Climate-Resilient City Pilot Programs and New-Quality Productivity: Causal Identification Based on Dual Machine Learning

Abstract

Climate change is a critical constraint on the development of new-quality productive forces (NQPFs), making it essential to clarify its relationship with urban development strategies to enhance productivity. Using panel data from 284 Chinese cities during 2010–2022, this study leverages the climate-resilient city pilot policy as a quasi-natural experiment and applies a double machine learning approach to estimate both the causal impact and underlying mechanisms of this policy on NQPFs. We further examine heterogeneous effects across geographic regions and city types. Our findings show that first, climate-resilient urban development significantly boosts NQPFs, with results remaining robust across multiple sensitivity tests. Second, this effect operates through three key channels—talent agglomeration, data flow enhancement, and infrastructure-related industrial upgrading. Third, the policy’s impact is stronger in western and coastal cities; resource-based cities and non-environmentally protected cities exhibit greater responsiveness, amplifying the positive outcomes. This study provides systematic empirical evidence on the nexus between climate resilience and high-quality development, offering actionable insights for designing localized strategies to advance climate-resilient urbanization and foster high-quality productive forces.

1. Introduction

Past industrial revolutions have consistently targeted “nature” as the primary object of transformation. However, this approach poses a risk of disrupting the normal functioning of ecosystems. The environmental issues arising from this “pulling up seedlings to help them grow” style of development ultimately threaten the intrinsic stability between humanity and the ecological environment, thereby jeopardizing long-term human progress. Traditional production methods primarily relied on factors such as land, capital, and labor to drive productivity growth [1]. Nevertheless, this extensive approach often resulted in low resource utilization efficiency and significant environmental pollution, hindering the construction of beautiful cities and high-quality economic development. In light of a new round of technological revolution and industrial transformation, the development of NQPFs has become imperative. The Third Plenary Session of the 20th CPC Central Committee emphasized that developing NQPFs must be grounded in reality and tailored to local conditions; it is positioned as a powerful driver for advancing national high-quality economic development. In this era characterized by digital intelligence, continuous iteration and innovation in science and technology—coupled with leaps in productivity—have created favorable conditions for realizing NQPFs. NQPFs are driven by novel labor forces, means of labor, and objects of labor; they represent a crucial safeguard for China to surpass its competitors in international competition [2].

However, within the framework of modern urban development, cities—as areas of high population and economic activity concentration—account for three-quarters of global energy consumption and greenhouse gas emissions. They are the primary culprits behind environmental degradation issues like climate change, significantly impacting normal production and operations worldwide [3,4]. As one of the main drivers affecting urban economic growth and social development, the consequences of climate change cannot be overlooked [5,6]. Faced with severe challenges from global climate change, such as frequent extreme weather events and rising sea levels, enhancing urban adaptability and resilience is no longer optional but an essential path to sustainable development. Against this backdrop, the European Union has established the European Carbon Market as part of its climate governance, significantly reducing greenhouse gas emissions [7]. The United States has enacted three major pieces of legislation—the Infrastructure Investment and Jobs Act (IIJA), the Inflation Reduction Act (IRA), and the CHIPS and Science Act—aimed at accelerating the energy transition. Full implementation of these policies is projected to drive U.S. greenhouse gas emissions down by over 40% from 2005 levels by 2030, accelerating the energy transition [8]. China’s climate change response has evolved over more than a decade, gradually forming a systematic framework spanning top-level design to local implementation, from pilot explorations to comprehensive deepening. In 2013, the State released the National Climate Change Adaptation Strategy, categorizing key national regions into three adaptation zones—urbanization, agricultural development, and ecological security—and deploying a series of safeguarding measures, marking the full launch of climate adaptation efforts at the national level. The 2024 Progress Report on China’s Climate Change Adaptation systematically summarizes policy outcomes, revealing the formation of a multi-dimensional adaptation network covering meteorological observation, disaster prevention, ecological restoration, and urban resilience. Notable achievements include collaborative disaster prevention in the Guangdong–Hong Kong–Macao Greater Bay Area and mangrove ecological restoration projects, signaling the advancement of climate adaptation into a high-quality development phase characterized by comprehensive implementation and targeted breakthroughs. Overall, the climate-resilient city policy represents a forward-looking, comprehensive urban governance system rather than passive environmental regulation. Its evolution reflects China’s shift in governance logic—from reactive responses to proactive adaptation, and from isolated measures to systemic integration. The policy’s core objective is to enhance the operational stability of cities under internal and external climate risks through a series of systemic measures, including strengthening infrastructure resilience and protecting ecosystems, thereby safeguarding urban productivity development.

The relationship between climate-resilient urban development and the advancement of new productive forces manifests primarily through concrete policy measures but more profoundly through deep strategic alignment and positive synergy. At the foundational level, the physical environment created by climate-resilient cities enhances the stability of various urban economic activities. These predictable operational conditions mitigate the negative impacts of climate change on emerging technologies, providing essential ground for technology R&D that requires long-term investment. They particularly create demonstration scenarios for cultivating and applying cutting-edge technologies like green low-carbon solutions and digital intelligence [9]. Simultaneously, the modern infrastructure networks established in climate-resilient cities—including alternative energy systems, smart transportation hubs, and digital climate governance platforms—function as advanced channels for the flow of key elements essential to the development of new-quality productive forces. These infrastructures not only accelerate the cross-regional circulation and transformation of knowledge, technology, and data but also promote the optimized allocation of resources within regional boundaries [10]. More importantly, the climate-resilient policy framework fosters a development environment focused on long-term value. By establishing comprehensive climate risk monitoring and early warning mechanisms, it bolsters market confidence in technological innovation investments and channels capital toward strategically significant emerging sectors [11].

Given China’s ongoing expansion and deepening of climate-resilient city pilot initiatives, it is imperative to deconstruct the system of NQPFs, measure their levels across cities under climate change pressures, and assess the impact of climate-resilient city policies on these forces. This represents a critical research priority.

