Search

International audience; 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.

The complex patterns observed in marine phytoplankton distributions arise from the interplay of biological and physical processes, but the nature of the balance remains uncertain centuries after the first observations. Previous observations have shown a consistent trend of decreasing variability with decreasing length-scale. Influenced by similar scaling found for the properties of the water that the phytoplankton inhabit, ‘universal' theories have been proposed that simultaneously explain the variability seen from meters to hundreds of kilometers. However, data on the distribution of phytoplankton alone has proved insufficient to differentiate between the many causal mechanisms that have been suggested. Here we present novel observations from a cruise in the North Atlantic in which fluorescence (proxy for phytoplankton), nitrate and temperature were measured simultaneously at scales from 10 m to 100 km for the first time in the open ocean. These show a change in spectra between the small scale (10–100 m) and the mesoscale (10–100 km) which is different for the three tracers. We discuss these observations in relation to the current theories for phytoplankton patchiness.

Phytoplankton are one of the most visible signs of life in our oceans. They also are a key component of the global carbon cycle and of the marine food web. Their complex patterns at the sea surface are routinely seen in satellite images, though the first observations go back centuries.The motivation of this thesis is to explore the spatial properties of phytoplankton. Inspired by ‘universal’ theories for the dynamics of turbulence, several ones have been proposed to explain phytoplankton patchiness as a balance between turbulent stirring by the water and biological processes involving the phytoplankton. The thesis examines the extant theories of plankton patchiness using a novel twin tracer approach, specifically using in situ simultaneous Chlorophyll-a and nitrate measurements from a cruise in the North Atlantic. A significant difference is observed between the variability spectra of the two biochemical variables, an outcome potentially explained only by one theory.More generally, although numerous observations testify to the existence of scaling behaviour of phytoplankton spatial variability, the collation of these studies indicates considerable variability, and hence uncertainty, in the power law behaviour, specifically the value of the spectral ‘slope’. The many different techniques used to evaluate the spectrum, the different sources of data, and the geographical and temporal limitations associated with the data all contribute to adding noise and uncertainties in the estimates for the slope and make a comparison between studies difficult. In this thesis, the existence of the universal scaling properties of phytoplankton are tested over a wide range of spatial (sub-regional and regional) and temporal (few days to a year) scales using in situ, satellite data and model output. For this purpose a robust method is developed that reliably evaluates the spectrum of phytoplankton. A power-law behaviour in the phytoplankton spectrum is consistently found across the sources of data used and the range of scales studied (from 10 m to 130 km). However, stronger universality for the phytoplankton spectrum, defined as constant or uniform slope, is undermined by the significant variability in spectral slope that is consistently demonstrated across the spatial and temporal scales studied.?????????????????????????????????????????????????????????????????????????????????????????????????