DEA, as a measure of the efficiency of complex systems, is highly applicable to the measurement of GTIE with multiple input and output indicators. However, the traditional DEA assumes that all outputs are desired outputs, so it is impossible to measure the efficiency value when there are undesired outputs. To overcome the above shortcomings, Tone (2001) proposed a non-radial, non-angular SBM model based on slack variables, which is able to quantify the effectiveness of decision units with undesired outputs, by considering the effect of slack on the basis of traditional DEA. Thereafter, Tone (2002) combined the SBM model with super-efficiency and proposed the S-SBM model. In this context, this paper uses the S-SBM model to measure GTIE in the CCEC.

The S-SBM model assumes that there are n decision-making units (DMUs) in the production system, each containing an input vector X , a desired output vector Y , and a undesired output vector Z . The vector elements can be represented as x ∈ R m , y ∈ R l , and z ∈ R h . Define the matrices X = [ x 1 , … , x n ] ∈ R m × n , Y = [ y 1 , … , y n ] ∈ R l × n , Z = [ z 1 , … , z n ] ∈ R h × n , where x i > 0 , y i > 0 , z i > 0 . The formula for S-SBM model that includes undesired outputs is shown in Equation (1).

where ρ * is the value of GTIE, x i m t , y i l t , z i h t represent the input, desired outputs, and undesired outputs of DMU, respectively, m, l, h represent the number of inputs, desired outputs and undesired outputs, respectively, s m x , s l y , s h z represent the slack vectors for inputs, desired outputs and undesired outputs, respectively, and λ i t is the weight of DMU.

The ESDA includes global and local spatial correlation analysis. To analyze the spatial correlation and distribution pattern of GTIE in CCEC, this paper adopts the global Moran’s I to test the global spatial agglomeration characteristics of GTIE, and its formula is shown in Equation (2).

where I is the global Moran’s I, n is the number of cities, x_i , x_j represent the GTIE of city i and city j, respectively, S² represents the sample variance, and w_ij represents the elements of the spatial weight matrix. The level of economic development of cities is closely related to the GTIE (Li, 2017), and the economic distance matrix based on the level of economic development is a more realistic reflection of the spatial correlation between cities than the traditional adjacency matrix. In this context, this paper refers to the method of Lin et al. (2005) to measure the economic distance between cities using the difference in per capita regional GDP of cities. The elements of the economic distance matrix between cities are w ij = 1 / | y ¯ i − y ¯ j | , where y ¯ is the city’s gross regional product per capita, and the economic distance between the city and itself is specified as zero. The global Moran’s I takes a value between −1 and 1. If I > 0, it means that there is a positive spatial correlation in the GTIE of cities. If I< 0, it means that there is a negative spatial correlation in the GTIE of cities. If I is close to zero, it means that the spatial distribution of GTIE is random and there is no spatial correlation.

The global spatial correlation analysis is difficult to reveal the location and local correlation degree of GTIE. Therefore, this paper uses the local Moran’s I to test the local spatial agglomeration characteristics of GTIE, and its formula is shown in Equation (3).

where I_i is the local Moran’s I and the rest of the symbols are consistent with Equation (2). If I_i > 0, it means that cities with similar GTIE characteristics are clustered together, which means that cities with high (low) GTIE are surrounded by cities with high (low) GTIE, defined as “high-high (low-low)” clustering. If I_i< 0, it means that cities with different GTIE characteristics are clustered together, which means that cities with high (low) GTIE are surrounded by cities with low (high) GTIE, defined as “high-low (low-high)” clustering.

The explained variable is GTIE. Based on the existing research (Shu et al., 2016; Zhu et al., 2018; Lv et al., 2019), this paper defines the connotation of GTI as a new innovation activity that takes into account technological innovation and ecological protection, which has the positive externality of knowledge spillover and the negative externality of environmental pollution, constructs the evaluation index system of GTIE from three aspects: factor input, desired output and undesired output (see Table 1 ), and uses the S-SBM model to measure the GTIE of each city in CCEC.

In the selection of input indicators, compared with the existing literature which only considers capital input and human input (Huang and Wu, 2019; Lv et al., 2019), this paper innovatively adds resource input, and measures the factor input from capital, human and resource, respectively. In the selection of desired output indicators, this paper not only covers general innovation outputs, but also adds environment-related outputs, and innovatively measures desired outputs in terms of economic benefits, technological benefits and environmental benefits. In the selection of undesired output indicators, this paper measures the undesired outputs in terms of air pollution, water pollution and dust pollution.

