This study takes the panel data of 285 cities in China as the research object, with a time span of 2008–2020. The data are derived from ‘China Urban Statistical Yearbook’, ‘National Statistical Yearbook’, ‘China Health and Family Planning Statistical Yearbook’, ‘China Energy Statistical Yearbook’, ‘China Labor Statistical Yearbook’, ‘China Social Statistical Yearbook’, ‘China Environmental Statistical Yearbook’, ‘China Population and Employment Statistical Yearbook’, provincial statistical yearbooks, municipal statistical yearbooks and statistical bulletins of national economic and social development of cities. Use the interpolation method of SPSS software to fill in missing values.

Sample selection in this study, in terms of regions, strictly adheres to the classification of China’s eastern, central, western, and northeastern regions by the National Bureau of Statistics. And the initial sample covers 293 prefecture-level cities across all regions of China. Considering the availability and continuity of data, this study excluded cities with substantial missing data, such as Lhasa City, Changdu City, Nyingchi City, Sansha City, Danzhou City, and Bijie City. The final sample comprises 285 prefecture-level cities in China, including 87 in the Eastern Region, 80 in the Central Region, 84 in the Western Region, and 34 in the Northeast Region. The complete list of cities is provided in Supplementary Appendix 1. Furthermore, based on the Official Notice on Adjusting the Criteria for Classifying City Sizes issued by the State Council of China and the Seventh National Population Census, this study classified the sample cities by size within the aforementioned research scope, as presented in Table 1.

In addition, sample characteristics are critical considerations when assessing spatial effects. To avoid their impact on the measurement of spatial effects, three weight matrices—ARWM, GDWM, and EDWM—were constructed for model estimation, ensuring the stability of spatial coefficients and coefficients of key independent variables. Second, control variables were incorporated to eliminate the influence of non-spatial effects. Third, a variance inflation factor was used to conduct a multicollinearity test, which mitigated pseudo-regression disturbances and further ensured result stability. Finally, the same sample was employed in previous research to verify the spatial heterogeneity of urban green productivity via group regression (Hou et al., 2024). Thus, based on the above work, the sample selection in this study does not affect the estimation of the overall spatial effect of urban green development.

From the perspective of productivity, urban green development pays more attention to the fundamental driving forces of urban development. For instance, the health level of the workforce directly affects production efficiency. This study starts from the perspective of productivity, based on the composite system theory of “economy society environment energy health” and sustainable development theory. Building on previous research, the fuzzy rough set theory and methods are used to quantitatively screen the measurement index system of urban green development from the economic, environmental, energy consumption, social security, and health levels, in order to avoid the influence of noise on the indicators, clarify the object membership relationship, and reduce information loss. At the same time, this method can more accurately capture the complexity and variability of urban green development. The specific steps include: Firstly, standardizing the data to avoid the influence of dimensionality on the indicators (Yuan et al., 2025). Two equations are defined. Positive indicator : E i j = X i j − min 1 ≤ i ≤ m X i j max 1 ≤ i ≤ m X i j − min X i j Negative indicator : E i j = max 1 ≤ i ≤ m X i j − X i j max 1 ≤ i ≤ m X i j − min X i j

