中国における炭素排出の省別差異とモデルベースの要因帰属：群知能最適化ランダムフォレストとシナリオ分析
Provincial differences and model-based driver attribution of carbon emissions in China: A swarm intelligence–optimized random forest and scenario analysis (原題)
Yiyang Luo, Jiameng Ren, Ruichen Wang, Chen Wang, Dongjie Niu, Jingyang Liu, Feilong Zhang
🤖 gxceed AI 要約
日本語
中国30省の炭素排出について、STIRPAT枠組みに群知能最適化ランダムフォレストを組み込み、人口・経済・技術・産業・エネルギーの12指標から要因を帰属分析した。AOAが選んだモデルは決定係数0.9784と高精度で、総エネルギー消費と化石エネルギー消費が最大の寄与要因だった。2023〜2060年の3シナリオ予測では、低炭素・強化削減シナリオで2060年に2022年比17.26%減・40.64%減となり、省ごとに変化率が異なる。
English
This study applies a swarm intelligence–optimized Random Forest within an extended STIRPAT framework to attribute provincial carbon-emission drivers across 30 Chinese provinces using 12 indicators. The AOA-selected model achieved R²=0.9784, with total and fossil energy consumption as top drivers. Scenario projections to 2060 show national emissions rising 7.38% under business-as-usual but falling 17.26% and 40.64% under low-carbon and enhanced-reduction scenarios, with heterogeneous provincial pathways.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
中国の省別炭素排出ドライバー分析と長期シナリオは、日本のSSBJ・有報でのScope3や気候関連リスク開示を検討する企業にとって、地域別排出要因の定量手法として参考になる。特にサプライチェーン上流の中国拠点を持つ日本企業の移行計画策定に示唆を与える。
In the global GX context
This paper contributes to global climate disclosure scholarship by demonstrating how machine-learning driver attribution and scenario analysis can support jurisdiction-specific transition planning, relevant to ISSB/TCFD-aligned reporting where companies must disclose climate scenarios and regional emission drivers. It offers a replicable framework for subnational carbon accounting that could inform CSRD and SEC climate disclosure practices.
👥 読者別の含意
🔬研究者:機械学習とSTIRPATを組み合わせた省別排出要因帰属と長期シナリオ予測の手法を、中国の実証データで検証した点が参考になる。
🏢実務担当者:中国に拠点を持つ企業は、省別の排出ドライバーと削減シナリオを自社の移行計画やScope3算定に活用できる。
🏛政策担当者:地域別の排出要因と削減ポテンシャルを定量的に示す手法は、地方自治体のカーボンニュートラル政策設計に有用である。
📄 Abstract（原文）
Assessing the long-term environmental implications of low-carbon development pathways requires methods that can capture nonlinear patterns and provincial differences. This study develops a swarm intelligence–optimized Random Forest model within an extended Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) framework for provincial carbon-emission driver attribution. Twelve indicators across five dimensions—population, economy, technology, industry, and energy—are incorporated. Six swarm intelligence algorithms optimize Random Forest hyperparameters, and Shapley additive explanations are used to interpret the model selected by the Arithmetic Optimization Algorithm (AOA). On the 2019–2022 temporal holdout, the selected model achieved a coefficient of determination of 0.9784 and a root mean squared logarithmic error of 0.1101. Total energy consumption and fossil energy consumption ranked first and second in mean absolute attribution values, while province-grouped attribution summaries showed differences in driver-attribution patterns across provinces. Three scenarios—business as usual, low carbon, and enhanced emission reduction—are used for conditional projections from 2023 to 2060 using a separate dynamic ridge-regression model developed within the same theoretical framework. Relative to 2022, national emissions in 2060 increase by 7.38% under the business-as-usual scenario but decline by 17.26% and 40.64% under the low-carbon and enhanced emission-reduction scenarios, respectively. All 30 provincial projections are below their 2022 levels under both emission-reduction scenarios by 2060, and the projected rates of change vary across provinces. By linking model-based driver attribution with conditional scenario projections, the framework provides a quantitative basis for comparing low-carbon pathways and provincial transition options.
🔗 Provenance — このレコードを発見したソース
- primary_source https://doi.org/10.1016/j.eti.2026.105243first seen 2026-09-27 23:54:00
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。