The explanatory variables are the main influencing factors of GTIE. According to the findings in the previous section of influence mechanism analysis, the main factors affecting urban GTIE of the CCEC are ED, EO, GS, and ER. Therefore, the above four variables are selected as explanatory variables in this paper. ED is measured by the city’s per capita gross regional product, EO is measured by the proportion of actual foreign investment used in the city to the total social fixed asset investment, GS is measured by the proportion of the city’s science and technology expenditure to the local fiscal expenditure, and ER is measured by the proportion of the city’s total investment in environmental pollution control to the regional gross product.

The control variables are other variables that affect urban GTIE. Referring to the findings of existing literature (Huang and Wu, 2019; Sun and Chen, 2021), this paper adds control variables such as industrial structure (IS), human capital (HC), and urbanization rate (UR) to the model. IS is measured by the share of value added of secondary industry in regional GDP in cities, HC is measured by the share of employment in secondary industry in total employment in cities, and UR is measured by the share of urban resident population in total population in cities at the end of the year.

Figure 3 shows that the GTIE of CCEC shows a growth trend with obvious characteristics of stage evolution, which can be divided into three stages during the sample period. The first stage is the “growth period” from 2006 to 2010, in which GTIE showed a trend of decreasing and then increasing, from 0.78 in 2006 to 0.85 in 2010. This stage is the early stage of the construction of the CCEC, technological innovation collaboration and joint construction of ecological barrier are the key tasks of Chengdu and Chongqing to promote the construction of the economic circle, and a series of policies are implemented to promote GTIE. The second phase is the “oscillation period” from 2010 to 2014, in which the GTIE shows a fluctuating trend and reaches the maximum value of 0.92 in the sample period in 2013. This stage is the alternate transition period of China’s “11th Five-Year Plan” and “12th Five-Year Plan”, and the pace of industrial restructuring is fast. Meanwhile, in 2011, China’s National Development and Reform Commission released the Regional Plan for Chengdu-Chongqing Economic Zone. The dual influence of industrial restructuring and government planning is an important reason for the fluctuation of GTIE. The third phase is the “revitalization period” from 2014 to 2020, in which GTIE shows an increase from 0.81 in 2014 to 0.83 in 2020. This stage is China’s “13th Five-Year Plan” period. Green development has become an important goal of China’s economic and social development, and the quality of the ecological environment has been widely valued by governments at all levels. The CCEC has actively promoted energy efficiency and emission reduction, and the implementation of environmental protection policies and the development of green industries have further promoted GTI.

In this paper, we classify and spatially visualize the time-series data of GTIE based on the natural fracture method for hierarchical classification and spatial visual representation. Figure 4 shows that the GTIE is spatially unbalanced, and that the “double core” position of Chengdu and Chongqing in the economic circle is gradually prominent. From 2006 to 2010, the higher-efficiency cities of GTIE are transformed from Chengdu to Chongqing, the GTIE in medium-efficiency cities is significantly improved, and the high-efficiency cities are transformed from Meishan, Ziyang and Chongqing to Mianyang, Chengdu, Meishan, Ziyang, Neijiang, Guang’an and Dazhou. From 2010 to 2015, Chongqing continued to maintain its status as a higher-efficiency city, and high-efficiency cities shifted from the north to the south of the economic circle, forming a cluster of low-efficiency and medium-efficiency cities in the north and a cluster of high-efficiency and medium-efficiency cities in the south, with an obvious north–south distribution difference in the spatial distribution of GTIE. From 2015 to 2020, the spatial distribution of GTIE changes from a single-core pattern in Chongqing to a double-core pattern centered on Chengdu and Chongqing, with a significant reduction in high-efficiency cities and a contiguous development of medium-efficiency and low-efficiency cities. It is worth noting that the GTI in the CCEC may fall into a “medium efficiency trap” in its future development. For low-efficiency and medium-efficiency cities, there is an invisible “ceiling” of GTIE. The “siphon effect” of higher-efficiency cities makes green innovation resources over-concentrated in a few cities, and the loss of green innovation resources in medium-efficiency and low-efficiency cities is serious, and their industrial structure has been dominated by secondary industries for a long time, which leads to insufficient endogenous power of green innovation. The combination of these factors hinders the progress of low-efficiency and medium-efficiency cities to higher gradient efficiency cities.

