Accéder directement au contenu Accéder directement à la navigation
Communication dans un congrès

Multi-modal deep learning models for ocean wind speed estimation

Abstract : Because it depends on multiple atmospheric and oceanographic variables interacting with each other at the sea surface, accurately forecasting offshore surface wind speed is challenging for oceanographers. However, with the expansion of today's Big Ocean Data, the same offshore site can now be monitored by multiple sensors such as hydrophones, satellites or weather buoys. These data are highly heterogeneous, but each of them can potentially bring complementary information on an ocean process. In this paper, a deep generative model is designed to jointly represent Underwater Passive Acoustics (UPA) and Synthetic Aperture Radar (SAR) images into the same latent space to describe surface wind speed located in the Ligurian Sea (North Western Mediterranean Sea). Qualitative and quantitative results obtained demonstrate that SAR images are able to refine the estimation of UPA for low wind speeds.
Type de document :
Communication dans un congrès
Liste complète des métadonnées
Contributeur : Marie Briec <>
Soumis le : vendredi 8 janvier 2021 - 17:12:25
Dernière modification le : lundi 15 février 2021 - 10:43:01
Archivage à long terme le : : vendredi 9 avril 2021 - 19:28:13


Fichiers éditeurs autorisés sur une archive ouverte


  • HAL Id : hal-03104246, version 1


Clémentin Boittiaux, Paul Nguyen Hong Duc, Nicolas Longépé, Sara Pensieri, Roberto Bozzano, et al.. Multi-modal deep learning models for ocean wind speed estimation. 2020 MACLEAN: MAChine Learning for EArth ObservatioN Workshop, MACLEAN 2020, Sep 2020, Virtual online, France. ⟨hal-03104246⟩



Consultations de la notice


Téléchargements de fichiers