Simultaneously detecting the intensity and position of Southwestern Atlantic Ocean Frontal Zones from satellite-derived SST by a multi-task deep learning model
Accurate detection of the Southwestern Atlantic Frontal Zones (SAFZ) requires high-precision detection of both its intensity and position, which is essential for understanding climate change and the thermohaline balance of the global ocean. The intensity of Frontal Zones (FZ) serves as the foundatio...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IOP Publishing
2025-01-01
|
| Series: | Environmental Research Communications |
| Subjects: | |
| Online Access: | https://doi.org/10.1088/2515-7620/adf313 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate detection of the Southwestern Atlantic Frontal Zones (SAFZ) requires high-precision detection of both its intensity and position, which is essential for understanding climate change and the thermohaline balance of the global ocean. The intensity of Frontal Zones (FZ) serves as the foundation for determining their position, making precise detection crucial. However, existing methods often overlook intensity detection and treat FZ intensity and position detection as separate tasks, leading to reduced computational efficiency and a lack of coupling between intensity and position detection results, which ultimately compromises accuracy. Additionally, traditional methods typically represent FZ position using frontal lines, which fail to capture the full width of the FZ, resulting in incomplete representations. To address these limitations, we propose a multi-task deep learning semantic segmentation model, named Multi-Task Attention D-LinkNet (MTAD-LinkNet), which utilizes D-LinkNet as the backbone. This model enables the simultaneous detection of both SAFZ intensity and position. We construct an intensity-based SAFZ dataset using daily sea surface temperature (SST) data from satellite observations spanning 2010-2019, ensuring comprehensive FZ representations. The dataset is then used to train and evaluate the model. Experimental results show that MTAD-LinkNet accurately detects both SAFZ intensity and position while significantly improving computational efficiency. Furthermore, a seasonal detection experiment confirms that the model maintains coupling between intensity and position detection, achieving high-precision results. |
|---|---|
| ISSN: | 2515-7620 |