Compared with existing literature, this paper’s potential contributions are as follows: First, it explores the role of climate adaptation pilot policies in fostering NQPFs within cities from a novel perspective—the impact of climate change on urban development. By treating these policies as exogenous shocks, it effectively mitigates endogeneity issues and expands research on NQPFs. Second, it employs a dual machine learning model to overcome the “curse of dimensionality” and model specification errors inherent in traditional causal inference methods, yielding an approximately unbiased estimator of the policy effects of climate-resilient cities and enhancing the reliability of research findings. Third, this study systematizes and refines existing measurement indicators for NQPFs. It examines the channels through which climate-resilient cities influence NQPFs across three dimensions—talent aggregation, data circulation, and infrastructure-driven industrial upgrading—providing theoretical support for advancing urban NQPFs development.

2. Literature Review and Theoretical Analysis

2.1. Literature Review

We can approach accurately understanding the concept of NQPFs by examining its attributes and technological elements. From the perspective of attributes, NQPFs emphasize two key aspects: “novelty” and “quality”. The term ‘novelty’ denotes a production approach that is distinctly different from traditional methods, which are often characterized by high energy consumption and low utilization rates. In contrast, “quality” indicates that NQPFs must prioritize disruptive technological innovation and breakthroughs in essential core technologies to continuously inject momentum into economic development [12]. From the standpoint of technological elements, high-quality technical talent—considered as new-quality labor—integrates into the system of NQPFs, thereby forming its developmental foundation [13]. The rapid and profound mining of data, combined with proactive deployment of emerging technologies such as artificial intelligence, quantum information technology, and humanoid robotics, has facilitated the creation of industrial chains and clusters exhibiting novel dynamics [14]. In summary, NQPFs—underpinned by innovative technologies, perspectives, and models—are aligned with the current national demand for high-quality economic development. They represent a crucial means to address the limitations inherent in traditional production methods while serving as a central engine for reshaping industrial competitive advantages and pioneering new pathways for development.

In this context, measuring and controlling these NQPFs has emerged as a critical issue. Within the framework of the productive forces system, Zhu et al. classify NQPFs into two primary categories: the first encompasses substantive elements, which include new-quality laborers, new-quality means of labor, and new-quality objects of labor; the second consists of pervasive elements that comprise new technologies, innovative production organization, and data components [15]. Gang and Zhao developed an evaluation index system that encompasses both technological productivity and green productivity, reflecting the dual dimensions of environmental sustainability and technological advancement within the context of NQPFs [16]. Xu et al. drew upon dynamic capability theory and evolutionary economics to construct a measurement framework comprising four dimensions: innovation-driven technological progress, green and low-carbon technologies, digital empowerment, and labor resource advantages [17]. Although measurement approaches have diversified, the core principle—ensuring technological innovation while emphasizing green production—remains consistent throughout the development of NQPFs. This principle aims to achieve two overarching objectives through policy guidance: technological innovation and ecological friendliness.

The development of new high-quality productive forces fundamentally relies on the guidance and support of relevant policies. From the perspective of positive policy effects, existing literature primarily focuses on technological innovation and green energy conservation in its research on new quality productive forces. In the realm of technological innovation, Yang and Zhang used the “Broadband China” policy as an exogenous shock to study the impact of digital infrastructure on new high-quality productive forces, finding it has a significant positive effect on productivity [18]. Wang et al. found that digital finance policies integrating new technologies like artificial intelligence, big data, blockchain, and cloud computing significantly promote regional development of new high-quality productive forces [19]. In environmental sustainability, Zhang et al. (2024) examined the impact of new energy policies on developing countries’ new high-quality productive forces, revealing a significant short-term positive effect [20]. Li et al. (2025) demonstrated that policies in green finance reform pilot zones substantially elevate regional new productive capacity by promoting energy conservation and emission reduction while restricting financing for polluting enterprises [21]. Li and Liu (2025) employed three dimensions of labor factors as indicators of new productive capacity, revealing that low-carbon city pilot policies effectively fostered corporate development of new productive capacity while significantly reducing carbon emissions [22]. From the perspective of policy instability, Ugur Korkut Pata (2024) found that climate policy uncertainty has led to increased carbon emissions in China and the United States, hindering the achievement of economic and ecological sustainability goals [23]. Hugo Morão’s (2025) research reveals that climate policy uncertainty has significant and heterogeneous impacts on Portugal’s energy sector, with varying effects across dimensions such as sales, prices, and labor, and these impacts exhibit time-varying characteristics [24]. Cem Işık and Serdar Ongan (2025) found that climate policy uncertainty delays the pace at which governments implement various emission reduction measures and affects business investment activities [25].

Current research primarily focuses on the impact of digitalization and green policies on the development of NQPFs, as well as the effects of climate policies on economic growth. However, as cities serve as vital carriers for developing NQPFs and key units for combating climate change, the intrinsic connection between their climate adaptation efforts and productivity enhancement has yet to be systematically explored. As China increasingly prioritizes climate governance under its dual carbon goals, deepening our understanding of how climate-resilient urban policies drive NQPFs will not only advance our knowledge of synergistic pathways between climate adaptation and economic growth but also provide crucial policy guidance for leveraging climate-resilient urban development to propel high-quality growth.

2.2. Theoretical Analysis

2.2.1. Climate-Adaptive Urban Policies and New Quality Productivity

The Pilot Policy for Climate-Adaptive Urban Development centers on systematically enhancing urban resilience and adaptive capacity. It aims to establish a sustainable support environment for cultivating and developing new productive forces, achieving a strategic shift from “risk mitigation” to “development empowerment.” This policy focuses on strengthening the climate responsiveness of urban physical infrastructure and ecosystems. By advancing disaster prevention information systems, green infrastructure, and intelligent monitoring systems, it reduces the impact of extreme weather events on urban economic systems and minimizes “disturbance costs,” thereby ensuring the continuity and stability of business operations [26]. This process not only optimizes the long-term investment environment but, more importantly, empowers new productive forces at a deeper level. On one hand, a highly stable urban operational foundation provides reliable testing grounds and application scenarios for critical technological innovations, facilitating the integration and breakthroughs of emerging technologies such as green technologies, information technologies, and manufacturing technologies [27]. On the other hand, secure and efficient energy networks, transportation systems, and digital information and communication facilities significantly enhance the allocation efficiency of advanced factors like data, talent, and capital, driving the intelligent and green transformation of industries [28]. Furthermore, the risk prevention and control system established by policies has reduced uncertainties in financial transactions, bolstering market entities’ confidence in investing in long-term innovative technologies [29]. Thus, climate-adaptive city pilot programs are not merely defensive strategies against climate risks. They serve as a lever for resilience-building, systematically unlocking the value chain linking “infrastructure resilience—economic stability—productivity leap.” This provides robust support for the continuous evolution of new productive forces within a secure, open, and efficient urban environment. Based on the above, this study proposes Hypothesis 1:

H1. 

