Scholarlink

<p>As global attention to climate change intensifies, China, as the world’s largest energy consumer and carbon emitter, faces the dual challenge of improving energy efficiency while the carbon rebound effect (CRE) counteracts efforts to reduce emissions. Based on panel data from Chinese prefecture-level cities from 2010 to 2021. It combines multiple linear regression with machine learning (ensemble and deep learning) to uncover the nonlinear drivers of CRE. Machine learning models capture high-dimensional interactions and threshold effects. SHAP values and ALE plots are employed to quantify the impact pathways of key variables, overcoming the limitations of traditional models. This study employs multiple linear regression, ensemble learning, and deep learning methods, combined with interpretable SHAP values and ALE charts, to systematically identify the core driving mechanisms behind the carbon rebound effect. The main conclusions of this paper are as follows: (1) The CRE exhibits an overall “M”-shaped fluctuation trend, with a spatial pattern of “East &gt; Central &gt; West” and “North &gt; South”; (2) Machine learning models accurately identify DEI, HTD, ER, Water, and UIS2 as the key drivers of CRE. These factors also exhibit significant and consistent threshold effects. This study provides a scientific basis for the formulation of differentiated ecological policies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning analysis of carbon rebound effect dynamics and drivers in Chinese prefecture-level cities

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月27日

查看全文
  • Shunyi Li,
  • Xingxing Li

摘要

As global attention to climate change intensifies, China, as the world’s largest energy consumer and carbon emitter, faces the dual challenge of improving energy efficiency while the carbon rebound effect (CRE) counteracts efforts to reduce emissions. Based on panel data from Chinese prefecture-level cities from 2010 to 2021. It combines multiple linear regression with machine learning (ensemble and deep learning) to uncover the nonlinear drivers of CRE. Machine learning models capture high-dimensional interactions and threshold effects. SHAP values and ALE plots are employed to quantify the impact pathways of key variables, overcoming the limitations of traditional models. This study employs multiple linear regression, ensemble learning, and deep learning methods, combined with interpretable SHAP values and ALE charts, to systematically identify the core driving mechanisms behind the carbon rebound effect. The main conclusions of this paper are as follows: (1) The CRE exhibits an overall “M”-shaped fluctuation trend, with a spatial pattern of “East > Central > West” and “North > South”; (2) Machine learning models accurately identify DEI, HTD, ER, Water, and UIS2 as the key drivers of CRE. These factors also exhibit significant and consistent threshold effects. This study provides a scientific basis for the formulation of differentiated ecological policies.