the overall economic performance of the region.②Education level (Edu): Measured as the share of education expenditure in regional GDP, this variable captures the region’s investment in education。③Social consumption level (SC): Expressed as the share of retail sales of consumer goods in GDP, it serves as a proxy for the consumption capacity of the region.④Fiscal dependence (CZ): Defined as the proportion of general public budget revenues in regional GDP, this variable indicates the region’s reliance on fiscal resources.⑤Financing level (Fin): Calculated as the ratio of the deposit and loan balances of financial institutions to GDP at the end of the year, it reflects the financial support available in the region.⑥Population concentration (PD): Characterized by population density, it captures the degree of population aggregation in the region.

Data description

This study selects panel data from 278 cities in China between 2005 and 2022. All data were sourced from authoritative publications, the” China Urban Statistical Yearbook”, the “China Environmental Statistical Yearbook”, the “National Bureau of Statistics”, the “China Urban Construction Statistical Yearbook”, and the “China Energy Statistical Yearbook”, ensuring the authenticity, scientific, and reliability of the data. To minimize data gaps and ensure consistency, missing values were addressed through linear interpolation. Table 3 presents the descriptive statistics for all variables included in this study.

Table 3.

Descriptive Statistics.

Variable	Obs	Mean	Std.	Min	Max
DGSP	5004	0.226	0.046	0.099	0.524
policy	5004	0.268	0.443	0	1
ES	5004	0.992	0.664	0.129	5.650
TP	5004	318.845	1068.186	0.000	18,959
GDP	5004	22,900,000	35,700,000	603,328	447,000,000
Edu	5004	0.032	0.016	0.003	0.148
SC	5004	0.367	0.103	0.000	1.013
CZ	5004	0.070	0.027	0.011	0.238
Fin	5004	2.336	1.171	0.508	21.301
FD	5004	437.377	336.509	0.64	3294.942

Empirical results

Factual analysis of characteristics

As shown in the 3D kernel density map in Fig. 1, the dynamic trend of coordinated urban digitalization and greening development from 2005 to 2022 reveals an overall trajectory of gradual improvement, moving from low to high levels, with distinct stage characteristics at various points. From Fig. 1a, the coupling coordination curve appears relatively flat and low between 2005 and 2010, indicating slow growth in coordination during this period. This suggests a weak alignment between digitalization and green development, likely due to the early stages of both technological advancements and low-carbon city pilot policies. During this time, the integration between digitalization and greening had not yet shown significant interaction. As shown in Fig. 1b, the coupling coordination degree increased significantly between 2011 and 2016, with greater volatility in the curve. This period marks the beginning of a new growth stage, where digitalization and green development show a more pronounced upward trend. This shift is likely associated with strengthened policies and the rapid development of technology. During this phase, digital technologies began to more closely integrate with green development strategies, leading to the gradual convergence of urban green strategies and digital transformation. Additionally, the more prominent peaks in the curve indicate that the coupling coordination between digitalization and green development reached higher levels, with synergistic effects becoming more apparent. Figure 1c illustrates the overall trend in the coupling coordination degree of digitalization and greening for 278 prefecture-level cities between 2017 and 2022. During this period, the coupling coordination degree peaked, showing a significant increase in the synergy between digitalization and green development. It can be inferred that, as digital technology and low-carbon city pilot policies advanced rapidly, the integration of urban digitalization and green development entered a mature phase. Green and low-carbon transition policies, along with technological innovation, facilitated the deep integration of digitalization and greening, driving a rapid rise in coupling coordination. At the same time, the growing global focus on Sustainable Development Goals further strengthened the synergy between digitalization and greening.

Fig. 1.

Trend of Urban Digital-Greening Synergies.

