Deep neural operator modeling frameworks have been recently developed for forecasting time series data. Such machine learning models may be useful in multiple domains including the atmospheric and oceanic ones, and in general, the larger fluids community. The present work investigates the effectiveness of deep neural operator models for reproducing and predicting fluid flows and realistic ocean simulations. We first evaluate the capabilities of such deep neural operator models when trained on simulated two-dimensional fluid flows. We then investigate their application to forecasting ocean circulation in the Gulf of Mexico, Middle Atlantic Bight, and Massachusetts Bay, learning from high-resolution data-assimilative simulations employed for real sea experiments. We confirm that trained deep neural operator models are capable of predicting idealized periodic eddy shedding. For realistic ocean surface flows, they can predict several of the features and show some skill, providing potential for applications.

Mesoscale features and their corresponding submesoscale structures can vertically transport heat, freshwater, and biogeochemical tracers (i.e., phytoplankton, oxygen, and carbon) from the surface to the stratified pycnocline or interior. This study examines the evolution of small-scale eddies formed by baroclinic instabilities of the Northern and Balearic Currents in the Western Mediterranean Sea. During the CALYPSO 2022 experiment, eight gliders were programmed to dive up to 700m from 25 March to 21 June 2022. The glider fleet measured temperature, salinity, velocity, chlorophyll fluorescence, oxygen, and acoustic backscatter. The data was mapped objectively in space and time on 10m vertical levels. Vertical and ageostrophic horizontal velocities were estimated from the omega equation. The analysis shows an uplift of the isopycnal surface 28.9 around 100m in 10km as a cyclonic eddy (CE) formed, which can nourish the euphotic layer through phytoplankton enhancement in the center. The CE has an asymmetric shape with ~25km width and ~35km length. Downward vertical velocities (w) ~20 m day^-1 developed around the eddy. As it developed, the CE axis shifted westward. After the first CE dissipated, the 28.9 kg/m³ isopycnal shoaled again in the east as a second CE formed, with the largest observed relative vorticity (~0.5f). The eddy axis shifted westward during growth. The largest downward vertical velocities during eddy intensification were 30 m day^-1, with the size ranging ~25km. Also, the positive values of the barotropic term show a transfer of mean kinetic energy to eddy kinetic energy on the Western side of the CE. Then, the new cyclonic feature spread over a few days before splitting into two 15km CEs on 2 May. The two smaller CEs proceeded north and west before leaving our study area. An anticyclonic eddy (AE) of ~20km formed during their separation. The observations show horizontal density gradients of 0.5kg/m³ over ~10km with 30cm/s maximum velocities. Upwelling and downwelling were also detected by biochemical tracers near the frontal interface. The glider sampling of the eddy field was adapted in real time based on multi-resolution forecasting and remote sensing. Data were assimilated in real-time so as to evaluate the vertical velocities developed by the eddy field and their ecological impact.

The Western Intermediate Water (WIW) is one of the most important water masses in the Western Mediterranean Sea (WMED), as it significantly impacts the general circulation and water mass exchange and contributes to the ventilation of WMED. It is formed in the Gulf of Lion and the Balearic shelf during the winter-to-spring transition period. Prior studies characterized the thermohaline characteristics of the WIW either using fixed limits (salinity < 38.3 and temperature < 13°C) or alternative detection approaches based on the geometrical analysis of temperature-salinity diagrams. In this study, we aim to better understand the formation, properties, and 3-dimensional pathways of the WIW in the western Mediterranean. We combined more than 20,000 in situ profiles (gliders, floats, UCTD) from the CALYPSO 2022 experiment with numerical simulations from the regional models. Our results show that the WIW mostly spreads between the isopycnals 28.8 and 29.0 kg m^-3 and is (now) warmer and saltier than previously reported. We found that WIW is formed at various sites in addition to the northern WMED; its formation is not limited to the winter season, and it can be detected at depths of up to 500 m in the Balearic Sea and other areas of WMED. Also, we used oxygen as a semi-conservative tracer to identify the WIW signal in the Balearic region. We showed that dissolved oxygen (DO > 210 μmol kg^-1) can be used as a tracer of WIW during its lateral advection by the northern current or mesoscale processes into the Balearic Sea. To understand better the mechanisms involved in the WIW formation, we developed a 1-D buoyancy loss model to simulate WIW formation at different sites. The model can simulate the deepest mixing using existing CTD profiles and atmospheric fluxes. We conclude by showing that this model could have an important role in understanding the 3-dimensional pathways of the WIW as well as in simulating and predicting the oxygen content of the WIW after leaving the surface.

Data-driven, deep-learning modeling frameworks have been recently developed for forecasting time series data. Such machine learning models may be useful in multiple domains including the atmospheric and oceanic ones, and in general, the larger fluids community. The present work investigates the possible effectiveness of such deep neural operator models for reproducing and predicting classic fluid flows and simulations of realistic ocean dynamics. We first briefly evaluate the capabilities of such deep neural operator models when trained on a simulated two-dimensional fluid flow past a cylinder. We then investigate their application to forecasting ocean surface circulation in the Middle Atlantic Bight and Massachusetts Bay, learning from high-resolution data-assimilative simulations employed for real sea experiments. We confirm that trained deep neural operator models are capable of predicting idealized periodic eddy shedding. For realistic ocean surface flows and our preliminary study, they can predict several of the features and show some skill, providing potential for future research and applications.

In December, we celebrated the end of the fall semester and holidays with a pizza party.
Now we kick off the spring semester with…another pizza party!
And this time, everybody was able to make it…and Ellen showed us how to be in two places at once!