Study on the Forecasting of Internal Solitary Wave Propagation in the Andaman Sea Using Joint Ascending-Descending Orbit Sentinel-1A Data and Machine Learning
Internal solitary waves (ISWs) are extensively present in the Andaman Sea. Their generation and propagation can cause damage to offshore engineering structures and pose a threat to the safe navigation of submarines. Consequently, the ability to accurately predict ISWs is of considerable significance...
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/10812683/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Internal solitary waves (ISWs) are extensively present in the Andaman Sea. Their generation and propagation can cause damage to offshore engineering structures and pose a threat to the safe navigation of submarines. Consequently, the ability to accurately predict ISWs is of considerable significance. This study addresses this challenge by analyzing the primary factors influencing ISWs propagation in the Andaman Sea, revealing that both water depth and stratification significantly affect propagation velocity. A CNN-LSTM hybrid neural network model, based on multiple features and the self-attention mechanism, has been constructed by incorporating the characteristics of ISWs (ISW-Attention-CNN-LSTM Net, IACL Net), aiming to predict the propagation of ISWs in the Andaman Sea. The dataset for this model was derived from Sentinel-1A ascending and descending orbit images of the Andaman Sea, spanning the years 2018–2023. From these, 864 images displaying clear ISWs characteristics were selected, resulting in the extraction of 5383 ISWs sample points along with their associated oceanic environmental parameters. The model utilizes the leading wavefront of the ISW wave packet as input to generate predictions of ISWs, which are subsequently compared with satellite observation data. The model's prediction results indicate that after one semidiurnal tidal cycle, the RMSE between the predicted dominant ISW wave crest and satellite observations is 3.891 km. Running the model over two semidiurnal tidal cycles produced similar results. Compared with other machine learning models, the prediction performance shows improvements across various metrics, demonstrating the model's robustness in predicting ISW. |
|---|---|
| ISSN: | 1939-1404 2151-1535 |