Upper ocean three-dimensional thermal structure inversion based on Transformer model

• Published in: 2024 5th International Conference on Computer Vision, Image and Deep Learning (CVIDL) pp. 475 - 478
• Main Authors: Zhao, Yang, Fan, Ruimin, Li, Ruotong, Yu, Fangjie, Chen, Ge
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.04.2024
• Subjects: Data models; Deep learning; Ocean thermal structure; Satellite remote sensing data; Sea surface; Surface reconstruction; Temperature distribution; Transformers; Tropical cyclones; Typhoon
• DOI: 10.1109/CVIDL62147.2024.10603754

The three-dimensional temperature structure under the influence of typhoon can deeply reveal the energy distribution and change mechanism of typhoon, but due to the complexity of the internal environment of typhoon and the limitation of observation conditions, etc., the three-dimensional thermal structure data of the ocean under the influence of typhoon are still insufficient. To address this problem, this paper is based on the deep learning Transformer model, in which a spatial module is introduced to reconstruct the ocean thermal structure dataset under the influence of the typhoon in a hierarchical manner by learning the relationship between the satellite remotely sensed sea surface data and the temperature profile within the typhoon influence area, with a total of 46 layers ranging from 5 meters to 100 meters. Meanwhile, the matched Argo data under the influence of the typhoon are utilized for accuracy assessment, and the results show that the RMSE is 0.92°C and the MAE is 0.77°C. This study is of great significance for improving the accuracy of typhoon forecasting and understanding the dynamics of the internal structure of the typhoon.