Climate-resilient city pilot policies can elevate the level of NQPFs in cities.

2.2.2. Talent Aggregation Effect

High-caliber talent, as the core vehicle for innovation activities, is a key element in advancing the development of new-quality productive forces and holds strategic significance for seizing the high ground in future development [30]. The continuous advancement of new-quality productive forces fundamentally relies on the full release of the innovative potential of high-quality talent pools. Climate-adaptive urban development creates favorable conditions for attracting and retaining talent by systematically enhancing urban livability and ecological quality. Specifically, policy-driven initiatives such as green space development, wetland ecological enhancement, and ecosystem restoration not only optimize urban spatial structures but also shape safe, healthy, and livable environments. This organic integration of ecological advantages and climate resilience significantly strengthens cities’ capacity to address climate change, manifesting in more reliable public services and superior ecological spaces. Moreover, the high-quality living environment and robust climate adaptation capabilities effectively meet talent’s pursuit of a high-quality life, fostering a synergistic development pattern between talent and the city [31,32]. When cities can provide resilient living environments, they not only help stabilize the local talent pool but also enhance their appeal to external top talent. This, in turn, lays a solid foundation for the sustained development of new productive forces, drives the aggregation and upgrading of innovation factors, and achieves a sustainable development path from environmental improvement to innovation-driven growth. Based on the above, this study proposes Hypothesis 2:

H2. 

Climate-adaptive city pilot policies attract talent aggregation by enhancing the livability of the urban ecological environment, thereby influencing the level of new-quality productive forces in cities.

2.2.3. Data Element Circulation Effects

As a growing factor of production, the importance of data elements in advancing NQPFs is becoming clearer. At the policy level, the government explicitly requires pilot cities to establish more comprehensive climate risk monitoring networks, data collection systems, and information-sharing platforms. This approach itself promotes the breaking down of data silos and facilitates information flow among different departments within cities, enhancing the digitalization and intelligence of urban governance. This enhanced government data governance and risk warning capability enables more precise climate risk information services for enterprises, optimizing their site selection, R&D, and supply chain management. Consequently, it indirectly supports the development of new productive forces. Climate data trading platforms and strengthened information infrastructure are being established to drive innovation in data circulation mechanisms through standardized systems. Digital platforms, as novel production tools, can effectively increase the digital coverage of urban public services, alleviate temporal and spatial constraints on climate warning information, and ensure smoother city operations and stable production by implementing efficient organizational structures and labor practices [33]. Within this concept, the development and refinement of digital platforms maximizes resource allocation efficiency through factor restructuring while lowering the marginal cost of data usage via economies of scale. This method enables the transformation and improvement of traditional aspects such as capital and labor. In this regard, the market for data elements enables the creation of new, high-quality productive forces via two key channels: “data trading platform activity” and “depth of data element utilization [34].” Based on these considerations, this study proposes Hypothesis 3:

H3. 

Pilot policies for climate-adaptive cities enhance NQPFs by improving the circulation of data elements.

2.2.4. Infrastructure Industry Upgrading Effects

Extreme climate events represent one of the disruptive factors affecting the stable operation and normal development of cities. As a “defensive wall,” the speed at which a city’s infrastructure systems recover effectively from natural and man-made disasters is particularly crucial [35]. The role of climate-resilient urban development in promoting infrastructure construction fundamentally reflects a policy-driven, demand-pull upgrade pathway. When cities build adaptive infrastructure such as smart flood monitoring systems and sponge cities to counter extreme climate impacts, this process does not directly intervene in specific industrial development. Instead, it stimulates market demand for high-performance building materials, smart monitoring equipment, and advanced technologies by establishing higher technical standards and construction requirements. As a vital component of new-type means of production and a prerequisite for developing new-type productive forces, the installation density of new-type intelligent infrastructure reaches a critical threshold. At this point, its scaled deployment not only enhances the climate resilience of urban systems but also fosters technological convergence and industrial synergy. This creates a modern industrial foundation supporting the development of new-type productive forces, ultimately achieving a virtuous cycle from infrastructure upgrading to comprehensive industrial structure optimization [36].

H4. 

Climate-adaptive city pilot policies influence the level of NQPFs in cities by upgrading the infrastructure industry.

The specific path mechanism diagram is shown in Figure 1.

3. Methodology

3.1. Identification Strategies

Given the close interconnection between urban new-quality productive forces and multiple dimensions, they are susceptible to various factors such as urban infrastructure development and production environments. Therefore, other relevant influencing factors should be controlled as much as possible. However, traditional causal inference methods exhibit significant limitations in such complex scenarios: classical regression models struggle to handle high-dimensional covariates and nonlinear relationships; the difference-in-differences approach relies heavily on the parallel trends assumption; and propensity score matching methods impose specific requirements on the distribution of the treatment group. In contrast, dual machine learning integrates modern machine learning techniques. It employs regularization algorithms for automatic variable selection and utilizes nonparametric modeling to capture complex relationships between variables, thereby overcoming the constraints inherent in traditional methods. The dual machine learning model offers three methodological advantages: First, it flexibly selects the optimal algorithm combination through cross-validation and ensemble learning, avoiding the bias inherent in single-model specifications. Second, it effectively controls multidimensional confounding factors by separating policy effects through sample partitioning and orthogonalization. Third, it incorporates hyperparameter tuning mechanisms like Bayesian optimization, enhancing predictive accuracy while ensuring model generalization capabilities [37,38].

Based on this, we first install necessary libraries such as scikit-learn, scipy, and numpy to configure the Python 3.9 environment. Subsequently, we install relevant machine learning packages like ddml in Stata 16 and locate the Python path within Stata to execute the program.

