The hypotheses of the study were formed by combining the theoretical framework (the STIRPAT model and Ecological Modernization Theory) with empirical evidence from previous studies and modified to fit the context of the BRI.

1. H1: Urban population growth positively affects CO₂ emissions.

A conceptual framework employed the presentation of a graphical model to determine the anticipated relation between the dependent variable and independent variables. The research process has identified the primary objective and systematically incorporated these objectives to derive the empirical outcomes. The explanatory (independent) variables include urban population growth, forest area, electricity production from oil sources, electricity production from nuclear sources, and average precipitation in depth, and the dependent variable is ecological performance. This study selects all variables based on theoretical and empirical relevance to ecological performance in CO₂ emissions in BRI countries. It includes urban population growth because their influence is an energy consumption driver and a major contributor to CO₂ emissions (Liu et al., 2021; Yan et al., 2023). Forest areas are considered Carbon Sinks, and the impact of deforestation on carbon emissions is considerable (Assis et al., 2022; Begum and Raihan, 2020). The study analyzed electricity production from oil sources to understand the environmental impacts of fossil fuel-based energy, as it is a major emitter (Voumik et al., 2022; Majumder et al., 2023) and nuclear energy is examined due to its potential to contribute to reducing emissions through low carbon electricity generation (Mahmood, 2022; Petruška and Litavcov, 2022). In order to include the average precipitation’s impact on ecosystems, agriculture, and water resource management, which are key for carbon sequestration, precipitation variability plays a crucial role in greenhouse gas emissions (Zhang et al., 2023; Ni et al., 2019). The conceptual framework (Figure 1) presents three emission pathways: (1) Urbanization (UPG) increases CO₂ by boosting energy needs; (2) Oil-based electricity (EPOS) worsens emissions, but the nuclear (EPNS) reduces them by substituting a fuel source; and (3) Forests (FA) and precipitation (APD) regulate emissions through carbon sequestration and climate-feedback activities. More importantly, the factors are interrelated; for example, urban development can lead to the depletion of forest cover, and changes in precipitation patterns can cause changes in energy and land-use patterns. Both must be analyzed together to prevent biased estimates.

The present study explores the influence of climatic factors on ecological performance in BRI countries. The data of 20 countries relating to BRI nations are collected from 2000 to 2022. The official website of the World Bank (WDI) was used to gather data on all variables. The broad scope of the data contains the potential for reporting inconsistency issues caused by different standards countries use to report their results. This study verified the data against multiple sources and implemented reliable statistical methods (removal of outliers) to address this issue. The method used to treat missing values (for instance, by interpolation, imputation, or exclusion, based on the amount of missingness). The study examines several variables, including ecological performance, annual urban population growth rate (%), percentage of forest area relative to total land area (%), electricity production from oil sources as a percentage of total production, electricity production from nuclear sources as a percentage of total production, and average annual precipitation measured in millimeters (mm). In this study, to ensure data reliability, values from the World Bank (WDI) were cross verified with secondary sources whenever discrepancies were identified. The missing values were handled through linear interpolation when gaps were minimal (1–2 years); for larger gaps, countries were excluded from those specific observations. To address reporting inconsistencies among BRI countries, this study applied standardization checks and outlier diagnostics to minimize bias and ensure accuracy. While all variables are based on established WDI definitions, this study recognize possible measurement errors due to inconsistent national reporting systems. To address potential measurement errors, this research applied logarithmic transformations and robustness checks using various estimation techniques (FMOLS, DOLS, and GMM) to ensure consistent relationships. Table 1 demonstrates the description and measurements of variables.

In particular, the two-step system Generalized Method of Moments (System GMM) estimator is used in this study to address several econometric issues, such as endogeneity, autocorrelation, and heteroscedasticity, which are commonly faced in panel data analysis. In contrast to conventional estimators, GMM is especially beneficial for dynamic models since lagged dependent variables are used as instruments to capture the persistence of CO₂ emissions over time (Arellano and Bond, 1991; Blundell and Bond, 1998). Two-step system GMM yields efficient and unbiased estimates, especially in when sample is small, while other traditional model (e.g., Fixed Effect Models) ignore that dynamic relationships. System GMM that may suffer from weak instrument bias show that two-step system GMM provides efficient estimates through robust instruments. This approach allows for robust and reliable results when assessing climatic factors’ impact on ecological performance in BRI countries.

