Ensemble 4DVAR (En4DVar) data assimilation in a coastal ocean circulation model. Part II: Implementation offshore Oregon-Washington, USA

Abstract The ensemble four-dimensional variational (En4DVar) data assimilation (DA) system introduced in Part I (Pasmans and Kurapov, 2019) is tested in the coastal waters offshore Oregon and Washington, U.S. West coast, during the spring and summer of 2011. The background error covariance B is derived from the forecast ensemble. Satellite sea-surface temperature (SST), sea-surface height (SSH), and daily-averaged radial surface currents from high-frequency radars (HFRs) are assimilated. The performance of the En4DVar system is compared with a “traditional” 4DVAR system using a static B . It is found that the presence of the Columbia River plume has a profound impact on the ensemble B . Near the plume front the SST-SSS covariance can be up to a factor 20 larger in magnitude than in the static B . This introduces large spatial and temporal variability in the ensemble B . The En4DVar system is more successful than the 4DVAR with the static B preserving the temperature-salinity properties when compared to glider data. The En4DVar system also produces more accurate forecasts and analyses for temperature in the subsurface below 30 m at a buoy location on the continental shelf. In comparisons with other surface and subsurface observations En4DVar shows consistent, albeit not significant, improvement over traditional 4DVAR. Large surface temperature-salinity covariances in combination with the episodic occurrence of large-scale errors in the SST observations lead to erroneous freshening in the centre of the model domain. Adding constraints on the surface salinity corrections based on the prior model reduces this effect.