Among them, E i j is the standardized value of the j-th index of the i-th evaluation object, and X i j is the value of the j-th index of the i-th evaluation object; m is the number of evaluated objects. Secondly, fuzzy similarity classes are calculated. The classical rough set requires the conversion of continuous data into discrete data when processing it, which leads to information loss. Although the fuzzy set itself retains information, the final decision-making process requires the “defuzzification” of fuzzy membership degrees; if this process is not properly executed, information loss will also occur. This study employs the fuzzy similarity relationship in fuzzy rough sets, which not only retains the complete information of continuous data—thus avoiding discretization—but also minimizes information loss. In this study, the similarity of 0.3 was selected as the critical value, which has been widely used in previous studies (Guo and Zhu, 2007). In theory, the setting of α refers to the extent to which a degree of membership qualifies as belonging or relevance. Setting α = 0.3 ensures the inclusion of the maximum amount of relevant information, thereby reducing the risk of omission; it also includes any object with a similarity level of at least 30% to the target in the initial consideration set. Thirdly, through the lower approximation operation, the neighbor information is aggregated to generate a new and more robust membership degree for each object, which can clarify the fuzzy relationship and generate a more reliable and more suitable measurement for decision-making, so as to reduce the influence of the model on noise sensitivity. In this study, the accuracy value of 0.9 was set as the critical value (Zhang and Chi, 2012), meaning that when the new classification results achieve a 90% confidence interval, the new classification is deemed to adequately reflect the information of the original classification. Theoretically, a 90% confidence level in statistics represents a critical point that achieves an optimal balance between rigor and information retention, which is more flexible than 95% or 99% (thereby preserving more useful information) while stricter than 80% or 70% (effectively filtering out most noise). This ensures that even if a small number of objects are misclassified due to noise interference, such objects can still be correctly assigned to the lower approximation set as long as most similar objects are classified correctly. This significantly enhances the model’s robustness and improves its adaptability to imperfect data. The setting of these two threshold parameters enables the use of a low standard to establish a broad indicator candidate pool and a strict high standard to extract indicators. Specifically, the low standard minimizes the risk of missing important indicators, while the high standard enhances the reliability of the final indicators; this dual-standard approach thus strikes an optimal balance between retaining valuable information and filtering out harmful noise. Finally, the approximate classification quality coefficient is calculated. 16 green development evaluation indicators were obtained, as shown in Figure 2. The specific steps can be found in relevant literature (Hou et al., 2024; Cheng et al., 2018).

Based on this, the entropy weight method was used to calculate the weights of the indicators in this study. Firstly, the proportion of the j-th index value of the i-th evaluation object is calculated. Secondly, the entropy value of the index is calculated. Finally, the entropy weight of the index is calculated.

From the perspective of adjacency relationship, geographical distance and economic distance, this study constructs the spatial weight matrix of urban green development on the basis of extracting urban coordinates. This matrix helps enhance the scientific rigor and accuracy of the evaluation results. It also offers targeted strategies and novel approaches for enhancing urban green development. Specifically, ARWM reflects the spatial adjacency relationship, which can appropriately reflect the spatial interaction relationship of green development among cities. It is set as a significant interaction between cities bordering each other, and the interaction between non-adjacent cities is not significant. Specifically, the matrix is set as a symmetric matrix with the main diagonal element being 0, and the elements in the matrix indicate whether two cities are geographically adjacent, if adjacent, then the weight is 1, if not, then the weight is 0. The GDWM is constructed based on the longitude and dimension of each city. D _ij represents the geographical distance between city i and city j. The closer the distance is, the stronger the spatial autocorrelation effect is. The EDWM is obtained by calculating the reciprocal of the mean GDP gap, and the greater the gap between two cities, the greater the weight. Q i j = 1   R e g i o n   i   i s   a d j a c e n t   t o   r e g i o n   j . 0   R e g i o n   i   i s   n o t   a d j a c e n t   t o   r e g i o n   j . Q i j = 1 d i j   i ≠ j 0   i = j Q i j = 1 G p i ¯ − G p j ¯   i ≠ j   0   i = j G p i ¯ is the value-added GDP of the i-province during the study period.

The global spatial autocorrelation reveals the spatial pattern of the entire study area, and the Moran’s I index is used to characterize the degree of autocorrelation in that area. The local spatial correlation index can reveal the similarity or correlation between the attribute feature values of spatial reference units and their adjacent spatial units, identify spatial clustering and spatial isolation features, and detect spatial heterogeneity. Therefore, this study selected Moran’s I index to measure the spatial autocorrelation effect of urban green development, and the spatial autocorrelation model is defined: G l o b a l   s p a t i a l   c o r r e l a t i o n : M o r a n ′ s I = ∑ i = 1 n ∑ j ≠ i r Q i j x i − x ¯ x j − x ¯ T 2 ∑ i = 1 n ∑ j = 1 r Q i j L o c a l   s p a t i a l   c o r r e l a t i o n : M o r a n ′ s I i = n x i − x ¯ ∑ j ≠ 1 Q i j x j − x ¯ ∑ i x i − x ¯ 2

Among them, x i and x j represents the urban green development index, x ¯ represents the mean of each city’s green development index, T 2 represents the mean square error, Q i j represents the spatial weight matrix, and I i represents the product of the observed value at spatial position i and the weighted values of adjacent observations. The range of Moran’s I values is [-1, 1]. When Moran’s I > 0, it indicates the existence of spatial positive correlation characteristics in urban green development; When Moran’s I < 0, it indicates that there is a spatial negative correlation in urban green development.

