Assessing the Impact of Innovation-Oriented Urban Policy on Urban Ecological Resilience: Empirical Evidence from 271 Cities in China

Abstract

Against accelerating urban expansion, climate risks, and green transition, fortifying urban ecological resilience (UER) has become central to global urban sustainability. However, there is as yet no systematic evidence to suggest how the National Innovative City Pilot Policy (NICPP) promotes green technology innovation (GTI) and strengthens UER. To address this gap, this paper builds upon the traditional resistance–adaptability–resilience (RAR) paradigm and its extension, the driving force–resistance–adaptability–resilience (DRAR) framework to formulate a comprehensive driving force–pressure–resistance–adaptability–resilience (DPRAR) assessment system. Moreover, this paper investigates a dataset of 271 cities in China spanning the period from 2006 to 2023. By treating the NICPP as a quasi-natural experiment, this paper integrates an entropy balancing technique into a staggered difference–in–differences (DID) model to identify its effect on UER. The results are outlined below. (1) From 2006 to 2023, UER across Chinese municipalities generally increased, though spatial imbalances persisted. (2) The NICPP enhanced urban UER, with effects characterized by time-lagged and cumulative patterns. (3) Both the quantity and quality structure of urban GTI served as channels through which the NICPP affected UER. (4) The NICPP effects are strongest in the central region, followed by those in the western and eastern regions, as well as in resource-dependent cities and cities with stricter environmental regulation. This research provides novel empirical perspectives linking innovation-driven development, GTI, and UER, with implications for NICPP optimization and UER enhancement in China.

1. Introduction

Cities now serve as the primary engines of economic expansion, industrial organization, and human settlement, steadily elevating their global importance. Demographic forecasts suggest urbanites will make up nearly 60% of humanity by late 2025—driving about 80% of worldwide GDP—and this figure will likely approach 70% by 2050 [1]. Nevertheless, this aggressive expansion inadvertently turns metropolitan areas into concentrated hotspots for compound climate risks, multimedia pollution, and ecological degradation [2,3]. While the UN Sustainable Development Goals (SDGs) deliver a universal governance blueprint, cities continue to struggle to reconcile economic expansion with long-term sustainability. The transboundary wastewater issue in the Tijuana River Basin, Mexico, has long threatened the health of 3 million urban residents in surrounding areas [4]. In Karachi, Pakistan, extreme heatwaves have in recent years driven a continuous spike in mortality [5]. These cases reveal the inherent limitations of traditional environmental governance tools, underscoring an urgent demand for novel regulatory paradigms.

China serves as an ideal setting to explore these transformative solutions. China is currently at a critical stage of synergistically advancing the comprehensive deepening of reform, high-quality development, and “dual-carbon” goals [6,7]. Cities serve not only as crucial carriers for innovation-driven development but also as essential spaces for ecological governance and green transition [8]. Therefore, how to coordinate innovative development, ecological governance, and risk response through institutional innovation has emerged as a key concern in promoting sustainable urban development across China.

Urban ecological resilience (UER) refers to the capacity of urban ecosystems to sustain its structural and functional stability when confronted with disturbances and unpredictable risks, ultimately achieving sustainable development through adaptation, recovery, and reorganization [9]. In contrast to traditional urban ecological studies, the theoretical framework for UER remains in its developmental stage [10]. Cities function as complex adaptive systems in which socioeconomic activities and natural ecological processes are closely interconnected. Accordingly, UER depends not only on the response capacity of ecosystems but also on development patterns, resource use, and governance behavior. Existing studies on UER can be broadly classified into three categories. The first strand focuses on conceptual clarification and the construction of measurement frameworks. Some studies take the urban ecosystem itself as the analytical object and measure UER from dimensions such as landscape patterns and ecosystem services. For example, Feng et al. [11] characterized UER through a vulnerability–sensitivity–self-organization theoretical framework. Additional research utilizes paradigms like the resistance–adaptability–resilience (RAR) model to construct UER evaluation systems from the perspective of resilience capacity, incorporating multidimensional indicators related to natural, ecological, and social conditions [12]. Moreover, the evaluation framework of driving force–pressure–state–impact–response has been introduced into UER assessment to partially bridge the gap among ecology, social sciences, and economics [13,14]. Building on this line of research, Lan et al. [15] further integrated the “driving force” component into the RAR framework and formulated the driving force–resistance–adaptability–resilience (DRAR) framework. This framework attempts to overcome the limitations of traditional frameworks that primarily focus on passive ecosystem responses. Second, researchers have increasingly tracked the spatiotemporal disparities and dynamic evolution of UER. This is typically achieved using spatial econometric tools, such as kernel density estimation for distribution patterns, the Dagum Gini coefficient for inequality decomposition, and Markov chains to capture transition probabilities [16,17]. Third, extensive scholarly inquiries have been directed toward unraveling the complex determinants that govern UER. It is widely recognized that green technology innovation, coupled with environmental regulations, critically shapes how UER improves and fluctuates across regions [15]. While prior research has enhanced our understanding of the connotation, measurement, and evolutionary patterns of UER, existing scholarship has yet to uncover the exact mechanisms through which institutional innovation policies drive the dynamic evolution of UER.

The National Innovative City Pilot Policy (NICPP) provides a useful context for examining how institutional innovation affects UER. China’s NICPP is a major pilot program implemented in successive phases to promote city-level innovation. By combining strategic planning, R&D funding, platform development, and support for high-tech enterprises, it effectively enhances the structure and integration of urban innovation systems [18]. Existing studies evaluated this policy primarily through the metric of raw innovation performance [19]. Subsequent research broadened this perspective, documenting the policy’s role in redistributing factor endowments, upgrading industrial bases, and fostering high-quality municipal expansion [20,21,22]. Emerging studies demonstrate that the NICPP achieves its environmental dividends primarily through the vital channel of green technology innovation (GTI) [8]. By stimulating GTI, the policy does more than just hit basic environmental performance targets and drive the low-emission development trajectories; it is instrumental in the rehabilitation and optimization of the urban ecological milieu [8,23].

Unfortunately, although substantial evidence on both the evolutionary patterns of UER and the policy effects of the NICPP has been accumulated, the following limitations still exist. First, the RAR framework focuses on the endogenous responses of urban ecosystems to disturbances. Its extensions introduce socioeconomic drivers, yet they lack the pressure transmission mechanism that connects these drivers to the ecosystem’s endogenous responses [13]. In particular, DRAR does not retain the DPSIR distinction between underlying socioeconomic drivers and the environmental pressures they generate, such as resource consumption, pollutant emissions, and land-use change [15,24]. This may blur the boundary between development capacity and ecological burden, limiting its representation of external constraints on UER. Furthermore, systematic evidence regarding the consequences of NICPP for overall UER is currently lacking, as existing research largely examines individual environmental indicators such as pollution reduction and carbon emissions [25,26]. Third, existing studies typically treat GTI as a homogeneous quantitative outcome, ignoring how the quantitative expansion and quality-structure optimization of GTI differentially affect UER [27,28]. In addition, existing difference–in–differences (DID) studies of the NICPP rarely account explicitly for the potential estimation bias arising from staggered policy adoption and heterogeneous treatment effects [26,29]. These gaps call for a unified analytical framework that integrates UER measurement, mechanism identification, and policy evaluation.

Therefore, the analysis is organized around three core issues: (1) Does the NICPP enhance or hinder UER in China? (2) What are the underlying channels through which the NICPP influences UER? (3) How does the effect of the NICPP on UER differ across regions, resource endowments, and levels of environmental regulation stringency?

To this end, we propose a holistic analytical approach. Specifically, this paper constructs a novel driving force–pressure–resistance–adaptability–resilience (DPRAR) evaluation system to gauge UER. Subsequently, treating the NICPP as a quasi-natural experiment and utilizing panel data from 271 Chinese cities for the period from 2006 to 2023, this paper combines entropy balancing with a staggered DID and modern DID estimators to estimate the effects of the NICPP and uncover its underlying transmission mechanisms. The contributions of this paper are threefold: First, building on the conventional RAR theory and extending the DRAR framework, this paper develops the DPRAR evaluation system. By incorporating a pressure dimension measured using indicators such as population density and energy consumption, the framework captures how socioeconomic drivers translate into observable environmental loads through demands on natural systems. Consequently, the DPRAR framework advances beyond DRAR by distinguishing socioeconomic development foundations from the resource-use and spatial pressures they generate. This distinction, which is conflated in DRAR, is essential for capturing the environmental costs of development and linking them to endogenous ecological responses. Second, this study moves beyond the limitations of existing NICPP research, which focuses solely on static ecological outcomes including environmental pollution control, carbon-emission mitigation, and green development performance. By combining entropy balancing with modern DID estimators, this study mitigates observable selection bias between treated and untreated cities and addresses potential estimation bias arising from staggered policy adoption and heterogeneous treatment effects. Therefore, this study offers fresh and more credible causal insights into how institutional innovation policies affect ecological resilience. Third, it further reveals the GTI mechanism underlying this effect and distinguishes the roles of green technology innovation quantity (GTIN) and quality structure (GTIQ) in shaping UER. Furthermore, this paper systematically examines how policy effects vary across different regions, resource endowments, and environmental regulation intensities. Based on these findings, it provides a more robust empirical foundation for innovation-oriented policy design and theoretical research on resilience in sustainable urban development.

2. Policy Context and Research Hypothesis

2.1. Policy Context

NICPP implementation began in 2008, with Shenzhen designated as the initial pilot city. The 2010 release of the Guiding Opinions moved the policy beyond an isolated trial and promoted its extension to multi-city implementation. Subsequently, related policy arrangements clarified construction objectives, key tasks, and monitoring requirements, placing innovative city construction within a more unified governance framework. The 2016 Working Guidelines further marked the transition toward institutionalized and standardized implementation. The NICPP broadened its scope in 2018 by incorporating an additional 17 urban centers, with Shaoxing, Xuzhou, and Jilin among them. Four years later, another 25 cities, including Baoding, Xiangtan, and Mianyang, were supported in carrying out pilot programs. Currently, a cumulative total of 103 innovative districts and cities have been designated across seven distinct rollout phases nationwide. Figure 1 delineates the phased implementation pattern of the NICPP.

2.2. Research Hypothesis

2.2.1. Direct Effects

Viewed through the lens of institutional logic, innovation policies can sustainably reshape urban development pathways only when they simultaneously affect factor mobility, organizational coordination, and the governance environment [30]. Cities are social-ecological systems with multiple stability domains, in which socioeconomic and natural elements interact dynamically through feedback [31]. Policy interventions can alter their evolutionary trajectories by influencing the system’s adaptive capacity and transformative potential [9,32]. Some scholars caution that innovation policies may produce rebound effects—where technological progress increases energy and resource consumption, thereby offsetting environmental gains [31]. Others point to information asymmetries that incentivize low-quality innovation at the expense of genuine environmental benefits [32]. However, whether these institutional provisions translate into actual ecological gains hinges on how local governments respond to the policy incentives embedded in the performance appraisal system. From the perspective of neoclassical economics, stringent regulatory pressures may impose compliance costs that crowd out firms’ R&D resources, potentially suppressing green technology innovation and, in turn, UER improvement [15]. The NICPP, however, operates differently: it prioritizes innovation incentives while embedding environmental performance into its appraisal system. It effectively weaves together the distribution of inventive assets, structural industrial shifts, cross-departmental administrative collaboration, and rigorous performance appraisals into a unified regulatory umbrella [20]. The publication of the “Guiding Opinions” in 2010 officially aligned the NICPP’s overarching goals with the objective of fostering synergistic and resilient socioeconomic advancement. This directive simultaneously introduced tangible administrative and environmental benchmarks, compelling assessments to factor in sewage treatment rates, shifts in the air quality index, total energy use, and reductions in pollutant discharges. Building upon this foundation, the subsequent 2016 “Working Guidelines” elevated “green and low-carbon” growth to the status of a fundamental building block for urban development. It incorporates metrics such as forest coverage and carbon dioxide emission intensity into a specific indicator system [29]. Such a trajectory reveals that boosting innovation output constitutes merely a fraction of the NICPP’s broader agenda. Under this integrated framework, the innovation compensation mechanism of the Porter hypothesis reflects compliance costs can be partially or fully offset through efficiency gains and technological improvements [33]. The resulting innovation gains, in turn, feed back into R&D investment and enhance ecosystems’ resource cycling efficiency and resistance to disturbances through technology spillover effects [34]. By improving operational efficiency and boosting R&D investment in emission reduction, this framework drives industrial transformation [35]. Through the two-way interaction between “industrial ecologization” and “ecological industrialization”, it strengthens the environmental adaptability of the economic system, as well as the material−energy cycling and self-regulation capacity of ecosystems, thereby providing structural support for UER [36]. Consequently, by aligning innovation incentives with ecological performance targets, the NICPP is expected to generate a net positive effect on UER. Accordingly, this paper puts forward Hypothesis 1.

H1. 

The NICPP enhances UER.

2.2.2. Green Innovation Effect

Given the twofold external advantages of knowledge diffusion and ecological improvement inherent in green technology innovation (GTI), its natural market undersupply renders targeted policy interventions indispensable [30,37]. Furthermore, drawing upon the paradigms of induced innovation and the Porter hypothesis, targeted institutional incentives and regulatory constraints may compel firms to undertake technological upgrades and green R&D. These actions subsequently improve resource efficiency and environmental performance through innovation offset effects [38]. Thus, the NICPP may enhance UER by optimizing green R&D allocation and promoting the accumulation and diffusion of GTI. Specifically, such GTI fosters emerging sectors, such as new energy equipment and green building materials. This strengthens the endogenous momentum for urban green development [39,40]. GTI applications in energy-efficient buildings and smart transportation reduce resource use and ease ecological pressure [41,42]. Under external shocks, clean energy substitution and emission-control technologies reduce pollutant emissions and ecological disturbances, thereby strengthening the resistance of urban ecosystems [43]. In terms of adaptability, innovative practices, such as sponge city development, smart grids, and solid waste recycling, improve a city’s dynamic response to environmental risks such as extreme rainfall [44,45]. Finally, regarding recoverability, green technological advancements in river and lake ecological restoration, soil pollution treatment, and postdisaster green infrastructure reconstruction accelerate the restoration of damaged ecological functions [46,47,48]. They also reduce secondary ecological risks following a disturbance. In summary, Hypothesis 2 is established as follows.

H2. 

Driven by the NICPP, advancing urban GTI would help to strengthen UER.

However, urban GTI has multidimensional attributes, which can be reflected in two dimensions: quantitative growth and quality–structure optimization [29]. By promoting R&D factor agglomeration, improving the innovation environment, and enhancing policy synergy, the NICPP lowers the barriers and trial–and–error costs of green R&D, thereby increasing the GTIN and facilitating green technology adoption [29,49,50]. Under the dual influence of innovation incentives and green constraints established by the NICPP, enterprises tend to prioritize green technology sectors characterized by shorter R&D cycles, lower implementation risks, and more mature commercialization conditions [51,52]. Ultimately, GTIN may enhance UER by broadening the portfolio of green technologies available for pollution control, resource conservation, and ecological risk response [27]. Accordingly, this paper proposes Hypothesis 2a.

H2a. 

NICPP enhances UER by increasing GTIN.

Beyond merely scaling up volume, the NICPP possesses the capacity to refine the structural quality of municipal green technology innovation (GTIQ), which in turn fortifies UER. Driven by high-caliber innovation, local economies can pivot away from resource–draining and emission–heavy paradigms toward a circular, low-carbon configuration. This shift naturally encourages green industrial clustering and the cohesive expansion of supply chains [53,54,55]. Such an endogenous adjustment curbs the presence of highly polluting industries and encourages a virtuous cycle in the ecological economy via the agglomeration of green industries [15]. Ultimately, this mechanism continuously fuels the dynamic adaptation and resilient trajectory of UER. We propose Hypothesis 2b.

H2b. 

The NICPP drives substantial improvements in UER via the optimization of urban GTIQ.

A visual summary of these interconnected hypotheses is shown in Figure 2.

3. Empirical Design and Data

The following details our empirical approach, covering the data, variables, and evaluation framework. Figure 3 provides a complete roadmap.

3.1. Methodology

3.1.1. Entropy Weight–TOPSIS Method

The entropy weight approach uses information–theoretic principles to assign objective weights to evaluation criteria, thereby mitigating subjective judgment bias. Conversely, the TOPSIS framework evaluates and orders these options according to their proximity to the theoretical optimum, thereby pinpointing the most suitable choice with greater precision [56]. Based on this procedure, the UER index is derived for 271 Chinese cities at the prefectural level during 2006–2023.

3.1.2. Kernel Density Estimation

To characterize how China’s UER shifts continuously over time, this paper applies the kernel density estimation (KDE) approach using Python (version 3.13.2) to conduct a nonparametric estimation of the temporal distribution of UER in China. Compared with parametric methods, KDE can more flexibly identify distributional location, dispersion, polarization, and multimodal structures [57].

Let $UE{R}_{it}$ denote the ecological resilience index for a given city $i$ at time $t$. Its kernel density function can then be expressed as Equation (1):

$${\widehat{f}}_t(UER)={\displaystyle \frac{1}{n_t{h}_t}}{\displaystyle \sum _{i=1}^{n_t}K}\left({\displaystyle \frac{UER-UE{R}_{it}}{h_t}}\right)$$

(1)

where $n_t$ indicates the total count of cities, $h_t$ defines the bandwidth, and $K\left(\cdot \right)$ represents the kernel function. In this paper, the Gaussian kernel function and the empirical rule proposed by Silverman [58], where $\widehat{\mathsf{\sigma }}$ denotes the sample standard deviation, are adopted:

$$h_t={\left({\displaystyle \frac{4{\widehat{\mathsf{\sigma }}}_t^5}{3n_t}}\right)}^{{\displaystyle \frac{1}{5}}}\approx 1.06{\widehat{\mathsf{\sigma }}}_t{n}_t^{-{\displaystyle \frac{1}{5}}}$$

(2)

3.1.3. Differences–in–Differences

By viewing the NICPP as a quasi-natural experiment, this paper applies a staggered DID design to assess its average treatment effect. The inclusion of city and time fixed effects can mitigate potential endogeneity concerns to some extent [59]. Following Beck et al. [60], this study constructs the following baseline regression specification:

$$UE{R}_{it}={\beta }_0+{\beta }_{1}NICP{P}_{it}+{\displaystyle \sum _{k=1}^K{\beta }_k}Control{s}_{kit}+{\mu }_i+{\nu }_t+{\epsilon }_{it}$$

(3)

where $i$ and $t$ represent city and year, respectively; $NICP{P}_{it}$ is a dummy variable indicating whether city $i$ is affected by the NICPP in year $t$; $Control{s}_{kit}$ represents the $k-th$ covariate; ${\mu }_i$ and ${\nu }_t$ represent city and year fixed effects, respectively; ${\epsilon }_{it}$ is the stochastic error; ${\beta }_1$ captures the impact of the NICPP on UER. All regression estimations are implemented in Stata (version 18.0).

3.1.4. Entropy Balancing Method

Because the NICPP was not implemented randomly, directly identifying its impact on UER may be affected by sample self-selection bias, thereby generating potential endogeneity concerns. Although conventional propensity score matching (PSM) helps alleviate some selection biases, it still faces three limitations: information loss caused by the exclusion of unmatched samples; imbalance in higher-order moments of covariates between the treated and untreated subsets after pairing; as well as the susceptibility of estimation results to the specification of the first-stage model and the selection of covariates [61].

Therefore, the current study applies the entropy balancing (EB) approach to preprocess the sample [62,63]. It applies reweighting coefficients to the comparison group so that the experimental and control groups meet established equilibrium constraints across the mean, variance, and skewness of the covariates [64]. The implementations are detailed below.

First, the dummy variable $treate{d}_i$ is used to identify the treatment status of each city. If city $i$ is selected into the NICPP, then $treate{d}_i=1$; otherwise, it is 0. For city $i$, let its observable covariate vector be as follows:

$$Z_i={\left(z_{i1},z_{i2},\dots ,z_{iJ}\right)}^{\prime }$$

(4)

where $j=1,\cdot \cdot \cdot ,J$ is the index covariate and $J$ denotes the number of covariates. The superscript ${\left(\cdot \right)}^{\prime }$ indicates vector transposition.

