Scholarlink

<p>Stringent emission controls inadvertently trigger an “aerosol-unmasking” response, where altered surface energy balances accelerate secondary aerosol formation, generating a nonlinear climate penalty (aerosol rebound). Traditional physical models struggle to resolve this across highly heterogeneous environments. To address this issue, we propose a physics-informed, causally weighted ensemble deep learning framework as a generalized modeling paradigm. By coupling PCMCI-based causal discovery with explicit dynamical features, the model incorporates neighboring spatial information and atmospheric-process signals, thereby improving AOD prediction under regionally heterogeneous atmospheric conditions. To unravel the underlying mechanisms and quantify radiation-driven aerosol rebound, we design a counterfactual inference pipeline. Global application reveals strong spatial selectivity in radiatively driven thermodynamic restructuring, with AOD rebounds mainly concentrated in Central Africa and Southeast Asia. We then extend the analysis to China by constructing a 2030 counterfactual scenario to evaluate deep-mitigation trajectories. The results indicate that seasonal precursor accumulation may modulate the climate penalty, with wintertime buildup and high spring radiative sensitivity associated with the strongest nonlinear rebounds, whereas summer rebounds are weaker under precursor-depleted conditions. These findings suggest that air quality management should shift from uniform mass-reduction strategies toward seasonally dynamic regulation during high-risk periods.</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

Quantifying radiative feedbacks on aerosol rebound: a causally informed deep learning framework

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月16日

查看全文
  • Gan Li,
  • Ming Zhang,
  • Lunche Wang,
  • Wenmin Qin

摘要

Stringent emission controls inadvertently trigger an “aerosol-unmasking” response, where altered surface energy balances accelerate secondary aerosol formation, generating a nonlinear climate penalty (aerosol rebound). Traditional physical models struggle to resolve this across highly heterogeneous environments. To address this issue, we propose a physics-informed, causally weighted ensemble deep learning framework as a generalized modeling paradigm. By coupling PCMCI-based causal discovery with explicit dynamical features, the model incorporates neighboring spatial information and atmospheric-process signals, thereby improving AOD prediction under regionally heterogeneous atmospheric conditions. To unravel the underlying mechanisms and quantify radiation-driven aerosol rebound, we design a counterfactual inference pipeline. Global application reveals strong spatial selectivity in radiatively driven thermodynamic restructuring, with AOD rebounds mainly concentrated in Central Africa and Southeast Asia. We then extend the analysis to China by constructing a 2030 counterfactual scenario to evaluate deep-mitigation trajectories. The results indicate that seasonal precursor accumulation may modulate the climate penalty, with wintertime buildup and high spring radiative sensitivity associated with the strongest nonlinear rebounds, whereas summer rebounds are weaker under precursor-depleted conditions. These findings suggest that air quality management should shift from uniform mass-reduction strategies toward seasonally dynamic regulation during high-risk periods.