Scholarlink

<p>The accelerating rise in atmospheric carbon dioxide (CO<sub>2</sub>) concentration poses a critical challenge for achieving China’s dual carbon goals. To uncover the complex mechanisms driving this trend, this study integrates high-resolution satellite observations with socioeconomic, ecological, and meteorological datasets to analyze the spatiotemporal dynamics of column-averaged atmospheric CO<sub>2</sub> (XCO<sub>2</sub>) across China from 2015 to 2020. Using an ensemble machine learning framework centered on the Random Forest (RF) model, this research moves beyond traditional linear regression to detect nonlinear thresholds, variable interactions, and spatial heterogeneity in the determinants of XCO<sub>2</sub>. Spatiotemporal analysis reveals a steady national increase in XCO<sub>2</sub>, accompanied by pronounced seasonal and regional fluctuations. The RF-based feature importance and partial dependence analyses expose several critical nonlinear relationships: XCO<sub>2</sub> increases sharply with Open Source Data Inventory of Anthropogenic Carbon Dioxide emissions up to 500,000&#xa0;kg m<sup>− 2</sup> h<sup>− 1</sup>; declines by 0.8 ppm for every 0.1 increase in Normalized Difference Vegetation Index beyond the 0.5 threshold; shifts from a negative to positive temperature correlation around 25&#xa0;°C; and follows a U-shaped pattern with relative humidity, decreasing below 72% and rising thereafter. Wind speed exhibits spatially heterogeneous effects, highlighting region-specific atmospheric circulation patterns. Building upon these machine learning insights, a scenario-based forecasting module is developed to project XCO<sub>2</sub> trajectories for the next decade under regulated versus unregulated emission pathways. The results demonstrate that targeted emission reduction and vegetation restoration strategies can substantially mitigate future CO<sub>2</sub> accumulation. By combining spatial analysis with nonlinear machine learning, this study not only advances the methodological frontier for atmospheric carbon research but also offers data-driven guidance for precision emission governance in China.</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 nonlinear drivers of atmospheric CO2 concentration in China: spatiotemporal patterns and policy implications

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月30日

查看全文
  • Wenxi Shan,
  • Mengshi Yan

摘要

The accelerating rise in atmospheric carbon dioxide (CO2) concentration poses a critical challenge for achieving China’s dual carbon goals. To uncover the complex mechanisms driving this trend, this study integrates high-resolution satellite observations with socioeconomic, ecological, and meteorological datasets to analyze the spatiotemporal dynamics of column-averaged atmospheric CO2 (XCO2) across China from 2015 to 2020. Using an ensemble machine learning framework centered on the Random Forest (RF) model, this research moves beyond traditional linear regression to detect nonlinear thresholds, variable interactions, and spatial heterogeneity in the determinants of XCO2. Spatiotemporal analysis reveals a steady national increase in XCO2, accompanied by pronounced seasonal and regional fluctuations. The RF-based feature importance and partial dependence analyses expose several critical nonlinear relationships: XCO2 increases sharply with Open Source Data Inventory of Anthropogenic Carbon Dioxide emissions up to 500,000 kg m− 2 h− 1; declines by 0.8 ppm for every 0.1 increase in Normalized Difference Vegetation Index beyond the 0.5 threshold; shifts from a negative to positive temperature correlation around 25 °C; and follows a U-shaped pattern with relative humidity, decreasing below 72% and rising thereafter. Wind speed exhibits spatially heterogeneous effects, highlighting region-specific atmospheric circulation patterns. Building upon these machine learning insights, a scenario-based forecasting module is developed to project XCO2 trajectories for the next decade under regulated versus unregulated emission pathways. The results demonstrate that targeted emission reduction and vegetation restoration strategies can substantially mitigate future CO2 accumulation. By combining spatial analysis with nonlinear machine learning, this study not only advances the methodological frontier for atmospheric carbon research but also offers data-driven guidance for precision emission governance in China.