The Impact of Big Data Pilot Zones on Urban Ecological Resilience: Evidence from a Machine Learning Approach

Abstract

Against the backdrop of the structural transition in China’s economic landscape, the implementation of digital economy policies—particularly through the Broadband China Demonstration Cities initiatives—has significantly enhanced urban ecological resilience. Based on panel data from 280 prefecture-level cities in China over the period 2013–2022, this study employs the national big data comprehensive pilot zone as a quasi-natural experiment and utilizes the dual machine learning method to examine how pilot zone construction influences urban ecological resilience. This analysis provides theoretical support for fostering green urban development. The results are summarized as follows. (1) The construction of national big data comprehensive pilot zones significantly enhances urban ecological resilience. The conclusion is robust to various tests, including the removal of outliers, changes in sample splitting ratios, and alterations in machine learning algorithms. (2) The construction of national big data comprehensive pilot zones indirectly improves urban ecological resilience through pathways of green innovation and energy efficiency. (3) This study assesses the heterogeneity of policy effects based on the generalized random forest (GRF) model to identify the sources of heterogeneity in policy effects, and conducts a comprehensive heterogeneity analysis from the three dimensions of resource endowments, geographical location characteristics, and the attributes of environmental protection zones. These findings enrich the analysis of the consequences of national big data comprehensive pilot zone policies and offer a theoretical basis and policy reference for how constructing big data pilot zones can better serve urban ecological development.

1. Introduction

In the context of the digital economy, global urbanization is accelerating, and building sustainable cities and communities has become a core component of achieving the United Nations’ 2030 Sustainable Development Goals (SDGs). However, cities worldwide face ecological and environmental challenges, including ecological damage, resource shortages, and environmental pollution. These problems weaken the ecological resilience of cities and seriously hinder their sustainable development. As the world’s largest developing country, China’s sustainable urban ecological development practice has significant global demonstration value. As China’s economy moves toward a high-quality development stage, the reform of the ecological civilization system is also steadily advancing. To seize the opportunities and meet the challenges brought about by the digital economy era, China has been building national-level comprehensive significant data pilot zones since 2015. As China’s first digital economy policy, building national comprehensive significant data pilot zones is based on digital information. The internet, big data, cloud computing, and artificial intelligence are the main digital carriers, technologies shown in the forms of digital industrialization and industrial digitalization. This strategy deepens the integration of the digital and real economies and provides a unique policy laboratory for global urban transformation. This strategy is an essential platform for promoting data elements’ integration, sharing, circulation, and application. It not only responds to the global governance consensus of the G20 meeting on the deep integration of digital and ecological development but also promotes the construction of China’s digital infrastructure by focusing on reducing carbon emissions, decreasing pollution, expanding green spaces, and fostering growth. It provides a clear direction for the simultaneous realization of industrialization and informatization in the new era and effectively promotes the development of green cities in China’s high-quality development stage. Based on this, as a quasi-natural experiment, the national-level comprehensive big data pilot zone has created a favorable research environment for this paper to study the relationship between the development of the digital economy and urban ecological resilience. This research is of enormous practical significance for alleviating global urban ecological and environmental problems and promoting the development of new forms of productivity.

In the 1970s, the ecologist Holling first proposed the concept of “ecological resilience” and defined it as the ability of an ecological environment to return to its original state after being polluted. This pioneering work marked the integration of the concept of resilience into ecology [1]. With the acceleration of urbanization, the idea of ecological resilience has gradually expanded from natural ecology to socio-ecology, and the concept of “urban ecological resilience” has been derived to describe the resistance, adaptability, and resilience of urban systems [2]. Scholars have conducted multidimensional research on urban ecological resilience. Some studies based on different geographical regions have constructed a theoretical framework and indicator system for essential resilience from the perspective of the conceptual evolution of urban ecological resilience and analyzed the measurement and impact mechanism of urban ecological resilience [3,4,5,6,7], providing crucial theoretical value for coordinated regional development. Other literature has explored the impact and mechanism of external factors on urban ecological resilience based on different models. First, the effect of interacting socio-economic factors on urban ecological resilience has been explored from the perspective of economic development [8,9,10]. Second, considering different driving factors, scholars have used the ecological resilience analysis framework and various models to measure and analyze the coupling and coordinated spatiotemporal evolution characteristics of driving variables and their impact on the comprehensive coupling degree of driving variables and ecological resilience. They have also explored regional differences and convergence characteristics [11,12]. In addition, to explore the development model of ecological civilization, some scholars have taken green space policies, urban structural transformation policies, and digital economy policies as exogenous institutional factors, and explored the effect of different economic policies on improving urban ecological resilience through double difference and spatial econometric methods [13,14,15,16]. Through a review of the literature on urban ecological resilience, it is clear that ecological resilience is closely related to urban economic development, and research on urban ecological resilience has made certain contributions to sustainable urban development.

The national-level big data comprehensive experimental zone represents the inaugural pilot policy for the digital economy. As early as 2015, the State Council promulgated the Action Outline for Promoting the Development of Big Data, which articulated the strategic vision for constructing national-level big data comprehensive experimental zones. In September of the same year, Guizhou province initiated the construction of the first pilot of a big data comprehensive experimental zone. The following year, China unveiled the list of the second batch of national-level big data comprehensive experimental zones, encompassing regions such as the Beijing–Tianjin–Hebei region, the Pearl River Delta, Shanghai, Henan, Chongqing, Shenyang, Inner Mongolia, and other regions. Empirical research has demonstrated that national-level big data comprehensive experimental zones have an impact on both the macroeconomy and microenterprises. Particularly with the rapid advancement of the digital economy, these zones, leveraging their inherent digital technology advantages, have optimized environmental monitoring and management to promote the green transformation of industries. Through the analysis of big data related to new energy resources, these zones can accurately forecast and manage energy demands, thus fostering the development and utilization of new energy. Consequently, as the pioneering pilot policy for the digital economy, the experimental zones provide an optimal research environment for examining urban ecological resilience from the perspective of the digital economy. Through a comprehensive review of the literature, it is evident that there is relatively abundant research on the policy effects of the pilot zone on the economy and ecology, mainly from the perspectives of economic consequences and environmental impact. In terms of financial impact, the pilot zone’s construction will promote the development of the city’s digital economy and the total factor productivity of enterprises [17,18]. Regarding environmental impact, the construction of the pilot zone will reduce urban carbon emissions and promote sustainable development [19,20,21]. Firstly, studies on urban ecological resilience predominantly focus on the spatiotemporal evolution or differentiation characteristics of regional urban ecological resilience. Furthermore, when examining the policy impacts of National Comprehensive Big Data Pilot Zones, there is a scarcity of literature that considers the perspective of urban ecological resilience. Thirdly, domestic research predominantly employs traditional econometric methods such as Difference-in-Differences (DiD), which have certain limitations. To address the shortcomings of current research, this paper primarily aims to validate two questions: Can the construction of pilot zones enhance urban ecological resilience? If so, what are the underlying mechanisms and pathways? In view of this, this paper selects 280 prefecture-level cities in China from 2013 to 2022 as samples. Leveraging the quasi-natural experiment of National Comprehensive Big Data Pilot Zones, this paper employs a double machine learning model to study the impact of pilot zone construction on urban ecological resilience, and conducts an in-depth analysis of its mechanism and heterogeneous effects, providing a reference for relevant policy formulation.