According to different representations of spatial overflow, spatial econometric models can be divided into Spatial Lag Models (SLM), Spatial Error Models (SEM), and SDM. The choice of different spillover models has a significant impact on the results. In order to more accurately verify the spillover effects of various driving factors on urban green development, we tested the models. Specifically, this study first uses the Lagrange Multiplier (LM) test in the Ordinary Least Squares (OLS) model to determine which spatial panel regression model is most suitable. Secondly, we used Hausman test to determine whether to use fixed effects or random effects, and verified the robustness of the spatial regression model through Likelihood Ratio (LR) test. We also determined that the SDM is the most suitable and cannot be converted into SEM or SLM. The model test results are shown in Table 2.

As shown by the LM test, its significance indicates that the presence of either a spatial error effect or a spatial lag effect. This warrants rejecting the null hypothesis of OLS, making it appropriate to adopt a spatial econometric model. Additionally, the LM test reveals that under the ARWM and GDWM, both the spatial error effect and spatial lag effect tests are significant, thus rejecting the “no spatial correlation” hypothesis. Although none of the tests under the EDWM passed the “no spatial correlation” test, the results of the Moran’s I index test confirm that urban green development exhibits both spatial error effects and spatial lag effects. We therefore conclude that the spatial panel regression model is the most appropriate for this study. Second, the Hausman test shows that the p-values for all weight matrices are significant. Accordingly, the fixed-effects design captures more spatial interaction effects and fixed-effects model yields the optimal performance. Thirdly, the Wald test and LR test results showed that the P values of LR passed the significance test (P = 0.000) regardless of the spatial weight matrix, indicating that SDM had higher goodness of fit and effect than SLM and SEM. The SDM model can provide more information and cannot be replaced by SLM and SEM. In addition, the SDM includes both the spatial lag term of the dependent variable and that of the independent variable, and exhibits superior adaptability and explanatory power across various spatial weight matrices. In contrast, the SEM is limited to the spatial autoregressive term for the error component, and the SLM is restricted to the spatial lag term of the dependent variable. Given the test results confirming the presence of spatial effects, this study integrated spatial correlation and spatial spillover effects, with particular consideration of the impacts from endogenous interaction, exogenous interaction, and error term interaction. Accordingly, a spatial Durbin model for urban green development was innovatively developed, which offers a comprehensive perspective for analyzing urban green development. Combined with the research object and related variables of this study, the SDM is defined: g p i t = ρ Q i j g p i t + β 1 e d l i t + β 2 h c i t + β 3 i s i t + β 4 i l i t + β 5 i c i t + β 6 g i i t + β 7 d f t i t + β 8 e r i t + λ Q i j X i t + μ i + σ i + ε i t

In the equation, g p i t represents the green development level of the i-th city in period t. Q i j g p i t represents the spatial lag term of the dependent variable, which describes the endogenous interaction effect between the dependent variables. ρ ∈ − 1 , 1 is a spatial autoregressive coefficient, which is used to describe the interaction between the dependent variables of the research object and the dependent variables of the adjacent object. Q i j is a spatial weight matrix, which quantifies the spatial dependence between different cities. This study mainly involves the ARWM, the GDWM and the EDWM. β is the estimated parameter of the independent variable, that is, the regression coefficient of the independent variable. X i t is the observation value of the independent variable in the t period of the i-th research object. Q i j X i t is the spatial lag term of the independent variable, which is used to describe the exogenous interaction effect between the independent variables of other adjacent objects and the dependent variables of the research object. λ is the spatial autoregressive coefficient of the independent variable of the research object. μ i is spatial fixed effect, σ i is time fixed effect, and ε i t is random error term. In addition, e d l i t h c i t i s i t i l i t i c i t g i i t d f t i t e r i t represents the independent variable value of the i-th city in period t, which are economic development level, human capital, industrial structure, innovation level, infrastructure conditions, government intervention, foreign trade openness and environmental regulation.

