GPU Ocean: Efficient simplified ocean models and non-linear data assimilation
Håvard Heitlo Holm

The predictions of ocean currents in a short time range are typically dominated by location of ocean eddies, which are rotating high and low pressure systems. Compared to their atmospheric counterparts, there are few direct measurements of these eddies, and because of their relative small sizes, a small misplacement of the eddies leads to large errors in predicted currents. Because of the large complexity of traditional ocean models, it is not feasible to cover the substantial uncertainties through large ensemble simulations.

Our approach is to study the uncertainty in the ocean through large ensembles of simplified ocean models, which can be efficiently simulated using Graphical Processing Units (GPUs). We aim on assimilating available observations of the ocean into the ensemble through non-linear data assimilation methods such as particle filters.