This study has several contributions to ecological resilience. First, based on the quasi-natural experiment of National Comprehensive Big Data Pilot Zones, this paper comprehensively explores and verifies the role of pilot zone construction in urban ecological resilience, enriching the research on influencing factors of urban ecological resilience. In the digital economy era, this paper provides theoretical support and beneficial exploration for accelerating pilot zone construction and promoting urban ecological environmental protection and green development. Second, from the perspective of external governance, this paper identifies important pathways influencing the relationship between pilot zone construction and urban ecological resilience. Previous research on urban ecological resilience has largely overlooked the crucial role played by external governance factors. As the first pilot policy for the digital economy, pilot zone construction serves as an institutional force that enhances urban ecological resilience. Therefore, the present paper selects pilot zone construction as a key external governance perspective, linking the logical chain between pilot zone construction and corporate investment efficiency. While improving the existing research on influencing factors of urban ecological resilience, this paper also expands the research on the policy effects of pilot zones. Third, different from previous studies that use traditional regression models to examine the relationship between the digital economy and corporate investment efficiency, this paper employs a double machine learning method based on a quasi-natural experiment to assess the impact of pilot zone construction on urban ecological resilience. This approach avoids the instability of estimators faced by traditional linear regression models, scientifically evaluates policy effects, and enhances the robustness of research results to a certain extent.

2. Theoretical Analysis and Hypotheses

2.1. Direct Impact of Pilot Zone Construction on Urban Ecological Resilience

By leveraging their unique advantages, pilot zones can effectively facilitate the circulation of data and information among cities, establish a comprehensive urban green development system, and enhance collaboration between urban governments and enterprises. This collaboration contributes to precise emission control and intelligent governance of industrial pollutants by governments and enterprises, thereby improving the ecological resilience of urban ecosystems in the face of destruction [22]. By integrating advanced digital technologies such as big data and cloud computing with environmental regulation, pilot zones facilitate the transition from traditional green regulatory models to digital ones, enhancing urban green regulatory capabilities and pollution treatment efficiency, which in turn boosts urban ecological adaptation resilience. Furthermore, pilot zones promote the digitization of industries and the industrialization of digital technologies, fostering the integrated development of the real economy and digital economy. Governments utilize the digital processing and regulatory advantages of pilot zones to ensure the rational implementation of ecological protection measures with the goal of protecting the ecological environment [23]. Pilot zones also employ digital technologies to measure and improve urban vegetation coverage, promoting green urban spaces. The construction of pilot zones helps optimize the urban ecological monitoring system, enabling monitoring and assessment of urban ecological conditions, detecting pollution sources, precisely treating pollutants, and conducting ecological restoration. With the assistance of digital green governance in pilot zones, the recovery efficiency of urban ecosystems is enhanced, contributing to the achievement of urban ecological civilization construction goals and sustainable development, and improving urban ecological recovery resilience. Based on the above discussion, the following hypothesis is proposed:

Hypothesis 1:

The construction of pilot zones has a positive effect on enhancing urban ecological resilience.

2.2. Urban Green Innovation, Pilot Zone Construction, and Urban Ecological Resilience

Endogenous technological innovation is a core concept of the new growth theory, which holds that knowledge, as an endogenous production factor of economic activity, is a prerequisite for technological innovation. The theory holds that knowledge is an asset derived from information and has the potential to grow exponentially. Therefore, urban green innovation has a technological spillover effect, can break down information barriers, overcome the temporal, spatial, and geographical obstacles in traditional information exchange, promote the interaction of innovation resources between different innovation entities in the city, and broaden the channels and scope of information dissemination [24]. Urban green innovation improves the application of advanced green technologies in the city and promotes the agglomeration of innovation factors.

Expressly, the construction of the pilot zone has provided a good research and development environment for the advancement of green technology in the region by leveraging its policy advantages, reducing the coordination and communication costs of scientific and technological researchers, and enabling innovation entities to quickly access relevant knowledge, resources and research and development momentum, thereby improving urban green innovation. It also improves the efficiency and resilience of urban ecological governance. At the same time, the pilot zone cities can promote the widespread application of digital green technology, such as smart city infrastructure, big data analysis, digital energy management and intelligent transportation, by improving urban green innovation, thereby reducing the environmental burden and improving ecological resilience. In addition, the construction of the pilot zone can promote real-time information sharing between different regions, effectively disseminate innovative green technologies, and assist enterprises in the region in energy conservation and emission reduction. In the wave of technological innovation, governments have widely adopted incentive policies for technological innovation to promote green technological innovation in enterprises [25]. Finally, practical green innovation is essential to improve environmental quality directly. Through the development and application of cleaner production technologies, waste recycling technologies, and pollution control technologies, emissions from industrial production and daily life can be effectively reduced [25], and cities’ ecological adaptability and resilience can be improved from the production source. Based on this, we propose the following hypothesis:

Hypothesis 2:

The construction of pilot zones has a positive effect on enhancing urban ecological resilience through urban green innovation

2.3. Energy Utilization Efficiency Improvement, Pilot Zone Construction, and Urban Ecological Resilience

The construction of pilot zones can improve market transparency and promote the flow of resource elements within the region, which can help reduce the welfare losses caused by inefficient energy use. Based on this theory, the construction of pilot zones can improve the transparency of the resource market and promote energy efficiency within the pilot zone, thereby helping to digitize and intelligently manage energy, promote the transformation of traditional industries [26], promote the substitution effect of clean energy, and ultimately reduce environmental pollution from multiple dimensions of the upstream and downstream of the industry and through urban ecological resilience.