Regarding the GMM system, Rodman (2009) argued that the instruments applied by Arellano and Bond (1991) are distinct from those formulated by Blundell and Bond. They also assumed that the implicit correlation between fixed effects and the initial differences used as instrumental was zero. Therefore, more instruments may be included in the model. The fundamental assumption is the absence of serial correlation. On the other hand, in the case of one-order serial correlation, yi may occur, cannot be used as an ancillary instrument, and requires lags for three or more periods. In the case of the second-order correlation, longer lags are needed (Roodman, 2009). Hence, Ullah et al. (2018) forced that this should be considered because the first difference transformation segregates the past observation from the current one, wherein larger data loss gaps occur. As a result, it affects the predicted value. System GMM alters instruments due to the fixed effects of these instruments, and their treatment is uncorrelated.

Furthermore, other instruments for the lagged dependent variable and any endogenous variables are incorporated to account for endogeneity by significantly improving efficiency. While Difference GMM eliminates the prior observation, system GMM subtracts their average overall future observations instead of the current values of observed variables (Roodman et al., 2009). Therefore, this paper used a two-step System GMM to test the correlation between dependent and explanatory variables. The basic structure of GMM, which addresses endogeneity, autocorrelation, and heteroscedasticity, bypasses testing for diagnostic purposes. However, we verified that our data was endogenous. The application of Arellano-Bond’s difference GMM is unsuitable when the endogenous variable is very persistent or closely follows a random walk. In this instance, the delayed levels exhibit a weak correlation with weak instruments when examining the initial differences among the variables. When this occurs, Blundell and Bond’s (1998) system GMM estimator is frequently more appropriate. System GMM enhances the differences between GMM by estimating levels and differences simultaneously. Consequently, a system GMM is the most effective means for detecting and rectifying endogeneity. Due to the persistent and time-dependent nature of CO₂ emissions, this study employs a dynamic panel specification. Incorporating lagged dependent variables enables the modeling of path-dependent structures in emissions, mitigating endogeneity arising from simultaneity and omitted variables. This approach is grounded in the STIRPAT framework and implemented via the two-step System GMM estimator (Arellano and Bover, 1995; Blundell and Bond, 1998), ensuring robust estimation under dynamic conditions. To structure the empirical model, the study uses the Cobb–Douglas production function, which is as follows: ln ⁡ E P i t = U P G i t β 1 F A i t β 2 E P O S i t β 3 E P N S i t β 4 A P D i t β 5 0   <   β 1 − 6   <   1 (1)

By transferring the data into a natural logarithm, the linear form of Equation 1 is shown as follows: ln ⁡ E P i t = β 0 + β 1 l n U P G i t + β 2 l n F A i t + β 3 l n E P O S i t + β 4 l n E P N S i t + β 5 A P D i t + ϵ i t (2)

To illustrate the dynamic relation, Equation 2 can be transferred as shown in Equation 3 as follows: ln ⁡ E P i t = β 0 + β 1 l n E P i t − 1 + β 2 l n U P G i t + β 3 l n F A i t + β 4 l n E P O S i t + β 5 l n E P N S i t + β 6 A P D i t + ϵ i t (3)

To illustrate the use of a system-GMM estimator in a model using dynamic panel data, Equation 4 is consider the following: ln ⁡ E P i t = β 0 + β 1 l n E P i t − 1 + β 2 l n U P G i t + β 3 l n F A i t + β 4 l n E P O S i t + β 5 l n E P N S i t + β 6 A P D i t + π t + μ i + σ t (4)

EP_it represents the ecological performance, E P i t − 1 is the lagged value of Ecological Performance, β 0 Represents the “constant term,” whereas β t Represents the coefficients for Urban Population growth (UPG), Forest area (FA), Electricity production from oil sources (EPOS), Electricity production from nuclear sources (EPNS) and Average precipitation in depth (APD), while εit (error term) is decomposed into three types. π t represents the time-specific effect, μ i is the country’s unobserved specific effect and σ t represents individual error terms. Ln represents the natural logarithms.