To mitigate the effects of multicollinearity and pseudo-regression, this study employs the variance inflation factor (VIF) to evaluate the panel data. Typically, a VIF threshold of 10 is used to judge the severity of multicollinearity: A VIF below 10 indicates an absence of multicollinearity, a VIF between 10 and 100 suggests strong multicollinearity and VIF of 100 or more indicates severe multicollinearity. The results of multicollinearity test are shown in Table 3.

The analysis reveals that the VIFs for all independent variables are below 3, and their tolerance levels are less than 1. Additionally, all variables passed the significance test at the 1% level. These results indicate that the model is free of multicollinearity issues, the data is robust, and the model construction process is sound.

This study uses Stata 16.0 to conduct a global Moran’s I analysis of the urban green development index from 2008 to 2020, and the results are shown in Figure 3.

From the distribution of Moran’s I index, the Moran’s I index for urban green development under the ARWM and GDWM is greater than 0, and the overall value is 0.028–0.193, and passes the significance test. This indicates that there is significant spatial correlation and agglomeration effect in the green development of adjacent cities and cities with similar geographical distances, which can drive the green development of local cities. It may be because the city’s borders are connected and the geographical distance is close, making it more vulnerable to the influence of surrounding cities. For example, the negative externality of environmental pollution. It may also be because convenient transportation closely links urban green development with the development of neighboring cities (Deng et al., 2024; Hysa et al., 2025). In addition, when formulating development policies, the government is also relatively easy to refer to and imitate the development policies of neighboring cities or regions, resulting in learning effect and policy interaction, this in turn makes their green development levels more likely to exhibit spatial agglomeration (Yang Y. et al., 2025). This enlightens us that in the practice of urban governance, the inter-urban synergy should be emphasized and the spatial spillover effect of urban green development should be fully leveraged. At the same time, a city-level platform for sharing data, knowledge, and technology dedicated to urban green development should be established, thereby allowing more advanced green development cities to drive the green transformation and development of surrounding cities. The Moran’s I index of urban green development under the EDWM was greater than 0 during the period of 2008–2013 (Moran’s I index is 0.163-0.300), and through significance testing, it was shown that cities with similar economic development exhibited spatial correlation and agglomeration of green development levels between 2008–2013. However, this spatial agglomeration gradually disappeared during the period of 2014–2020, and the spatial agglomeration effect no longer existed. One plausible explanation is that between 2008 and 2013, China’s economy was in a phase of rapid growth driven by resources and investment. During this period, the industrial structure—predominantly industrial—and energy consumption patterns of economically similar cities were similar, leading to the emergence of similarity and agglomeration in the level of urban green development. From 2014 to 2020, however, China’s economic development gradually entered a new normal, and the construction of ecological civilization was elevated to a national strategy (People’s Network, 2015). Differentiated urban development strategies and technological innovations enabled economically similar cities to pursue distinct paths of green development, thereby weakening or even eliminating the spatial correlation rooted in economic distance. Therefore, the government should promote the coordinated layout of green industries in the economic network, create cross regional green industry chain clusters, and set up regional green innovation community funds to transform competition from cities with similar economies into cooperation.

In addition, the local Moran’s I values of each city under different spatial weight matrices are shown in Figure 4.

On the whole, the trend line of urban green development scatter plot under different spatial weight matrices is located in the first quadrant and the third quadrant, which means that the local Moran’s I index is mainly characterized by ‘high-high’ clustering (hot spots areas, that is, cities with high levels of green development are surrounded by other high-level cities) or ‘low-low’ agglomeration (cold spot areas, that is, cities with low levels of green development are surrounded by other low-level cities), that is, the green development between cities affects or attracts each other. The local Moran’s I test further confirms that urban green development among the sample cities exhibits spatial correlation. Specifically, based on the local spatial autocorrelation analysis of urban green development under the ARWM, cities such as Dongguan City and Guangzhou City fall within high-high agglomeration clusters, exhibiting a high level of green development both locally and in their surrounding areas—classed as hot spots. In contrast, cities including Anshan City, Baishan City, and Wenzhou City are situated in low-low agglomeration clusters—classed as cold spots. The distribution of these hot and cold spots is closely linked to cities’ geographical location, industrial structure, and economic development level. Therefore, the spatial agglomeration of urban green development is prominent, which further confirms that research on urban green development requires the application of spatial econometric models.

