An advanced machine learning framework for predicting climate warming from greenhouse gas emissions

Abstract

The present research investigated the emissions of greenhouse gases (GHGs), namely carbon dioxide (CO₂), methane (CH₄), and nitrous oxide (N₂O), and their impact on the global mean surface temperature rise in India from 1851 to 2020. The emission data were derived from a combination of fossil fuel source emissions and emissions related to land use, land-use change, and forestry (LULUCF). Machine learning models including XGBoost, Random Forest (RF), LightGBM, and Nu Support Vector Regression (NuSVR) were employed to develop a regression models for predicting the total change in temperature based on GHG emissions data. A strong correlation was observed between these emissions and the global temperature rise, with CO₂ exerting the greatest impact. Fossil fuels constituted the primary source of CO₂ emissions, while LUCUCF was the major contributor to CH₄ and N₂O emissions. The results also indicated that these emission sources increased after 1950, possibly due to rapid industrialization, intensified agricultural practices, urbanization, and the greater use of fossil fuels as a major energy source. The Box–Cox transformation was applied to reduce skewness and kurtosis of the datasets. Model performance was evaluated using the correlation coefficient, mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE) on an 80:20 training-to-testing split. The results revealed that although all models performed well, the Random Forest and NuSVR models outperformed XGBoost and LightGBM. This work highlights the potential of machine learning for climate modeling and informs policy decisions aimed at mitigating climate change impacts in developing regions such as India.