In Table 2 , it is shown that the global Moran’s I of GTIE is significant at different test levels, indicating that there is a significant spatial correlation of GTIE. Specifically, from 2006 to 2007, the global Moran’s I of GTIE is positive, showing a significant positive spatial correlation, which is characterized by the spatial clustering of “high (low)-efficiency cities surrounded by high (low)-efficiency cities”. However, the absolute value of the Moran’s I gradually decreases, and the positive spatial correlation gradually decreases. From 2008 to 2015, the global Moran’s I of GTIE changed from positive to negative, showing a significant negative spatial correlation, which is characterized by the spatial agglomeration of “high-efficiency cities surrounded by low-efficiency cities”, and the spatial “polarization effect” is prominent. The possible reason is that the 2008 global financial crisis had a negative effect on the economic growth of cities in the CCEC, especially hitting the financial systems of each city, which led to a significant shortage of innovation capital investment in each city. However, cities with high GTIE are rich in green innovation resources, and green innovation factor inputs are less affected by the financial crisis, thus further increasing the green innovation development gap among cities in the economic circle, and GTIE shows a negative spatial correlation. From 2016 to 2020, the global Moran’s I of GTIE turns from negative to positive, showing a significant positive spatial correlation, which is manifested by cities with similar GTIE clustering with each other. The possible reason is the flow of capital, talent, technology and other innovation factors among cities in the economic circle accelerates with the gradual implementation of the government’s economic stimulus policies, driving the upgrading of urban industrial structure and the transformation of green low-carbon development, and the GTIE of each city improves significantly, and the gap of green innovation development among cities in the economic circle gradually narrows, making GTIE turn to be spatially positively correlated.

To analyze the local clustering dynamics of GTIE, this paper organized the distribution of each city in each quadrant of the local Moran’s I scatter plot. Table 3 shows that there are 10 cities located in the first and third quadrants in 2006, increasing to 11 in 2020, accounting for 62.5% and 68.8% of all cities, respectively. It shows that the GTIE in 2006–2020 is mainly spatially positively correlated, and its regional agglomeration pattern is mainly “high-efficiency type” and “low-efficiency type”.

In 2006, there are four cities located in the first quadrant, including Chengdu, Mianyang, Guang’an, and Chongqing. There are six cities in the third quadrant, including Zigong, Deyang, Leshan, Nanchong, Yibin, and Ya’an, which have relatively fewer green innovation resources and lower GTIE. There are relatively few cities in the second and fourth quadrants, accounting for only 37.5% of all cities. There are four cities in the second quadrant, including Luzhou, Suining, Neijiang and Dazhou, which have relatively low efficiency in GTI. Although they are adjacent to high-efficiency cities, they do not effectively undertake the technology spillover of green innovation from high-efficiency cities. There are two cities in the fourth quadrant, including Meishan and Ziyang, which have high GTIE but less green technology spillover to neighboring cities and less spatial radiation. In summary, the GTIE of CCEC is low in 2006, and the spatial agglomeration is mainly “high-efficiency type” and “low-efficiency type”.

In 2020, there are seven cities in the first quadrant, with Luzhou, Neijiang, Meishan and Ziyang added, and Chongqing reduced compared to 2006. Luzhou and Neijiang have changed from “hollow type” to “high-efficiency type”, and Meishan and Ziyang have changed from “polarization type” to “high-efficiency type”. There are four cities in the third quadrant, with Leshan and Yibin reduced compared to 2006. There are relatively few cities in the second and fourth quadrants, accounting for only 31.3% of all cities. There are two cities in the second quadrant, with Luzhou and Neijiang reduced compared to 2006. There are three cities in the fourth quadrant, including Leshan, Yibin, and Chongqing, with Chongqing changing from “high-efficiency type” to “polarization type”, while Leshan and Yibin change from “low-efficiency type” to “polarization type”.

To analyze the spatial leap path of GTIE of cities in the economic circle, this paper extends the scatter plot of the local Moran’s I by referring to the spatiotemporal analysis of Rey and Janikas (2006). Table 4 shows that the leap path of Type I is between “high-efficiency type” and “polarization type”, and the cities include Chongqing, Meishan and Ziyang. No cities leapfrogged in the path of Type II. The leap path of Type III is between “hollow type” and “high-efficiency type”, “low-efficiency type” and “polarization type”, and the cities include Luzhou, Neijiang, Leshan and Yibin. The rest of the cities in the economic circle belong to Type IV which no leap occurs. In summary, GTIE has obvious spatial agglomeration and low mobility characteristics, and its spatial leap shows obvious stability. More than half of the cities do not break away from the original spatial agglomeration pattern, and the double core spatial distribution pattern of the GTIE with Chengdu and Chongqing as the core has stability, and the “core-edge” structure of green innovation development in the CCEC tends to be stable.