Baseline regression analysis

This paper employs the Double Machine Learning approach to analyze the impact of low-carbon pilot policies on the synergistic development of urban digitalization and greening. The regression results are presented in Table 4. The regression coefficient for the overall sample is 0.012 and significantly positive, suggesting that the policy implementation led to an average annual increase of 1.2% in the index of coordinated development. Given that the average value of the synergistic development index over the sample period is around 0.226, this effect corresponds to a 5.3% improvement. This result has significant practical implications and indicates that the policy has a positive impact on the synergistic evolution of urban digitalization and greening. In the hierarchical diffusion model, the regression results for the first pilot cities show a significant positive impact, with a policy coefficient of 0.014, highlighting the key role of the central government in policy implementation and diffusion, which has led to an average annual increase in the synergistic development index of about 1.4%. Specifically, the central government has set clear low-carbon development goals, created a performance evaluation system, and linked evaluation results to fiscal incentives and subsequent project approvals. This approach generates mandatory and sustained external pressure, motivating local governments to actively implement policies aimed at achieving digitalization and green development goals, thereby fostering overall coordinated development (Zhou, 2010)⁶⁰. For example, in Shenzhen, supported by both central government finances and pilot policies, the city has quickly developed a smart energy management system and a carbon emissions monitoring platform that covers the entire city, ranking at the top in the national development and reform commission’s low-carbon pilot assessments. In contrast, the horizontal diffusion model adopted by the second and third pilot cities, while also showing a positive impact, has a regression coefficient of 0.010, meaning that the implementation of the policy has led to an average annual increase in the synergistic development index of about 1.0%, which is significantly lower than that of the hierarchical diffusion model. This difference may stem from the motivational bias of local governments when applying for pilot status. Although local governments show a willingness to participate actively, their motivation and the effectiveness of policy implementation are somewhat limited without direct support and incentives from the central government. Many local governments, when applying to become pilot cities, are primarily focused on securing central resources and policy concessions, rather than genuinely committing to achieving the low-carbon policy objectives. Additionally, the absence of an effective assessment mechanism is another key factor limiting the effectiveness of the horizontal diffusion model. The lack of assessment incentives leads to insufficient oversight and follow-up on policy implementation in some pilot cities, which reduces motivation and undermines the overall effectiveness of the policy.For example, although Beijing has also invested in green and digital projects, it relies more on local financial and technological resources, with relatively weak direct support from the central government. This has resulted in a relatively slow pace of policy implementation, particularly in the comprehensive rollout of smart energy platforms, which has taken a long time.

Table 4.

Baseline regression Results.

variable	(1) full sample	(2) hierarchical diffusion model	(3) horizontal diffusion model
Policy	0.012*** (5.96)	0.014*** (5.51)	0.010** (1.96)
Control variable	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes
City fixed	Yes	Yes	Yes
N	5004	4122	3708

In summary, the regression results demonstrate the positive impacts of low-carbon pilot policies on the synergistic development of urban digitalization and greening. The analysis also reveals notable differences in policy effects between the hierarchical and horizontal diffusion models. These findings offer a theoretical basis and practical insights for optimizing and implementing low-carbon policies in the future, contributing to more coordinated and sustainable urban development.

Robustness testing

Handling of anomalous samples

To remove the influence of outliers on the benchmark regression results, this study applies a 1% and 99% winsorization treatment to the sample variables. Specifically, observations below the 1 st percentile were replaced with the 1 st percentile values, and observations above the 99th percentile were replaced with the 99th percentile values (Angrist & Pischke, 2009)⁶¹. The processed data were then reintroduced into the baseline regression model for testing, and the results are presented in Table 5. The findings demonstrate that the regression outcomes after shrinkage treatment are nearly identical to those of the benchmark regression, reinforcing the conclusion that the low-carbon pilot policy has a positive impact and exhibits variation in its impact across different diffusion modes in promoting the synergistic development of urban digitalization and greening. This result not only validates the reliability of the previous analyses but also strengthens the robustness of the research conclusions.

Table 5.

Handling anomalous samples test Results.

variable	(1) full sample	(2) hierarchical diffusion model	(3) horizontal diffusion model
Policy	0.011*** (6.22)	0.013*** (5.89)	0.010*** (3.01)
Control variable	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes
City fixed	Yes	Yes	Yes
N	5004	4122	3708

Adjustment of the study sample

The first pilot cities for the low-carbon pilot policies were published in 2010, while the final list was released in 2017. To eliminate the potential impact of sample variability due to large differences in policy release times, this study restricts the analysis to the two years before and after the policy’s release, ensuring symmetry in the study period. The analysis includes data from cities between 2008 and 2019, with the results presented in Table 6. The regression analysis results align with those of the baseline regression, confirming the validity of the low-carbon pilot policy. Specifically, the implementation of the policy positively contributes to the synergistic development of urban digitalization and greening in the short term. By narrowing the study window to the two years surrounding the policy release, the analysis focuses on examining the immediate impact of the policy on urban development, thereby avoiding the interference of long-term trends and external factors. This allows for a clearer capture of changes in the policy’s effects. Importantly, the shortened timeframe did not diminish the significance of the policy impacts, suggesting that the positive effects of the low-carbon pilot policy were not only significant during the initial implementation phase but also remained stable in the short term following the policy’s release. This further demonstrates that the observed policy effects are robust and not incidental, strengthening the credibility of the research conclusions.

