Global greenhouse gas reduction forecasting via machine learning model in the scenario of energy transition.

Abstract

Global warming is becoming increasingly serious, with greenhouse gas (GHGs) emissions identified as a principal contributor. In response to the climate crisis, many countries are actively transitioning to renewable energy. Therefore, it is crucial to forecast GHGs emissions across different countries under varying degrees of energy transition to inform decision-making. Previous studies often focused on single regions and overlooked the developmental variance among countries. To address this problem, this study aims to project GHGs emissions in 39 major carbon-emitting countries globally, distinguishing between developed countries (DCs) and developing countries (LDCs). The results show that a 5.39% increase in global GHGs emissions from 2016 to 2021 and a 327.64% rise in the renewable electricity generation of LDCs. Additionally, this research develops various energy transition scenarios, employs Random Forest (RF) for feature selection, and utilizes an Extreme Gradient Boosting (XGBoost) model enhanced by Bayesian Optimization (BO) to forecast GHGs emission levels in DCs and LDCs. The performance test shows that RF-BO-XGBoost has higher stability and accuracy. The projection results indicate that the total emissions from all DCs and all LDCs will decrease as the scenario shifts from the baseline to the high energy transition scenario, by 1.22% and 5.23% respectively. Further, the study quantifies the impacts of energy transitions on GHGs emissions across individual countries, revealing that not all countries are likely to achieve optimal reduction under the high energy transition scenario. This study underscores the influence of transition costs and supports the climate policymaking.