Scholarlink

<p>Tropical cyclones (TCs) pose significant risks in coastal region; however, traditional numerical models have struggled with forecasting TC intensity, especially rapid intensification (RI). Here we develop an OWZP-Transformer model which incorporates 15 input variables including the Okubo-Weiss-Zeta Parameter (OWZP) as the structural parameter to predict TC track, intensity, and 24 h future intensity change simultaneously over the western North Pacific (WNP). The OWZP-Transformer achieves competitive performance for track prediction and has a 25.1–37.8% improvement in intensity and 37.4–54.8% improvement in 24 h intensity change prediction compared to existing models. It successfully identifies over 60% RI events from the test dataset during 2020–2023. We further perform ablation experiments to evaluate the impact of each predictor category on model performance and investigate the relative contribution of input variables using two explainable feature importance methods. This study highlights OWZP-Transformer as a reliable tool for enhancing both accuracy and efficiency of TC predictions.</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

Enhancing tropical cyclone track and intensity predictions with the OWZP-Transformer model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月3日

查看全文
  • Zihao Lin,
  • Jung-Eun Chu,
  • Yoo-Geun Ham

摘要

Tropical cyclones (TCs) pose significant risks in coastal region; however, traditional numerical models have struggled with forecasting TC intensity, especially rapid intensification (RI). Here we develop an OWZP-Transformer model which incorporates 15 input variables including the Okubo-Weiss-Zeta Parameter (OWZP) as the structural parameter to predict TC track, intensity, and 24 h future intensity change simultaneously over the western North Pacific (WNP). The OWZP-Transformer achieves competitive performance for track prediction and has a 25.1–37.8% improvement in intensity and 37.4–54.8% improvement in 24 h intensity change prediction compared to existing models. It successfully identifies over 60% RI events from the test dataset during 2020–2023. We further perform ablation experiments to evaluate the impact of each predictor category on model performance and investigate the relative contribution of input variables using two explainable feature importance methods. This study highlights OWZP-Transformer as a reliable tool for enhancing both accuracy and efficiency of TC predictions.