Table 6.

Adjustment of the study sample test Results.

variable	(1) full sample	(2) hierarchical diffusion model	(3) horizontal diffusion model
Policy	0.010*** (5.20)	0.011*** (4.12)	0.008** (2.15)
Control variable	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes
City fixed	Yes	Yes	Yes
N	3339	2748	2472

Changing the sample split ratio

To address the potential impact of the sample split ratio on the study’s outcomes, this paper modifies the sample split ratio in the Double Machine Learning model, adjusting it to 1:2 and 1:7 from the original 1:4.This adjustment purposes to explore the impact of different sample split ratios on the results, with the corresponding regression outcomes presented in Table 7. The results show that, whether the 1:2 or 1:7 split ratio is applied, the results match the baseline regression findings. This suggests that the benchmark regression results are not significantly affected by the sample split ratio, thus enhancing the reliability of the study’s conclusions and demonstrating the robustness and wide applicability of the policy effects.

Table 7.

Results of changing the split ratio test.

variable	(1)	(2)	(3)	(4)	(5)	(6)
Policy	0.014*** (7.02)	0.012*** (5.38)	0.008*** (2.69)	0.012*** (5.98)	0.013*** (5.38)	0.009*** (2.62)
Control variable	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes	Yes	Yes	Yes
City fixed	Yes	Yes	Yes	Yes	Yes	Yes
N	5004	4122	3708	5004	4122	3708

Reset the double machine learning model

To minimize the potential impact of model specification errors in the Double Machine Learning method, this study replaces the original random forest algorithm with lasso regression, gradient boosting, and neural network algorithms. This adjustment aims to evaluate the potential impact of different prediction methods on the study’s findings, with the corresponding regression results presented in Table 8. The outcomes show that the results obtained using lasso regression, gradient boosting, and neural networks align closely with those of the baseline regression. This indicates that the positive effect of low-carbon pilot policies on the synergistic development of urban digitalization and greening, as well as the differential effects under different diffusion models, are robust and unaffected by the choice of prediction algorithms. The use of different algorithms in the regression models does not substantially change the regression coefficients or their significance levels, suggesting that these prediction models are consistent in estimating policy effects. This means that the effectiveness of the policies is not influenced by the choice of specific algorithms, but rather reflects the true impacts of the low-carbon pilot policies.

Table 8.

Reset double machine learning model test results.

variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Policy	0.014*** (9.23)	0.014*** (7.63)	0.013*** (5.33)	0.013*** (7.91)	0.015*** (7.36)	0.011*** (4.02)	0.014*** (8.61)	0.015*** (7.84)	0.013*** (5.02)
Control variable	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City fixed	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	5004	4122	3708	5004	4122	3708	5004	4122	3708

Controlling for environmental factors

Environmental factors were added as control variables in the original model to examine their potential impact on the estimated policy effects. Specifically, regional carbon emissions and PM2.5 concentration were included to control the interference of regional environmental base differences on the regression results. The corresponding regression results are presented in Table 9. The findings indicate that the direction and significance of the core explanatory variables remain consistent with those from the baseline regression, after incorporating the environmental control variables. This suggests that the study’s results are robust to the inclusion of these controls. Moreover, the results confirm that the policy effects are not systematically affected by regional environmental differences, thereby strengthening the reliability of the causal identification and enhancing the explanatory power of the study’s conclusions.

Table 9.

Control of environmental factors test results.

variable	(1) full sample	(2) hierarchical diffusion model	(3) horizontal diffusion model
Policy	0.006*** (3.12)	0.007** (2.54)	0.007** (2.16)
Control variable	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes
City fixed	Yes	Yes	Yes
N	5004	4122	3708

Endogeneity Test

Given the potential for omitted variables due to data limitations and the possibility that core variables may be correlated with the error terms, this study conducts an Endogeneity test. Using a partially linear instrumental variable model with double machine learning, following Angrist and Pischke(2009)⁶¹, employing the interaction term between urban terrain undulation and the time trend term as an instrumental variable, which satisfies the exogeneity and correlation assumptions. The test results, presented in Table 10, align with those of the baseline regression, indicating that the policy effects are not influenced by omitted variables and that the correlation between the core variables and the error term has also been effectively excluded. This further confirms the robustness of the findings.