Next, this paper constructs and solves the entropy balancing optimization problem. By solving an entropy minimization problem, the optimal weight $w_i$ for each control unit can be obtained. The calculation is shown in Equation (5).

$$\underset{w_i}{\min }H\left(w\right)={\displaystyle \sum _{\left\{i| treate{d}_i=0\right\}}{w}_i}\ln \left({\displaystyle \frac{w_i}{q_i}}\right)$$

(5)

To ensure comparability in the covariate distribution, this paper requires the balancing conditions to match up to the third moment. Specifically, this paper constructs an extended vector of moment functions, $g_r\left(Z_i\right)$, which contains information on the mean, variance, and skewness of the covariates. Specific constraints apply as follows.

(1)

Moment balance constraint.

$${\displaystyle \sum _{\left\{i| treate{d}_i=0\right\}}{w}_i}{g}_r\left(Z_i\right)={\overline{g}}_r^{ treated},\hspace{1em}r=1,2,\dots ,R$$

(6)

(2)

Normalization constraint.

$${\displaystyle \sum _{\left\{i| treate{d}_i=0\right\}}{w}_i}=1$$

(7)

(3)

Nonnegativity constraint.

$$w_i\ge 0,\hspace{1em}\forall i\in \left\{i| treate{d}_i=0\right\}$$

(8)

where $n_0$ and $n_1$ represent the total observation counts for the non-treated and treated groups, respectively, $q_i=1/n_0$ is the base weight, and $R$ denotes the number of moment constraints.

After the optimal weights $w_i$ are obtained, they are used in the subsequent regression estimation to improve between-group comparability and alleviate the interference from observable selection bias when estimating treatment.

3.1.5. Callaway and Sant’Anna DID Method

The NICPP was implemented in multiple batches over time. Under staggered policy adoption timing and heterogeneous treatment effects, conventional two-way fixed effects (TWFE) event-study models could mix comparisons across different treatment cohorts, making it difficult to clearly identify dynamic average treatment effects [65]. To mitigate this potential estimation bias, this paper further employs the CS–DID method developed by Callaway and Sant’Anna [66] to carry out reliability checks and dynamic effect analysis. When controlling for covariates, this method combines inverse probability weighting with outcome regression adjustment to mitigate estimation bias under multiperiod staggered treatment settings [67].

Specifically, let $F_i$ denote the initial adoption year of the NICPP for city $i$. If a city never enters the NICPP, then $F_i=0$. Regarding the cohort of cities that first participated in year $g$, their corresponding average treatment effect during year $t$ can be expressed as:

$$ATT\left(g,t\right)=E\left[UE{R}_{it}\left(1\right)-UE{R}_{it}\left(0\right)| F_i=g\right],\hspace{1em}t\ge g$$

(9)

Furthermore, by aggregating according to the relative policy time $1=t-g$, the dynamic average treatment effect is obtained as follows:

$$AT{T}^{dyn}\left(l\right)={\displaystyle \sum _g{\omega }_{gl}}ATT\left(g,g+l\right),\hspace{1em}l=-6,-5,\dots ,5,6$$

(10)

where ${\omega }_{gl}$ denotes the corresponding aggregation weight. The hypothesis corresponding to the parallel trend evaluation is specified below:

$$AT{T}^{dyn}\left(l\right)=0,\hspace{1em}\forall l<0$$

(11)

If $AT{T}^{dyn}\left(l\right)$ is not statistically significant in each pre-policy period when $l<0$, the parallel trend assumption is also supported.

3.2. Definition and Measurement of Variables

3.2.1. Explained Variable

Existing research often measures UER using composite evaluation methods. Among them, the RAR framework emphasizes the endogenous response process of ecosystems under disturbances and shocks and has gained widespread adoption [10,68]. Extending the RAR framework, Lan et al. [15] developed the DRAR model by incorporating the “driving force” element derived from the DPSIR paradigm. This analytical scheme integrates socioeconomic factors into the evaluation of UER, expanding the explanatory scope of the RAR framework. However, the DRAR framework still inadequately captures disturbance processes such as resource consumption and ecological space squeezing. Consequently, its explanation of how external constraints affect UER remains relatively limited. Based on the DPSIR conceptual structure introduced by the European Environment Agency, driving forces reflect fundamental needs, such as economic growth, whereas pressures represent the human activities generated by these needs, including resource use or misuse, emissions, and land-use change [24]. Adopting this perspective, this paper further introduces the “pressure” dimension into the DRAR framework and constructs a DPRAR analytical framework. It enhances the original DRAR structure by deconstructing the overarching “driving force” category into distinct socioeconomic drivers and resource–environmental pressures. This refinement allows for a clearer identification of how innovation policies operate amidst the interplay of growth incentives and resource limitations, moving beyond the conceptual ambiguity of earlier models. Consequently, the “driving force–pressure–resilience–feedback” loop maps the transmission of exogenous drivers to internal ecosystem responses, establishing a rigorous theoretical foundation for UER quantification. The theoretical framework of DPRAR is presented in Figure 4.

The specific indicators considered for each dimension are as follows:

(1)

The driving forces dimension captures the antecedent socioeconomic conditions that initiate ecological change and reflects the source-level influence of human activities on urban ecosystems [15]. These indicators constitute the socioeconomic foundation for urban ecological development and governance, while the associated resource and spatial demands are captured separately by the pressure dimension. The rate of natural population increase [14] affects future population size and thereby alters resource demand and environmental loads. Urban residents’ disposable income per capita [14], per capita GDP [69], and retail sales of consumer goods per capita [15] characterize urban development foundations and potential governance capacity from the perspectives of individual purchasing power, regional economic development, and the scale of social consumption, respectively. The urbanization rat [13] mainly reflects urban development and the concentration of infrastructure and public services, although excessive land demand associated with rapid urbanization may exceed ecosystem carrying capacity.

(2)

In this study, the pressure dimension is defined as capturing the actual constraints imposed on urban ecosystems as socioeconomic drivers operate through population concentration, spatial expansion, and resource consumption. Population density [69] reveals the current population pressure on environmental carrying capacity. Compared with the urbanization rate, urban construction land area [70] more directly reflects the pressure exerted by spatial expansion on land resources and ecological space. Total energy consumption [69] indicates the scale of urban metabolism and the associated demand for resources, while water consumption in urban districts [13] reflects the direct intensity of water-resource consumption generated by urban socioeconomic activities. Accordingly, the pressure dimension reveals the pathways through which socioeconomic drivers are transformed into resource and spatial constraints and defines the external conditions facing the subsequent endogenous responses of urban ecosystems.

(3)

The resistance dimension represents the capacity of urban ecosystems to withstand disturbances and buffer or absorb external shocks under external pressure. This study measures resistance using pollutant and carbon-emission intensities per unit of economic output, thereby reflecting a city’s ability to control environmental loads and reduce disturbance intensity at a given level of economic activity. Industrial wastewater discharge per unit of GDP [71] captures the water-pollution load associated with each unit of output, and its variation may reflect cleaner production technologies, wastewater treatment capacity, and industrial restructuring. Industrial sulfur dioxide emissions per unit of GDP [15] and carbon emissions per unit of GDP [15] reflect the capacity to alleviate atmospheric pollution and carbon-emission pressures through cleaner energy structures and low-carbon transition, respectively. Industrial soot and dust emissions per unit of GDP [15] measure the capacity to limit the diffusion and deposition impacts of particulate pollution generated by industrial production. Lower values of the above indicators indicate that a city generates less environmental disturbance through technological regulation and structural optimization while maintaining economic output, thereby exhibiting stronger ecosystem resistance.

(4)

The adaptability dimension measures the capacity of urban ecosystems to maintain dynamic equilibrium under sustained external pressure through structural optimization, process regulation, and functional reorganization. The domestic wastewater treatment ratio [15] reflects the capacity of cities to reduce water pollution loads and control water environmental risks through wastewater collection and purification systems under sustained wastewater pressure. The integrated recycling rate of general industrial solid waste [72] indicates the capacity of industrial systems to reuse waste resources and optimize material-metabolism pathways. Per capita domestic waste collected and transported [10] reflects the volume of household waste generated per resident that enters formal collection and transportation systems; an increase may result from improved collection coverage and service capacity, but may also be affected by greater waste generation. The safe disposal rate of domestic waste [15] indicates the degree to which cities control environmental risks at the final disposal stage. Accordingly, adaptability is operationalized through improvements in pollution treatment, resource reuse, and waste management, capturing the capacity of cities to restore systemic balance under persistent external pressure.

(5)

The resilience dimension measures the capacity of urban ecosystems to restore ecological functions, improve environmental quality and re-establish a stable state after external disturbances. Land area per resident [71] represents the land-resource foundation available for ecological regulation, spatial reorganization, and functional recovery following disturbance. Park green area per resident [15] reflects the provision of public ecological space and provides important support for microclimate regulation, ecosystem-service recovery, and improvements in residents’ environmental well-being. The green coverage rate in built-up areas [16] captures the coverage and spatial continuity of urban green infrastructure, helping to strengthen landscape connectivity and promote the restoration of damaged ecosystems. Mean annual PM2.5 concentration [10] reflects the actual level of pollution-load reduction and environmental-quality recovery after disturbance. Together, these indicators characterize the comprehensive capacity of urban ecosystems for spatial restoration, green support, and environmental-quality recovery.

Within this framework, “driving forces” demarcate the foundational impact of socioeconomic conditions on ecological systems, whereas “pressures” aggregate the environmental stressors stemming from demographic concentration, urban sprawl, and resource consumption. The “resistance” signifies the inherent threshold for buffering environmental contaminants and disturbances. “Adaptability” signifies the systemic capacity to restructure via governance-led or indirect renewal following disturbances, facilitating its transition to new environmental and systemic states. Lastly, “resilience” encapsulates the ecosystem’s innate self-organizing strength to drive post-disturbance renewal, manifesting through spatial restoration, green support, and the recovery of environmental quality. These indicators constitute substantive dimensions of urban ecological resilience, and policy-induced changes in them may therefore represent genuine improvements or deterioration in the underlying urban ecological system. The metrics assigned to each criterion layer are reported in

Table 1. Indicator framework for measuring UER in China.

