Inferring oceanic vertical velocities from sea surface data with machine learning
Abstract
Vertical velocities in the ocean are crucial for the exchange of heat and biogeochemical properties such as nutrients, dissolved carbon, and oxygen. However, the vertical velocity is difficult to observe because it is orders of magnitude smaller than the horizontal velocities and we do not have methods to diagnose vertical velocity from remote sensing observations. In this study, we use supervised learning to examine the relationship between readily observable surface fields--such as horizontal surface velocity and sea surface temperature (SST)--and the depth-dependent vertical velocity in submesoscale-permitting simulations of an idealized coastal upwelling front. We train various machine learning models to predict the vertical velocity at different depths in the water column, comparing the results to a baseline estimate obtained by depth-integrating the surface horizontal velocity divergence. We further evaluate the sensitivity of vertical velocity predictions to various input variables and look at the depth-dependence of those sensitivities. Additionally, since surface current data is typically available at lower spatial resolution than SST, we analyze the effect of coarsening the spatial resolution of surface velocities for training. Linking surface observations with vertical velocity at depth can help extend the use of remote sensing data to infer the vertical transport of biogeochemical tracers, such as nutrients for phytoplankton productivity and carbon export.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMOS0230004H
- Keywords:
-
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICS;
- 4262 Ocean observing systems;
- OCEANOGRAPHY: GENERAL;
- 4299 General or miscellaneous;
- OCEANOGRAPHY: GENERAL;
- 4532 General circulation;
- OCEANOGRAPHY: PHYSICAL