Determining Tropical Cyclone Centers in Western North Pacific Basin Using an MBAF Model From Himawari-8 Satellite Images
Accurate localization of tropical cyclone (TC) centers is crucial for intensity estimation and track prediction. While traditional methodologies primarily rely on visible light and infrared (IR) imagery, the potential of multichannel data remains largely unexplored. In this article, a novel multiban...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11005392/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate localization of tropical cyclone (TC) centers is crucial for intensity estimation and track prediction. While traditional methodologies primarily rely on visible light and infrared (IR) imagery, the potential of multichannel data remains largely unexplored. In this article, a novel multiband adaptive fusion model is proposed to determine TC centers using Himawari-8 data in the western North Pacific basin. First, the model employs a parallel multibranch structure to selectively learn the distinctive features of each channel group, specifically IR, water vapor, and shortwave infrared channels. Then, to facilitate effective feature fusion across different branches, an adaptive fusion mechanism is introduced. Finally, the model incorporates a self-attention mechanism to capture global dependencies, thus enhancing complementary feature extraction and yielding the final localization results. The model was trained on datasets spanning 2015–2020 and evaluated on images from 2021 to 2022. Ablation experiments were conducted to analyze the contributory effects of different channel combinations and individual modules. The findings demonstrate that the incorporation of multichannel imagery significantly enhances the accuracy of center localization, achieving an overall mean absolute error of 32.42 km. This study leverages the information available from TCs across diverse band images, effectively extracts spatial features, and seamlessly integrates these features to compensate for the deficiencies present in previous models, thus markedly improving the positional accuracy of TC centers. |
|---|---|
| ISSN: | 1939-1404 2151-1535 |