Scholarlink

<p>Jiangsu Province is a pivotal economic and energy-consuming hub in China. To support its “dual-carbon” targets, this study reconstructs energy-related carbon emission (CE) dynamics from 2000 to 2020 using an XGBoost model integrated with 13 multi-source predictors, including nighttime lights, land surface temperature, population, and GDP. Following a comparative performance analysis, multi-dimensional spatial techniques were applied to characterize provincial emission patterns, while an interpretable XGBoost–SHAP framework quantified the nonlinear and interactive effects of key drivers. Results indicate that: (1) XGBoost achieved the highest CE reconstruction accuracy among all tested models; (2) CE followed a two-stage trajectory of rapid growth (2000–2010) and stabilization (2010–2020), with a persistent south-high, north-low spatial gradient; (3) significant spatial heterogeneity was confirmed, with high-value clusters aligned with major economic centers; (4) the CE centroid shifted slightly southward while spatial expansion slowed markedly after 2010, reflecting policy intervention effects; and (5) GDP, NTL intensity, and built-up land expansion were identified as dominant drivers, exhibiting distinct nonlinear thresholds and synergistic interaction effects. These findings provide a data-driven, spatially explicit foundation for evidence-based CE mitigation in rapidly developing regions.</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

Spatiotemporal Evolution and Determinant Analysis of Energy-Related Carbon Emissions in Jiangsu Province Using Interpretable Machine Learning

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • Ge Shi,
  • Lin Sun,
  • Qian Chayn Sun,
  • Jiantao Shi,
  • Chuang Chen,
  • Xinyi Sun,
  • Yilin Xiao

摘要

Jiangsu Province is a pivotal economic and energy-consuming hub in China. To support its “dual-carbon” targets, this study reconstructs energy-related carbon emission (CE) dynamics from 2000 to 2020 using an XGBoost model integrated with 13 multi-source predictors, including nighttime lights, land surface temperature, population, and GDP. Following a comparative performance analysis, multi-dimensional spatial techniques were applied to characterize provincial emission patterns, while an interpretable XGBoost–SHAP framework quantified the nonlinear and interactive effects of key drivers. Results indicate that: (1) XGBoost achieved the highest CE reconstruction accuracy among all tested models; (2) CE followed a two-stage trajectory of rapid growth (2000–2010) and stabilization (2010–2020), with a persistent south-high, north-low spatial gradient; (3) significant spatial heterogeneity was confirmed, with high-value clusters aligned with major economic centers; (4) the CE centroid shifted slightly southward while spatial expansion slowed markedly after 2010, reflecting policy intervention effects; and (5) GDP, NTL intensity, and built-up land expansion were identified as dominant drivers, exhibiting distinct nonlinear thresholds and synergistic interaction effects. These findings provide a data-driven, spatially explicit foundation for evidence-based CE mitigation in rapidly developing regions.