Target Layer	Dimension	Indicator	Unit	Effect	Weight	Reference
Urban ecological resilience	Driving forces	Per capita GDP	CNY per inhabitant	+	0.1054	Wei et al. [69]
		Urban residents’ disposable income per capita	CNY per inhabitant	+	0.0851	Xu et al. [14]
		Retail sales of consumer goods per capita	CNY per inhabitant	+	0.1023	Lan et al. [15]
		Rate of natural population increase	%	−	0.0315	Xu et al. [14]
		Urbanization rate	%	+	0.0522	Zhao et al. [13]
	Pressures	Population density	persons/km2	−	0.0125	Wei et al. [69]
		Urban construction land area	km2	−	0.0134	Li et al. [70]
		Total energy consumption	104 tce	−	0.0141	Wei et al. [69]
		Water consumption in urban districts	tonnes	−	0.0312	Zhao et al. [13]
	Resistance	Carbon emissions per unit of GDP	tons/100 million RMB	−	0.0189	Yin et al. [10]
		Industrial wastewater discharge per unit of GDP	tons/100 million RMB	−	0.0186	Li and Wang [71]
		Industrial sulfur dioxide emissions per unit of GDP	tons/100 million RMB	−	0.0158	Lan et al. [15]
		Industrial soot and dust emissions per unit of GDP	tons/100 million RMB	−	0.0159	Lan et al. [15]
	Adaptability	Domestic wastewater treatment ratio	%	+	0.0207	Lan et al. [15]
		Integrated recycling rate of general industrial solid waste	%	+	0.0279	Huang et al. [72]
		Per capita domestic waste collected and transported	104 tons/person	+	0.1368	Yin et al. [10]
		Safe disposal rate of domestic waste	%	+	0.0141	Lan et al. [15]
	Resilience	Park green area per resident	m2/person	+	0.0456	Lan et al. [15]
		Green coverage rate in built-up areas	%	+	0.0239	Lee et al. [16]
		Land area per resident	m2/person	+	0.1812	Li and Wang [71]
		Mean annual PM2.5 concentration	μg/m3	−	0.0328	Yin et al. [10]

. After data for the 21 indicators are collected, the present study utilizes an information entropy weighting technique to quantify the UER level for 271 Chinese prefectural units from 2006 to 2023. In the benchmark regression that follows, this index is treated as the central explained variable.

3.2.2. Core Explanatory Variables

The core explanatory variable is the policy dummy $NICP{P}_{it}$. It captures the treatment status of cities that were incorporated into the NICPP at different points in time.

3.2.3. Control Variables

To reduce bias arising from unobserved omitted factors, this paper follows prior studies and controls for a series of factors that might impact UER [15,73].

(1)

Human capital level (HCI)

HCI may affect UER by shaping regional competitiveness and facilitating labor mobility [74].

(2)

Degree of openness (OPEN)

Expanding export-oriented production can lead to a “scale effect” that intensifies local environmental stress [75]. They may also facilitate the introduction of clean technologies and the diffusion of environmental management experience, generating positive technological spillovers and institutional demonstration effects [76].

(3)

Government intervention (GOV)

GOV indicates how local governments manage urban resource allocation using fiscal policies [77]. It may affect UER by influencing ecological governance investment and the process of green transition [78].

(4)

Investment level (INV)

INV results in large-scale land cover shifts that frequently degrade habitats and constrain UER [79]. It may constrain UER through habitat destruction and ecological function degradation [80]. However, when capital is channeled into green infrastructure and restoration initiatives, it reverses this degradation and acts as a catalyst for enhanced UER [81].

(5)

Environmental regulation (ER)

ER influences UER through a complex mechanism. On the one hand, compliance pressures increase corporate costs, suppressing green technology innovation and thus reducing UER [15]. On the other hand, ER bolsters UER by coupling pollution mitigation with green infrastructure to upgrade socio-ecological adaptability [14].

(6)

Financial development level (FIN)

Financial development may affect UER by relaxing financing constraints for green projects, improving access to green credit, enhancing capital allocation efficiency and the availability of long-term capital [82,83].

Table 2. Variable description.

Variable Type	Metric	Abbreviation	Calculation Method	Unit
Explained variable	Urban ecological resilience	UER	Composite measurement	-
Core explanatory variable	National Innovative City Pilot Policy	NICPP	-	-
Control variables	Government intervention	GOV	Share of local government expenditure in the Gross Regional Product (GRP)	%
	Human capital level	HCI	Ratio of students enrolled in regular higher education institutions to registered population	%
	Trade openness	OPEN	Share of aggregate imports and exports in GRP	%
	Investment level	INV	Share of fixed asset investment in GRP	%
	Environmental regulation intensity	ER	Share of 15 eco-environmental keywords in total word frequency of government reports	%
	Financial development level	FIN	Proportion of year-end institutional credit and savings within the GRP	%
Mechanism variables	Quantity of green technology innovation	GTIN	Ratio of green invention patent applications to permanent population	Applications per 10,000 persons
	Quality structure of green technology innovation	GTIQ	Ratio of green invention patent applications to total green patent applications	%

reports the calculation methods for all the covariates.

Considering that ER, GOV, and FIN are not identical to the indicators used to construct UER, they may be conceptually related to its environmental-governance, adaptive-capacity, and resource-allocation dimensions. Moreover, these variables may themselves respond to the NICPP and may therefore absorb part of the policy-related variation when included as contemporaneous controls. To assess whether the baseline estimate is sensitive to their inclusion, we additionally report an EB–FE specification in Section 4.2 that jointly excludes ER, GOV, and FIN. The entropy balancing weights are re-estimated using the remaining covariates, namely HCI, OPEN, and INV.

3.2.4. Mechanism Variables

Existing studies commonly measure urban GTI solely through the volume of green patents, but this practice remains controversial [27,28]. The impact of policy shocks on urban GTI may differ, and a surge in green patent counts does not necessarily imply an improvement in innovation quality [27]. Moreover, relative to granted patents, patent applications provide a more timely and stable indication of innovative activity [84,85]. Invention patents also involve higher technological complexity and innovation thresholds than utility models do and can better capture substantive green innovation [29]. Accordingly, this paper measures urban GTI from two dimensions: urban GTIN and GTIQ [29].

(1)

GTIN is proxied by the per capita quantity of green invention patent filings. This indicator reflects the scale of invention activities while avoiding measurement bias caused by differences in city size [29,43].

(2)

GTIQ denotes the ratio of green invention patents relative to all green patent submissions. It reflects the structural transformation of urban GTI from scale expansion toward the accumulation of high-value innovation [29].

3.3. Data

Considering data availability, this research constructs its empirical analysis using city-level panel data 271 Chinese prefecture-level cities spanning the years 2006 to 2023. The starting year provides a five-year pre-policy window before the first major NICPP wave in 2010, while 2023 is the latest year with sufficient UER data. To examine the effect of the NICPP on UER, a staggered DID approach is implemented. The data required to calculate UER and the covariates are obtained mainly from the China Urban Construction Statistical Yearbook and the China City Statistical Yearbook. Minor data deficiencies are jointly supplemented by local municipal yearbooks alongside linear interpolation. To alleviate heteroskedasticity and mitigate the impact of outliers, this research converts control variables into their logarithmic forms [59]. Urban green patent data were gathered using the Green Patent Research Database and the China Patent Statistical Yearbook.

Table 3. Baseline characteristics and the VIF diagnostics.

Variable	Mean	SD	Min	Max	N	VIF	1/VIF
UER	0.2953178	0.1013528	0.1148678	0.6780835	4878	/	/
NICPP	0.1812218	0.3852413	0	1	4878	1.50	0.666159
lnHCI	−4.582678	1.132313	−10.89109	−1.689681	4878	1.95	0.511903
lnOPEN	−2.527983	1.43893	−7.988895	1.253828	4878	1.56	0.641482
lnGOV	−1.803308	0.4401029	−2.74432	−0.6192042	4878	2.19	0.455784
lnINV	−0.2652872	0.6879533	−5.785093	2.521394	4878	1.16	0.864100
lnER	−4.96166	0.4600748	−9.34243	−3.646118	4878	1.09	0.916384
lnFIN	0.7777335	0.4316762	−0.5312336	3.058775	4878	2.19	0.456501

details the baseline characteristics of the variables and checks for multicollinearity using the variance inflation factor (VIF). With all VIF scores ranging from 1 to 5, the model exhibits no severe multicollinearity.

4. Results

4.1. Spatiotemporal Dynamics of UER

4.1.1. Temporal Evolution Trends of UER

Figure 5 illustrates the temporal dynamics of China’s UER over the 2006–2023 period. Overall, the main peak of China’s UER KDE curve continued to shift rightward, indicating that the overall UER level was continuously enhanced. The decline in the main peak, along with the outward spread of the right tail, suggests that interregional disparities expanded to some extent, showing a certain pattern of gradient differentiation. In the eastern region, the KDE curve shifted toward the high-value range as a whole. The main peak was relatively wide and flat, indicating a higher proportion of cities in the middle-gradient range and a more dispersed regional distribution. The central region’s dominant peak showed the greatest rightward shift and the fastest decline, whereas the density of the right-side extension increased. The increase in UER in this region was reflected mainly in the rapid overall increase in low- and medium-gradient cities, exhibiting a pattern of internal diffusion. In the western region, the rightward shift of the dominant peak was relatively slow, and weak bimodality appeared during some periods, indicating relatively evident club-style differentiation. In addition, for visualization robustness, an alternative view of the three-dimensional kernel density estimates of UER is reported in Figure S1 of Supplementary Materials.

4.1.2. Spatial Distribution Patterns of UER

This research utilizes the comprehensive UER index selected for four benchmark years, namely 2006, 2011, 2017, and 2023. By applying the natural breaks classification, we categorize the index into five levels, and spatial visualization analysis is then conducted using ArcGIS Pro (version 3.6). The spatiotemporal patterns of UER in China are shown in Figure 6. In general, China’s urban UER pattern shifted from an earlier configuration in which northern cities held a relative advantage and inland hinterland areas lagged behind toward a more differentiated structure marked by contiguous enhancement along the eastern coast, high-value clustering in selected northern urban agglomerations, and continued catch-up in central urban agglomerations.