The estimation results of SDM are obtained based on the previous model settings (see Table 5 ). To compare the robustness of the model estimation results, this paper reports the estimation results of SLM and SEM. In Table 4 , it is shown that the significance of the parameters of models (2), (4), and (6) is significantly improved after controlling for city effect and time effect, which indicates that GTIE is influenced by individual city attributes and time trends. It is necessary for the model to control for individual and time effects so that the net effect of each influencing factor on the GTIE can be more accurately identified. In models (3) to (6), the spatial effect coefficients of GTIE are all significantly positive, which indicates that there is a significant positive spatial spillover effect of GTIE. The enhancement of GTIE in this city can lead to GTI in spatially adjacent cities. In models (1) and (2), the spatial effect coefficients of the spatial error terms are significantly positive, indicating that there is a spatial spillover effect of the spatially perturbed factors of the model, and the unobserved factors affecting GTIE in this city also have a positive effect on GTIE in spatially adjacent cities.

In addition, the results of the Likelihood Ratio (LR) test show that the test statistics of models (5) and (6) reject the original hypothesis that SDM can degenerate into SLM or SEM at the 1% significance level, which indicates that SDM is optimal. From the estimation results of the models, the spatial autoregressive coefficients of GTIE in models (5) and (6) are significantly positive at the 5% or 1% test level, which indicates that the direct and indirect effects cannot be identified through the estimated parameters in the models, and further partial derivative matrices need to be constructed to decompose the spatial effects. Therefore, this paper decomposes the spatial effects on the basis of model (6) (see Table 6 ).

The results in Table 6 show that the coefficient of the direct effect of ED is positive at the 5% level of significance, and the coefficient of its indirect effect is not significant. It indicates that ED has a positive effect on the GTIE of cities, but its spatial spillover effect on spatially adjacent cities is not significant. The improvement of the economic development level can stimulate the demand of urban GTI and provide more financial support for GTI activities, which in turn can promote the improvement of urban GTIE. However, Shi and Pan (2021) showed that the CCEC is at the stage of agglomeration development, and an orderly synergistic mechanism of industrial chain division has not yet been formed among cities in the economic circle, and there is even an obvious competition relationship. Therefore, the GTI dividends brought by urban economic development are more locally absorbed, while the promotion of GTI in spatially neighboring cities is limited.

The direct effect coefficient of EO is not significant, and the coefficient of its indirect effect is negative at the 1% level of significance. This indicates that the effect of EO on urban GTIE is not significant, but it shows a negative spillover effect on neighboring cities, which deviates from the outcome that was anticipated. The possible reason is that there is a substantial disparity between China’s degree of innovation in green technology and that of other countries, especially in the key core areas. The expansion of China’s opening to the outside world can increase foreign investment and bring advanced green innovation technologies and management models to cities, accelerating their green development process. However, the breadth and depth of the opening up of the CCEC is still low, and its resource-dependent economic growth model has not yet been completely changed, which leads to the insignificant incentive role of foreign investment in urban GTI. In addition, EO has a negative inhibitory effect on the GTIE of the neighboring cities. In the process of building the strategic highland of inland opening, cities in the CCEC are competing to introduce foreign investment, which to a certain extent leads to the tendency of bottom-up competition in the environment, and neighboring cities take over the low-end production capacity eliminated internationally, resulting in the phenomenon of “pollution paradise” (Huang and Wu, 2019), which inhibits the improvement of GTIE in neighboring cities.

The direct effect coefficient of GS is positive at the 1% significance level, and the coefficient of its indirect effect is not significant. This indicates that GS has a positive effect on the GTIE of the city, but it does not support the neighboring cities enough. The government has weakened the negative externalities and cost uncertainties of green innovation R&D activities mainly through R&D subsidies and low-carbon compensation, thus incentivizing companies to invest resources in GTI projects and enhancing GTIE of cities (Dong et al., 2021). In the process of actively creating China’s GTI center in the CCEC, the government’s financial support is an important guarantee to enhance GTIE. The economic growth model of the CCEC is changing from “factor driven” to “innovation driven”, and the guiding role of green innovation policy is increasingly prominent. The government has given policy and financial support in GTI activities of enterprises, which has improved the GTIE of cities. However, the policy and financial support of the local government are more inclined to local enterprises, which makes it difficult for the neighboring cities to obtain the policy dividend of the local government.

