Deep learning for Lagrangian drift simulation at the sea surface

TL;DR

This paper introduces DriftNet, a deep learning architecture inspired by the Fokker-Planck equation, to improve Lagrangian drift simulation at the sea surface, addressing computational and error issues of existing methods.

Contribution

The paper presents DriftNet, a novel fully-convolutional deep learning model for Lagrangian drift simulation, inspired by the Eulerian Fokker-Planck representation, with demonstrated advantages over existing approaches.

Findings

DriftNet outperforms state-of-the-art schemes in numerical experiments.

Fully-convolutional architecture enables effective neural inversion for velocity diagnosis.

Deep learning approach reduces computational complexity and error propagation.

Abstract

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.