The intensification of global climate change has exacerbated ecological and environmental challenges in urban areas (Tang et al., 2023; Zor, 2023). In China, a densely populated nation with highly concentrated economic activities, the impact of climate change on cities is particularly pronounced. Urban areas not only host large populations and economic activities, but are also vulnerable to disruptions in infrastructure, energy supply, and transportation networks caused by climate-related events. To address these challenges, the National Development and Reform Commission of China issued the PCRC in 2017, ³ which aims to enhance cities' capacity to adapt to climate change and ensure sustainable development. The PCRC designated 28 regions, including Hohhot in Inner Mongolia and Dalian in Liaoning Province, as pilot cities for climate adaptation. The distribution of the pilot cities is exhibited in Figure 1, which includes both provincial capitals and non-capital cities across eastern, central, and western China (Note: Fewer than 28 pilot regions are represented in Figure 1, as some of the pilot areas are counties, which fall outside the scope of this study). The PCRC represents not only a direct response to the growing significant impacts of climate change, but also a strategic initiative by the Chinese government to meet its climate commitments within the framework of international climate governance and advance its ecological civilization agenda (Huang and Yi, 2024; Wang and Chen, 2024).

Given the profound impact of climate change on urban development in China, there is a pressing need for strategic, top-down planning at the national level. The PCRC outlines a series of targeted actions aimed at enhancing cities’ ability to cope with climate change. These include strengthening climate monitoring and early warning systems, as well as improving meteorological disaster prediction and response mechanisms. In particular, the PCRC focuses on optimizing the layout of urban infrastructure to better respond to extreme weather events, such as heavy rainfall, heatwaves, and droughts. Additionally, the PCRC emphasizes the restoration and protection of urban ecosystems by increasing the presence of natural elements such as green spaces, forests, and lakes. This approach aims to enhance urban functions related to water conservation, temperature regulation, and soil retention, while mitigating urban heat island effects and flooding risks. Moreover, the PCRC encourages pilot cities to serve as policy experimentation hubs by promoting institutional innovation and public-private partnership models to attract private investment in climate adaptation projects. Such initiatives not only provide policy support and financial backing for pilot cities, but also offer valuable lessons and models for other cities to follow.

The PCRC provides a forward-looking framework for sustainable urban development through systematic policy planning and scientific implementation. First, the PCRC strengthens cities’ capacity to adapt to climate change by guiding policy formulation, allocating resources, and fostering technological innovation. Pilot cities are prioritized for the adoption of proactive adaptation measures, such as enhancing urban infrastructure, improving environmental monitoring systems, and promoting green building practices (Feng et al., 2024; Haarstad et al., 2024). These initiatives not only bolster cities’ resilience to extreme weather events, but also improve the overall urban ecological environment, thereby advancing urban ecological resilience (Liu et al., 2022; Wang and Chen, 2024). Second, the PCRC promotes a multi-level governance structure and cross-sectoral collaboration. Urban climate adaptation depends not only on technological and engineering solutions, but also on integrated governance and multi-stakeholder participation (Iturriza et al., 2020; Liao and Zhang, 2024). Pilot projects often involve extensive collaboration among government agencies, businesses, non-governmental organizations, and local communities. This multi-stakeholder model facilitates the integration of systematic adaptation strategies into urban planning, construction, and management, optimizing resource allocation and improving governance effectiveness (Huang and Yi, 2024). Furthermore, this collaborative mechanism strengthens societal capacity to address climate change challenges, contributing to the enhancement of ecological resilience.

Hypothesis 1

The PCRC can significantly improve urban ecological resilience.

Traditional urban development models have heavily relied on external resources, including energy, water, and construction materials, rendering cities highly susceptible to supply chain disruptions. The PCRC seeks to address this vulnerability by promoting the development of green infrastructure, energy-efficient buildings, and the widespread adoption of renewable energy sources. For example, pilot cities are encouraged to enhance localized energy production, such as solar and wind energy, thereby decreasing dependence on external energy supplies (Shi and Fan, 2024; Zheng et al., 2023). These strategies not only mitigate cities’ dependence on external resources, but also bolster urban capacity to respond to resource shortages or sudden climate events, ultimately strengthening ecological resilience (Cui et al., 2024). Furthermore, the PCRC emphasizes the importance of localized resource recycling and waste management (Liu et al., 2023). By fostering more autonomous and sustainable resource management systems, pilot cities can minimize carbon emissions associated with resource transportation and enhance their adaptive capacity in the face of resource shortages (Jiang et al., 2024; Zhao et al., 2022). Thus, by decreasing reliance on external resources, the PCRC enhances cities’ ability to cope with climate change and increases overall ecological resilience. Therefore, we propose Hypothesis 2:

Hypothesis 2

The PCRC can increase urban ecological resilience by reducing resource dependence.