Specifically, cities transform traditional energy through digital methods such as artificial intelligence and 5G. In particular, by real-time monitoring the energy consumption data of manufacturing enterprises in the city and using a big data management platform for analysis and visual management, the government can identify high energy–consuming enterprises and key time nodes, thereby improving energy efficiency, reducing energy waste, and promoting the city’s ecological recovery potential [27]. In addition, in the context of integrating the digital and real economies, the construction of the pilot zone improves energy efficiency. It reduces energy transaction costs through market supply and demand matching, thereby reducing excess energy consumption. Through big data analysis, the pilot zone can identify key industries and regions with energy waste, optimize resource allocation to reduce environmental pollution caused by energy waste, and promote the ecological resilience of the city. Finally, the pilot area monitors the deviation between current energy use and the benchmark in real time. It adjusts strategies accordingly, improves energy efficiency, promotes the transformation of traditional industries, and forces high-polluting enterprises to use clean energy [28], thereby improving the ecological resilience of the city. Based on these insights, Hypothesis 3 is posited:

Hypothesis 3:

The construction of pilot zones has a positive effect on enhancing urban ecological resilience through energy utilization efficiency improvement.

Therefore, this paper argues that the construction of pilot zones can significantly enhance cities’ ecological resilience and empower urban ecological resilience through two paths: improving the level of urban green innovation and energy efficiency. The specific mechanism process is shown in Figure 1.

3. Research Design

3.1. Model Design

Most existing studies employ the difference-in-differences (DID) approach as the primary model for evaluating policy effects. However, the application of the DID model relies on the parallel trends assumption, which is inherently untestable. Direct application of this model may lead to biased estimation results. Additionally, traditional models struggle with challenges such as uncertainty in the functional form of confounding factors, underfitting caused by regularization, the curse of dimensionality, and issues of overfitting or oversimplification stemming from an excessive focus on “consistency.” To address these limitations in model specification and linear assumptions, Chernozhukov et al. [29] proposed the Double Machine Learning (DML) framework. By combining propensity score methods with linear regression, the DML approach decomposes causal inference into two independent prediction steps: Neyman orthogonalization and cross-fitting. This method mitigates the stringent and specific assumptions of traditional models, leveraging machine learning algorithms to enhance the accuracy and robustness of causal effect estimation and thus providing a novel tool for policy effect evaluation.

This paper employs a partially linear DML model to analyze the impact of experimental zone construction on urban ecological resilience. The following is the model (Equation (1)) and its auxiliary equations (Equation (3)):

$${\mathrm{Y}}_{\mathrm{i}\mathrm{t}}={\theta }_0{\mathrm{D}}_{\mathrm{i}\mathrm{t}}+\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)+{\epsilon }_{\mathrm{i}\mathrm{t}},\mathrm{E}\left({\epsilon }_{\mathrm{i}\mathrm{t}}|{\mathrm{X}}_{\mathrm{i}\mathrm{t}},{\mathrm{D}}_{\mathrm{i}\mathrm{t}}\right)=0$$

(1)

In this paper; i represents the city; t represents the year; ${\mathrm{Y}}_{\mathrm{i}\mathrm{t}}$ represents urban ecological resilience; ${\mathrm{D}}_{\mathrm{i}\mathrm{t}}$ is the core explanatory variable, namely, the binary variable indicating whether the city is part of a national big data comprehensive experimental zone; ${\theta }_0$ is the regression coefficient of focus; ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ is a set of multidimensional control variables; $\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)$ needs to be estimated using machine learning methods; ${\epsilon }_{\mathrm{i}\mathrm{t}}$ is the error term; and the conditional mean is 0. We directly estimated Equation (1) to obtain the regression coefficient $\widehat{{\theta }_0}={\left({\displaystyle \frac{1}{\mathrm{n}}}\sum {\mathrm{D}}_{\mathrm{i}\mathrm{t}}^2\right)}^{-1}{\displaystyle \frac{1}{\mathrm{n}}}\sum \widehat{{\sigma }_{\mathrm{i}\mathrm{t}}}\left({\mathrm{Y}}_{\mathrm{i}\mathrm{t}}-\widehat{\mathrm{g}}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)\right)$, However, since ${\theta }_0$ is difficult to converge to $\widehat{{\theta }_0}$, there is an estimated bias that needs further investigation and correction. The specific method is as follows:

$$\sqrt{n}(\widehat{{\theta }_0}-{\theta }_0)={\left({\displaystyle \frac{1}{\mathrm{n}}}\sum {\mathrm{D}}_{\mathrm{i}\mathrm{t}}^2\right)}^{-1}{\displaystyle \frac{1}{\sqrt{\mathrm{n}}}}\sum _{\mathrm{i}\in \mathrm{I}}{\epsilon }_{\mathrm{i}\mathrm{t}}+{\left({\displaystyle \frac{1}{\mathrm{n}}}\sum {\mathrm{D}}_{\mathrm{i}\mathrm{t}}^2\right)}^{-1}{\displaystyle \frac{1}{\sqrt{\mathrm{n}}}}\sum _{\mathrm{i}\in \mathrm{I}}{\epsilon }_{\mathrm{i}\mathrm{t}}[\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)-\widehat{\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)}]$$

(2)

In Equation (2), let x =${\left({\displaystyle \frac{1}{\mathrm{n}}}\sum {\mathrm{D}}_{\mathrm{i}\mathrm{t}}^2\right)}^{-1}{\displaystyle \frac{1}{\sqrt{\mathrm{n}}}}\sum _{\mathrm{i}\in \mathrm{I}}{\epsilon }_{\mathrm{i}\mathrm{t}}$ and follow a normal distribution with a mean of 0, and let y =${\left({\displaystyle \frac{1}{\mathrm{n}}}\sum {\mathrm{D}}_{\mathrm{i}\mathrm{t}}^2\right)}^{-1}{\displaystyle \frac{1}{\sqrt{\mathrm{n}}}}\sum _{\mathrm{i}\in \mathrm{I}}{\epsilon }_{\mathrm{i}\mathrm{t}}[\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)-\widehat{\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)}]$. Dual machine learning models use machine learning and regularization algorithms to estimate $\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)$. Although this can avoid the problem of overestimation of the variance, the introduction of the regularization term can lead to estimation bias. $\widehat{\mathrm{g}}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)$ converges slowly to $\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)$, $n^{-{\phi }_g}>n^{-1/2}$, when $\mathrm{n}\to \mathrm{\infty },\mathrm{y}\to \mathrm{\infty },\widehat{{\theta }_0}$ is difficult to converge to ${\theta }_0$. Therefore, this paper uses an orthogonal method to correct this bias:

$${\mathrm{D}}_{\mathrm{i}\mathrm{t}}=\mathrm{m}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)+{\sigma }_{\mathrm{i}\mathrm{t}},\mathrm{E}\left({\sigma }_{\mathrm{i}\mathrm{t}}|{\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)=0$$

(3)

