Scholarlink

<p>Artificial Intelligence (AI) holds transformative potential for Earth System Science (ESS), yet its adoption is hindered by the interpretability gap of black-box models. Explainable AI (XAI) addresses this limitation by enhancing model transparency and enabling human-AI collaboration. This review provides an accessible introduction to XAI for ESS researchers, aiming to promote broader AI adoption by clarifying interpretability and addressing common misconceptions. We survey core XAI methodologies and their applications across ESS, highlighting their roles in model interpretation, model development, and scientific discovery. We further identify key challenges, including methodological limitations, compatibility issues, inadequate validation, and propose integrative solutions. To bridge the XAI-ESS gap, we advocate for co-designed frameworks that incorporate domain knowledge, causal inference, and human-centered interfaces. We anticipate that future advances will yield more faithful, efficient, and deeply integrated XAI tools, ultimately strengthening ESS research through reliable AI-enabled insights.</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

Bridging the gap: Explainable AI for advancing Earth System Science

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Feini Huang,
  • Shijie Jiang,
  • Wei Shangguan,
  • Lu Li,
  • Ye Zhang,
  • Ruqing Zhang,
  • Qingliang Li,
  • Danxi Li,
  • Yongjiu Dai

摘要

Artificial Intelligence (AI) holds transformative potential for Earth System Science (ESS), yet its adoption is hindered by the interpretability gap of black-box models. Explainable AI (XAI) addresses this limitation by enhancing model transparency and enabling human-AI collaboration. This review provides an accessible introduction to XAI for ESS researchers, aiming to promote broader AI adoption by clarifying interpretability and addressing common misconceptions. We survey core XAI methodologies and their applications across ESS, highlighting their roles in model interpretation, model development, and scientific discovery. We further identify key challenges, including methodological limitations, compatibility issues, inadequate validation, and propose integrative solutions. To bridge the XAI-ESS gap, we advocate for co-designed frameworks that incorporate domain knowledge, causal inference, and human-centered interfaces. We anticipate that future advances will yield more faithful, efficient, and deeply integrated XAI tools, ultimately strengthening ESS research through reliable AI-enabled insights.