Table 10.

Results of the endogeneity test.

variable	(1) full sample	(2) hierarchical diffusion model	(3) horizontal diffusion model
Policy	0.049* (1.95)	0.078** (2.51)	0.067* (1.73)
Control variable	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes
City fixed	Yes	Yes	Yes
N	5004	4122	3708

Analysis of the regulatory mechanisms

Building on the above research, this paper examines the moderating role of industrial structure upgrading under different policy diffusion modes. The specific regression results are presented in Table 11, where column (1) tests the moderating effect of industrial structure upgrading across the full sample, and columns (2) and (3) test the moderating effect of industrial structure upgrading in the hierarchical and horizontal diffusion modes, respectively. The findings indicate that industrial structure upgrading significantly moderates the synergistic development of urban digitalization and greening in both the full-sample and hierarchical diffusion modes. This suggests that the central government has effectively guided industrial transformation in pilot cities through clear policy frameworks and resource allocation mechanisms. The top-down approach not only enhances the efficiency of resource allocation but also supports the development of emerging green industries, enabling cities to achieve low-carbon transition goals more rapidly. The upgrading and optimization of industrial structures have improved the industrial ecology of cities, promoting the concentration of high-tech and high-value-added industries, thereby boosting their innovation capacity and competitiveness. Furthermore, the policy implementation has facilitated the digital transformation of enterprises, creating a positive feedback loop that promotes sustainable business development, injects new momentum into urban economic growth, and fosters a win-win situation in both ecological and economic terms. In contrast, the moderating effect of industrial structure upgrading is not significant under the horizontal diffusion model. This may reflect the resource constraints faced by local governments when they proactively apply for pilot city status. Although local governments actively attempt to promote industrial upgrading through local policy incentives, the lack of central government support undermines the effectiveness of the horizontal diffusion model in driving industrial restructuring.

Table 11.

Regulatory mechanism test Results.

variable	(1) full sample	(2) hierarchical diffusion model	(3) horizontal diffusion model	(4) full sample	(5) hierarchical diffusion model	(6) horizontal diffusion model
Policy	0.010*** (3.75)	0.009** (2.38)	0.009** (1.97)	0.011*** (5.11)	0.011*** (4.64)	0.008*** (2.83)
Policy*ES	0.025** (2.05)	0.048** (2.37)	0.001 (0.39)
Policy*TP				0.108*** (3.71)	0.142*** (3.24)	0.107*** (2.82)
Control variable	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes	Yes	Yes	Yes
City fixed	Yes	Yes	Yes	Yes	Yes	Yes
N	5004	4122	3708	5004	4122	3708

Meanwhile, this paper explores the moderating role of green technology innovation under different diffusion modes. The specific regression results are presented in Table 11, with column (4) testing the moderating mechanism of green technology innovation for the full sample, and columns (5) and (6) testing the moderating effect in the hierarchical and horizontal diffusion modes, respectively. The results indicate that green technology innovation significantly enhances the synergistic development of urban greening and digitalization in both the full-sample and hierarchical diffusion modes. However, its moderating effect is significant but relatively weaker in the horizontal diffusion mode. This suggests that, within the hierarchical diffusion model, the central government’s policy framework and incentive mechanisms effectively promote the diffusion of green innovation, enhance the technological capabilities of enterprises, and facilitate the efficient use of resources and the development of environmentally friendly products. Furthermore, strong government guidance fosters a positive innovation ecosystem, encourages companies to invest in R&D for green technologies, and accelerates the transformation of cities towards low-carbon and digitalization goals. In contrast, the relatively weak effect of green technology innovation under the horizontal diffusion model may stem from the insufficient innovation incentives provided by local governments in the absence of centralized support. This limits the motivation of firms to invest in green technology. While local governments have made some efforts to promote technological innovation, these initiatives do not reach the significant levels observed in the hierarchical diffusion model.

Heterogeneity analysis

A test of heterogeneity based on cities in the north and south

Based on the Qinling-Huaihe River line, this paper divides prefecture-level cities into northern and southern regions and analyzes the heterogeneous impacts of low-carbon pilot policies and their various diffusion modes on the synergistic development of digital greening in these regions. The results are presented in Table 12. The full-sample regression results show that low-carbon pilot policies have a significant positive impact on both southern and northern cities, with the estimated coefficients being larger in southern cities. This suggests that the low-carbon pilot policy has a stronger effect on promoting the synergistic development of digitalization and greening in southern cities. This may be due to southern cities being better equipped for digitalization and green transformation, with more favorable industrial structures and higher levels of economic development, allowing the policy’s effects to be more fully realized.