This study employs the Granger non-causation test (Dumitrescu and Hurlin, 2012) to assess endogeneity. The test is applied to validate whether the independent variables are Granger causes of CO₂ emissions. The test is general for panel data heterogeneity and cross-sectional dependence.

Table 2 presents the descriptive statistics of the study variables, which describe the distribution and variability of these variables. The mean values show that the Ecological Performance (Ln EP) is 1.952 and Urban Population Growth (Ln UPG) is negative, −0.118, suggesting that in some areas, there are declines. Forest Area (Ln FA) is relatively high in mean (3.135), implying high forest coverage. Electricity Production from Oil (Ln EPOS) is the most highly standardized variable among the electricity production measures, which also includes Electricity Production from Nuclear (Ln EPNS), with the highest standard deviation, making one notice evident cross-country variation. Urban Population Growth (−6.098) and Forest Area (−1.698) are minimum values indicating significant differences between urban expansion and deforestation. Forest Area (−2.263), Urban Population Growth (−1.649), and Electricity Production from Oil (−0.600) are the most negatively skewed variables, whereas Electricity Production from Oil (−0.600) is close to normal distribution. Values of Kurtosis for Urban Population Growth (8.306) and Forest Area (8.914) are extreme, whereas Ecological Performance (3.639) and Electricity Production from Oil (3.306) are normally distributed. With these in mind, further econometric analysis of these data must account for these key data characteristics (variability and skewness). The trends in the variables are described as shown in the appendix (Appendix Figure A1).

Table 3 shows the correlations between the study variables, and the correlation heat map is shown in the appendix (Appendix Figure A2). Urban Population Growth (Ln UPG) has a negative correlation with Ecological Performance (Ln EP) (−0.273), Forest Area (Ln FA) (−0.130), and Electricity Production from Oil (Ln EPOS) (−0.057); therefore, it argued that increases in these variables are likely related to lower ecological performance (higher CO₂ emissions). However, electricity production from nuclear (Ln EPNS, 0.135) has a weak positive correlation with ecological performance, which indicates a weak correlation with lower emissions. A negative correlation between Ln EP and Ln APD (−0.693) indicates that higher precipitation levels can significantly affect ecological outcomes. Also, Ln FA and Ln EPOS correlate moderately positively (0.319), which implies that countries with more forest coverage will also rely more on oil-based electricity. Most independent variables show weak correlations, which means there is little to no multicollinearity problem that should affect the rest of the econometric analysis.

Table 4 presents cross-sectional dependency test findings, indicating a larger chance of cross-sectional interdependence. The study distinguishes between two generations of unit root testing to verify cross-sectional dependence. The second-generation CIPS tests are used to investigate stationary variables. Table 5 presents findings from the current study’s examination of CIPS for unit roots. The first difference may reject the unit root null hypothesis, demonstrating that variables remain stationary at first difference at 1% significance level.

Table 6 presents the results of the Pedroni, Kao, and Westerlund cointegration test and indicates long-run equilibria between the variables. The results for the Pedroni test show strong evidence of cointegration, with the Panel V Statistic being 2.654 (p < 0.01), Panel PP Statistic being −3.842 (p < 0.01), and Panel ADF Statistic being −3.103 (p < 0.01) rejecting the null hypothesis that there is no cointegration at the 1% significance level. Furthermore, the Panel rho Statistic (−1.932, p < 0.05) also shows cointegration at a 5% level. These findings are further supported by the Kao test, which has a statistic of −2.893 (p < 0.01), indicating a long-run relationship. As with the Westerlund test, the statistics for the Westerlund test is −2.893 (p < 0.01), thus adding robustness to the conclusion that the variables are cointegrated.

