A retrieval strategy for interactive ensemble data assimilation

Author: Ross N. Hoffman, and Thomas Nehrkorn
Date: 
August 2, 2011
Type: 
Presentation
Venue: 
15th Symposium on Integrated Observing and Assimilation Systems for the Atmosphere, Oceans and Land Surface (IOAS-AOLS)
Citation: 

Hoffman, R. N., M. J. H. Janusz Eluszkiewicz, Steven J. Greybush, E. Kalnay, T. Nehrkorn, and R. J. Wilson. A retrieval strategy for interactive ensemble data assimilation. 15th Symposium on Integrated Observing and Assimilation Systems for the Atmosphere, Oceans and Land Surface (IOAS-AOLS), American Meteorological Society, Boston, MA, Seattle, Washington, January 2011, abstract submitted 8/2/2010.

We convert standard retrievals into "observations" with expected errors that should be zero mean, uncorrelated, and unit variance, and define a corresponding obs-function (or H-operator) that is a weighted sum of the temperatures on the radiative transfer model vertical grid. Our approach follows some ideas from Rodgers' book, "Inverse Methods for Atmospheric Sounding: Theory and Practice". Effects of smoothing and the prior are removed from the "observation". The weights are determined from various inputs and outputs of the retrieval process, including the so-called Jacobian of the retrieval, which is the matrix of sensitivity of radiance to state vector, evaluated at the retrieval solution. No changes to the assimilation method are needed, except to interpolate to the radiative transfer model vertical grid and to calculate the weighted sum. The weights could also be useful in the vertical localization for data selection. We test this approach in our Mars data assimilation system using Mars Global Surveyor (MGS) Thermal Emission Spectrometer (TES) radiances and retrievals.

There is a clear and feasible path to interactive retrievals in which the prior (mean and covariance) come from the background ensemble. With a few extra steps in the preprocessor to calculate the weights, we could make use of the EOFs used by the retrieval to reduce the number of observations, without changing anything else.