Furthermore, this study uses the econometric model to calculate the spatial correlation degree of urban green development under different weight matrices, and the results are shown in Table 4.

The data demonstrates that urban green development is affected by at least 13% of the green development of cities with similar economic levels. The impact of cities with close geographical distance reached 84%, and the impact of green development of adjacent cities reached 63%. Hypothesis 1 is verified. Therefore, urban green development has spatial correlation and agglomeration. This pattern echoes Li’s (2019) findings concerning spatial correlations environmental quality, environmental governance, development quality, and green living across various provinces and cities, but the study only analyzes the spatial correlation of ecological civilization construction in each province under the ARWM in 2016. This study constructs the spatial weight matrix of adjacency relationship, geographical distance and economic distance, reveals the spatial effect of urban green development from 2008 to 2020, and can comprehensively reflect the spatial impact of urban green development. In addition, some studies only analyzed and verified the spatial autocorrelation of urban green transformation under the ARWM and the GDWM (Li, 2019; Wang D. et al., 2023), while this study included the EDWM, which could more comprehensively analyze the spatial effect of urban green development. In addition, this conclusion is also similar to the research conclusions of Lin et al. (2023) and Ding et al. (2025). That is, the urban green development level of the Yangtze River Economic Belt has spatial dependence and spatial spillover effects. In contrast to previous studies, this study scientifically quantifies the spatial correlation of urban green development and derives specific quantitative outcomes, which explicitly reveals the agglomeration characteristics of urban green development, thereby extending and complementing existing research. Moreover, it can effectively and targetedly enhance urban green development levels. For policy implications, policymakers should prioritize the establishment of intercity cooperation platforms, encourage neighboring cities to accelerate the flow of factors such as information and technology, and thereby leverage the spatial spillover effects of cities with high green development levels.

Considering that the spatial econometric model is easily affected by the spatial weight matrix, this study uses the ARWM, the GDWM and the EDWM to test the robustness of the spatial regression results. The results of econometric regression analysis of urban green development are shown in Table 5.

The findings reveal that the regression results of urban green development under the ARWM and the GDWM are largely consistent, confirming that the model results are highly robust and reliable. In addition, results from the ARWM and GDWN indicate that urban green development has a spatial spillover effect, suggesting that urban green development is affected by neighboring cities and cities with close geographical distances. These effects likely stem from the rapid circulation of production factors facilitated by proimity and interconnected urban areas, forming a “convenience circle” that promotes mutual urban developmemt (Yang X. L. et al., 2024; García-Chan et al., 2025). This is similar to the research conclusion of Xiao et al. (2022), who used SDM to analyze the spatio-temporal evolution pattern and spatial spillover effect of green development level in Hubei Province, and found that there is a significant spatial spillover effect of green development, and the regional spillover effect is larger than the intra-regional spillover effect. However, the research only analyzes the spillover effect of green development level in Hubei Province. This study expands the research scope to 285 prefecture-level cities in China, and finds that the spatial spillover effect of urban green development under the geospatial weight matrix is higher than that under the adjacency weight matrix, which makes the research results universal. Additionally, this study indicates that the spillover effects of urban green development are more substantial under the GDWM than the ARWM. This supports the view that spatial correlation of urban green development is stronger when measured by geographical distance rather than mere adjacency. At the same time, this conclusion also verifies that the flow of factors brought about by the rapid development of regional economic integration will promote the agglomeration of industries and economies. Therefore, it is crucial to consider spatial spillover effects in establishing urban sustainable development strategies.

