Our research relies on complementarities between numerical approaches (modelling, assimilation, inversion) and observational approaches (spatial and in situ data): simulations help the interpretation of observed variability and the detection/attribution of long-term trends; synthetic data (e.g. SWOT) from high-resolution numerical simulations feed OSSEs guiding the optimization of observing systems, analysis and assimilation of future data; reanalyses and forecasts estimate the past and future 3D state of the ocean variability constrained by observations.
Our methods are based on probabilistic or Bayesian approaches describing the uncertainty of oceanic states (physical and biogeochemical, simulated and observed), in the current context of increasingly complex data flow and the emergence of artificial intelligence methods in oceanography. They are developed in a non-Gaussian multivariate framework, in order to take into account the specific dynamics and the multiscale character of the variability. The long-term objective is to build a more generic framework for solving inverse problems, with the double concern of describing the uncertainty on the produced solutions and controlling the computational costs, in particular via the development of emulators and simplified dynamical models by statistical learning.
These developments aim at several types of applications: (i) ensemble estimation of uncertainties and their temporal evolution, (ii) assimilation of observations into models, in particular for the improvement of CMEMS operational systems; (iii) reconstruction of surface layer dynamics (in the context of nadir, wide-swath and Doppler radar altimetry, and optical imagery missions); (iv) quantification of the intrinsic chaotic and deterministic parts of the observed ocean variability (heat and volume transport, heat content, sea level), (v) interpretation of the observed evolution of key ocean variables based on ensemble simulations, and (vi) detection and attribution of interannual to decadal fluctuations and observed trends.
Our research in data assimilation further targets other application domains (hydrology, glaciology, etc.) in collaboration with the key players in Grenoble.