19-country G20 study forecasts divergent CO2 pathways through 2035

A peer-reviewed study of 19 G20 economies finds that energy consumption was the dominant statistical driver of carbon dioxide emissions in 17 countries, with the strongest positive effects in China, the United States and India. The research combines machine-learning analysis, an assessment of decoupling between economic growth and emissions, and forecasts extending from 2023 to 2035. Its results indicate that advanced economies may stabilise or reduce emissions, while several emerging economies are projected to continue increasing emissions as economic growth and energy demand rise.

The model’s attribution analysis identified energy consumption as the leading emissions driver in 17 of the 19 economies. GDP and trade per capita were positively associated with emissions, while renewable-energy use was correlated with lower emissions. These are statistical associations within the analysed data rather than proof that any one factor alone caused emissions changes.

The researchers separately applied Tapio’s decoupling index across six five-year windows between 1991 and 2022. Seven countries showed strong decoupling, six showed weak or unstable decoupling, and six remained closely coupled. The study therefore identified substantial variation in how economic growth and emissions developed across the selected economies.

This comparison describes patterns within the historical dataset; it does not establish that a particular policy produced the different decoupling outcomes, and the supplied findings do not define the index categories in further detail. Four machine-learning models were compared for emissions forecasting from 2023 through 2035. A feed-forward neural network had the highest short-run accuracy, reporting an R-squared value of 0.996 and a mean absolute percentage error of 5.44%.

A long short-term memory network generated the most stable long-run projections. Based on the combined driver, decoupling and forecasting results, the authors project that advanced economies will likely stabilise or reduce emissions. China, India, Indonesia and Türkiye are identified as emerging economies expected to face continued emissions increases, with economic growth and rising energy demand identified as important drivers.

These are model-based estimates for future years, not observed emissions outcomes. The authors argue that the results support differentiated emissions strategies rather than a single uniform approach. For high-growth, energy-intensive economies, they identify renewable-energy expansion and energy-efficiency investment as priorities, while already-decoupled nations may need continued support for decarbonisation.

The evidence remains limited to the selected 19-country dataset, aggregate historical relationships and predictive models, so the projections should not be interpreted as direct estimates of the effect of any individual policy intervention.

Subject of Research: Machine-learning analysis of CO2 drivers, decoupling and forecasts across 19 G20 economies

Article Title: Explaining CO2 emission pathways in G20 nations through SHAP analysis decoupling assessment and deep learning forecasting

Article References: Vaddiraju, S. A., & Parwez, S. (2026). Explaining CO2 emission pathways in G20 nations through SHAP analysis decoupling assessment and deep learning forecasting. Discover Sustainability. https://doi.org/10.1007/s43621-026-04471-4

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04471-4

Keywords: CO2 emissions, G20 economies, energy consumption, machine learning, XGBoost, SHAP analysis, decoupling, renewable energy, emissions forecasting, deep learning, Explaining, emission