Specifically, taking the level of economic development and industrial structure as examples, the estimated coefficients of the level of economic development and industrial structure are significantly positive under the ARWM and the GDWM, indicating that the rapid development of urban economy and the adjustment of industrial structure can promote the green development of the city. It is worth noting that this study found that after considering the spatial lag term, the spatial effect interaction coefficient of economic development level and industrial structure is negative, indicating that the green development of the city has a significant negative impact on the green development of other cities. In order to further reveal the reasons behind this phenomenon, this paper analyzes the action path of the spillover effect of each driving factor, so as to more accurately reflect the relationship between each variable and urban green development. This study uses Stata16.0 software to decompose the spatial effects of urban green development under different weight matrices, as shown in Table 6.

Table 6 reveals that, from the perspective of spillover effect, the level of economic development has a significant negative spillover effect on adjacent cities and cities with close geographical distances, but has a significant positive spillover effect on cities with similar levels of economic development. Urban siphon and competition effects may cause economically developed cities to absorb high-quality resources from surrounding regions; alternatively, these effects may constrain the improvement of adjacent cities’ green development levels through industrial competition (Duan et al., 2025; Aydin et al., 2025). This is similar to the research conclusion of Duan et al. (2025), that is, due to the existence of inter city competition and pollution shelter effect, developed cities may use their good economic conditions to provide more preferential policies to attract foreign investment in order to maintain their economic growth advantages after development and expansion, which will form huge competitive pressure on surrounding cities while enhancing their own strength. In addition, developed cities attract high-quality production factors such as capital, talents and technology from surrounding areas, which will also produce a “siphon effect”, thus promoting the green development of the city and inhibiting the green development ability of surrounding cities (Yang Y. et al., 2025). Therefore, the government should focus on the construction of regional coordinated development mechanisms, such as ecological compensation and enclave economy, to avoid the plunder of surrounding resources by core cities. The industrial structure and infrastructure level have a significant negative spillover effect on the growth of green development in adjacent cities. In addition, cities with similar economic development levels are more likely to learn from each other’s experiences and adopt effective practices, and subsequently form industrial alliances or regional cooperation bodies—such as those within the Yangtze River Delta and the Guangdong-Hong Kong-Macao Greater Bay Area. Through these bodies, initiatives like the co-construction of green standards and joint research and development of green technologies have achieved the scale effect and synergy effect, thereby generating the positive spatial spillover effect (National Development and Reform Commission, 2022). Therefore, it is necessary to give full play to the positive spillover effect of economic level and improve the level of green development in cities and regions.

This study reveals that the industrial structure significantly impacts improvements in green development within cities and exerts a considerable negative spillover effect on adjacent cities. The reason is that the upgrading of industrial structure can directly promote the green development of the city by reducing pollution sources, improving efficiency and introducing technology (Hou, 2025; Erdogan et al., 2025). The negative spillover effect of industrial structure upgrading reveals that a city’s green development improvement comes at the cost of greater environmental burden on adjacent cities (Chen B. et al., 2025). Specifically, following a city’s industrial structure upgrading, existing polluting enterprises choose neighboring cities as ideal locations to sustain their operations, which in turn hampers the green development of these cities. This conclusion is also similar to the research findings of Xiao et al. (2022), Khan et al. (2022) and Sadiq et al. (2024), that is, industrial structure has a negative spillover effect on the green development of cities and industrial structure also shed inverse impacts on environmental quality. Taking 285 cities as the research object, this study comprehensively clarifies the negative spillover effect of industrial structure on urban green development, which is an in-depth supplement to the existing research. This suggests that government should prioritize industrial division of labor and collaboration among adjacent cities, avoid homogeneous competition, guide the orderly flow and optimal allocation of production factors within the regional cluster, and work to convert the “siphon effect” into the “diffusion effect”, so as to achieve the common green transformation of these cities. A city’s innovation level drives sustainable economic growth and profoundly influences its development of green development (Alnafrah et al., 2023). The level of innovation has a significant improvement effect on the urban green development, and has a significant positive spillover effect on the green development of adjacent and geographically close cities. This is similar to the research conclusion of Liao et al. (2024) and Degirmenci et al. (2025), that is, green innovation is conducive to improving the gross economic product of local and neighboring cities and innovation can also promote the sustainable development of Organization for Economic Co-operation and Development (OECD) countries. The researches of Duan et al. (2025) and Zhu et al. (2025) confirm that the quality of green innovation is crucial for the sustainable development of urban agglomerations and reveal that it can also promote green development in neighboring cities. This also enlightens that we should not ignore the role of spatial dependence or spatial spillover in research and development (R&D) activities, especially the spatial spillover effect and diffusion effect caused by R&D personnel flow and technical information exchange (Anselin, 2013; Chen L. et al., 2025). Current trends indicate that R&D activities in China are experiencing unprecedented growth marked by increases in expenditures and patent filings. Therefore, city policymakers are encouraged to bolster support for R&D, optimize the allocation of innovation resources, and implement scientifically informed strategies to enhance green development in urban areas.