Specifically, in 2006, cities with high UER values were concentrated mainly in northern provinces such as Inner Mongolia, Heilongjiang, and Liaoning. Ecological cities with relatively low levels of urbanization, such as Ordos, Hulunbuir, and Zhangye, showed particularly strong performance. By 2011, regional central cities that implemented the NICPP relatively early, such as Shanghai, Wuhan, and Guangzhou, had taken a leading position in green innovation and had begun to enhance UER earlier than other cities did. High-value areas expanded from the northern advantaged regions and the eastern coast to the surrounding areas, while urban areas within eastern coastal provinces including Jiangsu, Zhejiang, and Guangdong experienced an overall increase in UER. By 2017, with the continued support of the NICPP, the eastern coastal region maintained its leading position in UER. Inland central cities such as Hefei, Zhengzhou, and Chengdu also became new growth poles of UER. In 2023, UER was further enhanced nationwide, but intraregional disparities remained evident. A relatively clear high-value UER corridor emerged along the eastern coast. In particular, medium- to high-value zones were generally shaped by the hub cities across three urban agglomerations: the Yangtze River Delta (YRD), the middle reaches of the Yangtze River (MRYZR), and the Guangdong Fujian Zhejiang Coastal (GFZC). Meanwhile, their outlying counterpart cities remained relatively weak. In contrast, except for a few nodal cities such as Chengdu and Chongqing, most cities in Southwest China had lower UER because of delayed policy adoption, limited innovation resources, and insufficient capacity for green technology R&D and transformation.

4.2. Baseline Regression Results

To minimize initial variations across the experimental and control groups, this paper applies entropy balancing before the baseline regression by imposing constraints on sample covariates, thereby improving between-group comparability.

Table 4. Covariate balance before and after entropy balancing.

Variables	Mean			Variance			Skewness
	Treated	Control		Treated	Control		Treated	Control
		Before	After		Before	After		Before	After
lnHCI	−3.775	−5.026	−3.775	0.8948	0.9403	0.8949	−0.08421	−0.6956	−0.08418
lnOPEN	−1.71	−2.977	−1.71	1.223	1.967	1.224	−0.3623	0.0399	−0.3663
lnGOV	−2.071	−1.657	−2.071	0.09714	0.1859	0.09717	0.07314	0.1032	0.07384
lnINV	−0.3214	−0.2345	−0.3214	0.4058	0.5078	0.4058	−1.097	−1.233	−1.097
lnER	−4.937	−4.975	−4.937	0.1483	0.2459	0.1483	−0.7241	−2.298	−0.7241
lnFIN	0.9343	0.6918	0.9343	0.2155	0.1496	0.2156	0.2987	0.3031	0.2995

reports the entropy balancing results.

The findings from the benchmark regression regarding the NICPP’s effect on UER are displayed in

Table 5. Baseline regression and other robustness test results.

Variables	TWFE	TWFE	EB–FE	EB–FE	EB–FE	Tobit	RE
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
NICPP	0.0346 ***	0.0302 ***	0.0244 ***	0.0229 ***	0.0150 ***	0.0307 ***	0.0309 ***
	(0.0036)	(0.0033)	(0.0060)	(0.0036)	(0.0021)	(0.0013)	(0.0033)
lnHCI		−0.0021		0.0059	0.0023	−0.0011	−0.0008
		(0.0020)		(0.0062)	(0.0035)	(0.0009)	(0.0020)
lnOPEN		0.0003		0.0033	0.0160 ***	0.0007	0.0008
		(0.0016)		(0.0035)	(0.0033)	(0.0006)	(0.0015)
lnGOV		−0.0421 ***			−0.0603 ***	−0.0427 ***	−0.0429 ***
		(0.0061)			(0.0067)	(0.0026)	(0.0058)
lnINV		0.0071 ***		0.0099 ***	0.0103 ***	0.0069 ***	0.0068 ***
		(0.0018)		(0.0027)	(0.0015)	(0.0007)	(0.0018)
lnER		−0.0020			−0.0006	−0.0020 **	−0.0019
		(0.0014)			(0.0023)	(0.0009)	(0.0014)
lnFIN		−0.0196 ***			−0.0280 ***	−0.0177 ***	−0.0170 ***
		(0.0056)			(0.0054)	(0.0024)	(0.0053)
Constant	0.2890 ***	0.2123 ***	0.3365 ***	0.3588 ***	0.2768 ***	0.1178 ***	0.1191 ***
	(0.0006)	(0.0182)	(0.0015)	(0.0252)	(0.0234)	(0.0111)	(0.0209)
City FE	√	√	√	√	√	√	√
Year FE	√	√	√	√	√	√	√
Obs.	4878	4878	4878	4878	4878	4878	4878
R2	0.9550	0.9619	0.9638	0.9635	0.9733
Wald test						1514.93 ***	215.442 ***

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects are included.

. The first two specifications report the TWFE estimates, where the former omits control variables. To further improve covariate balance, the next three specifications re-estimate the model based on entropy balancing weights (EB–FE), with Column (3) excluding all covariates and Column (4) jointly excluding ER, GOV, and FIN. The last two columns offer robustness assessments from the tobit specifications model and random effects, respectively.

Across all specifications, the coefficient for NICPP is persistently positive, reaching significance at the 1% threshold. In the preferred EB–FE specification in Column (5), the coefficient indicates that the NICPP significantly enhances UER by 0.015 units in China at the 1% level. Moreover, the coefficient in Column (4) remains significantly positive after excluding the potentially overlapping controls, confirming that the result is robust to alternative control-variable specifications. In contrast to other specifications, the estimates in Columns (3)–(5) are more conservative. This suggests that estimates obtained from unbalanced samples may be confounded by observable heterogeneity between treated and control cities. Overall, H1 is preliminarily supported.

4.3. Parallel Trend Test

To verify the benchmark findings and assess the validity of the DID specification, Figure 7 reports the parallel trend analysis through an event–study design. For the years preceding NICPP adoption, the statistical insignificance of coefficients prior to the NICPP implementation indicates a lack of pre-existing trends. After the policy takes effect, the estimated effects gradually become significant. Such results uphold the fundamental parallel trend assumption, thereby warranting the application of the DID framework in assessing the impact of the NICPP.

4.4. Testing Heterogeneous Treatment Effects Using Modern DID Methods

4.4.1. Dynamic Effects Based on CS–DID

This paper further employs the CS–DID event–study approach to address concerns regarding estimation bias caused by heterogeneous treatment. In this method, the counterfactual is constructed using only cities that have not yet entered the NICPP by the current period as the comparison group. Those that have already entered the NICPP are no longer used as controls for comparison. Estimates for the parallel trends assumption and dynamic impacts results based on the CS–DID are presented in Figure 8. Pre-policy coefficients exhibit no systematic deviation, indicating the absence of significant pretrends. In contrast, the impact in the year of policy implementation remains statistically negligible. Starting from the first period after policy implementation, the corresponding $AT{T}^{dyn}\left(l\right)$ exhibits a notable upward trend, progressively increasing throughout the subsequent five-year period. This finding indicates that NICPP significantly enhances UER, with certain time-lagged and cumulative effects.

4.4.2. Robustness Checks Using Other Modern DID Estimators

Following the results of the CS–DID event study, this paper further tests the robustness of the overall average treatment effect using the imputation-based DID estimator of Borusyak et al. [86] and the two-stage DID estimator of Gardner [87]. The imputation DID estimator estimates the average treatment effect after imputing counterfactual outcomes on the basis of untreated observations. The two-stage DID estimator first eliminates variations specific to groups and time periods followed by calculating the differential impact across the treated and control groups. Such methodologies mitigate identification bias arising from heterogeneous treatment effects, thereby enabling more precise estimation of treatment effects.

The first two specifications in

Table 6. Robustness checks using modern DID estimators.

Variables	CS–DID (Not-Yet-Treated)	CS–DID (Only Never-Treated)	Interpolated DID	Two-Stage DID
	(1)	(2)	(3)	(4)
NICPP	0.0157 ***	0.0165 ***	0.0123 ***	0.0350 ***
	(0.0051)	(0.0054)	(0.0020)	(0.0040)
Control variables	√	√	√	√
City FE	√	√	√	√
Year FE	√	√	√	√
Clustered SE	City	City	City	City
Obs.	4878	4878	4090	4878

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects or controls are included.

indicate that the NICPP boosts UER regardless of the control group specification (not-yet-treated vs. never-treated), yielding results substantially in line with the benchmarked evidence. Results from Columns (3) and (4) additionally validate that the promotion of UER by the NICPP is insensitive to the choice of modern DID estimators.

4.5. Placebo Test

A further robustness check is conducted to address potential omitted-variable bias by conducting a placebo analysis following Hagemann [88]. Specifically, while keeping the actual size of the treatment group unchanged, this research randomly selects sample of equal size to serve as the pseudotreatment group, while the rest serves as the pseudocontrol group. The first treatment years of the actual NICPP cities are then randomly assigned to the pseudotreated cities. Subsequently, we rerun the baseline model to capture the range of placebo estimates.

The data in Figure 9 are obtained from 1000 placebo test simulations. The coefficients on the randomly constructed placebo indicators cluster around zero, with most estimates failing to reach statistical significance. By comparison, the true policy coefficient clearly deviates from the center of the placebo distribution. These results confirm that the significant treatment influence established herein is not an artifact of randomized assignment or sample coincidence. Therefore, the baseline regression conclusions remain valid.

4.6. PSM–DID

Given the widespread use of the PSM–DID in assessing policy impacts, this paper further uses this method as an additional robustness check. Propensity scores are estimated using all baseline controls, and the sample is matched through nearest-neighbor, radius, and kernel matching.

Table 7. Robustness checks based on PSM–DID.

Variables	K-Nearest Neighbors Matching	Radius Matching	Kernel Matching
	(1)	(2)	(3)
NICPP	0.0146 ***	0.0108 ***	0.0128 ***
	(0.0021)	(0.0025)	(0.0023)
lnHCI	0.0052	0.0053	0.0048
	(0.0037)	(0.0042)	(0.0039)
lnOPEN	0.0167 ***	0.0180 ***	0.0169 ***
	(0.0034)	(0.0036)	(0.0035)
lnGOV	−0.0628 ***	−0.0650 ***	−0.0648 ***
	(0.0069)	(0.0082)	(0.0075)
lnINV	0.0096 ***	0.0113 ***	0.0107 ***
	(0.0016)	(0.0020)	(0.0018)
lnER	−0.0015	−0.0029	−0.0014
	(0.0024)	(0.0027)	(0.0025)
lnFIN	−0.0246 ***	−0.0215 ***	−0.0221 ***
	(0.0055)	(0.0068)	(0.0058)
Constant	0.2739 ***	0.2672 ***	0.2687 ***
	(0.0241)	(0.0288)	(0.0261)
City FE	√	√	√
Year FE	√	√	√
Obs.	4303	3091	3648
R2	0.9748	0.9787	0.9771

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects are included.

shows that the NICPP coefficients remain significantly positive across specifications, suggesting that the initial results are insensitive to the choice of matching method or sample composition.