Continuously, a regression estimate is performed on Equation (3) to obtain an estimate of the residual term ${\sigma }_{\mathrm{i}\mathrm{t}}={\mathrm{D}}_{\mathrm{i}\mathrm{t}}-\mathrm{m}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)$. ${\sigma }_{\mathrm{i}\mathrm{t}}$ is regarded as the instrumental variable of ${\mathrm{D}}_{\mathrm{i}\mathrm{t}}$ and is regressed to obtain an unbiased estimate of ${\theta }_0$ as shown below:

$$\widehat{{\theta }_0}={\left({\displaystyle \frac{1}{\mathrm{n}}}\sum \widehat{{\sigma }_{\mathrm{i}\mathrm{t}}}{\mathrm{D}}_{\mathrm{i}\mathrm{t}}\right)}^{-1}{\displaystyle \frac{1}{\mathrm{n}}}\sum \widehat{{\sigma }_{\mathrm{i}\mathrm{t}}}\left({\mathrm{Y}}_{\mathrm{i}\mathrm{t}}-\widehat{\mathrm{g}}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)\right)$$

(4)

This study further examines its estimation bias:

$$\sqrt{n}(\widehat{{\theta }_0}-{\theta }_0)={\left({\displaystyle \frac{1}{\mathrm{n}}}\sum {\mathrm{D}}_{\mathrm{i}\mathrm{t}}^2\right)}^{-1}{\displaystyle \frac{1}{\sqrt{\mathrm{n}}}}\sum _{\mathrm{i}\in \mathrm{I}}{\sigma }_{\mathrm{i}\mathrm{t}}{\mathrm{D}}_{\mathrm{i}\mathrm{t}}+{\left({\displaystyle \frac{1}{\mathrm{n}}}\sum \mathrm{E}\left({{\sigma }_{\mathrm{i}\mathrm{t}}}^2\right)\right)}^{-1}{\displaystyle \frac{1}{\sqrt{\mathrm{n}}}}\sum _{\mathrm{i}\in \mathrm{I}}[\mathrm{m}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)-\widehat{\mathrm{m}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)}][\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)-\widehat{\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)}]$$

(5)

At this point, the convergence rate of ${\theta }_0$ will depend on the convergence rates of $\widehat{\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)}$ to $\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)$ and $\widehat{\mathrm{m}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)}$ to $\mathrm{m}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)$. On the one hand, the two machine learning estimates help to exclude the influence of the confounding variable set ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ in the treatment variable ${\mathrm{D}}_{\mathrm{i}\mathrm{t}}$. On the other hand, they can accelerate the convergence rate of $\widehat{{\theta }_0}$, thereby obtaining accurate estimates with a finite sample. In addition, this paper uses a 5-fold cross-validation method to process the training samples in the double machine learning regression process to improve the stability and reliability of the model estimation.

3.2. Variables

3.2.1. Dependent Variable

Urban ecological resilience (RES). Drawing on the environmental quality performance assessment methods in the existing literature [10,11,15,16] and considering the socio-economic characteristics of the city, this study decomposes urban ecological resilience into three sub-indices: resistance resilience, recovery resilience, and adaptive resilience. These sub-indices are measured by 14 tertiary indicators (see

Table 1. Urban ecological resilience indicator system.

Primary Indicator	Secondary Indicator	Tertiary Indicators	Indicator Directionality
Urban Ecological Resilience	Resistance Resilience Index	Per capita industrial wastewater discharge	Negative indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Per capita industrial sulfur dioxide emissions	Negative indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Per capita industrial smoke and dust emissions	Negative indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Per capita industrial nitrogen oxide emissions	Negative indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Annual average PM2.5 emissions	Negative indicator
	Adaptive Resilience Index	Industrial sulfur dioxide treatment	Positive indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Industrial smoke and dust treatment	Positive indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Harmless disposal of domestic waste	Positive indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Centralized treatment of sewage in treatment plants	Positive indicator
	Resistance Resilience Index Adaptive Resilience Index Recovery Resilience Index	Comprehensive utilization of industrial solid waste	Positive indicator
	Recovery Resilience Index	Per capita water resources availability	Positive indicator
		Green coverage rate in built-up areas	Positive indicator
		Per capita green space in built-up areas	Positive indicator
		Per capita built-up area	Positive indicator

).

Considering that different indicators have both positive and negative impacts on the overall resilience index, this study employs dimensionless normalization methods and uses the entropy weighting method to comprehensively calculate the ecological resilience index of cities. The following are the specific formula steps for calculating the urban ecological resilience index using the entropy weight method.

Standardize the Extreme Values According to the Direction of the Indicator

Positive indicator (the larger the better, e.g., Per capita water resources availability):

$$x_{ij}^{\prime }={\displaystyle \frac{x_{ij}-min\left(x_j\right)}{max\left(x_j\right)-min\left(x_j\right)}}$$

(6)

Negative indicator (the smaller the better, e.g., Per capita industrial wastewater discharge):

$$x_{ij}^{\prime }={\displaystyle \frac{max\left(x_j\right)-x_{ij}}{max\left(x_j\right)-min\left(x_j\right)}}$$

(7)

In this study, the sample size is n, the number of evaluation indicators is m, and the original data matrix is X =$x_{ij}$.

Entropy Method Application

Calculate weight of indicator:

$$p_{ij}={\displaystyle \frac{x_{ij}^{\prime }+\epsilon }{{\sum }_{i=1}^n(x_{ij}^{\prime }+\epsilon )}}$$

(8)

Calculate information entropy:

$${entropy}_j=-k{\sum }_{i=1}^n{p}_{ij}In(p_{ij}),k={\displaystyle \frac{1}{In\left(n\right)}}$$

(9)

Determine the weight of the indicator:

$${weight}_{ij}={\displaystyle \frac{1-{entropy}_j}{{\sum }_{j=1}^{m}1-{entropy}_j}}$$

(10)

Calculate the comprehensive ecological resilience score for city i:

$$urban \mathrm{e}\mathrm{c}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{e}={\sum }_{j=1}^m{x_{ij}^{\prime }\times weight}_{ij}$$

(11)

3.2.2. Core Explanatory Variable

The core explanatory variable in this study is the construction of national big data comprehensive experimental zones (${\mathrm{D}}_{\mathrm{i}\mathrm{t}}$). A policy dummy variable (${\mathrm{T}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{t}}_{\mathrm{i}\mathrm{t}}$) is constructed to indicate whether a city is designated as a pilot city for these zones. The variable takes the value of 1 if the city is a pilot city and 0 otherwise. A time dummy variable (${\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}}_{\mathrm{i}\mathrm{t}}$) is also constructed, where the starting year is 2015 for the first batch of zones and 2016 for the second batch. If a city was established as or began to be designated as a zone in or after these years, the variable takes the value of 1; otherwise, it is 0.${\mathrm{D}}_{\mathrm{i}\mathrm{t}}={\mathrm{T}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{t}}_{\mathrm{i}\mathrm{t}}\ast {\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}}_{\mathrm{i}\mathrm{t}}$ is a dummy variable used to measure the construction of the pilot area.