As shown in Table 7, the Fully Modified Ordinary Least Squares (FMOLS) estimation results show a significant long-run relationship between the independent variables and CO₂ emissions. The coefficients imply that Urban Population Growth (0.456, p < 0.01) and Electricity Production from Oil (0.302, p < 0.01) are positively and significantly contributing to CO₂ emissions. On the other hand, Forest Area (−0.123, p < 0.05) and Electricity Production from Nuclear Sources (−0.212, p < 0.05) have the opposite effects on CO₂ emissions and thus serve as a mitigating factor to environmental degradation. Weakly significant at p < 0.10, Average Precipitation (0.098) indicates that changes in precipitation levels might have a moderate impact on emissions.

As shown in Table 8, the Dynamic Ordinary Least Squares (DOLS) results also indicate that Urban Population Growth (0.472, p < 0.01) and Electricity Production from Oil (0.285, p < 0.01) have a very strong positive relationship with emissions, while Forest Area (−0.115, p < 0.05) and Electricity Production from Nuclear Sources (−0.198, p < 0.05) remain a negative and significant relationship. Both estimation methods show a consistent but weak relationship, with the coefficient for Average Precipitation (0.102, p < 0.10) remaining marginally significant. Overall, the FMOLS and DOLS results are very close and confirm the robustness of the findings. Therefore, the conclusion is that urbanization and oil-based electricity production drive high emissions, whereas forest expansion and nuclear energy adoption are important in reducing environmental impact.

To ascertain the existence of autocorrelation and heteroscedasticity, we employ the Wooldridge test and the Wald test. The findings presented in Table 9 demonstrate the presence of autocorrelation and heteroscedasticity within our model. To remedy this, we implement robust standard errors in the variance-covariance matrix estimator.

Table 10 displays the results of the multicollinearity employed by the VIF test. The VIF test’s results LnUPG (2.647), LnFA (2.746), LnEPOS (2.242), LnEPNS (1.781), LnAPD (1.668), show that all the VIF values are less than 5. Therefore, these findings indicate that there is no multicollinearity among the variables. Therefore, the mean value of VIF is 2.216. Additionally, the moderate strength of correlations shown in the correlation matrix between variables supports the conclusion that intense multicollinearity is absent because not pairwise correlation measure exceeds 0.7. The findings based on these results demonstrate both reliable regression estimations and valid conclusions.

Table 11 demonstrates that the Dumitrescu-Hurlin test rejects the null hypothesis of no Granger causality for all independent variables (Z-bar p < 0.05), confirming endogeneity. Therefore, the GMM system is employed to mitigate bias.

Table 12 displays the outcomes of system GMM estimation and fixed-effect regression results. The results show that both two models maintain the same sign and significance for key variables, verifying consistency. However, System GMM provides more efficient and unbiased estimates by addressing endogeneity, autocorrelation, and heteroscedasticity through instrumental variables. Unlike Fixed Effects, GMM incorporates lagged dependent variables, making it ideal for dynamic panel data analysis. Therefore, the Lag of ecological performance has a statistically significant and positive impact on ecological performance in BRI countries. The first hypothesis is accepted as per the result (coefficient = 0.067, t-statistic = 3.157, and p < 0.1) at 1% level of significance. Urban population growth has a statistically significant and positive impact on BRI countries’ ecological performance (CO₂). The study’s findings are consistent with the prior research (Liu et al., 2021; Luqman et al., 2021; Yan et al., 2023). Moreover, the forest area has a significant negative impact on ecological performance (CO₂) in BRI countries. The outcomes (coefficient = −0.075, t-statistic = -4.231, and p < 0.01) indicate the second hypothesis has been accepted at 1% level of significance, and this finding is aligned with the previous research (Begum and Raihan, 2020; H. Li and Zhang, 2022; M. Y; Raza et al., 2021; Tourism and Management, 2023). Electricity production from oil sources has a significant negative impact on the ecological performance of BRI nations. The finding (coefficient = 0.040, t-statistic = 3.379, and p < 0.01) shows that the third hypothesis is accepted at a 1% level of significance. The results are aligned with the prior research (Majumder et al., 2023; Voumik et al., 2022; 2023; Yilmaz and Sensoy, 2022). Therefore, electricity production from nuclear sources has a significant negative impact on ecological performance (CO₂) in BRI countries. The outcomes (coefficient = −0.015, t-statistic = -2.245, and p < 0.05) indicate the second hypothesis has been accepted at 5% level of significance, and this finding is aligned with the previous research (Mahmood, 2022; Mohamed and Razek, 2023; Petruška and Litavcov, 2022). Finally, the average precipitation in depth has also significantly and negatively impacted BRI nations’ ecological performance (CO₂). The finding (coefficient = −0.828, statistic = -22.349, and p < 0.1) shows that the third hypothesis is accepted at a 1% level of significance. The results are aligned with the prior research (Mahmood, 2022; Mohamed and Razek, 2023; Ni et al., 2019; Petruška and Litavcov, 2022; Vidon et al., 2015; Zhang et al., 2023).