In addition, to ensure reliability, detailed matching balance diagnostics are reported in Figure S2, Tables S1 and S2 of Supplementary Materials.

4.7. Comparative Validation and Sensitivity Analysis of the DPRAR Framework

Given that DPRAR extends DRAR by incorporating a pressure dimension, we conduct three analyses to assess its validity and robustness. First, we re-estimate the baseline model using a DRAR-based UER index constructed by excluding population density, urban construction land, energy consumption, and urban water use, while holding the sample, normalization procedure, and entropy-weighted TOPSIS method constant. Second, the weights of the four pressure indicators are decreased and increased by 20%, respectively, followed by renormalization and recalculation of UER. Third, the internally weighted pressure-dimension score is examined separately to determine whether the added dimension captures information that is not explicitly represented in DRAR and whether the composite result is mechanically driven by its constituent indicators.

Table 8. Comparative robustness and weight-sensitivity tests of the DPRAR framework.

Variables	DRAR–EB–FE	DP−20%RAR–EB–FE	DP+20%RAR–EB–FE
	(1)	(2)	(3)
NICPP	0.0154 ***	0.0157 ***	0.0158 ***
	(0.0041)	(0.0040)	(0.0039)
lnHCI	0.0037	0.0033	0.0030
	(0.0060)	(0.0060)	(0.0059)
lnOPEN	0.0167 **	0.0162 **	0.0158 **
	(0.0081)	(0.0079)	(0.0077)
lnGOV	−0.0620 ***	−0.0608 ***	−0.0595 ***
	(0.0134)	(0.0132)	(0.0130)
lnINV	0.0107 ***	0.0102 ***	0.0097 ***
	(0.0034)	(0.0033)	(0.0033)
lnER	−0.0010	−0.0012	−0.0013
	(0.0034)	(0.0034)	(0.0033)
lnFIN	−0.0271 ***	−0.0278 ***	−0.0283 ***
	(0.0085)	(0.0083)	(0.0081)
Constant	0.2716 ***	0.2745 ***	0.2781 ***
	(0.0469)	(0.0464)	(0.0459)
City FE	√	√	√
Year FE	√	√	√
Obs.	4878	4878	4878
R2	0.9749	0.9747	0.9743

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects are included.

reports the results using the DRAR-based measure and two weight-perturbation scenarios. All specifications are estimated using the baseline EB–FE model. When the pressure dimension is excluded, the estimated NICPP coefficient remains positive and statistically significant at the 1% level. The estimates under the ±20% weight perturbations are also highly consistent with the baseline DPRAR result. These results confirm that the DPRAR-based estimates of the NICPP effect are robust to alternative framework specifications and moderate perturbations of the pressure-indicator weights.

Table 9. Incremental information captured by the pressure dimension.

Variables	Pressure–TWFE	Pressure–TWFE	Pressure–EB–FE	Pressure–EB–FE
	(1)	(2)	(3)	(4)
NICPP	−0.0414 ***	−0.0374 ***	−0.0203 **	−0.0227 ***
	(0.0061)	(0.0056)	(0.0089)	(0.0060)
lnHCI		0.0132 ***		0.0237 ***
		(0.0030)		(0.0090)
lnOPEN		−0.0003		0.0106
		(0.0023)		(0.0107)
lnGOV		0.0156 *		−0.0019
		(0.0086)		(0.0166)
lnINV		0.0071 **		0.0166 **
		(0.0032)		(0.0066)
lnER		0.0066 ***		0.0069
		(0.0023)		(0.0053)
lnFIN		0.0093		0.0343 **
		(0.0086)		(0.0158)
Constant	0.8643 ***	0.9790 ***	0.7996 ***	0.9109 ***
	(0.0011)	(0.0252)	(0.0023)	(0.0731)
City FE	√	√	√	√
Year FE	√	√	√	√
Obs.	4878	4878	4878	4878
R2	0.9593	0.9613	0.9592	0.9637

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects are included.

reports separate regressions using the pressure subindex as the dependent variable to assess the incremental information contributed by the pressure dimension. Because higher pressure scores indicate lower resource and environmental pressure, the consistently negative and statistically significant NICPP coefficients suggest that the policy intensifies pressures related to population concentration, construction-land expansion, energy consumption, and water demand. The pressure dimension therefore captures a countervailing development cost not explicitly represented in the conventional DRAR framework, thereby indicating that the composite UER result is not mechanically generated by all constituent indicators moving uniformly in a favorable direction.

4.8. Additional Robustness Tests

To validate our core findings, we conduct supplementary robustness analyses by excluding potential policy-related confounding effects and replace the measurement for key indicators.

4.8.1. Excluding the Influence of Contemporaneous Policies

The NICPP was not implemented in isolation. Existing studies show that the Smart City Pilot Policy (SCPP) strengthens the adaptability of cities and catalyzes eco-friendly and low-carbon development via digital transformation [89]. Meanwhile, the Low-Carbon City Pilot Policy (LCCP), Carbon Emissions Trading Pilot Policy (CETPP), and National Ecological Civilization Pilot Zone (NECPZ) policy have been associated with low-carbon innovation, industrial structure upgrading, and improved ecological performance, and may therefore influence urban ecological resilience through partially shared channels [90,91,92]. These contemporaneous policies may overlap with the NICPP in terms of both implementation periods and treated cities. Consequently, failing to account for these policies may cause their effects to be incorrectly attributed to the NICPP.

To address this concern, we separately control for the SCPP, LCCP, CETPP, and NECPZ policies and then include all four policies simultaneously in the baseline model. All specifications are estimated using the EB−FE framework. As shown in

Table 10. Results after excluding the influence of other policy shocks.

Variables	SCPP−EB−FE	LCCP−EB−FE	CETPP−EB−FE	NECPZ−EB−FE	All−EB−FE
	(1)	(2)	(3)	(4)	(5)
DID	0.0149 ***	0.0149 ***	0.0153 ***	0.0162 ***	0.0160 ***
	(0.0040)	(0.0039)	(0.0040)	(0.0038)	(0.0036)
SCPP	0.0045				0.0049
	(0.0055)				(0.0055)
LCCP		0.0009			0.0011
		(0.0045)			(0.0043)
CETPP			0.0076		0.0021
			(0.0081)		(0.0073)
NECPZ				0.0207 **	0.0199 **
				(0.0089)	(0.0088)
lnHC	0.0024	0.0023	0.0027	0.0023	0.0025
	(0.0062)	(0.0061)	(0.0064)	(0.0062)	(0.0063)
lnOPEN	0.0157 **	0.0160 **	0.0159 **	0.0150 **	0.0146 **
	(0.0074)	(0.0076)	(0.0075)	(0.0075)	(0.0070)
lnGOV	−0.0601 ***	−0.0604 ***	−0.0606 ***	−0.0596 ***	−0.0596 ***
	(0.0131)	(0.0133)	(0.0129)	(0.0130)	(0.0130)
lnINV	0.0102 ***	0.0103 ***	0.0100 ***	0.0095 ***	0.0095 ***
	(0.0034)	(0.0033)	(0.0033)	(0.0032)	(0.0032)
lnER	−0.0007	−0.0005	−0.0005	−0.0014	−0.0015
	(0.0034)	(0.0033)	(0.0033)	(0.0033)	(0.0033)
lnFIN	−0.0281 ***	−0.0277 ***	−0.0248 ***	−0.0219 ***	−0.0210 **
	(0.0081)	(0.0084)	(0.0080)	(0.0082)	(0.0084)
Constant	0.2752 ***	0.2761 ***	0.2731 ***	0.2648 ***	0.2616 ***
	(0.0467)	(0.0469)	(0.0466)	(0.0466)	(0.0475)
City FE	√	√	√	√	√
Year FE	√	√	√	√	√
Obs.	4878	4878	4878	4878	4878
R2	0.9733	0.9733	0.9734	0.9739	0.9740

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects are included.

, the estimated coefficients of the NICPP remain positive and statistically significant at the 1% level across all specifications. In particular, after simultaneously controlling for all four contemporaneous policies, the NICPP coefficient remains significantly positive at 0.0160. Overall, these findings confirm that the estimated effect of the NICPP is not driven by other policies implemented during the same period.

4.8.2. Alternative Variable Measurement Method

To assess whether the benchmark findings are sensitive to alternative UER indicators, UER is reconstructed in this paper using principal component analysis, and the alternative indicator is denoted as UER_PCA. The estimations presented in

Table 11. Robustness checks using an alternative measure of UER.

Variables	UER_PCA–FE	UER_PCA–FE	UER_PCA–EB–FE	UER_PCA–EB–FE
	(1)	(2)	(3)	(4)
NICPP	0.0113 ***	0.0077 **	0.0160 **	0.0113 ***
	(0.0041)	(0.0036)	(0.0063)	(0.0041)
lnHCI		0.0075 **		0.0105
		(0.0032)		(0.0087)
lnOPEN		−0.0005		−0.0039
		(0.0022)		(0.0049)
lnGOV		−0.0416 ***		−0.0430 ***
		(0.0080)		(0.0108)
lnINV		0.0045 **		0.0056
		(0.0021)		(0.0041)
lnER		0.0034		0.0060 *
		(0.0029)		(0.0034)
lnFIN		−0.0411 ***		−0.0606 ***
		(0.0077)		(0.0186)
Constant	0.4579 ***	0.4671 ***	0.5458 ***	0.5793 ***
	(0.0007)	(0.0284)	(0.0016)	(0.0624)
City FE	√	√	√	√
Year FE	√	√	√	√
Obs.	4878	4878	4878	4878
R2	0.9694	0.9740	0.9796	0.9849

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects are included.

indicate that UER_PCA consistently yields significantly positive impacts across different specifications with and without control variables. This demonstrates that the conclusions of the baseline regression model do not depend on the specific measurement method of UER.

5. Further Analysis

5.1. Mechanism Analysis

The benchmark estimation reveals that NICPP enhances UER. Furthermore, urban GTI reflects both the quantitative growth of innovation outputs and the qualitative upgrading of innovation structures [27]. Drawing on Jiang’s [93] mechanism analysis framework, this paper further examines whether GTI constitutes a potential channel through which NICPP improves UER, thereby testing H2. However, green patent activity may also be shaped by a city’s pre-existing innovation capacity, industrial structure, and other local policy interventions. To mitigate potential contemporaneous reverse causality among the variables, this study uses UER, GTIN, and GTIQ measured one period after NICPP implementation in the mechanism analysis.