Partially linear dual machine learning models are as follows:

$$Y_{it}={\theta }_{o}Even{t}_{it}+g(X_{it})+U_{it}$$

(1)

$$E(U_{it}|Even{t}_{it},X_{it})=0$$

(2)

Cities and years are denoted by a and t, respectively. Y_it represents the dependent variable—new quality productive forces. Event_it denotes the policy variable for “climate-adaptive cities”, taking the value 1 if a city is designated as a pilot and 0 otherwise. X_it is a set of high-dimensional control variables that influence NQPFs; these variables affect the dependent variable Y_it through the function $g(X_{it})$. The functional form of $g(X_{it})$ is estimated using machine learning techniques to derive its corresponding value $\widehat{g}(X_{it})$. The model also includes a random error term U_it, which has a conditional mean of 0. Based on the direct regression estimation of Equations (1) and (2), the following results can be derived:

$${\stackrel{\wedge }{\theta }}_o={({\displaystyle \frac{1}{n}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}^2})}^{-1}{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}(Y_{it}-\stackrel{\wedge }{g}(X_{it}))}$$

(3)

n represents the sample size. Building upon Equation (3), a further exploration of its estimation bias can be carried out:

$$\sqrt{n}({\stackrel{\wedge }{\theta }}_o-{\theta }_o)={({\displaystyle \frac{1}{n}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}^2})}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}{U}_{it}+(\frac{1}{n}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}^2})}}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}[g(X_{it})-\stackrel{\wedge }{g}}(X_{it})]$$

(4)

$$a={({\displaystyle \frac{1}{n}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}^2})}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}{U}_{it}}$$

(5)

$$b={({\displaystyle \frac{1}{n}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}^2})}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\displaystyle \sum _{i\in I,t\in T}Even{t}_{it}[g(X_{it})-\stackrel{\wedge }{g}}(X_{it})]$$

(6)

The parameter a follows a normal distribution with mean 0. Within the double machine learning framework, the regularization term introduced to estimate the function $\widehat{g}(X_{it})$ effectively controls variance but inevitably introduces estimation bias. The key issue arises from the slow convergence of $\widehat{\mathrm{g}}(X_{it})$ to $g\left(X_{it}\right)$, particularly when $n^{-{\phi }_g}>n^{-1/2}$. which leads to the bias b also tending towards infinity as the sample size n approaches infinity, ultimately preventing the treatment effect ${\widehat{\theta }}_0$ from converging to the true parameter value ${\theta }_0$. To expedite the convergence process, an orthogonalization approach is employed to correct the estimation bias, and the corresponding auxiliary regression model is formulated as follows:

$$Even{t}_{it}=m(X_{it})+V_{it}$$

(7)

$$E(V_{it}|X_{it})=0$$

(8)

The function $m\left(X_{it}\right)$ denotes the regression relationship between the treatment variable and high-dimensional control variables. Its specific form, $\widehat{m}(X_{it})$, must also be estimated using a machine learning model. The error term V_it is assumed to have a conditional mean of 0. In practical implementation, the residuals ${\widehat{V}}_{it}=Even{t}_{it}-\widehat{m}(X_{it})$ from the auxiliary regression are first computed; these residuals are then utilized as an instrumental variable for Event_it and substituted into Equation (3) to yield an unbiased estimator.

$${\stackrel{\wedge }{\theta }}_0={({\displaystyle \frac{1}{n}}{\displaystyle \sum _{i\in I,t\in T}{\stackrel{\wedge }{V}}_{it}Even{t}_{it}})}^{-1}{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i\in I,t\in T}{\stackrel{\wedge }{V}}_{it}(Y_{it}-\stackrel{\wedge }{g}(X_{it}))}$$

(9)

The estimation bias associated with Equation (9) is:

$$\sqrt{n}({\stackrel{\wedge }{\theta }}_o-{\theta }_o)={[E(V_{it}^2)]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{{\displaystyle \sum _{i\in I,t\in T}{V}_{it}{U}_{it}+[E(V_{it}^2)]}}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\displaystyle \sum _{i\in I,t\in T}[m(X_{it})-\stackrel{\wedge }{m}}(X_{it})][g(X_{it}-\stackrel{\wedge }{g}(X_{it})]$$

(10)

$$c={[E(V_{it}^2)]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\displaystyle \sum _{i\in I,t\in T}{V}_{it}{U}_{it}}$$

(11)

$$d={[E(V_{it}^2)]}^{-1}{\displaystyle \frac{1}{\sqrt{n}}}{\displaystyle \sum _{i\in I,t\in T}[m(X_{it})-\stackrel{\wedge }{m}}(X_{it})][g(X_{it}-\stackrel{\wedge }{g}(X_{it})]$$

(12)

The random disturbance term c follows a normal distribution with mean 0. Because the double machine learning procedure is applied twice, the overall convergence rate of d is determined by two factors: the rate at which the estimator $\widehat{m}(X_{it})$ converges to $m\left(X_{it}\right)$, and the rate at which the estimator $\widehat{g}(X_{it})$ converges to $g\left(X_{it}\right)$, denoted as $n^{-({\phi }_m+{\phi }_g)}$. Even if the convergence rates of these two function estimations are relatively slow, as long as the aforementioned conditions hold, the cross-product term of the error term $\sqrt{n}({\widehat{\theta }}_0-{\theta }_o)$ will converge to 0 at a faster rate. This ensures the unbiasedness of the treatment coefficient estimator ${\widehat{\theta }}_0$ and, consequently, allows for an unbiased estimation of the treatment effect even when the functional form of the covariates is unknown.

3.2. Variable Selection

3.2.1. Dependent Variable: New-Quality Productive Forces (NQPFs)

Although there is currently no unified standard for measuring NQPFs, based on their connotation and contemporary characteristics, and drawing on the research of Han et al. [39] and Zhou et al. [40], this study employs the entropy value method to construct an evaluation indicator system from three dimensions: new-quality labor force, new-quality objects of labor, and new-quality means of labor. This multidimensional measurement framework not only reflects issues in factor allocation but also reveals the synergistic evolution patterns of talent, resources, and technology, providing scientific support for the layout of NQPFs in cities. Specific indicators are shown in

Table 1. Indicator System for the Level of New Quality Productivity.