According to the National Development and Reform Commission’s list, the national-level comprehensive pilot zones for big data include two cross-regional comprehensive pilot zones (the construction area comprises Beijing, Tianjin, Hebei Province, and the cities in the Pearl River Delta), four regional demonstration comprehensive pilot zones (the construction area comprises Shanghai, Henan Province, Chongqing, and Shenyang), and one comprehensive pilot zone for the coordinated development of big data infrastructure (built in the Inner Mongolia Autonomous Region). For specific cities, please refer to Figure 2.

3.2.3. Control Variables

Based on existing literature [11,13,14,15,16] and data availability, this study selects the following control variables:

• Industrial structure;
• Level of economic development;
• Informationization level;
• Financial development level;
• Technological level;
• Urbanization level.

For a detailed explanation of the variables, see

Table 2. Variable indicators.

Variable Type	Variable Description	Calculation Methodology
Dependent Variable	Urban Ecological Resilience	See detailed calculations above
Core Explanatory Variable	The Construction of National Big Data Comprehensive Experimental Zone $({\mathrm{D}}_{\mathrm{i}\mathrm{t}}$)	See detailed calculations above
Control Variables	Industrial Structure	Ratio of tertiary industry output to total GDP
	Level of Economic Development	Logarithm of GDP per capita
	Informationization Level	Ratio of internet users to total population at year-end
	Financial Development Level	Ratio of financial institution deposits to GDP
	Technological Level	Ratio of R&D expenditure to GDP
	Urbanization Level	Logarithm of urban population at year-end

.

3.2.4. Mechanism Variable

According to the theoretical analysis in the previous section, the construction of the pilot zone can promote the improvement of urban ecological resilience through two paths: improving the level of urban green innovation and energy efficiency. Therefore, this paper selects urban green innovation and energy efficiency as the mechanism variables. The

Table 3. Mechanism variable.

Mechanism Variable	Measurement Method	Notation
urban green innovation	the logarithm of the number of green utility patents filed in a given year	GI
urban energy efficiency	the SBM super-efficiency model’s energy efficiency data	SBM

explains these two variables.

3.3. Data Sources

This study uses data from 280 cities in China between 2013 and 2022. The data primarily come from the National Bureau of Statistics, the China Urban Statistical Yearbook, the China Energy Statistical Yearbook, and the China Industrial Statistical Yearbook. Missing data were supplemented using interpolation methods, resulting in a final sample of 2766 observations.

4. Empirical Analysis

4.1. Benchmark Regression Analysis

Using the random forest algorithm, this paper randomly splits the dataset into five sub-samples (with a sample split ratio of 1:4). It then uses a partially linear double machine learning model to check the policy impact of building the pilot zone on urban ecological resilience. Furthermore, time and city fixed effects are added, and then the main regression for prediction (Equation (1)) and the auxiliary regression (Equation (3)) are solved. The specific regression results are shown in

Table 4. Benchmark regression.

Title 1	Title 2	Title 3	Title 4
	(1) RES	(2) RES	(3) RES
The construction of national big data comprehensive experimental zones	0.0010 *** (0.0002)	0.0015 *** (0.0005)	0.0016 *** (0.0005)
The first-order term of control variables	No	Yes	Yes
The second-order term of control variables	No	No	Yes
Time fixed effects	Yes	Yes	Yes
City fixed effects	Yes	Yes	Yes
N	2766	2766	2766

*** indicate significance at the 1% levels, respectively. The robust standard errors clustered at the city and time levels are in parentheses.

. Under the condition of adding the fixed effect of year and city, the construction of the experimental area has a positive impact on the ecological resilience of the city, and it is significant at the level of 1%. The results show that the construction of a national big data comprehensive pilot zone can significantly improve the ecological resilience of cities. After adding the control variable (Column 2) and the quadratic term to the control variable (Column 3), the regression coefficient results were still significant. In each regression, the dual machine learning framework robustly estimates the impact of the construction of the experimental area on the regression coefficient (${\theta }_0$) by adjusting the dimension of the control variable (${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$). The significance and coefficient stability (similar to the coefficients in the three columns of Table 4) verify the reliability of Hypothesis 1; that is, the construction of experimental areas can improve the ecological resilience of cities.

4.2. Robustness Tests

4.2.1. Instrumental Variable Method

To better avoid the problem of endogeneity and satisfy the correlation between endogenous variables and instrumental variables, this paper selects the multiplier term of the number of calls per 100 people and the number of internet users in the previous year as the instrumental variables. To a certain extent, the number of telephone users and the number of internet users in the city reflect the level of urban digitalization, which is closely related to the construction of the national big data comprehensive pilot zone. The number of telephone usage and the number of internet users in the city are determined by the needs of enterprises and individuals, and there is no relevant evidence that the multiplication term between the number of telephone usage and the number of internet users in regional cities is correlated with urban ecological resilience, which is in line with the exogenous hypothesis. In this paper, a dual machine learning partial linear tool variable model is constructed to test the endogeneity, and the model settings are as follows: the regression results are shown in column (1) of

Table 5. Robustness test.

	(1) RES	(2) RES	(3) RES	(4) RES	(5) RES	(6) RES	(7) RES
The construction of national big data comprehensive experimental zones	0.0325 *** (0.0147)	0.0151 *** (0.0043)	0.0053 *** (0.0015)	0.0009 * (0.0004)	0.0010 ** (0.0004)	0.0010 *** (0.0004)	0.0011 *** (0.0004)
The first-order term of control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes
The second-order term of control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Province–time	No	No	No	No	No	Yes	No
City–time	No	No	No	No	No	No	Yes
N	2766	2766	2766	2766	2766	2766	2766

*, **, *** indicate significance at the 10%, 5% and 1% levels, respectively. The robust standard errors clustered at the city and time levels are in parentheses.

, and the regression coefficient of the construction of the experimental area significantly indicates that the results are reliable.

