A deep learning approach to extract internal tides scattered by geostrophic turbulence

COLLOQUIUM TALK

Title: A deep learning approach to extract internal tides scattered by geostrophic turbulence

Venue: Online mode (Zoom meeting details attached)

Abstract: Internal tides (ITs) are inertia-gravity waves generated by large-scale oceanic tidal currents flowing over topography, important to oceanographers due to their roles in problems such as deep/upper ocean mixing. Conventionally, for altimetric observations of Sea Surface Height (SSH) data, ITs have been extracted by harmonically fitting over observed time sequences. However, in presence of strong time-dependent phase shifts induced by interactions with mean flows or changes in stratifications, harmonic fits do not work well for data with coarse temporal sampling. Such problems would be exacerbated in the Surface Water Ocean Topography (SWOT) satellite mission due to the finer spatial scales to be resolved. However, SWOT’s wide swaths unprecedentedly produce SSH snapshots that are spatially two-dimensional, which allows us to treat tidal extraction as an operation on two-dimensional images. Here, we regard tidal extraction purely as an image translation problem. We design and train a deep learning algorithm, which, given a snapshot of raw SSH, generates a snapshot of the embedded tidal component. The presentation will introduce in detail a recent work where we train and test a  conditional Generative Adversarial Network (cGAN) on a set of idealized numerical eddying simulation. No temporal information or physical knowledge is required for the cGAN to work in this scenario. The cGAN is tested on data whose dynamical regimes are different from the data provided during training. Despite the diversity and complexity of data, it accurately extracts tidal components in most individual snapshots considered and reproduces physically meaningful statistical properties. Predictably, the cGAN’s performance decreases with the intensity of the turbulent flow. Ongoing work where we simplify and improve the algorithm will be discussed too.