December Webinar presented by Ivo Pasmans, Postdoc from the University of Reading working on WP4, and Chaired by Alberto Carassi, WP4 lead PI’s from the university of Bologna.

Popular data assimilation (DA) methods, designed to improve estimates of the truth by incorporating information from observations, typically assume that errors are Gaussian and model are linear. However, these assumptions fail in the context of complex sea ice models. In this presentation, we will outline the fundamentals of DA and demonstrate how machine learning techniques can help address its limitations. Specifically, we will show that applying DA in the latent space of a variational autoencoder enables the generation of post-DA states that adhere to the physics embedded in the model.

Whatch the recording :