Scholarlink

<p>Leveraging the Conditional Nonlinear Optimal Perturbation (CNOP) approach, this study introduces an innovative adjoint-free methodology to discern optimal initial errors (OIEs) that influence the prediction of the 2020–2022 multi-year La Niña events in a data-driven transformer model. Both being characterized by three-dimensional (3D) structures and capable of inducing maximum error, two distinct types of OIEs are identified: one with positive OIE in the upper eastern Pacific (POIE) and another with a similar structure but negative OIE (NOIE). Dynamically, these OIEs trigger a strong Bjerknes feedback mechanism, thus leading to rapid error growth. Conducting target observing experiments intensified in these regions with large values of OIEs can further improve the 2020–2022 La Niña events by reducing the prediction errors to 80% at most. From predictability point of view, compared to 2020 and 2022 La Niña, the prediction of the 2021 La Niña is more sensitive to initial conditions, underscoring the insight that the predictability of the consecutive La Niña events is of complexity in nature. The innovative integration of the CNOP approach into a deep learning model offers a new framework for improving ENSO prediction and predictability.</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

Integrating an CNOP analysis into a deep learning model to identify optimal initial errors for 2020–2022 La Niña prediction

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Lingjiang Tao,
  • Rong-Hua Zhang,
  • Wansuo Duan,
  • Lu Zhou,
  • Tiaoye Li

摘要

Leveraging the Conditional Nonlinear Optimal Perturbation (CNOP) approach, this study introduces an innovative adjoint-free methodology to discern optimal initial errors (OIEs) that influence the prediction of the 2020–2022 multi-year La Niña events in a data-driven transformer model. Both being characterized by three-dimensional (3D) structures and capable of inducing maximum error, two distinct types of OIEs are identified: one with positive OIE in the upper eastern Pacific (POIE) and another with a similar structure but negative OIE (NOIE). Dynamically, these OIEs trigger a strong Bjerknes feedback mechanism, thus leading to rapid error growth. Conducting target observing experiments intensified in these regions with large values of OIEs can further improve the 2020–2022 La Niña events by reducing the prediction errors to 80% at most. From predictability point of view, compared to 2020 and 2022 La Niña, the prediction of the 2021 La Niña is more sensitive to initial conditions, underscoring the insight that the predictability of the consecutive La Niña events is of complexity in nature. The innovative integration of the CNOP approach into a deep learning model offers a new framework for improving ENSO prediction and predictability.