FY 2003

Temperature data assimilation with salinity corrections: Validation for the NSIPP Ocean Data Assimilation System in the tropical Pacific Ocean, 1993–1998

Troccoli, A., M.M. Rienecker, C.L. Keppenne, and G.C. Johnson

Technical Report Series on Global Modeling and Data Assimilation, NASA/TM-2003-104606, Vol. 24, NASA/Goddard Space Flight Center, Greenbelt, MD, 23 pp (2003)

The NASA Seasonal-to-Interannual Prediction Project (NSIPP) has developed an ocean data assimilation system to initialize the quasi-isopycnal ocean model used in our experimental coupled-model forecast system. Initial tests of the system have focused on the assimilation of temperature profiles in an optimal interpolation framework. It is now recognized that correction of temperature only often introduces spurious water masses. The resulting density distribution can be statically unstable and also have a detrimental impact on the velocity distribution. Several simple schemes have been developed to try to correct these deficiencies. Here the salinity field is corrected by using a scheme which assumes that the temperature-salinity relationship of the model background is preserved during the assimilation. The scheme was first introduced for a z-level model by Troccoli and Haines (1999). A large set of subsurface observations of salinity and temperature is used to cross-validate two data assimilation experiments run for the 6-year period 1993–1998. In these two experiments only subsurface temperature observations are used, but in one case the salinity field is also updated whenever temperature observations are available.

The effectiveness of the Troccoli and Haines scheme is reflected not only in a better salinity field but also in an improved temperature field. The root-mean-square difference (RMSD) between the assimilation analyses and observations in the equatorial Pacific shows an average improvement in the upper 900 m of 20% in the salinity field and of 6% in the temperature field. The impact of the subsurface assimilation has been also assessed via data retention experiments (simulated forecasts). The RMSD diagnostic for these "forecasts" increases only moderately up to a 6-month lead, showing the retention for several months of information from the assimilation.