Currently, global climate issues are becoming increasingly severe, and climate governance has become a focal point of international attention. Countries worldwide are under immense pressure to reduce carbon emissions and promote green development. Against this backdrop, China has actively responded to the global call for climate governance by proposing the "dual carbon" goals of "carbon peaking and carbon neutrality." This goal imposes new and higher requirements on China's economic development model, energy structure adjustment, and ecological and environmental protection, becoming an important practical need to drive China's economic green transformation.

Green technology innovation plays a strategic role in achieving the "dual carbon" goals. The development and application of green technologies such as photovoltaic, hydrogen, and carbon capture are considered key means to optimize energy structures and reduce carbon emissions. As clean energy sources, photovoltaic and hydrogen can replace traditional fossil fuels, lowering carbon emissions during energy consumption. Carbon capture technology effectively captures and treats Carbon Dioxide (CO2) generated in industrial production processes, reducing carbon emissions per unit of Gross Domestic Product (GDP) from an end-of-pipe treatment perspective. This argument provides technological support for regional low-carbon development. Emerging productive forces, centered on technological innovation, represent an efficient form of productivity. Their theoretical framework encompasses multiple dimensions, including technological advancement, enhanced resource allocation efficiency, and management innovation. In the context of green development, cultivating and upgrading these productive forces has become a crucial driver of coordinated economic growth and carbon-reduction targets. By improving production efficiency and optimizing resource utilization, they provide sustained momentum for regional low-carbon transformation.

This study uses a province-Fixed Effects (FE) model to analyze the mediating effect of TFP on CO2 emission intensity in green technology innovation from 2017 to 2021. The reason the time fixed effect was not introduced is mainly due to the following two considerations. On the one hand, this study focuses on the heterogeneity of the mediating path through which green patent impact affects TFP and CO2 emission intensity across different provinces. The FE of provinces can effectively control for characteristics that do not change over time, such as geographical location, resource endowment, and industrial foundation. On the other hand, from 2017 to 2021, there were no macroeconomic policy changes at the national level that had a significant impact on the mediating relationship between green patents, TFP, and CO2 emission intensity. Additionally, the sample provinces did not show significant national time trend changes in economic development models, energy structure adjustments, etc. Therefore, introducing time FE may make the model overly complex and not significantly improve estimation performance.

This study employs panel data from 30 Chinese provinces spanning 2017-2021 (excluding Hong Kong, Macau, Taiwan, and the Tibet Autonomous Region due to data availability constraints), with primary sources including the National Bureau of Statistics, the China Deep Data Repository, and the China Energy Statistical Yearbook.

Fig. 2 shows the correlation between the core variables and control variables in this study. The scatter plot matrix intuitively reflects the distributional characteristics of each variable and the correlation trends between pairs. The results showed a significant negative correlation between green technology innovation and carbon emission intensity, indicating that green technology innovation plays a key role in promoting regional carbon reduction. Meanwhile, the new quality productivity TFP is also negatively correlated with carbon emission intensity, confirming the positive role of productivity improvement in reducing carbon emissions. In addition, there is a clear positive correlation between green technology innovation and new quality productivity TFP, which preliminarily confirms that green technology innovation is an important driving factor in improving new quality productivity.

In terms of controlling variables, the proportion of coal and the increase in the secondary industry are positively correlated with carbon emission intensity, which aligns with the reality that energy-intensive industries and dependence on fossil fuels are the main sources of carbon emissions. However, per capita GDP is negatively correlated with carbon emission intensity, suggesting that improvements in economic development may be accompanied by reductions in carbon emissions driven by industrial structure optimization or technological progress. These preliminary observations are highly consistent with the theoretical assumptions and subsequent regression analysis results of this study, laying a solid data foundation for empirical analysis.

To initially investigate the direct impact of green technology innovation on regional carbon emissions, this study first conducts a baseline regression analysis. The results are shown in Table 4: Column (1) indicates that without adding control variables, the regression coefficient of the core explanatory variable, "Number of Patent Applications," is -2.642, and it is highly significant at the 0.001 statistical level. It demonstrates that green technology innovation has a significant negative impact on regional carbon emissions. In Column (2), after adding "Coal Consumption Share," "Per Capita GDP," and "Value-Added of the Secondary Industry" as control variables, the control variables do not show a significant impact because during the study's sample period (2017–2021), the studied regions actively promoted energy structure diversification; the substitution effects of natural gas and renewable energy gradually emerged, weakening the marginal impact of coal consumption on carbon emissions, thus leading to an insignificant relationship in the regression.

Furthermore, as the study regions are in a period of economic transition, the regulatory effects of green technology innovation on economic growth and industrial structure have, to some extent, offset the traditional "economy-emission" linkage. It makes it difficult to identify the direct impact of Per Capita GDP and the Secondary Industry's Value-Added on carbon emissions using a linear model, rendering them insignificant. However, the total effect coefficient of "number of patent applications" on carbon emissions is -2.081 and is significant at the 5% level (p<0.05). It implies that green technology innovation can, to some extent, directly reduce carbon emissions [8]. H1 is thus verified.

Based on the preceding mechanism analysis and variable settings, this study conducts a stepwise regression analysis using a mediation effect model to explore the mediating role of TFP in the relationship between green technology innovation and carbon emissions. The results are shown in Table 5: Column (1) represents the baseline impact of green technology innovation on carbon emissions. Column (2) shows the impact of green technology innovation on TFP; the coefficient of "number of patent applications" on the mediating variable TFP is 1.31e-06, and it is significant at the 10% level ($p<0.1$), indicating that green technology innovation can significantly enhance TFP. Column (3) adds the mediating variable, TFP, to the total effect model to test the independent variable's direct effect on the dependent variable, controlling for the mediator. The data show that the coefficients for both "number of patent applications" and TFP on carbon emissions are significantly negative and significant at the 10% level (p<0.1). Since the direct effect of the "number of patent applications" remains significant after introducing the mediating variable, and the mediator TFP's coefficient is also significant, this indicates that TFP partially mediates the impact of green technology innovation on carbon emissions. That is, green technology innovation not only directly reduces carbon emissions but also indirectly promotes carbon abatement by enhancing TFP. H2 is thus verified [9].

Given that heterogeneity may exist in the relationship between green technology innovation and carbon emissions across regions with different levels of economic development, this study divides the sample into a "high per capita GDP" group and a "low per capita GDP" group based on Per Capita GDP for sub-sample regression analysis. The results are shown in Table 6.

The heterogeneity test results reveal disparities in the emission-reduction effects of green technology innovation across regions with different levels of economic development. In the high per capita GDP group, the regression coefficient for "number of patent applications" is -2.706, and it is significant at the 10% level (p<0.1), indicating that green technology innovation has a significant inhibitory effect on carbon emissions in regions with high economic development. However, in the Low Per Capita GDP group, the regression coefficient for "Number of Patent Applications" is -2.153, but it is not statistically significant. This result implies that in regions with higher economic development, green technology innovation may be more easily translated into actual emission reductions. It could be attributed to more developed technology transfer mechanisms, greater economic affordability, and a more mature green industrial base. Conversely, regions with lower economic development may lag in R&D investment for green technology, technology diffusion and application, and the green transformation of their industrial structure, thereby preventing the full realization of the emission-reduction effects of green technology innovation.

To ensure the core conclusions are not affected by model specification, this study compares results across the FE, Random Effects (RE), and Pooled OLS models. It uses a Hausman test to verify model suitability. This argument validates whether the core conclusion depends on a specific model choice. The results are shown in Table 7.

To validate model performance and identify the most suitable analytical model based on the preceding definitions and criteria, we utilize empirical data. The results are as follows.