Primary Indicators	Secondary Indicators	Third-Level Indicators	Direction	Weighting
New quality labor force	The number of employees in emerging industries	The number of employees in listed companies of emerging industries	+	10.08%
	Individual employee capabilities	Average wage of on-the-job employees	+	0.98%
	High-quality staff level	The number of regular institutions of higher learning	+	5.08%
New quality labor objects	Infrastructure	Internet broadband access users	+	3.17%
		Total volume of telecommunications services	+	4.29%
	Future development	Robot installation density	+	2.47%
	Ecological environment	Investment in environmental pollution control and treatment	+	5.01%
		Carbon trading, energy consumption rights trading, and pollutant discharge rights trading	+	5.10%
		The harmless treatment rate of domestic waste	+	0.13%
New quality labor materials	Technology research and development	The proportion of scientific expenditure in local fiscal expenditure	+	2.58%
	Innovation output	The number of inventions applied for that year	+	8.87%
		The number of utility model inventions applied for that year	+	7.63%
	Intelligentization	The number of artificial intelligence enterprises	+	12.09%
	Greening	The number of green inventions applied for that year	+	9.64%
		The number of green utility model applications filed that year	+	7.98%
	Data elements	The mean of the natural logarithm of (the frequency of listed companies’ data assets + 1)	+	1.37%
		Collect data from the data exchanges of each prefecture-level city, marking 1 if there is a value and 0 if there is none.	+	13.53%

.

3.2.2. Explanatory Variable: Pilot Policies for Climate-Adaptive Cities (Weather)

Match the list of climate adaptation pilot cities released by China’s National Development and Reform Commission and Ministry of Housing and Urban-Rural Development with corresponding city data (Figure 2). Construct a climate adaptation policy dummy variable based on the designation timeline for pilot cities. The variable is set as follows: if a city is designated as a pilot city, the policy dummy variable is set to 1 for 2017 and subsequent years; otherwise, it is set to 0.

3.2.3. Control Variables

This paper draws upon relevant materials from Zhang et al. [41] and Yin and Kuang [42]. It selects a series of factors that may influence the level of NQPFs in cities as control variables for inclusion in the regression, thereby enhancing the model’s precision. Specific variables include (1) Urban Economic Density (Density), measured by the ratio of regional GDP to the land area of the administrative region; (2) Degree of Foreign Openness (FOpenness), measured by the ratio of actual foreign capital utilization to regional GDP; (3) Urbanization Rate (URate), measured by the ratio of non-agricultural population to registered population; (4) Human Capital Level (HCapital), measured as the proportion of full-time students in regular higher education institutions relative to the year-end total population; (5) Population Size (PSize), measured as the logarithm of the year-end registered population; (6) Education Expenditure (EExpenditure), measured as the proportion of education spending relative to general government fiscal expenditure; (7) Science and Technology Expenditure (TExpenditure), measured by the proportion of science and technology expenditure in general government fiscal expenditure; (8) Market Size (MSize), measured by the proportion of total retail sales of consumer goods in regional GDP; (9) Fiscal Decentralization (FDecentralization), measured by the proportion of government fiscal revenue relative to government fiscal expenditure; (10) Economic Development Level (EDevelopment), measured by the logarithm of per capita GDP; and (11) Fiscal Investment Intensity (FInvestment), measured by the proportion of fixed asset investment relative to general government fiscal expenditure.

3.3. Data Sources and Descriptive Statistics

Given the availability, completeness, and update frequency of urban data, this study utilizes a panel data sample covering 284 Chinese cities from 2010 to 2022, excluding Hong Kong, Macau, and Taiwan. All data were downloaded and compiled in December 2024. The research data primarily originates from the annual China Urban Statistical Yearbook, China Urban Construction Statistical Yearbook, and official municipal websites across various prefecture-level cities. Additionally, individual cities with excessive missing values or renamed entities were excluded, while linear interpolation was applied to address minor data gaps in remaining cases. Descriptive statistics for the variables are presented in

Table 2. Descriptive Statistics of Main Variables.

Variable	Name (Symbol)	Obs	Mean	Se (Mean)	Std	Min	Max
Explanatory variable	Pilot policies for climate-resilient cities (Weather)	3692	0.462	0.00821	0.499	0.000	1.000
Dependent variable	new quality productive forces (NQPFs)	3692	0.051	0.00123	0.075	0.002	0.618
Control Variables	Urban Economic Density (EDensity)	3692	7.261	0.02110	1.282	3.039	11.996
	Degree of Foreign Openness (FOpenness)	3692	0.002	0.00004	0.003	0.000	0.029
	Urbanization Rate (URate)	3692	0.392	0.00343	0.208	0.075	1.000
	Human Capital Level (HCapital)	3692	0.020	0.00041	0.025	0.000	0.151
	Population Size (PSize)	3692	5.891	0.01148	0.698	2.970	8.136
	Education Expenditure (EExpenditure)	3692	0.176	0.00064	0.039	0.018	0.356
	Technology Expenditure (TExpenditure)	3692	0.017	0.00028	0.018	0.001	0.207
	Market Size (MSize)	3692	0.380	0.00189	0.115	0.000	2.351
	Fiscal Decentralization (FDecentralization)	3692	0.452	0.00364	0.221	0.056	1.541
	Economic Development Level (EDevelopment)	3692	10.742	0.00966	0.587	8.618	12.456
	Fiscal Investment Intensity (FInvestment)	3692	4.397	0.04941	3.002	−22.238	17.759

.

4. Results

4.1. Benchmark Regression Results

To assess the impact of climate-resilient city pilot policies, this study employs a dual machine learning model to measure changes in urban NQPFs. Multiple regressions are conducted in the baseline regression, incorporating both first-order and second-order control variables. Second, we separately tested fixed effects for both year and city. The regression results are presented in

Table 3. Benchmark Regression Results.

Variable	(1) NQPFs	(2) NQPFs	(3) NQPFs	(4) NQPFs
Weather	0.015 ***	0.016 ***	0.014 ***	0.006 ***
	(0.003)	(0.003)	(0.003)	(0.001)
Constant	0.000	0.000	0.000	0.000
	(0.000)	(0.000)	(0.000)	(0.000)
Control	Yes	Yes	Yes	Yes
Control2		Yes	Yes	Yes
Year FE	Yes	Yes	Yes
City FE	Yes	Yes		Yes
Obs.	3692	3692	3692	3692

Note: t statistics in parentheses, *** p < 0.01.