The level of infrastructure has a significant role in promoting the improvement of green development in the city, and has a significant negative spillover effect on the growth of green development in adjacent cities. The reason is that the convenience and accessibility of transportation are conducive to the flow of resources from adjacent cities into the city, thus improving the economic development level of the city and promoting the improvement of people’s living standards (Abdolpour and Marut, 2025). However, evacuating the development resources of adjacent cities and possibly shifting the environmental costs lead to the shortage of resources in adjacent cities and hindering the improvement of green development in adjacent cities. This also enlightens government that when planning infrastructure projects, it should start from the perspective of regional coordination, build a networked, multi-center pattern, and enable surrounding cities to share convenience. The results of this study are different from those of Wang Y. J. et al. (2023), which emphasizes that information infrastructure has a significant positive spillover effect on the development of urban green intelligence. This also enlightens us to further analyze the differences in the impact of different infrastructures (such as new infrastructure, green infrastructure, etc.) on the level of urban green development in the future. Government intervention tends to have a positive spillover effect on green development in adjacent cities, indicating that the relevant policies and measures of the city government can play a positive demonstration effect (Yi et al., 2025). This is similar to the conclusion of Kwilinski et al. (2024) that investment of government funds can promote industrial technology innovation, so as to promote sustainable development. However, different from the research conclusion of Liu and Han (2024), They made contrasting observations in examining the influencing factors of green innovation efficiency in Guangdong-Hong Kong-Macao Greater Bay Area based on the SDM, where government support exerted a negative spillover effect on efficiency values in neighboring areas. The disparity between Liu and Han’s results and the findings of this analysis may be that the Guangdong-Hong Kong-Macao has particularly high green innovation efficiency as China’s most open and strongest market economy.

Foreign trade openness also exhibits a positive spillover effect on adjacent cities, indicating that the knowledge, technologies, management, and standards derived from foreign trade openness break through administrative boundaries, drive the flow of talents and technology, and facilitate the diffusion of advanced green production concepts and technologies to these cities (Wang and Zhao, 2025). Zamani and Tayebi (2021), observation that trade and foreign direct investment exert a significant spillover effect on economic growth across OECD member countries. Meanwhile, this conclusion is consistent with Chen’s (2024) finding that open and shared development inhibit carbon emission intensity. This may be because neighboring cities absorb the spillover of a city’s foreign-funded enterprises, driving the development of green development in those surrounding areas. This suggests that government should encourage opening-up, plan regional open economy, strengthen industrial division and cooperation among cities, and maximize the effect of knowledge and technology spillover. In addition, environmental regulation has a significant positive spillover effect on the level of green development in cities with similar levels of economic development. It may be that cities with similar levels of economic development have similar industrial structure, pollution types, and financial capabilities. Therefore, the relevant policies for the green improvement of a city can provide high reference value and operability for cities with similar levels of economic development. In addition, when multiple cities with similar economic development levels collectively adopt strict regulations, this thereby jointly fosters the formation of a sizeable regional green technology market and enhances the collective green development level of these agglomerations (Gong and Zhang, 2020; Feng, 2024). The results of this study are different from those of Aydin et al. (2025), which emphasizes that strict environmental regulation can promote green growth and the development of low-carbon energy. Therefore, in the process of urban green development, a normalized environmental policy exchange platform between cities with similar economic development levels should be established. At the same time, for urban agglomerations with similar economic levels, the government should actively explore and formulate relatively unified environmental standards to avoid vicious competition. Hypothesis 2 is verified.