The direct effect coefficient of ER is positive at the 10% level of significance, and its indirect effect coefficient is negative at the 10% level of significance. It shows that ER can improve GTIE of cities, but it shows a negative spillover effect on neighboring cities. Dong et al. (2021) argue that environmental regulation enhances urban GTIE through the “innovation compensation” effect. In the process of jointly building the ecological barrier in the upper reaches of the Yangtze River, the CCEC has continuously strengthened environmental protection standards, which forces enterprises to innovate production technology, creates the “innovation compensation” effect to compensate for the production burden of enterprises under high environmental protection standards, and promotes GTI of enterprises. However, the strict environmental protection standards of cities will not only restrict the environmental pollution of local enterprises, but also force local pollution-oriented enterprises to transfer to neighboring cities with relatively loose environmental protection supervision (Dong and Wang, 2019), and hinder the GTI of neighboring cities.

In order to test the robustness of the spatial effect decomposition results, this paper re-estimates the model parameters in terms of replacing the spatial weight matrix and the measurement indicators of the explained variable, respectively. Firstly, this paper adopts the binary adjacency matrix of cities in the CCEC as the new spatial weight matrix. Secondly, this paper collects green patent application data of cities in the CCEC from 2006 to 2020 based on the International Green Patent Classification List released by the World Intellectual Property Organization in 2010 and the patent search and analysis system of the State Intellectual Property Office of China with reference to the processing methods of existing studies (Zhang C. et al., 2022), and re-measures the urban GTIE by using green patent application data as a measure of desired output. Based on the above two approaches, this paper re-estimates the model (6) in Table 5 and decomposes its spatial effects. The results in Table 7 show that after the introduction of the new spatial weight matrix and the replaced urban GTIE, the absolute values of the direct effect coefficients and indirect effect coefficients of each variable increase or decrease to some extent, and the significance of some variables increases or decreases, but the influence direction and significance of each variable are basically consistent with the results in Table 5 . To sum up, the estimation results of the spatial econometric model in this paper are robust.

The cities in the CCEC are at different stages of industrial transformation and upgrading, and their capacities for GTI and economic development are different. Therefore, this paper takes the quantile of the mean value of GTIE of each city from 2006 to 2020 as the basis for classification, and classifies those with the mean value after the 75% quantile as high-efficiency cities, those with the mean value before the 25% quantile as low-efficiency cities, and the rest as medium-efficiency cities. Then, this paper uses the SDM and partial differential method to analyze the spatial heterogeneity of the influencing factors of GTIE in different types of cities.

The results in Table 8 show that, from the perspective of direct effect, EC is significantly positive in high-efficiency and low-efficiency cities, and its absolute value of coefficient in low-efficiency cities is large. EO is significantly positive only in medium-efficiency cities. GS is significantly positive only in low-efficiency cities. ER is significantly positive only in high-efficiency cities. It shows that ED significantly promotes GTIE in both high-efficiency and low-efficiency cities, and the positive effect is stronger in low-efficiency cities. EO promotes GTIE in medium-efficiency cities. GS promotes GTIE in low-efficiency cities. ER improves GTIE in high-efficiency cities. The economic growth and green development of high-efficiency cities are coordinated with each other, and the stronger economic strength enhances their “siphon effect” on green innovation resources, thus stimulating the vitality of urban GTI, and the “innovation compensation” effect of urban environmental regulation is more obvious. The breadth and depth of opening to the outside world in the CCEC is still low, and increasing foreign investment can improve GTIE in medium-efficiency cities to a certain extent. Low-efficiency cities are limited by the backward development of green industries, the cost burden of enterprises to carry out GTI relies on government financial support, and the incentive effect of economic development level on GTIE is more obvious.

From the perspective of indirect effect, EO is significantly negative in high-efficiency and medium-efficiency cities. GS is significantly positive only in high-efficiency cities. ER is significantly negative only in high-efficiency cities. It shows that EO of high-efficiency and medium-efficiency cities has significant spatial negative spillover effects on neighboring cities, GS of high-efficiency cities have significant spatial positive spillover effects on neighboring cities, and ER of high-efficiency cities have significant spatial negative spillover effects on neighboring cities. High-efficiency and medium-efficiency cities have obvious advantages in green innovation capacity and green industry clustering, and are more attractive to green and innovative foreign enterprises. The cities’ environmental standards are also more stringent, prompting high-polluting enterprises to move to neighboring cities with lower environmental standards. Neighboring cities are hindering their own GTI by taking on relatively low-end local spillover capacity and low-end international capacity. At the same time, in the process of industrial transformation and upgrading in high-efficiency cities, the government has played an important supporting role in building a basic green innovation platform and green innovation R&D investment, which has strengthened the green innovation links within the region, and the neighboring cities have improved GTIE under the radiation of high-efficiency cities.