. Findings indicate that regardless of whether fixed effects are included, the regression coefficients for climate-adaptive pilot cities remain statistically significant at the 1% level. This confirms that climate-adaptive pilot cities contribute to enhancing urban NQPFs levels.

4.2. Robustness Test

4.2.1. Adjust the Research Sample

Considering that the first batch of pilot cities commenced in 2017, this study narrows the sample timeframe to 2013–2021 to ensure consistent pre- and post-policy implementation periods, thereby enabling more precise examination of policy effects. As shown in the first column of

Table 4. Robustness Test (1).

Variable	Adjusted Sample	Delete Outlier		Interaction	Other Policy Implication
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Weather	0.008 ***	0.010 ***	0.018 ***	0.022 ***	0.015 ***	0.012 ***	0.013 ***
	(0.002)	(0.002)	(0.004)	(0.008)	(0.003)	(0.002)	(0.002)
Constant	0.001 **	0.000	0.000	0.000 *	0.000	0.000	0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Control	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Control2	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	2556	3692	3692	3692	3692	3692	3692

Note: t statistics in parentheses, * p < 0.1, ** p < 0.05 *** p < 0.01.

, even after narrowing the sample, the regression coefficient for climate-adaptive city policies remains statistically significant at the 1% level.

4.2.2. Eliminate the Influence of Outliers

To mitigate the interference of extreme values in the data on the overall estimation accuracy, the control variables in the baseline regression were individually subjected to 1% and 5% trimmed tailing. Data points exceeding the highest quartile and falling below the lowest quartile were replaced, and regression analysis was conducted accordingly. The specific results are presented in Table 4 (2) and (3). The outcomes after removing outliers did not exhibit significant changes.

4.2.3. Consider the Interaction Between Provinces and Time

As the core tier within China’s administrative system, provinces play a pivotal role in the governance framework. This often leads to significant convergence among cities within the same province across dimensions such as policy orientation, geographical conditions, and cultural heritage. Therefore, this study incorporates a province–time interaction fixed effect into the baseline regression to control for the impact of provincial variations over time. Column 4 of Table 4 indicates that, even after accounting for the province–time interaction, climate-adaptive urban policies remain significantly influential on cities’ NQPFs at the 1% significance level.

4.2.4. Consider Other Concurrent Policies

Following the implementation of the “Smart City” and “Broadband China” pilot policies in 2013 and 2014, climate adaptation pilot initiatives inevitably faced interference from other policy initiatives. To ensure the accuracy of policy effect estimation, this paper incorporates policy dummy variables for “Smart City” and “Broadband China” into the regression analysis. The specific regression results are shown in Table 4 (5)–(7). The policy effects for climate-adaptive cities remain statistically significant, and the original conclusion still holds.

4.2.5. Change the Sample Split Ratio

To avoid the influence of sample division ratios on research conclusions during cross-validation, the sample division ratios were adjusted to 1:7 and 1:2. The specific regression results are presented in

Table 5. Robustness Test (2).

Variable	Cutting ratio		Change DML		Endogeny
	(1)	(2)	(3)	(4)	(5)
Weather	0.011 ***	0.014 ***	0.087 ***	0.009 ***	0.058 ***
	(0.002)	(0.004)	(0.012)	(0.000)	(0.014)
Constant	0.000	0.000 *	0.000		0.000 **
	(0.000)	(0.000)	(0.000)		(0.000)
Control	Yes	Yes	Yes	Yes	Yes
Control2	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes
Obs.	3692	3692	3692	3692	3692

Note: t statistics in parentheses, * p < 0.1, ** p < 0.05 *** p < 0.01.

(1) and (2). The regression coefficients for policy variables remained statistically significant at the 1% level, indicating that the benchmark regression results remained robust.

4.2.6. Reset the Machine Learning Model

To avoid the impact of dual machine learning model algorithm biases on the benchmark regression conclusions, the random forest algorithm previously used for prediction was replaced with gradient boosting (gradboost). Furthermore, some linear models were subsequently replaced with general interactive models to investigate the effects of different algorithms and model settings on the conclusions of this paper. The changes applied to the primary and auxiliary regression analyses are as follows:

$$Y_{it+1}=g(Even{t}_{it},X_{it})+U_{it}$$

(13)

$$Even{t}_{it}=m(X_{it})+V_{it}$$

(14)

The estimated coefficient of the treatment effect derived from the interactive model is:

$${\stackrel{\vee }{\theta }}_1=E[g(Even{t}_{it}=1,X_{it})]-g(Even{t}_{it}=0,X_{it})]$$

(15)

The results are shown in Table 5 (3) and (4). Whether the double machine learning algorithm is replaced or a general interactive model is adopted instead, the research conclusion remains robust.

4.2.7. Endogeneity Test

To mitigate endogeneity issues arising from omitted variables, this study adopts the instrumental variables approach, drawing upon the research of Zhang and Li [43], to address endogeneity concerns as effectively as possible. The interaction term between urban terrain roughness and the time trend variable is employed as an instrumental variable for climate-adaptive city pilot projects, incorporated into a partial linear instrumental variables model. The specific specification is as follows:

$$Y_{it+1}={\theta }_{0}Even{t}_{it}+g(X_{it})+U_{it}$$

(16)

$$Instrumen{t}_{it}=m(X_{it})+V_{it}$$

(17)

Specifically, Instrument_it serves as the instrumental variable for Event_it. The analytical results following the aforementioned processing are reported in Table 5, column (5), demonstrating that the robustness of the research conclusions remains unaffected.

4.3. Mechanism Test

4.3.1. Test of the Talent Agglomeration Effect

This study references the research by Yang et al. [44], selecting talent agglomeration (TA) as a proxy variable to measure the talent agglomeration effect. Specifically, it calculates the proportion of employees in scientific research, technical services, geological exploration, information transmission, computer services, and software industries relative to the total number of urban unit employees. Column (1) in

Table 6. Mechanism Verification.

Variable	(1)	(2)	(3)	(4)
	TA	Phone	ISS	UIS
Weather	0.005 **	0.135 **	0.141 ***	0.042 ***
	(0.002)	(0.054)	(0.044)	(0.012)
Constant	0.000 *	0.001	0.006	−0.000
	(0.000)	(0.003)	(0.004)	(0.001)
Control	Yes	Yes	Yes	Yes
Control2	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes
Obs.	3692	3692	3692	3692

Note: t statistics in parentheses, * p < 0.1, ** p < 0.05, *** p < 0.01.

indicates that climate-adaptive pilot cities enhance urban livability, attracting more high-caliber talent from outside to settle. This, in turn, elevates the level of NQPFs in the labor dimension.

