Supporting Oman's Decarbonization Strategic Planning: Tree-Based Machine Learning Approaches for Forecasting Carbon Dioxide Emissions

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Provides a comparative evaluation of tree-based algorithms for CO2 forecasting, useful for ML applications in environmental science.

🏢実務担当者:The developed GUI allows practical simulation of emission scenarios, helping energy managers identify priority control measures.

🏛政策担当者:Shows how predictive models can underpin evidence-based decarbonization strategies, especially for oil-dependent economies.

Abstract The escalating population growth and rising energy demand mainly govern greenhouse gases (GHGs) emissions, which pose a serious threat to the global environment. Carbon dioxide (CO2) majorly contributes to GHGs due to its heat-trapping capacity. At COP28 meeting, countries emphasized collaborative actions on environmental issues, and Oman vowed to attain net-zero emissions by 2050. Therefore, it is indispensable to critically analyse vital sources of CO2 emissions in Oman to assist evidence-based and rational decarbonization planning. Historical data (1965-2022) were sourced from the BP Statistical Review of World Energy. We employed a two-stage modelling approach. Initially, using historical data, a Recurrent Neural Networks (RNNs) model was applied to forecast features, including primary energy consumption, oil production, oil consumption, natural gas consumption, coal consumption and electricity generation and a target variable (CO2 emissions) through 2027. Afterwards, tree-based algorithms- Decision Tree Regressor (DTR), Random Forest Regressor (RFR), AdaBoost Regressor (ABR), XGBoost Regressor (XGBR), LightGBM (LGBMR), and Extremely Randomized Trees (ETR)- were used to develop a predictive CO2 emissions model using six features. Results showed a consistent rise in all forecasted inputs during 2023-2027, which can be attributed to Oman's economic development and rapid population growth. Consider forecasting, primary energy consumption and oil production escalated by ~4% and ~2.49%, respectively. Correspondingly, oil consumption, natural gas consumption, coal consumption, and electricity generation increased by ~4.63%, ~3.87%, ~5.46%, and ~2.27%, respectively. Target variable soared from 77.6301 to 80.2573 million tonnes (~3.38%). Model evaluation illustrated R2 and MAE values: ETR (0.9989, 0.8545), RFR (0.9986, 0.9248), XGBR (0.9958, 1.4888), LGBMR (0.9933, 1.9989), ABR (0.9889, 2.5087), and DTR (0.4592, 19.3184). Best-performing tree-based algorithm was ETR followed by RFR, XGBR, LGBMR, ABR, and DTR. All algorithms showed phenomenal performance except DTR, which failed to capture the underlying trend in the dataset. The features’ importance is arranged in descending order: natural gas consumption (0.2191), primary energy consumption (0.1981), electricity generation (0.1921), oil consumption (0.1811), coal consumption (0.1601) and oil production (0.0494). A Graphical User Interface (GUI) was developed based on the best-performing model (ETR) to assist policymakers in evaluating the key contributors to CO2 emissions. Consequently, this study empowers Oman's authorities to make a strategic, targeted CO2 mitigation plan through effective control measures of the dominant emission drivers. Although Machine Learning (ML) has widespread applications in predicting environmental-based parameters, no study has assessed the performance of tree-based algorithms in predicting CO2 emissions in Oman. This work, for the first time, evaluates tree-based algorithms’ performance based on historical and forecasted features and the target variable. A GUI built on the best-performing model can be a decisive tool for Oman's authorities to examine the primary drivers of CO2 emissions and assist in planning CO2 mitigation strategies with sustainable development.

• crossref https://doi.org/10.2118/232651-msfirst seen 2026-05-18 05:28:42 · last seen 2026-05-20 05:49:35