Table 12.

Results of heterogeneity test for cities in the North and south.

variable	(1) full sample		(2) hierarchical diffusion model		(3) horizontal diffusion model
	the southern city	the northern city	the southern city	the northern city	the southern city	the northern city
Policy	0.013*** (4.57)	0.006** (1.95)	0.018*** (5.12)	0.003 (0.75)	0.008* (1.81)	0.012** (2.12)
Control variable	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes	Yes	Yes	Yes
City fixed	Yes	Yes	Yes	Yes	Yes	Yes
N	2772	2232	2160	1962	1944	1764

Further analysis of the heterogeneity impact in the hierarchical diffusion model reveals significantly positive coefficients for cities in the South, while the coefficients for cities in the North are not significant. This suggests that, in the hierarchical diffusion model, where the central government drives policy from the top down, Southern cities receive a stronger policy push for the synergistic development of digitalization and greening. This may be because Southern cities have greater flexibility and adaptability in policy implementation and resource allocation, enabling them to respond more effectively to the central government’s policy guidelines for industrial upgrading and green technology adoption. In contrast, Northern cities may have a weaker economic base and more significant resource constraints, which could lead to a less effective policy response under the horizontal diffusion model. Under the horizontal diffusion mode, the impact of low-carbon pilot policies is significant in both northern and southern cities, but the policy effect is more pronounced in northern cities, with a higher coefficient than in southern cities. This suggests that the synergistic development of digitalization and greening is more strongly promoted in northern cities under the horizontal diffusion mode, which is initiated by local governments themselves. Compared to the hierarchical diffusion model, the horizontal diffusion model provides greater autonomy to northern cities, stimulates local governments’ initiative in policy implementation, and may consequently lead to more significant policy effects. For cities in the South, while the policy effects remain significant, the marginal push is relatively weaker due to their higher digitalization and greening base.

A test of heterogeneity based on cities in eastern, central, and western

This paper categorizes prefecture-level cities into eastern, central, and western regions based on geographical location and national policy orientation. The eastern region includes economically developed coastal provinces and some inland cities with advanced economies, while the central and western regions encompass cities with relatively slower economic development. According to the heterogeneity test results in Table 13, the impact of low-carbon pilot policies varies between eastern and central/western cities. The full sample regression results show that the low-carbon pilot policy has a significant positive effect on the coordinated development of digitalization and green initiatives in eastern cities. In the eastern region, the low-carbon pilot policy has significantly enhanced the coordinated development of digitalization and greenization. In contrast, the regression coefficient for central and western cities is 0.005, which is significant but relatively weaker. This may be due to the lower level of economic development in these regions and the greater challenges they face in industrial transformation.

Table 13.

Results for heterogeneity among cities in eastern, central, and western.

variable	(1) full sample		(2) hierarchical diffusion model		(3) horizontal diffusion model
	Eastern cities	Central and western cities	Eastern cities	Central and western cities	Eastern cities	Central and western cities
Policy	0.016*** (4.03)	0.005** (2.25)	0.029*** (5.83)	0.001 (0.51)	0.002 (0.22)	0.017*** (3.52)
Control variable	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes	Yes	Yes	Yes
City fixed	Yes	Yes	Yes	Yes	Yes	Yes
N	1548	3456	1152	2970	1098	2610

Under the hierarchical diffusion model, the regression coefficient for eastern cities is significantly positive, suggesting that under a top-down policy implementation model, eastern cities can effectively respond to central government initiatives aimed at promoting the coordinated development of digitalization and green development. However, the regression coefficient for central and western cities in this model did not reach statistical significance, indicating that policy effects in these regions are weaker under the hierarchical diffusion model. This may be due to more significant difficulties in resource allocation and policy implementation in these regions, making it difficult to achieve substantial effects through top-down policy promotion. Under the horizontal diffusion model, the low-carbon pilot policy has a significantly positive impact on central and western cities. This suggests that, in the horizontal diffusion model, where local governments proactively apply for and promote policies, central and western cities can adjust policy implementation plans more flexibly based on local conditions, thereby fostering the coordinated development of digitalization and green initiatives. However, the regression coefficient for eastern cities under this model is positive but not significant. A possible explanation for this is that eastern cities, with more developed economies, implement low-carbon pilot policies from a higher starting point, which results in limited marginal gains in policy effectiveness.