Weighted average of carbon -0.312 -2.462 Not significant in either emission intensity in (1.384) (1.824) neighboring provinces (spatial lag/error term) Constant term -0.014 0.030 Not significant in either (0.140) (0.124) Spatial effects Spatial lag/error coefficient -0.312 -2.462 Stronger in SEM Model Statistics Model explanatory power 0.272 0.320 Superior for SEM Adjusted explanatory power 0.185 0.239 Superior for SEM Adjusted F-statistic Overall significance 3.114 3.925 More significant for SEM (p=0.044) (p=0.020)

Notes: Standard errors in parentheses; *** p<0.01, ** p<0.05.

The model results indicate that the absolute value of the spatial error coefficient in the SEM ( =2.462) is significantly greater than that of the spatial lag coefficient in the SAR (=0.312), suggesting stronger spatial error dependence. Furthermore, the core explanatory variable (green innovation) is more significant in the SEM, indicating its superiority over the SAR model. From the perspective of model fit, both the R2 (0.320) and the adjusted R2 (0.239) of the SEM are higher than those of the SAR, and its F-statistic is more significant (p = 0.020), demonstrating that the SEM possesses greater explanatory power and robustness compared to the SAR model. In conclusion, the SEM should be adopted. The impact of the core variables on CO2 emission intensity within the SEM framework is presented below:

The coefficient for green innovation (P) is negative, but its p-value (0.3951) exceeds the 5% significance level, indicating no statistically significant relationship between green innovation and carbon emission intensity. This lack of significance may stem from the fact that, during the sample period, the technological transformation and scale of green innovation had not yet fully realized their emission-reduction potential. Alternatively, insufficient synergistic efforts in green innovation across regions may have prevented the formation of an effective collective force for emission reduction.

In contrast, the coefficient for TFP is -2.5947 with a p-value (0.0000) far below the 5% significance level, demonstrating that TFP growth significantly reduces carbon emission intensity. TFP advancement contributes to emission reduction by optimizing production technologies and enhancing resource-use efficiency, thereby lowering carbon emissions per unit of output. This process facilitates the synergistic achievement of economic development and emission-reduction goals, positioning TFP growth as a crucial driver of regional low-carbon transition.

Given the imbalanced regional development within China, the regression results at the national level may mask underlying spatial heterogeneity. To thoroughly investigate the differential impacts of green innovation and TFP on carbon emissions across regions, this study further conducts empirical tests by dividing the sample into three major areas: the eastern, central, and western regions. The results are presented in the table below.

Note: *** denotes statistical significance at the 1% level.

In the eastern region, higher investment in technology R&D and stronger transformational efficiency of green innovation lead to significant reductions in emissions. The western region, however, demonstrates a pronounced driving effect of TFP growth (e.g., through industrial upgrading) on emission reduction. In contrast, green innovation has not yet yielded significant results due to its underdeveloped foundation. In the central region, likely constrained by a heavier industrial structure, neither variable shows a significant effect.

In the robustness checks, the SAR model was employed to verify the core relationships. The results show that the regression coefficients for both the short-term and long-term effects of green technological innovation (P) on carbon emission intensity are negative and statistically significant in most scenarios. The coefficient signs for new quality productivity TFP are consistent with those in the baseline model SEM. Although the level of significance varies slightly under certain conditions, the overall trend indicates a suppressive effect on carbon emissions. The interpretation of other parameters in this model (such as the effects of control variables, such as the share of coal consumption and GDP per capita) is presented in the empirical results section above and will not be reiterated here.

Therefore, the inhibitory effect of green technological innovation on carbon emission intensity is robust. The emission-reduction trend of New Quality Productivity aligns with the baseline conclusion. It demonstrates that, whether the SEM or the SAR Model is used, the direction of impact and the key significance characteristics of the core explanatory variables on carbon emissions remain stable, indicating that the regression results are highly reliable.

The results show that Green technology innovation significantly reduces carbon emissions. In both the baseline model and the robustness test, the number of green patents is significantly negatively correlated with carbon emission intensity.

5.1.1|New productive forces exhibit a mediating effect.

Green technological innovation not only directly reduces emissions but also indirectly enhances TFP, thereby validating the transmission mechanism of 'technological innovation → productivity improvement → carbon emission reduction'.