Understanding the role of urban population growth, forest area, electricity production, and average precipitation in CO₂ emissions is crucial for policy-making and urban planning. The strategy also encourages the sustainable development of cities through integrated city designs, green infrastructure, and efficient transport systems. Protection and reforestation of forests also contribute to mitigating CO₂ and biodiversity conservation. Energy transition strategies are shaped by the knowledge of the emissions intensity of different generation sources of electricity, promoting the adoption of renewable energy and fueled phase-out of fossil fuel-based electricity generating units. Climate-resistant structures and water management techniques cannot be developed, provided the linkage between the mean precipitation patterns and CO₂ emissions is understood. This information provides the basis for integrating strategies to implement the complexity of the linkages between urbanization, land use changes, energy production, precipitation patterns, and CO₂ emissions, facilitating the implementation of more efficient mitigation and adaptation actions.

Based on these findings, it is recommended that BRI governments adopt specific policy implications to enhance ecological sustainability. Dedicated finance could be raised by levying a carbon sink on deforestation-affected infrastructure projects, creating an opportunity for community-led reforestation funds. For instance, dedicated environmental financing mechanisms, such as green bonds aligned with the EU Taxonomy, can be mandated for BRI infrastructure projects that emit above a certain threshold. Governments could introduce carbon-offset levies on projects that significantly reduce forest cover, redirecting those funds toward community-led reforestation initiatives. Moreover, establishing a BRI-wide Environmental Scorecard would enable annual assessments of project-level emissions, thereby linking future funding to performance. Urban planning could adopt ‘transit-oriented development’ models, ensuring that at least 50% of urban residents are located within a 1-km radius of public transport hubs, thereby reducing emissions from private vehicles. These policy actions reflect scalable, evidence-based strategies suitable for BRI nations at varying income levels.

The study on the effect of urban population growth, forest area, electricity production, and average precipitation on CO₂ emissions is limited. The complexity of these interactions makes it very hard to trace the specific effects of each factor on CO₂ emissions. Moreover, Future research must utilize detailed datasets combined with refined modeling processes like machine learning to better represent non-linear data patterns. While the study addresses endogeneity through GMM estimation, this study acknowledges that omitted variable bias may still be present, particularly for variables related to technological innovation, energy subsidies, or governance quality, which are challenging to capture uniformly across BRI countries. Unobserved heterogeneity, despite the inclusion of country and time-fixed effects, may also influence the results due to structural differences in institutional capacity or environmental regulation. Furthermore, the dataset spans the COVID-19 pandemic period (2020–2021), which introduced disruptions in economic activity and emissions. Although the panel structure partially absorbs such shocks, their unique effects may influence short-term dynamics that this research model does not fully capture. Analyzing BRI sub regions through longitudinal studies and interdisciplinary research units combining environmental science elements with economics and policy analysis would create a holistic understanding of sustainable development matters.