Variational Assimilation of XBT Data. Part II. Sensitivity Studies and Use of Smoothing Constraints

View More View Less
  • 1 Hooke Institute for Atmospheric Research and Department of Atmospheric, Oceanic and Planetary Physics, Ciarendon Laboratory, Oxford, U.K.
© Get Permissions
Full access

Abstract

A linear reduced-gravity model of the tropical pacific is used to assimilate XBT data. The model cannot fit the data in the eastern equatorial Pacific for the whole assimilation period. Several experiments with real and simulated data are performed to investigate the source of this deficiency, which may be in the model or the wind stress used to force the model. It is shown that on the basis of the simple model physics we cannot unambiguously partition the error between model and forcing in the real data assimilation experiments although simulated data experiments do permit discrimination between model and forcing errors. Because the data is incomplete and does not permit a unique determination of the initial state, the use of prior information in the form of first-guess fields and/or smoothing constraints is examined. The filtering characteristics of the optimization algorithm are also discussed by looking at the evolution of the initial conditions as a function of the iteration number.

Abstract

A linear reduced-gravity model of the tropical pacific is used to assimilate XBT data. The model cannot fit the data in the eastern equatorial Pacific for the whole assimilation period. Several experiments with real and simulated data are performed to investigate the source of this deficiency, which may be in the model or the wind stress used to force the model. It is shown that on the basis of the simple model physics we cannot unambiguously partition the error between model and forcing in the real data assimilation experiments although simulated data experiments do permit discrimination between model and forcing errors. Because the data is incomplete and does not permit a unique determination of the initial state, the use of prior information in the form of first-guess fields and/or smoothing constraints is examined. The filtering characteristics of the optimization algorithm are also discussed by looking at the evolution of the initial conditions as a function of the iteration number.

Save