Scholarlink

<p>Summer extreme hot events led to catastrophic consequences on North America’s public health, the economy, and crop production. While a significant advance has been made in applying machine learning (ML) to weather forecasts, whether ML is a winner in seasonal climate prediction of extreme hot events over North America is still uncertain. Here, we analyze the spatiotemporal characteristics of the leading modes of extreme high temperature days (EHDs) over western North America (WNA) and create a precursory predictor library for each of the leading EHDs modes. We then construct ML-based prediction models using the library. Although the ML-based models performed perfectly during training, their performance during independent prediction were less satisfactory. In comparison, a physics-based empirical (PE) model using six physical meaningful predictors showed better prediction skills than the ML models. Notably, the PE model can predict abnormal WNA-EHDs in 2021, while all the ML-based models failed. These results highlight the limitations of pure data-driven approaches without the physical constraints, and emphasize the continued value of physically grounded models in seasonal climate prediction.</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

Seasonal prediction of extreme high temperature days in western North America: a comparison of physics-based and machine learning models

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Hui Tan,
  • Zhiwei Zhu,
  • Tim Li,
  • Bin Wang

摘要

Summer extreme hot events led to catastrophic consequences on North America’s public health, the economy, and crop production. While a significant advance has been made in applying machine learning (ML) to weather forecasts, whether ML is a winner in seasonal climate prediction of extreme hot events over North America is still uncertain. Here, we analyze the spatiotemporal characteristics of the leading modes of extreme high temperature days (EHDs) over western North America (WNA) and create a precursory predictor library for each of the leading EHDs modes. We then construct ML-based prediction models using the library. Although the ML-based models performed perfectly during training, their performance during independent prediction were less satisfactory. In comparison, a physics-based empirical (PE) model using six physical meaningful predictors showed better prediction skills than the ML models. Notably, the PE model can predict abnormal WNA-EHDs in 2021, while all the ML-based models failed. These results highlight the limitations of pure data-driven approaches without the physical constraints, and emphasize the continued value of physically grounded models in seasonal climate prediction.