Heterogeneity test based on city type

Prefecture-level cities were categorized into resource-based and non-resource-based cities, based on China’s National Sustainable Development Plan for Resource-Based Cities (2013–2020). This classification was used to test the heterogeneous effects of low-carbon pilot policies and their different diffusion modes on the synergistic development of digitalization and greening in both types of cities. The results are presented in Table 14. The full-sample regression results show that the coefficients are positive, meaning the low-carbon pilot policies have a notable influence on both resource-based and non-resource-based cities. Interestingly, the estimated coefficients for non-resource-based cities are higher than those for resource-based cities, indicating that the low-carbon pilot policy has a stronger positive impact on the synergistic development of digitalization and greening in non-resource-based cities.

Table 14.

Results of the test for heterogeneity of urban types.

variable	(1) full sample		(2) hierarchical diffusion model		(3) horizontal diffusion model
	Resource city	non-resource city	Resource city	non-resource city	Resource city	non-resource city
Policy	0.008** (2.42)	0.014*** (5.05)	0.008*** (3.18)	0.017*** (6.85)	0.010* (1.77)	0.012*** (2.85)
Control variable	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed	Yes	Yes	Yes	Yes	Yes	Yes
City fixed	Yes	Yes	Yes	Yes	Yes	Yes
N	1998	3006	1764	2358	1602	2106

Under further analysis of heterogeneity within the hierarchical diffusion model, it is found that the effect of the low-carbon pilot policy remains both significant and effective in non-resource-based cities. This suggests that these cities have made more notable progress in the integrated development of digitalization and the environment, largely driven by the central government’s guidance. This may be attributed to the easier adoption and spread of digital low-carbon technologies in non-resource-based cities, resulting in more effective policy implementation. Resource-based cities experience weaker policy effects due to their industrial structures, which heavily depend on traditional resource-based industries that are more difficult to transform. In the horizontal diffusion model, the low-carbon pilot policies positively impact both resource-based and non-resource-based cities, with a stronger effect observed in non-resource-based cities. In this model, where local governments actively seek participation in the pilot program, non-resource-based cities are more proactive in using policy opportunities to accelerate the mutual development of digitalization and greening. In contrast, while the policy effects in resource-based cities remain significant, they are less pronounced compared to those in non-resource-based cities, primarily due to the greater challenges these cities face in transforming their industrial structures.

Conclusions and policy recommendations

This paper analyzes how low-carbon pilot policies and their different diffusion modes impact the synergistic development of urban digital greening. Using panel data from 278 cities in China between 2005 and 2022, the study uses a more reliable method called double machine learning. The main conclusions are as follows: First, from 2005 to 2022, low-carbon pilot policies significantly promoted the synergistic development of urban digital greening, with notable variations observed between different diffusion modes. Specifically, the hierarchical diffusion mode demonstrated a stronger effect on promoting urban dualization compared to the horizontal diffusion mode. Second, the regulatory analysis reveals that under the hierarchical diffusion model, both industrial structure upgrading and green technology innovation play important moderating roles in how low-carbon pilot policies affect the dual development of cities. In contrast, under the horizontal diffusion model, only green technology innovation plays a positive moderating role. Third, the heterogeneity analysis demonstrates that the effects of low-carbon pilot policies and their diffusion modes differ across southern and northern cities, as well as between resource-based and non-resource-based cities. Based on these findings, the paper proposes the following recommendations:

Optimising policy diffusion models and adapting to local conditions for hierarchical and horizontal diffusion

