CMCC Seminar
May 20th at 14:30 in the Geophysics Library (first floor, Viale Carlo Berti Pichat 8)

Ocean currents and atmospheric winds transport a variety of natural (e.g., aerosols, pathogens, water masses, plankton, sediments, etc.) and artificial (e.g., pollutants, floating debris, search and rescue, etc.) Lagrangian materials. We present principled PDE-based methods for the probabilistic prediction, Bayesian estimation, optimal sampling, and machine learning of stochastic flow maps and Lagrangian transport in geophysical fluid flows. We obtain super-accurate schemes for advective transport through flow map composition, achieving minimal errors and strong theoretical guarantees. We derive dynamically orthogonal (DO) equations for stochastic Lagrangian fields and develop Bayesian Lagrangian data assimilation methods. We then apply and showcase information-theoretic decision schemes for Lagrangian adaptive sampling. To characterize cohesion and mixing, we derive objective criteria that predict and classify sets of fluid parcels that remain most coherent/incoherent throughout an extended time interval. We also utilize 3D flow map forecasts to extract subduction regions and processes. We finally develop generative learning methods to infer and predict Eulerian and Lagrangian fields directly from observational data. Results are presented for simulated geophysical flows and real-time at-sea experiments with autonomous sensing platforms in diverse ocean regions and dynamical regimes including the Southern Pacific Ocean – Palau Island region, New England Shelf, Alboran Sea, and Balearic Sea.