Scholarlink

<p>This research focuses on the impact of seasonal segmentation on cyclone landfall prediction for cyclones formed in the Bay of Bengal. Using historical cyclone data from 1983 to 2023, including key meteorological parameters such as atmospheric wind speed, wind direction and sea surface temperature, the study aims to highlight the importance of seasonal patterns in improving prediction accuracy. Unlike previous studies that primarily used non-seasonally segmented data, this research analyzes how breaking down the data into specific seasons enhances the model's ability to predict cyclone landfall with greater precision. Artificial Neural Network (ANN) with backpropagation algorithm, combined with the generalized delta rule, was employed to predict the landfall outcomes. The latitude longitude error (ANN-1) and haversine distance error (ANN-2) are the two different errors used as the loss function for the different ANN models in this study to evaluate their performances. The results demonstrate that the seasonally segmented ANN model achieves an 18.5% reduction in average landfall distance error at a 24-h lead time compared to the non-segmented data, using haversine distance error (ANN-2) as the loss function. The study highlights the crucial role of season-specific Artificial Intelligence-based models in enhancing disaster preparedness and strengthening community resilience against cyclones. By incorporating seasonal variations, these models improve early warning systems, allowing for more accurate predictions that help protect communities and support sustainable infrastructure.</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

A neural network model for cyclone landfall prediction in the Bay of Bengal

  • 期刊论文

  • 发表日期：

    2025年10月4日

查看全文
  • J. Dhayanand,
  • T. R. Neelakantan

摘要

This research focuses on the impact of seasonal segmentation on cyclone landfall prediction for cyclones formed in the Bay of Bengal. Using historical cyclone data from 1983 to 2023, including key meteorological parameters such as atmospheric wind speed, wind direction and sea surface temperature, the study aims to highlight the importance of seasonal patterns in improving prediction accuracy. Unlike previous studies that primarily used non-seasonally segmented data, this research analyzes how breaking down the data into specific seasons enhances the model's ability to predict cyclone landfall with greater precision. Artificial Neural Network (ANN) with backpropagation algorithm, combined with the generalized delta rule, was employed to predict the landfall outcomes. The latitude longitude error (ANN-1) and haversine distance error (ANN-2) are the two different errors used as the loss function for the different ANN models in this study to evaluate their performances. The results demonstrate that the seasonally segmented ANN model achieves an 18.5% reduction in average landfall distance error at a 24-h lead time compared to the non-segmented data, using haversine distance error (ANN-2) as the loss function. The study highlights the crucial role of season-specific Artificial Intelligence-based models in enhancing disaster preparedness and strengthening community resilience against cyclones. By incorporating seasonal variations, these models improve early warning systems, allowing for more accurate predictions that help protect communities and support sustainable infrastructure.