4.3.2. Testing the Circulation Effects of Data Elements

Against the backdrop of emphasizing the development of NQPFs, data elements have emerged as a superior means of production compared to traditional factors. In the process of data circulation, data carriers serve as indispensable tools. Mobile phones, as a transmission medium with extremely high per capita ownership today, possess a certain degree of universality. Therefore, drawing upon the research of Wang et al. [45], this paper adopts mobile phone penetration rate (Phone) as a proxy variable to measure data circulation. Specifically, it calculates the average number of mobile phones per person. As shown in Column (2) of Table 6, in the development of climate-adaptive cities, measures such as building digital platforms facilitate the circulation of various types of data, thereby driving the growth of NQPFs in urban areas.

4.3.3. Testing the Upgrading Effect of Infrastructure Industry

Accordingly, this study adopts two indicators to quantify urban climate adaptation intensity: the overall industrial structure upgrade (UIS), following the methodology of Xu and Jiang [46], and industrial sophistication (ISS), based on Zhao et al. [47]. The UIS index is calculated using the formula $UIS=1\times {\kappa }_1+2\times {\kappa }_2+3\times {\kappa }_3$, which denotes the share of the added value of the industry in GDP. ISS is measured by the ratio of the added value of the tertiary sector to that of the secondary sector. As presented in columns (3) and (4) of Table 6, the pilot policy for climate-resilient cities significantly promotes industrial upgrading, thereby facilitating the growth of NQPFs.

4.4. Heterogeneity Analysis

4.4.1. Heterogeneity of Urban Geographical Locations

China’s vast territory spans distinct geographical regions—the East, Central, and West—resulting in heterogeneous climatic conditions. These regional climate environments exert varying influences on the level of new-type productive forces. Therefore, this study divides the overall sample into three regions—East, Central, and West—to examine the impact of climate-adaptive urban development on the level of new-type productive forces in cities.

As shown in

Table 7. Analysis of Geographical Location Heterogeneity.

Variable	NQPFs
	Urban Area			Maritime and Continental Position
	(1) East	(2) Middle	(3) West	(4) Coastal	(5) Inland
Weather	0.038 *	0.009	0.013 ***	0.007 *	0.293 ***
	(0.020)	(0.008)	(0.005)	(0.004)	(0.104)
Constant	−0.000	0.000	0.000	−0.000	0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Control	Yes	Yes	Yes	Yes	Yes
Control2	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes
Obs.	1300	1300	1092	2210	1482

Note: t statistics in parentheses, * p < 0.1, *** p < 0.01.

(1)–(3), the effect of building climate-resilient cities is most robust and significant in the western region, followed by the eastern region, while the effect is weakest in the central region. A possible explanation is that the western region generally exhibits arid conditions with low rainfall and large diurnal temperature variations, resulting in a relatively fragile ecological environment. Developing climate-resilient cities by constructing climate-adaptive infrastructure and ecosystems can effectively mitigate the impact of climate disasters like heavy rainfall and drought on urban production and daily life. This prevents production stagnation and supply chain disruptions, safeguards stable regional economic operations, and provides a reliable environmental foundation for the development of NQPFs.

Furthermore, as a country with both maritime and continental characteristics, China exhibits distinct climatic differentiation between coastal cities dominated by maritime influences and inland cities significantly affected by continental climates. This study categorizes cities into coastal and inland groups based on their geographical location to examine the role of climate-resilient city development. As shown in Table 7 (4) and (5), the implementation of climate-adaptive city pilot policies in inland cities demonstrates significantly greater effectiveness for fostering NQPFs compared to coastal cities. A plausible explanation lies in the extreme climate threats faced by inland regions, such as drought and desertification, which drive comprehensive technological innovation and industrial restructuring. Unlike coastal regions’ incremental improvement paths focused on defensive engineering, inland areas leverage external environmental challenges to drive breakthrough innovations. By transforming ecological governance demands into vehicles for green technology practices, they rapidly bypass constraints inherent in traditional development stages. During resource-economy transitions, they cultivate entirely new industrial systems spanning eco-friendly materials and intelligent equipment manufacturing, propelling resource-based economies toward emerging industrial chains like advanced environmental materials and smart devices. Simultaneously, the policy flexibility and digital infrastructure development potential in inland regions provide testing grounds for innovative technologies. This enables them to achieve breakthroughs in areas such as converting ecological resources into capital and establishing climate adaptation standards. In doing so, they are redefining the rules of the green economy and forming new centers of productive forces with greater growth potential.

4.4.2. Heterogeneity of Urban Types

To further investigate the heterogeneous impacts of climate-adaptive city pilot policies on urban NQPFs levels, this study refines the classification of cities based on two criteria. First, cities are categorized as resource-based or non-resource-based according to the State Council’s “National Sustainable Development Plan for Resource-Based Cities (2013–2020).” Second, cities are classified as key environmental protection cities or non-key environmental protection cities based on the State Council’s “National Environmental Protection Plan for the 11th Five-Year Plan Period”.

The grouped regression results for resource-based cities, presented in

Table 8. Heterogeneity Analysis of Urban Types.

Variable	NQPFs
	The Abundance of Resources		Intensity of Environmental Regulation
	(1) Resource-Based	(2) Non-Resource-Based	(3) Key Protection	(4) Non-Key Protection
weather	0.025 **	0.015 ***	0.012 ***	0.074 **
	(0.010)	(0.002)	(0.002)	(0.033)
Constant	0.000	0.000	−0.000	−0.000
	(0.000)	(0.000)	(0.000)	(0.001)
Control	Yes	Yes	Yes	Yes
Control2	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes
Obs.	2210	1482	2249	1443

Note: t statistics in parentheses, ** p < 0.05, *** p < 0.01.