The PCRC actively cultivates an environment conducive to green innovation by optimizing infrastructure, enhancing technical support, and improving legal and market frameworks. For example, pilot cities may prioritize the development of smart grids, green buildings, and clean energy infrastructure, providing the necessary physical support for the application and dissemination of green technologies. Additionally, the pilot policies promote the establishment of intellectual property protection mechanisms and green technology standards, ensuring that innovations are adequately protected and can be widely adopted (Wen et al., 2024). These measures foster a robust ecosystem favorable to green innovation, enabling innovators to conduct research and development in a more stable and supportive environment. Simultaneously, the PCRC stimulates market demand for green products and technologies through policy incentives and demonstration projects. This growing demand not only drives the further development and application of green technologies, but also significantly enhances the competitiveness of firms in the green innovation sector (Huang and Yi, 2024). As demand for low-energy buildings, water-saving technologies, and ecological restoration solutions continues to rise, businesses and research institutions are increasing their investments in these areas, thereby accelerating the pace of green innovation. By leveraging the catalytic effect of market demand, pilot cities have significantly strengthened their capacity for green innovation, providing a solid technological foundation for enhancing urban ecological resilience. So, we propose:

Hypothesis 3

The PCRC can improve urban ecological resilience by increasing green innovation capacity.

As shown in Figure 2. The PCRC enhances urban ecological resilience by addressing both the systemic vulnerabilities of cities and their adaptive capacities in the face of climate change. The PCRC enables the implementation of proactive measures, including the optimization of urban infrastructure, establishment of robust environmental monitoring systems, and promotion of green building practices. These interventions reduce cities’ exposure and sensitivity to climate-related risks, such as extreme weather events, while simultaneously enhancing their capacity to absorb, recover, and adapt. Moreover, the PCRC fosters a multi-stakeholder governance framework that integrates public institutions, private enterprises, and local communities, ensuring the efficient allocation of resources and the adoption of holistic adaptation strategies (Jiao et al., 2023). This collaborative approach strengthens the overall resilience of urban systems by embedding climate adaptation measures into urban planning, management, and daily operations, thereby safeguarding ecological functionality and long-term sustainability (Wang and Chen, 2024).

The PCRC plays a crucial role in enhancing ecological resilience by reducing resource dependence and promoting green innovation. By focusing on sustainable resource management, the PCRC helps cities shift away from over-reliance on finite or environmentally harmful resources. Such a transition is particularly important in the context of climate change, as it alleviates pressure on ecosystems and strengthens their ability to recover from shocks. Policies that promote renewable energy, sustainable land use, and efficient waste management reduce the dependency on non-renewable resources and reduce environmental degradation. Furthermore, the PCRC fosters green innovation, which includes the development and application of environmentally friendly technologies. This can involve advances in energy efficiency, water conservation, and sustainable urban infrastructure. Green innovation enables cities to enhance their resilience by facilitating adaptive measures to cope with climate impacts. In this way, the PCRC improves ecological resilience not only through resource efficiency, but also by creating a conducive environment for sustainable technological and organizational innovation that supports long-term environmental sustainability.

The impact of climate-adaptive urban pilot policies on ecological resilience is heterogeneous, influenced by several key factors such as resource base classification, ecological vulnerability, and locational heterogeneity. First, the classification according to resource base distinguishes between regions with abundant versus scarce natural resources, which in turn affects the extent to which these policies can be effective. In resource-based cities, the urgency to minimize resource consumption may drive more immediate adoption of efficient practices and green technologies. Second, ecological fragility plays a critical role in determining the impact of the PCRC. Ecologically fragile areas, which are more susceptible to environmental degradation, benefit more significantly from climate adaptation measures, as these areas often face higher vulnerability to climate-related risks. In contrast, non-ecologically fragile areas may experience less immediate need for such measures. Lastly, locational heterogeneity, including geographical and socio-economic differences, influences how these policies are implemented and their outcomes. Urban areas in diverse locations, whether due to climate variability or socio-economic conditions, will experience differing effects based on their local adaptation capacities, institutional frameworks, and public engagement.

