To subscribe to the ORE seminar mailing list, click here.
For a (working draft) list of upcoming seminars, click here.
If you are interested in giving an ORE seminar, please contact us at nosal [at] hawaii [dot] edu.
- This event has passed.
Seminar: WVNet: A SAR Foundational Model with Applications for Air-sea Interaction Research
Yannik Glaser
PhD Candidate in Information and Computer Science
Department of Atmospheric Sciences
University of Hawai’i at Mānoa
Justin Stopa
Associate Professor
Department of Ocean and Resources Engineering
University of Hawai’i at Mānoa
Location Information
**This seminar will be held both in person (Marine Sciences Building 100) and over Zoom**
Zoom Invitation Link: https://hawaii.zoom.us/j/98919200762
Meeting ID: 989 1920 0762
Passcode: 201326
Synthetic aperture radars (SAR) aboard space-borne satellites measure sub-mesoscale oceanic and atmospheric phenomena at a global scale. This work uses the European Space Agency (ESA) Sentinel-1 (S-1) SAR mission’s sea surface roughness to study sub-mesoscale phenomena such as turbulence in the atmosphere, rain, and slicks. To utilize the large S-1 archive (>1Pb), we develop automatic image detection methods using deep learning. We developed a foundational contrastive model trained solely on millions of S-1 images. This model, called WVNET, exceeds the performance of other models such as ImageNet in a variety of tasks including regression and classification. With the improved model performance, we have more confidence in estimating the climatology of the sea surface roughness morphology. We find the atmosphere dominates the SAR imagery. The time-space mapping of WVNet’s predictions is relevant for the study of air-sea interactions.