MetaCarbon: a provincial carbon emission peak path prediction method based on dynamic spatiotemporal partitioning
Year: 2026
Type: article
Abstract: Accurate provincial carbon-emission forecasts are essential for planning China’s carbon-peaking and carbon-neutrality pathways. This study develops a forecasting framework that incorporates regional heterogeneity, temporal information, and uncertainty in future driving variables. Feature selection uses the average normalized SHAP importance across the four base learners of the Stacking ensemble—XGBoost, Random Forest, gradient boosting decision trees, and AdaBoost—to identify GDP, secondary industry share, energy intensity, energy consumption ratio, and nighttime lights as predictors. TS-EKC–K-means then divides 30 provinces into four groups according to composite GDP and CO₂ growth factors that incorporate temporal and spatial information. An ALA–Stacking–ARIMA model combines hyperparameter optimization, multivariate prediction, and time-series adjustment, while Monte Carlo simulations propagate uncertainty in future drivers under business-as-usual, mitigation, and high-growth scenarios. Compared with basic Stacking, the proposed model reduces RMSLE by 22.7%–33.3% across the four clusters and achieves a maximum R 2 of 0.9912. The scenario projections show substantial differences in provincial peak timing and emission levels. Mitigation-oriented assumptions generally lead to earlier peaking and lower peak emissions, whereas faster economic expansion tends to increase peak levels and delay peaking. Scenario rankings also vary across provinces, indicating the need to consider regional growth patterns and emission characteristics when developing mitigation strategies. The framework provides conditional projections of provincial carbon-peaking pathways to inform planning under alternative economic and energy-development assumptions.
Source: Carbon Balance and Management
Institutions: Chinese Academy of Sciences, Institutes of Science and Development, China University of Labor Relations, Chongqing University of Posts and Telecommunications, Shanxi University, Shanxi University of Electronic Science and Technology, Taiyuan University of Science and Technology, Institute of Industrial Science, The University of Tokyo, The University of Tokyo
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $1,890
OpenAlex ID: W7233568010
FWCI:
Cites:
Cited by:
Related to:
