Ensemble data assimilation for idealized California current system with ROMS-LETKF

A local ensemble transform Kalman filter (LETKF) is applied to the Regional Ocean Modeling System (ROMS) with the idealized California coast setup at a 7-km resolution (ICC6 configuration by Capet et al. 2008); data assimilation experiments are performed with simulated observations under the perfect model assumption. First, bred vectors (BVs) are computed to understand better the dynamical nature of the evolving model and to have implications to the observing network design. The small scale instabilities shown by BVs near the eastern coastal boundary imply that denser observations are desirable in those regions. The BVs indicate larger scales in the offshore area farther than about 200 km from the coast, suggesting sparser observations would suffice there. The vertical structure suggests high error correlation in surface mixing layer which suddenly changes near the thermocline depth. BV has stronger signal at the thermocline, which becomes smaller in deeper levels. Thus, more independent observations are desirable near the thermocline.

A forecast/assimilation cycle experiment with a regular observing network is performed to ensure that the LETKF works appropriately with ROMS. Then, several ROMS-LETKF cycle experiments are performed with different observing networks including simulated real observations such as sea-surface temperature (SST) and height (SSH) from satellites, surface current by ocean radars located near the coast, and some sporadic ocean gliders to observe vertical profiles as deep as 1000 m. These observations are major sources of the operational three-dimensional variational (3D-Var) data assimilation system for California coast ocean prediction by Jet Propulsion Laboratory (JPL). The impact of these observations is investigated with the idealized data assimilation experiments.