The regression equation is expressed as:

$$UE{R}_{i,t+1}={\theta }_0+{\theta }_{1}NICP{P}_{it}+{\displaystyle \sum _{k=1}^K{\theta }_k}Control{s}_{kit}+{\mu }_i+{\nu }_t+{\epsilon }_{i,t+1}$$

(12)

$$GT{I}_{i,t+1}={\gamma }_0+{\gamma }_{1}NICP{P}_{it}+{\displaystyle \sum _{k=1}^K{\gamma }_k}Control{s}_{kit}+{\mu }_i+{\nu }_t+{\eta }_{i,t+1}$$

(13)

where $GT{I}_{it+1}$ includes GTIN and GTIQ. Equation (12) estimates the effect of NICPP on one-period-ahead UER, whereas Equation (13) estimates its effect on one-period-ahead GTI. The remaining indicators are identical to the baseline.

Table 12. Test results for the mechanism effect.

Variables	UERt+1–FE	GTINt+1–FE	GTIQt+1–FE	UERt+1–EB–FE	GTINt+1–EB–FE	GTIQt+1–EB–FE
	(1)	(2)	(3)	(4)	(5)	(6)
NICPP	0.0301 ***	0.5099 ***	0.0529 ***	0.0156 ***	0.3637 ***	0.0461 **
	(0.0033)	(0.0684)	(0.0108)	(0.0043)	(0.0689)	(0.0182)
lnHCI	−0.0020	−0.2212 ***	−0.0263 *	0.0016	−0.7042 ***	−0.0900 **
	(0.0020)	(0.0569)	(0.0136)	(0.0063)	(0.1453)	(0.0369)
lnOPEN	0.0000	−0.0694 **	0.0132	0.0142 *	−0.1196	0.0266 *
	(0.0016)	(0.0301)	(0.0088)	(0.0078)	(0.0813)	(0.0149)
lnGOV	−0.0399 ***	−0.0648	0.0185	−0.0563 ***	0.0289	−0.0399
	(0.0059)	(0.0886)	(0.0245)	(0.0120)	(0.2221)	(0.0442)
lnINV	0.0081 ***	−0.0902 **	0.0169 **	0.0117 ***	−0.2118 **	0.0313 **
	(0.0018)	(0.0348)	(0.0078)	(0.0036)	(0.1003)	(0.0132)
lnER	−0.0012	−0.1221 ***	−0.0213 **	0.0027	−0.3895 **	−0.0212
	(0.0014)	(0.0322)	(0.0095)	(0.0032)	(0.1938)	(0.0151)
lnFIN	−0.0238 ***	−0.3251 ***	−0.0245	−0.0345 ***	−0.6569 ***	−0.0445
	(0.0056)	(0.0891)	(0.0269)	(0.0091)	(0.1887)	(0.0330)
Constant	0.2279 ***	−1.4123 ***	0.2762 ***	0.3064 ***	−3.4441 ***	−0.0131
	(0.0179)	(0.4747)	(0.0938)	(0.0431)	(1.1529)	(0.2125)
City FE	√	√	√	√	√	√
Year FE	√	√	√	√	√	√
Clustered SE	City	City	City	City	City	City
Obs.	4607	4607	4607	4607	4607	4607
R2	0.9647	0.7563	0.4167	0.9747	0.8703	0.5693

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects are included.

reports the results obtained from the TWFE specifications in Columns (1)–(3) and the EB–FE specifications in Columns (4)–(6). The estimates show that NICPP has a positive effect on GTIN at the 1% significance level in both specifications. Its effect on GTIQ remains positive and is significant at the 1% level in the TWFE model and at the 5% level in the EB–FE model.

NICPP can shape UER in multiple ways through GTI. From the production side, the expansion of urban green technology supply and application reserves allows technologies, such as waste heat recovery, process energy-efficiency retrofitting, and industrial carbon-reduction technologies to be widely incorporated into production processes [94,95]. Such technologies are effective in helping lower energy use and the carbon intensity of production, thereby enhancing the resistance capacity of UER [15]. In terms of governance, scaling up the GTIN makes it easier for cities to transform green innovation into sustained ecological governance capacity. It enhances the stability and adaptability of UER by refining the performance of restorative interventions and pollutant containment, while effectively neutralizing the threats posed by the spatial propagation of contaminants. [96]. From the structural perspective, the increase in GTIQ helps promote the restructuring of industrial chains toward low-carbon, circular, and clean development, thereby optimizing the distribution efficiency of ecological resources [97]. The knowledge spillovers and resource allocation effects generated by green industrial agglomeration further amplify ecological benefits and enhance the long-term adaptive and recovery capacities of UER [98,99]. Therefore, Hypothesis H2 is verified, and Hypotheses H2a and H2b are also supported.

5.2. Heterogeneity Analysis

The results of this paper show that the NICPP significantly enhances UER, and these findings are supported by various robustness checks. However, the UER of China and its driving factors exhibit significant regional differences [17]. This paper further examines heterogeneity along three dimensions: geographical location, resource endowment, and environmental regulation intensity.

5.2.1. Regional Heterogeneity

China has a vast territory, and different regions vary in their capacity for innovation transformation and ecological governance foundations. Columns (1)–(4) of

Table 13. Results of the heterogeneity analysis.

Variables	Region				Resource Endowmen		Environmental Regulation
	Eastern	Central	Western	Northeast	Resource-Dependent	Non-Resource-Dependent	ER ≥ 50%	ER < 50%
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
NICPP	0.0135 ***	0.0218 ***	0.0194 **	0.0113	0.0160 ***	0.0142 ***	0.0182 ***	0.0157 **
	(0.0044)	(0.0048)	(0.0076)	(0.0073)	(0.0059)	(0.0048)	(0.0041)	(0.0060)
Constant	0.2738 ***	0.2385 ***	0.3495 ***	0.2991 ***	0.2270 ***	0.2843 ***	0.3162 ***	0.2478 ***
	(0.0536)	(0.0807)	(0.0665)	(0.0502)	(0.0774)	(0.0561)	(0.0441)	(0.0663)
Control variables	√	√	√	√	√	√	√	√
City FE	√	√	√	√	√	√	√	√
Year FE	√	√	√	√	√	√	√	√
Obs.	1530	1422	1368	558	1944	2934	2448	2430
R2	0.9808	0.9586	0.9814	0.9827	0.9658	0.9735	0.9822	0.9684

Notes: Heteroscedasticity-robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. √ indicates that the corresponding fixed effects or controls are included.

report the estimation results for the four geographical regions. We observe that UER improves significantly following NICPP implementation in all regions except the Northeast. The magnitude of this enhancement is most pronounced centrally, followed by the western and eastern regions. Historically burdened by path dependence on heavy manufacturing, demographic shrinkage, and subdued market activity, Northeastern China faces severe structural hurdles. These obstacles fundamentally hinder the policy’s innovation incentives from translating into actual ecological upgrades. Overall, the evidence confirms stark regional heterogeneity in how the NICPP shapes UER.

5.2.2. Heterogeneity by Urban Resource Endowment

Resource endowments shape urban development trajectories and may influence how effectively innovation policies are converted into ecological outcomes. Columns (5) and (6) of Table 13 demonstrate that the NICPP generates a stronger marginal ecological governance effect in more resource-dependent areas. Driven by historical inertia, resource-based cities often experience structural lock-in, resulting in continued dependence on capital-intensive and resource-intensive industries. As a result, their green transition depends more on external institutional incentives and technological substitution [100].

5.2.3. Heterogeneity by Environmental Regulation Intensity

Environmental regulation intensity may alter the ecological effect of the NICPP by influencing the allocation direction of innovation resources. In this paper, cities are grouped according to their baseline environmental regulation intensity, and the 50th percentile is used as the cutoff to classify cities into cities into groups with stronger and weaker environmental regulation to examine the NICPP effect under different regulatory contexts. Columns (7)–(8) of Table 13 reveal that the positive effect of the NICPP is substantially more pronounced in cities subject to stricter environmental oversight. These empirical outcomes align with the Porter hypothesis as well as the theory of induced innovation. Stronger environmental regulation increases the cost of pollutant emissions, strengthens constraints on firms’ green R&D and clean production, and guides more innovation resources toward energy conservation, green process substitution, and pollution control. Therefore, where environmental enforcement is more rigorous, the NICPP integrates more effectively with broader ecological frameworks, ultimately bolstering the stability and sustained adaptive capacity of urban ecosystems [23].

6. Discussion

6.1. Spatiotemporal Differences in the Evolution of Urban Resilience

Empirical analysis reveals that UER in China continued to improve from 2006 to 2023. However, this process did not show simple convergence. Instead, it exhibited an evolutionary pattern in which overall improvement coexisted with gradient differentiation. Zhong et al. [17], Zhou et al. [101], and Lan et al. [15] reached similar conclusions on the basis of the PSR, RAR, and DRAR frameworks, respectively. Among them, the DRAR framework strengthens the explanation of cities’ socioeconomic development foundations. It is therefore more likely to identify common advantages among regions with similar development bases, but it may also weaken the representation of differences caused by resource environmental constraints within highly developed areas. For example, Lan et al. [15] reported that the YRD urban agglomeration in 2018, alongside the peripheral areas of the MRYZR and the GFZC urban agglomeration in 2022, showed continuous and relatively balanced high-value distributions. In contrast, on the basis of the five-dimensional DPRAR framework with 21 indicators, the YRD urban agglomeration in 2017 showed mainly clustered agglomeration and partial contiguous expansion driven by core cities rather than a clearly balanced distribution. By 2023, the peripheral areas of the MRYZR and the GFZC urban agglomeration mainly exhibited belt-like extension along regional development axes and partial contiguous clustering, while clear internal hierarchical differences remained. These results imply that incorporating a pressure metric better captures the growth bottlenecks and disparities in high-UER cities, thereby further improving the explanatory power of the UER measurement framework.

6.2. Impact of the NICPP on UER and Its Mechanism Explanation

In this paper, the NICPP significantly enhances urban UER by 0.0150 units. This finding indicates that its benefits extend beyond innovation output itself and further reach the stable operation and risk response of urban ecosystems. Through encouraging green technology accumulation and advancing pollution control and green transition, the NICPP gradually enhances the UER.