4.2.2. Dynamic Effect Analysis and Placebo Test

First, the benchmark regression analyzes the average treatment effect of the construction of the pilot zone on urban ecological resilience. This paper considers the time heterogeneity of the construction of the pilot zone. It explores the long-term dynamic performance of the policy effect of the construction of the pilot zone in empowering urban ecological resilience. The following dynamic effect analysis model is introduced based on the baseline regression:

$${\mathrm{Y}}_{\mathrm{i}\mathrm{t}}={\theta }_{\mathrm{j}}{\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{\mathrm{j}}+\mathrm{g}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)+{\epsilon }_{\mathrm{i}\mathrm{t}},\mathrm{E}\left({\epsilon }_{\mathrm{i}\mathrm{t}}|{\mathrm{X}}_{\mathrm{i}\mathrm{t}},{\mathrm{D}}_{\mathrm{i}\mathrm{t}}\right)=0$$

(12)

$${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{\mathrm{j}}=\mathrm{m}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)+{\sigma }_{\mathrm{i}\mathrm{t}},\mathrm{E}\left({\sigma }_{\mathrm{i}\mathrm{t}}|{\mathrm{X}}_{\mathrm{i}\mathrm{t}}\right)=0$$

(13)

In Formula (12), ${\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{\mathrm{j}}$ is a dummy variable for each year after the construction of the experimental area, ${\theta }_{\mathrm{j}}$ represents the policy effect of the construction of the experimental area in year j, and other variables are defined in the same way as in Formula (1). The results of the analysis of the dynamic effects of the construction of the experimental area are shown in

Table 6. Dynamic effect analysis and placebo test.

	Average Effect	Dynamic Effect	Placebo Test
	(1) RES	(2) RES	(3) RES	(4) RES
The construction of national big data comprehensive experimental zones $({\mathrm{D}}_{\mathrm{i}\mathrm{t}}$)	0.0016 *** (0.0005)
${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{2016}$		0.0008 ** (0.0002)
${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{2017}$		0.0010 *** (0.0003)
${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{2018}$		0.0007 *** (0.0002)
${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{2019}$		0.0013 *** (0.0004)
${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{2020}$		0.0017 *** (0.0005)
${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{2021}$		0.0015 *** (0.0005)
${\mathrm{D}}_{\mathrm{i}\mathrm{t}}\times {\mathrm{D}\mathrm{y}\mathrm{n}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{c}}_{2022}$		0.0019 *** (0.0006)
${\mathrm{D}}_{2014}$			−0.0011 (0.0007)
${\mathrm{D}}_{2013}$				0.0009 (0.0006)
The first-order term of control variables	YES	YES	YES	YES
The second-order term of control variables	YES	YES	YES	YES
Time fixed effects	YES	YES	YES	YES
City fixed effects	YES	YES	YES	YES
Observations	2766	2766	2766	2766

**, *** indicate significance at the 5% and 1% levels, respectively. The robust standard errors clustered at the city and time levels are in parentheses.

column (1) and (2). It can be seen that after the implementation of the policy, the impact of the construction of the experimental area on the ecological resilience of the city tends to increase.

Second, this paper advances the construction of the experimental area by one (${\mathrm{D}}_{2014}$) to two years (${\mathrm{D}}_{2013}$) to rule out the impact of other policies and unobservable variables on urban ecological resilience. If the coefficient significance of the construction of the experimental area at this time is the same as the benchmark regression result, it indicates that other policy variables may affect urban ecological resilience. On the contrary, it indicates that urban ecological resilience is affected by the construction of the experimental area. Table 6 lists the regression estimates for column (3) and column (4), which are one year and two years ahead of the construction year of the experimental area, respectively. The results show that after the construction of the experimental area is advanced, the regression coefficients are insignificant and pass the placebo test.

4.2.3. Exclude the Impact of Other Pilot Policies

During the sample period of the construction of the pilot zone, the smart city policy introduced in 2013 and the broadband China pilot policy implemented in 2013 may have caused bias in the policy effects of the construction of the pilot zone. Therefore, dummy variables for implementing smart city and broadband China pilots are added to the baseline regression model to control for their effects. The specific results are shown in columns (2) and (3) of Table 5. After controlling for the impact of the three policies separately, the coefficient of the pilot zone construction is still significantly positive, which indicates that the construction of pilot zones has a significant effect on urban ecological resilience at the same time. The results of this paper are still robust.

4.2.4. Eliminate the Influence of Outliers

In order to avoid the presence of outliers in the sample that may affect the unbiased regression results, all variables are tailed by 1% and 5% samples. As can be seen from columns (4) and (5) of Table 5, the regression coefficients of the national big data comprehensive experimental area on urban ecological resilience are still significant after the 1% and 5% samples are tailed, which indicates that the benchmark regression results in this paper are still robust.

4.2.5. Consider the Province–Time and City–Time Effects

Provinces and cities serve as a bridge between the central and local governments, responsible for the implementation of the pilot zone construction and making appropriate adjustments based on actual conditions. Considering provincial and city factors in the study of urban ecological resilience in the construction of experimental areas is helpful for addressing concerns about unobserved heterogeneity over time and better evaluating the effectiveness of policies. Provinces and cities have their own characteristics in government governance, and jointly promote the modernization of the national governance system and governance capacity. Therefore, on the basis of benchmark regression, this paper adds the fixed effects of province–time and city–time interaction to control the impact of different provinces and cities over time. The specific regression results are shown in columns (6) and (7) of Table 5, and the impact of the construction of the national big data comprehensive pilot zone on urban ecological resilience is still significantly positive at the level of 1% after controlling for the fixed effects of province–time and city–time interaction, and the conclusion of this paper is still valid.

4.2.6. Reset the Split Ratio

In order to explore the influence of the sample segmentation ratio on the conclusion of this paper, this paper considers that the segmentation ratio K of the main sample and the auxiliary sample in the benchmark regression is set to 4 with the value of 5; that is, the sample segmentation ratio is 1:3. The regression results are shown in column 6 (1) of Table 7, and the regression coefficient of the construction of the experimental area is still significantly positive, which means that the change of the sample segmentation ratio does not affect the research conclusion of this paper.

4.2.7. Reset the Dual Machine Learning Algorithm

In order to avoid the impact of the bias of the double machine learning algorithm on the conclusion and test the robustness of the conclusion, this paper adopts the method of replacing the machine learning algorithm by replacing the random forest algorithm previously used for prediction with gradient boosting, lasso regression, and support vector machine to alleviate the impact of the algorithm as much as possible. As can be seen from columns (2), (3), and (4) of Table 7, after replacing the machine learning algorithm, the regression coefficient of the construction of the experimental area is still significant, indicating the reliability of the conclusion that the construction of the experimental area improves the ecological resilience of the city.

4.2.8. Reset the Machine Learning Model

In this paper, the machine learning model is replaced to test the robustness of the conclusion to avoid the impact of the double machine learning model setting bias on the conclusion, and the previous partial linear model is replaced with an interactive model and the regression results are shown in column 6 (5). The regression coefficient of the construction of the national big data comprehensive pilot area is still significantly positive, which means that the replacement of the machine learning model does not affect the robustness of the conclusions of this paper.