In 2007, China’s State Council issued the National Program for Addressing Climate Change, ⁴ marking a turning point in the government’s increased focus on climate governance. Given that city-level data for some variables is only available up to 2021, we have selected the study period from 2008 to 2021. After excluding cities with severe data deficiencies, we construct a balanced panel dataset covering 284 cities. Data on urban green patent applications are obtained from the CNRDS database, while other variables are sourced from the China City Statistical Yearbook, the EPS database. All continuous variables are winsorized at the 1% level.

The application of the DML method offers significant advantages in evaluating the effect of the PCRC on ecological resilience. First, ecological resilience is influenced by a multitude of complex, interrelated economic, social, and environmental factors, which often exhibit highly nonlinear relationships (Hansen and Siggaard, 2024; Liu et al., 2024). Traditional causal inference models, such as the DID approach, while widely used in policy evaluation, face several challenges when dealing with high-dimensional data. For instance, the DID model requires the parallel trends assumption, which is difficult to satisfy in practice, especially when evaluating complex environmental policies. In contrast, the DML employs regularization algorithms to automatically select relevant control variables from high-dimensional data, effectively mitigating the “curse of dimensionality” and enhancing estimation precision and reliability (Gao et al., 2024; Wang et al., 2024a). This method not only overcomes the limitations of traditional models in handling complex datasets, but also accurately captures the true effects of policy interventions without relying on strict linear assumptions.

The DML approach also excels in addressing nonlinear relationships, which is crucial when evaluating the effect of the PCRC on ecological resilience. The relationship between ecological resilience and policy interventions is likely nonlinear, a complexity that traditional linear regression models may struggle to capture, potentially leading to unreliable estimates. The DML is capable of more accurately estimating causal effects in the presence of high-dimensional data and nonlinear relationships (Liu et al., 2021). Specifically, the DML utilizes cross-fitting and orthogonalization techniques, which reduce regularization bias and overfitting issues commonly encountered in machine learning methods, thereby enhancing the robustness and accuracy of estimates (Zhao et al., 2024). The DML not only better handles the complexities of data processing, but also provides policymakers with a more scientific and reliable foundation for decision-making.

In view of this, referring to Chernozhukov et al. (2017), we design a partially linear DML model to evaluate the effect of the PCRC: Y i t + 1 = θ 0 D i d i t + g X i t + ε i t (1) E U it | D i d i t , X i t = 0 (2) among them, i represents cities, t represents years; Y _it+1 is the dependent variable; Did _it is the treatment variable, indicating whether city i is selected for the PCRC in year t (with a value of 1 if selected, and 0 otherwise); θ 0 is the disposition coefficient, which is the policy effect; X _it is a collection of high-dimensional control variables; ε i t is error term. We estimate Equation 1 and Equation 2, which leads to the following estimate of the disposition coefficient. θ ^ 0 = 1 n ∑ i ϵ I , t ϵ T D i d i t 2 − 1 1 n ∑ i ϵ I , t ϵ T D i d i t Y i t + 1 − g ^ X i t (3) In Equation 3, where n is the sample size.

The estimation bias is then checked. n θ ^ 0 − θ 0 = 1 n ∑ i ϵ I , t ϵ T D i d i t 2 − 1 1 n ∑ i ϵ I , t ϵ T D i d i t U i t + 1 n ∑ i ϵ I , t ϵ T D i d i t 2 − 1 1 n ∑ i ϵ I , t ϵ T D i d i t g X i t − g ^ X i t (4) In Equation 4, a = 1 n ∑ i ϵ I , t ϵ T D i d i t 2 − 1 1 n ∑ i ϵ I , t ϵ T D i d i t U i t follows the normal distribution with a mean of 0; b = 1 n ∑ i ϵ I , t ϵ T D i d i t 2 − 1 1 n ∑ i ϵ I , t ϵ T D i d i t g X i t − g ^ X i t , b is the canonical bias term, which is usually divergent.

Auxiliary regressions are as follows: D i d i t = m X i t + V i t (5) P V i t | X i t = 0 (6) In Equation 5, m X i t is the regression function of the disposition variable on the high-dimensional control variables, again requiring the machine learning algorithm to estimate its specific form m ^ X i t . In Equation 5, V i t is the error term with a conditional mean of 0.