These findings extend existing NICPP studies. Current research on the NICPP focuses mainly on innovation output, technological spillovers, and economic growth effects, while discussions of environmental and ecological dimensions remain relatively limited [26,29,102]. Although recent scholarship has tentatively addressed its effects on ecological sustainability, they focus mostly on identifying static outcomes [8]. Insufficient attention has been given to the dynamic evolutionary processes that underpin the development of resilience in urban ecosystems facing external shocks and disturbances. The pressure-subindex results show that the NICPP does not eliminate development-induced resource and spatial pressures. Nevertheless, unlike studies emphasizing the rebound effects of innovation policies, our findings suggest that the NICPP partially contains such pressures through performance appraisal, green innovation incentives, and interdepartmental coordination. Consistent with the Porter hypothesis, these institutional arrangements facilitate innovation compensation and efficiency gains, enabling technological progress to generate a positive net effect on UER despite the accompanying expansion in resource demand [103]. Overall, this paper further extends both the UER measurement framework and the policy identification strategy, thereby generating more detailed empirical evidence on the ecological consequences of the NICPP.

Unlike existing studies that mostly explain the GTI mechanism of the policy from the perspective of green patent application counts [104,105], this paper divides GTI, one of the most important innovation outputs of the NICPP, into two dimensions: urban GTIN and urban GTIQ. According to our estimations, the NICPP’s capacity to improve UER is largely realized by the expansion and accelerated diffusion of the GTIN. In contrast, improving the GTIQ usually requires a longer period of R&D accumulation, factor agglomeration, and institutional support. Its effect is more likely to serve as medium- and long-term support for UER [106,107]. These findings echo those of this paper’s dynamic effect analysis, suggesting that the lagged and cumulative effect of the NICPP on UER is most likely associated with the continued accumulation and diffusion of GTI.

6.3. Heterogeneity of UER Across Different Types of Cities

Existing studies have not sufficiently discussed how the NICPP affects UER differently across urban contexts. In this paper, subgroup regressions are categorized according to spatial geography, resource endowment, alongside environmental regulation intensity. Fu et al. [108] demonstrated that major urban centers with superior administrative status situated east of the Hu Huanyong Line reap the most pronounced resilience benefits from the NICPP. In contrast, we divide the sample into four regions and identifies higher marginal returns of the NICPP for UER in the central region. This may be because urban resilience emphasizes the comprehensive pressure-bearing and recovery capacity of urban systems, whereas UER focuses more on the regulation, restoration, and reorganization of urban ecosystems under sustained pressure. UER is more sensitive to the ecological transformation of innovation outputs. These findings can also be explained by gradient development theory and core–periphery theory. The eastern region occupies a “core” position, with abundant innovation resources and a high concentration of advanced factors. It also has a stronger foundation for green transition. NICPP can steadily increase UER in this region, but its marginal gains are limited [109]. The central region absorbs some of the industrial transfer and carbon spillover, but it also has strong locational advantages, industrial support capacity, and technological absorption capacity [110]. Moreover, its original UER level and green governance capacity still have considerable room for improvement. These conditions make it easier for NICPP-induced innovation incentives to generate ecological benefits and enhance UER within Central China. Western China has absorbed more transfers of high-carbon and pollution-heavy industries [111]. Moreover, the imperfect mechanism for allocating cross-regional environmental responsibilities has led to the redistribution of pollution burdens across regions, exacerbating the overall vulnerability of ecosystems in the western region [15,112]. Therefore, the western region’s relatively low baseline UER and greater ecological transition needs leave more room for NICPP-induced improvements than in the eastern region, although the effect remains weaker than that observed in central China. This may be because constraints on talent, capital, and innovation networks limit the full translation of policy incentives into UER gains.

For resource-dependent cities, long-term dependence on mining activities, energy development, and the petrochemical sectors translates into a greater accumulation of ecological pressure, stronger resource consumption constraints, and tighter pollution emission constraints [113]. For instance, in cities dominated by metal mining and smelting, such as Jinchang, as well as petroleum-dependent cities like Dongying and Daqing, the secondary sector has consistently accounted for over 50% of GDP in recent years. However, these cities face a distinct “upgrading dilemma” under stringent environmental regulations: compliance pressures force capacity cuts, while outdated production technologies hinder green transitions, leading to chronic underinvestment in ecological restoration and rising unemployment [114]. Weak regulatory enforcement further incentivizes superficial compliance—firms either cut environmental spending or shift pollution to less monitored areas, exacerbating ecological degradation. By contrast, although non-resource-dependent cities have more diversified economies, they remain dominated by conventional manufacturing. Rising environmental compliance costs compress profit margins, crowding out R&D budgets for green technologies [115]. Meanwhile, under persistent pressure to maintain growth, local governments tend to relax environmental enforcement, further undermining regulatory effectiveness. These cities can obtain more significant marginal pollution-reduction effects through the NICPP, thereby alleviating pressure on UER and strengthening its resistance capacity. Their more diversified industrial structure, however, means that innovation demand and investment are more dispersed, so the NICPP may not necessarily direct innovation activities toward ecological governance-related fields. Consistent with previous studies, within cities enforcing stricter environmental oversight, municipal administrations are more inclined to leverage stimulus measures to foster innovation and secure major strides in green sustainability [8].

7. Conclusions and Implications for Policy

7.1. Main Findings

On the basis of the DPRAR framework, this research formulates a holistic UER measurement matrix encompassing five core components: driving factors, pressure, resistance, adaptability, and resilience. Using a panel dataset of 271 cities spanning the period from 2006 to 2023, we evaluate the impact and dynamic effects of the NICPP on UER and applies several modern DID methods suitable for staggered policy evaluation to conduct key robustness checks. In addition, this study further explores the potential mechanism roles of GTIN and GTIQ, and then evaluates how the policy impacts vary among municipalities with distinct attributes. The primary findings are summarized below:

(1)

From 2006 to 2023, UER in Chinese cities continued to increase overall, but regional differentiation remained evident. The spatial evolution pattern was characterized by coastal enhancement, nodal agglomeration, and catch-up in the central region.

(2)

The NICPP significantly enhances urban UER. Dynamic effect analysis further revealed that this effect exhibited certain time-lagged and cumulative characteristics.

(3)

The mechanism analysis reveals that the NICPP can further positively affect UER by increasing GTIN and improving GTIQ.

(4)

A further exploration of heterogeneity reveals that the positive contribution of the NICPP toward UER is stronger across Central and Western China, resource-dependent areas, and cities with stronger environmental regulation.

7.2. Policy Implications

On the basis of the above findings, several policy suggestions are put forward:

(1)

Policy makers should refine the policy design and performance targets of the NICPP. Pilot implementation should avoid placing excessive emphasis on innovation output, economic growth, and industrial expansion, while overlooking urban ecosystem carrying capacity and the quality of green development. In addition to existing indicators of innovation capacity and development performance, the assessment system could incorporate more binding indicators related to resource use efficiency, pollutant emission intensity, green and low-carbon transition outcomes, ecological space protection, and environmental governance performance. Moreover, integrating the NICPP more closely with broader ecological goals is essential. This involves weaving the “dual carbon” agenda, ecological civilization construction, and the push for new quality productive forces into the very fabric of innovative city design. Such alignment would help innovative city construction better coordinate innovation-driven development, green transition, and ecological improvement, which contributes institutional support to the long-term improvement of UER. More broadly, developing countries may draw lessons from the NICPP when designing innovation-oriented institutions to reduce the risk of resource-intensive rebound effects while balancing innovation-driven development with ecological sustainability.

(2)

The support structure of the NICPP should be optimized with green technological innovation (GTI) as a key transmission channel, coordinating improvements in both GTIN and GTIQ. GTI should not be assessed by patent volume alone, but also by the share of high-quality invention patents, technology transformation efficiency, and real-world ecological outcomes. Policy support should therefore target key scenarios such as pollution control, energy efficiency improvement, ecological restoration, and green infrastructure construction to strengthen the linkage between green R&D, technology transformation, and urban governance needs, thereby increasing the share of invention patents within overall green technological innovation. By encouraging original R&D over incremental modifications, this approach improves the structure of green innovation and enhances cities’ capacity to recover from and adapt to ecological shocks.

(3)

Given the heterogeneity of urban conditions in China, the NICPP should strengthen differentiated policy implementation and targeted support. The government should classify pilot entry thresholds, assessment indicators, and dynamic exit mechanisms according to cities’ innovation foundations, ecological pressures, and governance capacities to avoid simple expansion and homogenized implementation. At the regional level, the eastern region should shift from “incremental expansion” to “quality spillovers”, with a focus on strengthening breakthroughs in key green technologies, cross-city technology commercialization, and experience diffusion. The central region should continue to maintain the policy advantages of the NICPP. Given that the western region faces substantial resource and environmental constraints while increasingly engaging in industrial gradient relocation, local governments should strengthen green innovation infrastructure and improve mechanisms for technology transfer and commercialization. Relevant authorities should promote policy synergy among the NICPP, environmental access regulation, green finance, and industrial upgrading policies, thereby enhancing the GTI and the capacity for green industrial transformation. In the northeastern region, the NICPP should be integrated with the green renewal of old industrial bases and the reconstruction of innovation systems, while obsolete capacity should be withdrawn through stricter regulatory measures. On the basis of the development characteristics of different cities, NICPP policy resources should be prioritized for resource-dependent cities and cities with stronger environmental regulation. The former should focus on clean production transformation, resource recycling, and ecological restoration in mining areas, whereas the latter should strengthen the synergy between innovation incentives and green constraints. For cities with weak environmental regulation foundations that have already been included in the pilot program, shortcomings in governance standards, monitoring capacity, and coordinated enforcement should be addressed as soon as possible.

Despite these contributions, there are still some limitations in the research. Primarily, the empirical lens remains fixed at the macroscopic level. Although we analyzed 271 cities from 2006 to 2023, the absence of county-level metrics precludes a deeper exploration of localized spatial disparities. Additionally, the UER evaluation system is constrained by data availability, as several sub-indicators required for the DPRAR index remain unpublished for 2024 and 2025, limiting the timeliness of our empirical findings. Methodologically, while the staggered DID design effectively addresses observable selection bias and heterogeneous treatment effects, it relies on linear functional-form assumptions that may oversimplify the complex, nonlinear interactions inherent in urban social-ecological systems. Future research could therefore introduce machine learning approaches—such as random forest or gradient boosting—to complement the traditional DID framework, while also refining the UER indicator system and extending the temporal coverage as more data become available, so as to more fully capture the dynamics of UER evolution.