Finescale abyssal turbulence : sources and modeling

Abstract

A detailed understanding of the intensity and three-dimensional spatial distribution of diabatic abyssal turbulence is germane to understanding the abyssal branch of the global overturning circulation. This thesis addresses the issue through 1) an investigation of the dynamics of an abyssal boundary layer and through 2) the construction of a probabilistic finescale parameterization using mixture density networks (MDNs). A boundary layer, formed by the interaction of heaving isopycnals by the tide and viscous/adiabatic boundary conditions, is investigated through direct numerical simulations (DNS) and Floquet analysis. Turbulence is sustained throughout the tidal period in the DNS on extra-critical slopes characterized by small slope Burger numbers, leading to the formation of turbulent stratified Stokes-Ekman layers. Floquet analysis suggests that the boundary layers are unstable to disturbances to the vorticity component aligned with the across-isobath tidal velocity on extra-critical slopes. MDNs, trained on microstructure observations, are used to construct probabilistic finescale parameterization dependent on the finescale vertical kinetic energy (VKE), N² f⁻² , and both variables. The MDN model predictions are as accurate as conventional parameterizations, but also predict the underlying probability density function of the dissipation rate as a function of the dependent parameters.