Speaker: Aviv Solodoch, Ph.D.
Atmospheric and Oceanic Sciences,
University of California Los Angeles
Location: Virtual (Click here for the recording)
Abstract: The meridional overturning circulation (MOC) in the oceans is a fundamental circulation pattern whereby surface water cool and densify in polar regions, and subsequently sink to great depths. These dense waters then spread horizontally at depth to cover virtually all deep ocean basins globally. The MOC has critical roles in the climate system, including influencing global circulation patterns and heat fluxes, and regulating the amount of anthropogenic heat and CO2 that is absorbed into the deep ocean, buffering the advance of climate change. Therefore, monitoring MOC variability and its interaction with climate change are of fundamental importance. In-situ monitoring of the MOC presents significant technological and logistical challenges due to the global extent of this circulation pattern. However, some aspects of ocean circulation are now regularly measured via satellite remote sensing, e.g., sea surface elevation and ocean bottom pressure. Therefore, we develop a methodology to monitor MOC variability based on machine learning of satellite-measured ocean properties. We test this methodology within a data-constrained numerical simulation of the oceans, i.e., using its output “satellite-observable’’ variables and MOC strength series as the ocean “truth’’.
We find that, using a simple 1-layer feed-forward Neural Network (NN) with Bayesian regularization, the MOC time-variability across most latitudes can be reconstructed with high skill. The reconstruction skill is higher than that of previously published dynamically based methods. To gain insight into the relations learned by the NN we use machine-learning interpretability techniques, showing for example that most of the Southern Ocean MOC reconstruction skill is due to data from just a few key locations (mainly large seabed ridges), qualitatively consistent with fundamental physical theory. We further examine which satellite observables hold the most potential for MOC reconstruction. Finally, we evaluate the robustness of the methodology and discuss a roadmap for implementing the method with real satellite data.
About speaker: Aviv Solodoch obtained a BSc in Math and Physics from Tel Aviv University, and a MSc in Physics from the Weizmann Institute of Science in Israel. He later completed a PhD in Atmospheric and Oceanic Sciences at UCLA, where he is currently a postdoctoral researcher. During his MSc, Aviv investigated air-sea interaction and heat exchange. During his PhD, Aviv investigated processes causing instability, offshore material exchange, and vortex formation in oceanic currents, using both numerical simulations and theory, with a focus on currents which form part of the overturning circulation in the North Atlantic. Aviv also conducted observational research with UCLA Marine Operations, studying coastal circulation dynamics in the Gulf of Mexico. He is presently studying the overturning circulation in the Southern Ocean, as well as the dynamics of transport of material between the coastal and deep ocean regions.