Table 7. Machine learning model tests.

	(1) RES	(2) RES	(3) RES	(4) RES	(5) RES
The construction of national big data comprehensive experimental zones	0.0016 *** (0.0005)	0.0013 *** (0.0004)	0.0009 ** (0.0005)	0.0032 *** (0.0004)	0.0007 *** (0.0002)
The first-order term of control variables	Yes	Yes	Yes	Yes	Yes
The second-order term of control variables	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes
City fixed effects	Yes	Yes	Yes	Yes	Yes
N	2766	2766	2766	2766	2766

**, *** indicate significance at the 5% and 1% levels, respectively. The robust standard errors clustered at the city and time levels are in parentheses.

Table 7. Machine learning model tests.

	(1) RES	(2) RES	(3) RES	(4) RES	(5) RES
The construction of national big data comprehensive experimental zones	0.0016 *** (0.0005)	0.0013 *** (0.0004)	0.0009 ** (0.0005)	0.0032 *** (0.0004)	0.0007 *** (0.0002)
The first-order term of control variables	Yes	Yes	Yes	Yes	Yes
The second-order term of control variables	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes
City fixed effects	Yes	Yes	Yes	Yes	Yes
N	2766	2766	2766	2766	2766

**, *** indicate significance at the 5% and 1% levels, respectively. The robust standard errors clustered at the city and time levels are in parentheses.

4.3. Heterogeneity Analysis

4.3.1. Heterogeneity Analysis Based on the Generalized Random Forest Model

In the analysis of policy evaluation, treatment effects often vary due to the characteristics of individuals or groups. This study adopts the generalized random forest (GRF) model proposed by Bedane [30] to construct an estimation framework for the conditional average treatment effect (CATE). By using nonparametric methods, it breaks through the prior constraints on the functional form of the treatment effect in traditional regression models, enhancing the flexibility and adaptability of the model. Furthermore, the GRF model captures the heterogeneity patterns in the multidimensional feature space through a recursive partitioning algorithm, which is suitable for exploring the nonlinear interaction between the policy effects of big data pilot zones and urban characteristics.

• Distribution of Treatment Effects

The distribution characteristics of the treatment effects can be seen in Figure 3, the treatment effect values mainly concentrate around 0.005, and the distribution shows a right-skewed trend. This indicates that the construction of the pilot zones has a significant positive impact on the ecological resilience of most cities, and the impact effects mainly lie in the positive value area. The wide distribution range of the treatment effects shows that there are significant differences in policy effects among different cities. Such differences may stem from the disparities in urban characteristics such as economic levels, informatization levels, and urbanization levels. In addition, the extreme-value characteristics on the right side of the distribution suggest that there may be a super-linear enhancement effect in the policy response mechanism of specific cities, meaning that the construction of the pilot zones has a particularly significant effect on improving the ecological resilience of these cities.

2.

Degree of Feature Contribution

In this article, the generalized random forest model is constructed using Python(version 3.9)’s GRF library, and the feature importance is obtained after the model training is complete. The feature importance score reflects the magnitude of the role played by each feature in the model prediction. Based on the feature importance analysis in this study (Figure 4), the marginal contribution of the urbanization level is significantly higher than that of other variables. Cities with a high urbanization level may possess stronger resource-integration capabilities and adaptability, and can respond more effectively to the construction of the pilot zones and enhance ecological resilience. In addition, the informatization level and scientific and technological level also significantly contribute to urban ecological resilience. It may be that technological and information infrastructure plays a key role in enhancing urban resilience, especially in promoting urban intelligence, digitization, and sustainable development. In contrast, the contribution of the industrial structure to urban ecological resilience is relatively low, and its importance value is significantly smaller than that of other features. This indicates that, under the current policy framework, the marginal effect of industrial structure adjustment on urban resilience improvement is relatively limited, or its role in urban resilience construction has not been fully released.

3.

Treatment Effects and Urbanization Level (Rate)

Since the urbanization level contributes the most to the policy effect of the experimental area in promoting the improvement of urban ecological resilience, this paper groups the sample according to the estimated effect of policy treatment in the pilot area, and divides the sample into a “low” effect group and a “high” effect group according to the median of the treatment effect to further analyze the possible nonlinear relationship between the level (rate) of urbanization and the effect of policy treatment in the pilot area. A scatter plot is created using the Python data visualization library to display the results and show the inverted U-shaped relationship between the level (rate) of urbanization and the treatment effect in different effect groups (Figure 5). The treatment effect generally shows an upward trend with the increase in the urbanization level, but it decreases slightly when the urbanization level reaches 3.0. Specifically, when the urbanization level is in the range of 1.0 to 2.5, the positive impact of the policy on urban ecological resilience gradually increases, indicating that cities with high urbanization have significant advantages in resource integration, infrastructure improvement, and policy-implementation capabilities, and thus are more likely to benefit from the policy. However, when the urbanization level reaches 3.0, the treatment effect decreases slightly. This may be because the ecological carrying capacity constraints caused by the congestion effect may offset the policy benefits.

4.3.2. Group Heterogeneity Analysis

• Heterogeneity in Resource Endowments

Resource-based cities are defined as those whose principal economic pillars rest upon the exploitation and processing of natural resources, and they are often accompanied by the characteristics of ecological environment pollution. Enhancing urban ecological resilience constitutes one of the objectives in the construction of experimental zones, which aligns with the long-term goal of resource-based cities in their pursuit of transformation and development. Consequently, referring to the “National Sustainable Development Plan for Resource-based Cities (2013–2020)” issued by the State Council in 2013, in order to examine whether there exists heterogeneity in resource endowments regarding the impact of the construction of experimental zones on urban ecological resilience, cities are categorized into resource-based cities and non-resource–based cities. The regression results are presented in columns (1) and (2) of

Table 8. Group heterogeneity analysis.

	(1) RES	(2) RES	(3) RES	(4) RES	(5) RES	(6) RES	(7) RES
The construction of national big data comprehensive experimental zones	0.001 ** (0.001)	0.001 (0.000)	0.000 (0.000)	0.000 (0.001)	0.003 *** (0.001)	0.001 ** (0.001)	0.001 * (0.000)
The first-order term of control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes
The second-order term of control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	2766	2766	2766	2766	2766	2766	2766

*, **, *** indicate significance at the 10%, 5% and 1% levels, respectively. The robust standard errors clustered at the city and time levels are in parentheses.

. It is demonstrated that the construction of big data comprehensive experimental zones is conducive to the ecological resilience of the cities within the experimental zones, and its promotional effect on the ecological resilience of resource-based cities is even more pronounced.

