Scholarlink

The damage caused by hurricanes has made them one of the worst natural disasters in recent times. This has made the research on Hurricane Intensity (HI) crucial in order to minimize the impact of this natural disaster. Recent advancements in AI have demonstrated the superior ability of Transformers to perform a variety of tasks, including forecasting based on both time series and imagery data. Since the intrinsic characteristics of HI estimation are analyzing time series numeric data and imagery data, e.g., satellite-captured images, etc., we heuristically aim to apply a transformer model to help forecast the HI. Considering this, in this study, we present a novel Transformer model based on a multi-step approach to forecast 24-h intensity based on the previous 10 time steps. Specifically, we adopt real-world data, including features such as landfall time, local wind speeds, damages, deaths, and cyclone size, in a period of ten years (2010 to 2020). The simulation results demonstrate the effectiveness of the proposed method.

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

Transforming Hurricane Intensity Prediction: Leveraging Transformers for Enhanced Forecasting Accuracy

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Rose Atuah,
  • Khalil Ibrahim Almakrami,
  • Qianlong Wang,
  • Michael McGuire

摘要

The damage caused by hurricanes has made them one of the worst natural disasters in recent times. This has made the research on Hurricane Intensity (HI) crucial in order to minimize the impact of this natural disaster. Recent advancements in AI have demonstrated the superior ability of Transformers to perform a variety of tasks, including forecasting based on both time series and imagery data. Since the intrinsic characteristics of HI estimation are analyzing time series numeric data and imagery data, e.g., satellite-captured images, etc., we heuristically aim to apply a transformer model to help forecast the HI. Considering this, in this study, we present a novel Transformer model based on a multi-step approach to forecast 24-h intensity based on the previous 10 time steps. Specifically, we adopt real-world data, including features such as landfall time, local wind speeds, damages, deaths, and cyclone size, in a period of ten years (2010 to 2020). The simulation results demonstrate the effectiveness of the proposed method.