Our SASIP Webinar #4 with Tobias Finn on ‘Making the explicit link between Gaussian process regression and ensemble data assimilation’ took place on Wednesday 24 November at 3pm CEST.

Check the recording of the meeting :

In his talk, Tobias showed how Gaussian process regression, also called Kriging, might help us to learn sub-grid-scale dynamics of sea ice. Starting with linear regressions, he explained in a stepwise manner the reasoning behind this type of machine learning. This reasoning will bring him to the approximation of Gaussian processes with random fourier features. These random fourier features will help us to understand Gaussian processes more into detail and can be used to scale this regression. On the basis of these fundaments, Tobias has explicitly stated the link between Gaussian process regression and feature-based data assimilation with ensemble Kalman filters. This additionally shows how ensemble data assimilation can be understood from a machine learning point of view. In the end, he gave a short outlook on how these ideas can be used to incorporate information from ensemble predictions into deep learning.