2.

Heterogeneity in Geographical Location

China boasts a vast territory. Due to diverse geographical distributions, different cities exhibit significant disparities in terms of economic development levels, cultural customs, and so forth. The impact of the construction of experimental zones on urban ecological resilience may vary depending on geographical locations. Hence, cities are divided into three parts—namely, the eastern, western, and central regions according to their geographical positions—and the impacts of these different regions on urban ecological resilience are investigated separately. The results are shown in columns (3), (4), and (5) of Table 8, indicating that the construction of experimental zones has a more conspicuous impact on the ecological resilience of cities in the western region. This might be attributed to the relatively weaker natural environment foundation in the western region compared to that in the eastern and central regions, rendering the ecosystem more vulnerable to damage. After the implementation of effective protection measures in the construction of the experimental zones, the recovery effect of the ecological environment and the enhancement of ecological resilience in the western region will be more evident.

3.

Environmental Protection Areas

Environmental protection areas typically regard environmental protection as one of the goals of urban development and adopt a series of effective measures to safeguard and improve the ecological environment. These cities usually possess relatively high ecological environment quality, abundant natural resources, as well as a comprehensive system of environmental protection regulations and policies. The impact of the construction of experimental zones on urban ecological resilience may be subject to differences depending on whether a city is an environmental protection area city or not. Therefore, 272 cities are classified according to whether they are environmental protection area cities. If a city belongs to an environmental protection area, it is assigned a value of 1; otherwise, it is assigned a value of 0. The results are presented in columns (6) and (7) of Table 8. The regression coefficient of the construction of experimental zones within environmental protection area cities is more significant, suggesting that the construction of experimental zones is more capable of promoting the urban ecological resilience of environmental protection area cities.

5. Further Analysis

According to the theoretical analysis in the previous section, the construction of pilot zones may promote the improvement of urban ecological resilience by improving the level of urban green innovation and energy utilization efficiency. Therefore, this section draws on Jiang Ting’s ideas and methods to verify the causal mediation effect.

Referring to the practices of Xu and Dai [26,28], the green innovation level (GI) and energy efficiency (SBM) of cities are measured using the logarithm of the number of green utility patents filed in a given year and the SBM super-efficiency model’s energy efficiency data. A random forest algorithm is employed, incorporating time and city fixed effects. Column (1) of

Table 9. Further analysis.

	(1) GI	(2) SBM
The construction of national big data comprehensive experimental zones	525.496 *** (158.515)	0.063 *** (0.010)
The first-order term of control variables	YES	YES
The second-order term of control variables	YES	YES
Time fixed effects	YES	YES
City fixed effects	YES	YES
N	2766	2766

*** indicate significance at the 1% levels, respectively. The robust standard errors clustered at the city and time levels are in parentheses.

uses the green innovation level (GI) as the dependent variable.

The regression coefficient of pilot zone construction is significantly positive, indicating that the establishment of pilot zones significantly improves urban green innovation levels. The improvement in the level of green innovation has promoted the development of green technology and the application of technological innovation in environmental protection. In particular, the accelerated growth of green patents has promoted the research, development, and application of energy-saving technologies, emission reduction technologies, and resource recycling technologies, alleviating the burden on the environment [25,26], improving the adaptability of cities to climate change, and shortening the recovery period after ecological disasters while enhancing adaptive and restorative resilience.

Column (2) of Table 9 presents the regression results for urban energy efficiency (SBM). The regression coefficient of pilot zone construction is significantly positive, suggesting that the establishment of pilot zones significantly enhances urban energy efficiency. Improvements in energy efficiency not only substantially reduce pollutant and greenhouse gas emissions [31], enhancing urban ecological resistance resilience, but also drive the optimization of energy structures and sustainable urban development capabilities [32,33], ultimately improving urban ecological resilience.

These findings validate Hypotheses 2 and 3, demonstrating that pilot zone construction enhances urban ecological resilience through improvements in urban innovation levels and energy efficiency.

6. Conclusions

In the context of advancing the construction of a modern socialist state in China, deepening the reform of the ecological civilization system has emerged as a critical step, with the development of the digital economy garnering significant societal attention. As the inaugural pilot policy for the digital economy, national big data comprehensive pilot zones provide a new perspective for achieving urban green development. Using a sample of 280 Chinese prefecture-level cities from 2013 to 2022, this study adopts a quasi-natural experiment framework centered on the creation of national big data comprehensive pilot zones. Employing a double machine learning model based on random forests, the research investigates the impact and underlying mechanisms of the digital economy on urban ecological resilience. Furthermore, this study delves into heterogeneities associated with resource endowments, urban geographical locations, and environmental protection zones. The empirical results reveal the following:

• The digital economy significantly enhances urban ecological resilience. This conclusion remains robust after a series of robustness checks.
• Heterogeneity analysis indicates that urban ecological resilience varies across different resource endowments, geographical locations, and environmental protection zones. Meanwhile, the urbanization level makes the greatest contribution to the policy effectiveness of the pilot zone construction.
• Mechanism analysis demonstrates that the digital economy enhances urban ecological resilience through improvements in urban innovation levels and energy efficiency.

These findings provide important policy implications for the impact of the development of the world’s digital economy on the ecological resilience of cities:

• Promote the construction of national comprehensive pilot zones for big data and stimulate the development momentum of the digital economy: Give full play to the pilot zone’s leading and exemplary effect and promote the pilot zone’s successful and replicable experience to other developing countries. This will share the results of the construction of the pilot zone and expand its impact on the ecological resilience of cities. Continue to promote the integration of the digital and real economies, leverage the data flow advantages of the pilot zone, strongly support innovation in the global digital economy, and promote the empowering role of the development of the digital economy in the field of international ecological sustainability.
• Accelerate technological innovation empowerment in the pilot area and enhance urban green innovation and energy efficiency: Drawing on the Chinese government’s strategic initiatives, other governments can provide institutional guarantees for green technological innovation and incentivize innovative enterprises and scientific research institutions to increase investment in green technology research and development. In addition, by leveraging the advantages of data circulation, they can optimize resource allocation, improve urban energy efficiency, and promote urban green development.
• Adapt to local conditions and leverage the advantages of pilot regions: In the case of China, for resource-based cities, cities in the western region, and eco-friendly cities, the guiding role of pilot region construction should be deepened, and the application potential and contribution of pilot regions in enhancing urban ecological resilience should be enhanced. Globally, governments can conduct case studies on regional sustainable construction based on their resource endowments and economic growth, focusing on enhancing urban ecological resilience, summarizing and sharing successful experiences, and providing a reference for global sustainable development.