Ensembles of Eddying Ocean Simulations for Climate : Modeling Techniques, Diagnostics, First Results.

Unlike laminar Ocean General Circulation Models (OGCMs), eddying OGCMs spontaneously generate a strong chaotic variability, not only at mesoscale but also up to basin- and multidecadal-scales. This component of the variability is largely sensitive to initial states and locally accounts for most of the low-frequency (interannual and slower) variance found in fully-forced oceanic hindcasts. Climate-oriented (multi-decadal) high-resolution ocean simulations therefore require ensemble approaches to disentangle the atmospherically-forced (ensemble mean) and the chaotic (ensemble spread) ocean variability components, and to quantify the uncertainty due to non-linear ocean dynamics.

The OCCIPUT project is currently performing the first 50-member ensemble of 1/4° global ocean/sea-ice hindcasts driven by the same realistic forcing over the last 57 years. The ensemble spread, produced through a stochastic parameterization during the first year, subsequently grows and cascades toward long space and time scales. The NEMO model has been adapted to provide the following features : (1) simultaneous integration of the 50 members as one executable over several thousands of processors, so that ensemble (inter-member) statistics can be computed online as well ; (2) online production of ensemble synthetic observations allowing the use of probabilistic metrics for model assessment ; (3) flexible output archiving strategy through the use of 50 parallelized I/O servers.

We will present our modeling and processing approaches, some probabilistic products derived from this global ensemble run and from its regional (20-year 10-member North Atlantic) version, hence providing insights into the actual imprint of the atmosphere on the ocean variability.