Between 2005 and 2022, low-carbon pilot policies have had a more significant impact on promoting the synergistic digitalization and greening of cities under the hierarchical diffusion model. Therefore, when formulating and implementing policies, targeted incentives and resource support mechanisms should be designed according to the characteristics of different diffusion modes to enhance the effectiveness and adaptability of policy implementation. For the hierarchical diffusion model, the central government should fully leverage its coordinating role in top-level design and resource allocation, and strengthen the vertical policy transmission mechanism. A “central–local low-carbon responsibility list” could be developed, drawing on the allocation framework of the EU Recovery Fund, with funds released in phases based on target achievement. On this basis, a “low-carbon transition special fund,” modeled after Norway’s sovereign wealth fund, could be established to provide targeted support for pilot cities in areas such as digital infrastructure development, green technology R&D, and low-carbon project operations. The division of responsibilities between the central and local governments should be clearly defined, creating a policy chain in which the central government sets the overall framework and local governments handle implementation. The vertical evaluation mechanism for policy implementation outcomes should be strengthened, with financial preference and support given to cities that demonstrate significant results, thereby creating positive incentives for policy implementation. For the horizontal diffusion model, the focus should be on fostering local governments’ initiative and capacity for institutional innovation, as well as promoting the sharing and dissemination of policy experience among regions. A “regional low-carbon technology sharing platform” could be established, modeled after the cross-city cooperation framework of the U.S. Clean Energy Alliance, to promote interoperability of technical standards, joint procurement of equipment, and the mobility of specialized talent—thereby reducing the costs of adopting and promoting low-carbon technologies. The “Digitalization-Greenification Synergy Index” should be incorporated into local governments’ performance evaluation systems as a supplementary indicator for measuring progress in ecological civilization construction, thus enhancing their intrinsic motivation. Cities that achieve or significantly improve their synergy index should receive special fiscal rewards to incentivize innovation and practical efforts in low-carbon transformation. In addition, pilot cities should be encouraged to create local autonomous low-carbon funds and develop green financial products to attract social capital investment in green infrastructure construction and low-carbon industries.In addition, during the policy design process, we can draw on Germany’s experience in coordinating the advancement of “Industry 4.0” and “energy transition” to incorporate the construction of “digital-green” integrated industrial parks. These parks serve not only as hubs for technology and industry but also as testing grounds for institutional innovation and resource allocation. Their main goal is to drive the deep integration of smart manufacturing and clean energy, creating a dual engine for digital transformation and green, low-carbon growth. The central government could introduce a dedicated plan to define the functions, industrial chain layout, and energy optimization pathways of such parks, while encouraging local governments to explore tailored implementation approaches that help build replicable “digital + green” demonstration models.

Strengthening the regulatory role of industrial structure upgrading and green technology innovation to enhance policy effectiveness

The analysis of the regulatory mechanism shows that industrial structure upgrading and green technology innovation in the hierarchical diffusion model significantly enhance the role of low-carbon pilot policies in promoting the synergistic development of digitalization and greening in cities. To further realize the synergistic effects of these policies, a hierarchical and quantifiable system of policy instruments should be implemented for different diffusion modes. Under the hierarchical diffusion model, the central government’s coordination capacity and resource allocation advantages can be utilized to accelerate the greening and upgrading of traditional industries and promote the clustering of new industries. Specifically, the first step is to establish an “incentive mechanism for the exit of high-energy-consuming industries,” offering tax exemptions and financial compensation to enterprises that complete clean transformations or shut down outdated production capacity ahead of schedule. The second step involves setting up a “green industry development guidance fund” to support high-value-added and low-energy-consumption enterprises in fields such as new energy, new materials, and green manufacturing. This could involve attracting social capital through equity investments and subsidized loans. The third step is to implement a “low-carbon industry cluster cultivation plan” for key city-level parks, aiming to enhance the overall competitiveness of the green industry through spatial clustering, energy efficiency coordination, and integration of upstream and downstream industries. In terms of green technology innovation, enterprises can be systematically promoted in green innovation through a comprehensive support mechanism covering “R&D – transformation – application.” One hand, the establishment of a “special fund for green technology R&D” would prioritize core technologies with emission reduction potential and industrialization prospects. On the other hand, the promotion of a “green technology achievements inventory and evaluation system” would provide mature technologies with channels for promotion and priority procurement. Simultaneously, using the carbon market mechanism, enterprises could be guided to combine the results of low-carbon technology application with carbon asset management, thus realizing both technological and economic benefits. Under the horizontal diffusion model, local governments need to focus on policy guidance and innovative mechanism design in the absence of centralized resource support. Local governments could set up “green technology application incentive programs” to provide financial subsidies or government procurement preferences for specific projects in green renovation, energy-saving equipment, digital management, and more. At the same time, they could promote the establishment of “local green technology service platforms” to offer technology assessment, intellectual property consulting, and industry-academia-research collaboration services, lowering the barriers to green technology application for SMEs.By building a diversified system of policy tools by mode, link, and subject, it will be possible to more effectively promote the optimization of industrial structure and the linked development of green technological innovation. This will ultimately realize the sustainable synergistic path of “digital + green” in the low-carbon transformation of cities.