In Equation 7, first, the auxiliary regression m ^ X i t is estimated using a machine learning algorithm to take its residuals V i t ^ = D i d i t - m ^ X i t ; second, the same machine learning algorithm is used to estimate g ^ X i t , changing the main regression form to Y i t + 1 − g ^ X i t = θ 0 D i d i t + ε i t ; finally, V i t ^ is an instrumental variable for D i d i t . θ ˇ 0 = 1 n ∑ i ϵ I , t ϵ T V ^ i t D i d i t − 1 1 n ∑ i ϵ I , t ϵ T V ^ i t Y i t + 1 − g ^ X i t (7)

As showed in Table 2. The mean of ecological resilience (Er) is 0.466, with a minimum of 0.309 and a maximum of 0.705, indicating substantial variation in ecological resilience levels across cities. The average resource dependence (Re) is 0.120, suggesting that, on average, the ratio of employees in the extractive industry relative to the secondary industry workforce is approximately 0.12%. Additionally, the wide range between the minimum and maximum values of green patent applications (Gi) highlights significant disparities in green innovation capacity across cities.

We create spatial distribution maps of ecological resilience, to provide a visual representation of its spatiotemporal variation. Figure 5 presents the results for the years 2008, 2012, 2017, and 2021. The maps reveal that areas with higher ecological resilience are primarily concentrated in eastern China, particularly in regions such as the Shandong Peninsula, Jiangsu Province, and Zhejiang Province. These areas exhibit higher resilience levels due to their strong economic foundations and robust environmental governance capacities. In contrast, regions with lower ecological resilience are predominantly located in northwestern and southwestern China, where limited economic development and challenging natural conditions constrain resilience levels. Notably, in 2008, both Guangxi and Guangdong provinces had relatively low levels of ecological resilience. However, by 2021, their ecological resilience had improved significantly.

The DML model is employed in this study. Specifically, we utilize a random forest algorithm for prediction, with a sample split ratio of 1:4. As exhibited in Table 3. Column (1) controls for the linear terms of other urban control variables. The result indicates that the pilot program’s coefficient on urban ecological resilience is positive (Huang and Yi, 2024; Wang and Chen, 2024). Building on this, column (2) further includes the quadratic terms of urban control variables. The coefficient remains significantly positive, with little difference from the results of column (1). Thus, Hypothesis 1 is confirmed. The paper also explores the impact of the PCRC on the ecological resilience sub-dimension. The dependent variables in columns (3)–(5) are condition resilience, pressure resilience and response resilience, respectively. The results show that the PCRC significantly improves condition resilience and pressure resilience, especially condition resilience, in the pilot cities, while the effect on response resilience is insignificant.

The baseline regression results confirm the effectiveness of the PCRC. To further validate this conclusion, we conducted a series of robustness checks. First, we consider the unique administrative status of the four directly governed municipalities, namely, Beijing, Tianjin, Shanghai, and Chongqing, by excluding these cities from the analysis and re-estimating the model. The result is presented in column (1) of Table 4, remain consistent with our initial findings. Secondly, as exhibited in column (4), following Chen et al. (2018), we incorporate the interaction fixed effects of province and year to account for regional temporal features. Third, during the period 2008–2021, the government implemented a range of environmental regulatory measures, and these environmental governance tools are likely to affect ecological resilience. In view of this, we include policies such as pilot low-carbon cities, pilot carbon emissions trading, air pollution prevention and control action plans, pilot zones for green financial reform and innovation, and pilot smart cities as control variables in the model. The result is presented in column (3).

Given the potential subjectivity in model specification, we next perform robustness checks by adjusting the model design. We change the initial sample split ratio from 1:4 to 1:2, with the results shown in column (4). This adjustment has minimal impact on the baseline conclusions. In column (5), we then replace the random forest algorithm with a gradient boosting algorithm. Lastly, while the baseline regression employs a partially linear model based on the DML, we substitute this with a more general interactive model to examine the sensitivity of our findings to model specification. The results, presented in column (6), remain consistent with the baseline conclusions.

Compared to the traditional three-step approach, the two-step method offers greater advantages in addressing endogeneity issues (Baron, 2022). Therefore, this section employs the two-step method to examine the mechanisms through which the PCRC enhances urban ecological resilience. ⁵