A Deep Learning approach to spatiotemporal SSH interpolation and estimation of deep currents in geostrophic ocean turbulence
|Author(s)||Manucharyan Georgy E.1, Siegelman Lia2, Klein Patrice2, 3, 4|
|Affiliation(s)||1 : School of Oceanography, University of Washington, Seattle, WA, USA
2 : Jet Propulsion Laboratory, California Institute of Technology Pasadena CA, USA
3 : Laboratoire de Météorologie Dynamique, CNRS Ecole Normale Supérieure, Paris, France
4 : Laboratoire d'Oceanographie Physique et Spatiale, IFREMER,CNRS, Brest, France
|Source||Journal of Advances in Modeling Earth Systems (1942-2466) (American Geophysical Union (AGU)), 2021-01 , Vol. 13 , N. 1 , P. e2019MS001965 (17p.)|
|Keyword(s)||sea surface height interpolation, Deep Learning, state estimation, mesoscale eddies, baroclinic instability, deep ocean flows|
Satellite altimeters provide global observations of sea surface height (SSH) and present a unique dataset for advancing our theoretical understanding of upper ocean dynamics and monitoring its variability. Considering that mesoscale SSH patterns can evolve on timescales comparable to or shorter than satellite return periods, it is challenging to accurately reconstruct the continuous SSH evolution as currently available altimetry observations are still spatially and temporally sparse. Here we explore the possibility of SSH interpolation via Deep Learning by using synthetic observations from an idealized quasigeostrophic (QG) model of baroclinic ocean turbulence. We demonstrate that Convolutional Neural Networks with Residual Learning are superior in SSH reconstruction to linear and recently developed dynamical interpolation techniques. Also, the deep neural networks can provide a skillful state estimate of unobserved deep ocean currents at mesoscales. These conspicuous results suggest that SSH patterns of eddies might contain substantial information about the underlying deep ocean currents that are necessary for SSH prediction. Our training data is focused on highly idealized physics and diversification of processes needs to be considered to more accurately represent the real ocean. In addition, methodological improvements such as transfer learning and implementation of dynamically‐aware loss functions might be necessary to consider before its ultimate use with real satellite observations. Nonetheless, by providing a proof of concept based on synthetic data, our results point to deep learning as a viable alternative to existing interpolation and, more generally, state estimation methods for satellite observations of eddying currents.
Plain Language Summary
Satellite observations of sea surface height (SSH) are widely used to derive surface ocean currents on a global scale. However, due to gaps in SSH observations, it remains challenging to retrieve the dynamics of rapidly evolving upper‐ocean currents. To overcome this limitation, we propose a Deep Learning framework that is based on pattern recognition extracted from SSH observations. Using synthetic data generated from a simplified model of ocean turbulence, we demonstrate that deep learning can accurately estimate both surface and sub‐surface ocean currents, significantly outperforming the most commonly used techniques. By providing a proof of concept, our study highlights the strong potential of deep learning for estimating ocean currents from satellite observations.