, indicate that climate-adaptive city development exerts a stronger positive effect on NQPFs in resource-based cities compared to non-resource-based cities. This disparity likely stems from the fact that resource-based cities typically rely heavily on single-resource industries and exhibit relatively homogeneous economic structures. Climate adaptation initiatives can facilitate the green transformation of traditional resource-based industries—such as through clean coal utilization technologies—and simultaneously stimulate the emergence of new economic sectors like ecological industries. In contrast, non-resource-based cities generally have more diversified economic structures, where climate adaptation measures primarily yield localized optimization effects. These cities lack the same capacity as resource-based cities to catalyze large-scale productivity transformations through industrial restructuring and policy synergies, resulting in weaker overall promotional effects.

Regression analyses grouped by the intensity of environmental regulation further reveal that climate adaptation initiatives significantly enhance the development of new productive forces in non-environmentally protected cities. This outcome may be attributed to the fact that environmentally protected cities primarily prioritize the creation of favorable ecological environments, which constrains industrial development. In contrast, non-environmentally protected cities possess greater potential for traditional industrial upgrading. Climate adaptation initiatives, by promoting green infrastructure upgrades and facilitating industrial green transformation, more effectively activate urban productive capacity. Moreover, the pressing development needs of these cities make them more likely to secure policy support and resource investments, thereby enabling better alignment between climate adaptation measures and broader development strategies. This synergy fosters virtuous cycles of sustainable urban development.

5. Discussion

The theoretical significance of this study is as follows: First, it establishes a theoretical analytical framework linking climate-adaptive cities with new-quality productive forces in urban settings. Existing research predominantly examines the causal relationship between climate adaptation and new-quality productive forces through traditional lenses such as technological innovation upgrades and green transition policies, which inherently carry certain limitations in research perspective. This paper innovatively incorporates the perspective of climate adaptation policies into the analytical framework, demonstrating that climate-adaptive urban policies are not only a key catalyst for the emergence of new productive forces but also further expand the theoretical relationship between climate adaptation policies and the development of new productive forces. This provides a new perspective for understanding the driving role of policies in fostering new productive forces. Second, it explores the influence mechanism between climate adaptation policies and new productive forces. By constructing a theoretical analytical framework, this study introduces talent aggregation, data flow, and infrastructure industry upgrading as mediating variables. It builds a three-channel transmission mechanism model to further explore the complex relationship between climate-resilient urban policies and new productive forces, shedding light on the “black box” of their interaction.

The practical significance of this study lies in providing policymakers with a reference “policy blueprint”. In climate-resilient urban development, high-quality talent, data elements, and smart infrastructure all possess immense growth potential and serve as crucial anchors for NQPFs. Therefore, government departments can strengthen the application pathways for climate-resilient cities and establish a multidimensional urban productivity development system. For instance, governments should enhance incentives in talent recruitment policies by offering research start-up funds, housing subsidies, and other support packages to attract high-level climate adaptation experts. Simultaneously, universities should introduce specialized courses in climate-resilient urban planning and ecological restoration technologies to cultivate professionals with multidisciplinary competencies in climate monitoring, risk assessment, and emergency management. Collaborating with research institutions, governments can build a climate talent ecosystem integrating industry, academia, research, and application; accelerate adaptive infrastructure projects like drainage network upgrades and smart grids, guiding private capital participation to strengthen physical urban resilience foundations; establish cross-departmental climate data sharing platforms leveraging cloud computing, IoT, and AI technologies to break down data silos across meteorology, transportation, water resources, and other sectors; and building on this foundation, promote the transformation and upgrading of traditional infrastructure industries through climate resilience standards while simultaneously fostering emerging sectors like new energy and smart equipment, thereby continuously strengthening the development momentum of new urban productive forces.

6. Conclusions

In the context of increasingly frequent global extreme weather events—such as floods, droughts, and heatwaves—climate change significantly threatens the stability and sustainability of urban economic development. As a strategic policy instrument to address climate risks, climate-resilient city pilot programs not only mitigate the adverse impacts of extreme climatic events but also enhance the development level of urban new-quality productive forces. This study analyzes panel data from 284 Chinese prefecture-level cities between 2010 and 2022, employing a double machine learning model to rigorously assess the causal effect of climate-resilient urban policies on NQPFs. The findings are threefold: First, climate-resilience pilot policies significantly boost urban NQPFs. This result remains robust across a series of sensitivity checks—including outlier removal, control for parallel policy interference, and model re-estimation—providing strong support for Hypothesis 1. Empirical examples include Haikou’s Energy Trading Center in Hainan Province and Guangyuan’s Sponge City initiative in Sichuan Province, both of which have strengthened urban climate resilience through enhanced infrastructure durability and ecosystem stability, thereby safeguarding urban productive competitiveness and long-term sustainable development potential. Second, mechanism analysis confirms that these policies promote NQPFs through three key pathways: talent agglomeration, data flow enhancement, and infrastructure-related industrial upgrading—validating Hypotheses 2, 3, and 4, respectively. A representative case is Shenzhen’s climate-resilient community pilot project, where interdepartmental data platforms enable precise urban climate risk profiling, AI-driven ensemble forecasting models are iteratively refined, and sponge green corridors and ecological infrastructure are scientifically restored. This integrated approach not only improves local resilience but also fosters a high-quality environment conducive to the emergence of new productive forces. Third, heterogeneity analysis reveals significant variation in policy effectiveness. Geographically, western and inland cities exhibit stronger policy impacts than central, eastern, and coastal regions. In terms of city type, resource-based cities and non-environmentally protected cities demonstrate greater responsiveness in enhancing NQPFs. Despite its contributions, this study has several limitations pointing to future research directions. First, the sample period (2010–2022) is relatively limited; extending the time frame could better capture the dynamic evolution of climate policy effects. Second, the current indicator system measures NQPFs primarily based on labor inputs, offering only a preliminary assessment. Future studies should incorporate broader dimensions—such as technological innovation and green transformation—to construct a more comprehensive, multidimensional index. Third, while the double machine learning model offers robust causal inference, its performance depends on appropriate model specification. Complementary methods—such as qualitative comparative analysis and in-depth case studies—could strengthen the validity and interpretability of results. Finally, micro-level investigations into sectoral and firm-level heterogeneity—particularly across industries with differing technological intensities—are warranted. Incorporating moderating climatic variables like temperature and precipitation would further refine the explanatory framework for understanding how climate-resilient urban development drives productivity growth.