On Dynamical /Statistical Initialization for Numerical Weather Prediction

View More View Less
  • 1 Astro-Geophysics Department, University of Colorado, Boulder 80309
© Get Permissions
Full access

Abstract

A dynamical/statistical approach to initialization that is compatible with the dynamics of potential vorticity conservation is proposed. This approach consists of combining weighted assimilation, which minimizes the analysis error by means of linear regression, with a dynamical constraint imposed by this conservation principle. As a consequence, the initial analysis is shown to be optimal and dynamically compatible with the forecast model used in the present study.

Two situations that contribute to error growth in numerical prediction models are considered. 1) differences in the phase propagation speed of the model disturbance relative to that of the true or control state and 2) distortion of the initial error field by nonlinear wave interactions. In each case results obtained with the proposed dynamical/statistical initialization are compared with results from weighted assimilation using uncorrelated observational errors and from initialization by direct use of error-contaminated observations. These comparisons demonstrate the theoretical advantage of using an initialization scheme that is compatible with the model dynamics. However, it is pointed out that practical aspects involving additional computations and use of data from mixed observing systems have not been taken into account.

Abstract

A dynamical/statistical approach to initialization that is compatible with the dynamics of potential vorticity conservation is proposed. This approach consists of combining weighted assimilation, which minimizes the analysis error by means of linear regression, with a dynamical constraint imposed by this conservation principle. As a consequence, the initial analysis is shown to be optimal and dynamically compatible with the forecast model used in the present study.

Two situations that contribute to error growth in numerical prediction models are considered. 1) differences in the phase propagation speed of the model disturbance relative to that of the true or control state and 2) distortion of the initial error field by nonlinear wave interactions. In each case results obtained with the proposed dynamical/statistical initialization are compared with results from weighted assimilation using uncorrelated observational errors and from initialization by direct use of error-contaminated observations. These comparisons demonstrate the theoretical advantage of using an initialization scheme that is compatible with the model dynamics. However, it is pointed out that practical aspects involving additional computations and use of data from mixed observing systems have not been taken into account.

Save