Implementing differentiated policy support in response to regional differences

Heterogeneity analysis shows that the effectiveness of low-carbon pilot policies varies significantly between northern and southern cities, eastern and central-western cities, as well as resource-based and non-resource-based cities. To better enhance the effectiveness of the policy, it is recommended that regional differentiation of policy support be implemented, and targeted support measures be taken according to the economic foundation, industrial structure, and policy adaptability of each region. For southern and eastern cities, they should give full play to their digital foundation and technological advantages, focusing on promoting the pilot ‘digital twin city’, promoting the deep integration of the Internet of Things and energy management systems, and realising the efficient synergy of digitisation and greening; for northern and central-western cities, they should focus on the reconstruction of the energy structure and the remediation of high-carbon industries, and enhance the construction of renewable energy substitution projects and the construction of clean heating systems, and improve the quality of the energy supply system. For resource-based cities, it is recommended to set up ‘industrial transformation subsidies’ to support the introduction of intelligent and environmentally friendly equipment in traditional industries, promote green production methods to replace resource-dependent paths, and strengthen the construction of green finance and technology transfer platforms to help them achieve structural transformation. For non-resource cities, they can be used as ‘pilot test zones’ for green systems and technological innovation, focusing on supporting green industrial parks, green transformation of small and medium-sized enterprises, and the construction of regional green technology alliances, to enhance the effectiveness of green development and demonstration and leading capacity. By building a multi-dimensional and precise regional differentiated policy system, the comparative advantages and functional positioning of various types of cities can be brought into full play in a more targeted manner, promoting the formation of a synergistic development pattern of digital greening across the country with a synergistic division of labour, diversified paths and complementary mechanisms.

Enhancing dynamic management of policy implementation and impact assessment

To ensure the optimal effectiveness of low-carbon pilot policies across different regions and diffusion modes, it is essential to establish a robust mechanism for policy implementation and effectiveness assessment. The implementation process should rely on a real-time data monitoring platform to dynamically track and provide immediate feedback on the policy’s progress in each region. This will help identify and address implementation challenges promptly. Using big data analysis and IoT technology, the policy’s effectiveness, particularly in terms of energy use, carbon emissions, and green technology adoption, should be regularly evaluated. Policy adjustments should then be made based on the specific needs of each region. This approach ensures the accuracy of policy execution while also allowing for adjustments in policy intensity and support in line with the economic development, industrial structure, and resource availability of each region. Furthermore, the creation of a data-driven evaluation system provides a scientific foundation for policy optimization and adjustments. This will enhance the policy’s adaptability and flexibility, ensuring that it effectively promotes the synergistic development of digitalization and greening in cities.

Limitations

Based on this study, future research could expand in the following areas: First, the selection of pilot cities may be influenced by economic or environmental factors. To mitigate sample selection bias and improve the robustness of the results, future studies could use techniques such as propensity score matching (PSM) or dynamic panel models. Second, the timeliness of the data used in this study has some limitations. Future research could incorporate data from 2023 onwards to analyze the impact of “new infrastructure” policies (such as 5G base station construction) on the synergistic effects of digitalization and green development. This would help assess how policy adjustments in the post-pandemic era affect the relationship between digitalization and green development across cities, particularly in the transformation process driven by new infrastructure.Additionally, this study uses a DML model based on linear assumptions. Future research could explore the application of nonlinear DML models or causal forest methods, which may better capture complex nonlinear relationships and, as a result, improve the model’s predictive and explanatory power. Finally, industry heterogeneity analysis should be further explored. Future studies could segment industries (e.g., manufacturing and services) to examine how policy effects vary across different sectors. External factors, such as fluctuations in international energy prices, may also influence the effectiveness of pilot low-carbon city policies. Therefore, future research could consider incorporating these external factors to reduce potential biases and enhance the accuracy and reliability of the findings.These extensions will offer a more comprehensive and precise evaluation of the impact of policy implementation and different diffusion patterns on digitalization and green transformation, providing policymakers with more evidence-based recommendations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Zhang.JY: Conceptualization, Data curation, Methodology, Writing-original draft; Guo.JG: Supervision, Software, Funding acquisition.

Funding

This work was supported by the National Natural Science Foundation of China (72274029)、Humanities and Social Sciences Planning Fund of the Ministry of Education of China(24YJA630022) and Basic Research Project of Liaoning Provincial Department of Education(JYTZD2023051、JYTMS20230641).

Data availability

The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Competing interests

The authors have no competing interests to declare that are relevant to this article.