Deep learning for Lagrangian drift simulation at the sea surface

Daria Botvynko (1, 2, 3, 4) , Carlos Granero-Belinchon (1, 2) , Simon Van Gennip (3) , Abdesslam Benzinou (4, 1) , Ronan Fablet (1, 2)

1 Lab-STICC_OSE - Equipe Observations Signal & Environnement
2 IMT Atlantique - MEE - Département Mathematical and Electrical Engineering
3 Mercator Océan
4 ENIB - École Nationale d'Ingénieurs de Brest

We address Lagrangian drift simulation in geophysical dynamics and explore deep learning approaches to overcome known limitations of state-of-the-art model-based and Markovian approaches in terms of computational complexity and error propagation. We introduce a novel architecture, referred to as DriftNet, inspired from the Eulerian Fokker-Planck representation of Lagrangian dynamics. Numerical experiments for Lagrangian drift simulation at the sea surface demonstrates the relevance of DriftNet w.r.t. state-of-the-art schemes. Benefiting from the fully-convolutional nature of Drift-Net, we explore through a neural inversion how to diagnose modelderived velocities w.r.t. real drifter trajectories.

Mots clés: Lagrangian; Deep Learning; Ocean; Trajectory Generation; Inversion

Domaines: Océan, Atmosphère; Intelligence artificielle [cs.AI]; Traitement du signal et de l'image [eess.SP]

Daria Botvynko, Carlos Granero-Belinchon, Simon Van Gennip, Abdesslam Benzinou, Ronan Fablet. Deep learning for Lagrangian drift simulation at the sea surface. ICASSP 2023: IEEE International Conference on Acoustics, Speech, and Signal Processing, Jun 2023, Rhodes, Greece. ⟨10.1109/ICASSP49357.2023.10094622⟩. ⟨hal-03852489⟩