GMAO-Ocean Modeling and Data Assimilation

NASA Goddard Space Flight Center
GLOBAL MODELING AND ASSIMILATION OFFICE

OCEAN LINKS

+ Ocean Home

+ ODAS Home

+ ODAS Intranet (Internal)

+ Ocean Modeling

Collarborations

● ODASI Consortium

● USGODAE

● ECCO GODAE

● JCSDA

Publications

● Papers

● Annual Reports

● Presentations

Temperature and Altimetry Assimilation into a High Resolution OGCM with a Localized Ensemble Kalman Filter

Author: Christian Keppenne - AGU Ocean Sciences Meeting 2006 Honolulu, Hawaii, (Power Point Presentation)
Web Adaptation: Robin Kovach

The GMAO CGCMV1 coupled forecasting system consists of the NSIPP1 AGCM (2 x 2.5 x L34), the Mosaic LSM, and the Poseidon V4 OGCM (1/3 x 5/8 x L27). The Ocean Data Assimilation System (ODAS) is run prior to coupling to the AGCM and LSM.

The multivariate update scheme uses compact support, updates T, S, u, and v, has layer thickness ajustments between analyses, incremental updates, and processes SSH and T observations separately. Online bias estimation (OBE) is used in the SSH assimilation. Experiments consist of runs with no-assimilation (Control) and the production ODAS OI+S(T) (Optimal Interpolation for temperature plus a salinity correction). These are compared with 9, 17, and 33 member EnKF runs.

Image to Left: The first plot shows that the control run is mostly biased. The T-OI decreases the SSH bias significantly even though the SSH is not processed. The 33-member EnKF run shows no noticable SSH bias at all. Red is RMS OMA, orange is mean OMA, blue is RMS OMF, and cyan is mean OMF. Results for temperature are similar.

The image below shows that the 17-member and 33-member EnKF runs have similar performance. The hybrid covariances used in the T-assimilation with EnKF result in stronger data weights than OI+S(T).

The covariance structures for the EnKF17 and EnKF33 are also very consistent for both T and S while the EnKF9 results are spurious. The marginal temperature kalman gain for a one-sigmaT T innovation at 125W, 0N, and 100m is shown in the image below.

A history of May-start hindcasts were used to assess the impact of assimilation on SST hindcast skill. The EnKF-intialized hindcasts (image below) can miss or undersestimate El Nino events but result in fewer false El Nino or La Nina alerts than production forecasts.