One of the outstanding problems in data assimilation has been, and continues to be, how best to utilize satellite data while balancing the trade-off between accuracy and computational cost. A number of weather-prediction centres have recently achieved remarkable success in improving their forecast skill by changing the method in which satellite data are assimilated into the forecast model from the traditional approach of assimilating retrieved products to the direct assimilation of radiances in a variational framework. Although there are. clear theoretical advantages to the direct radiance-assimilation approach, it is not obvious at all to what extent the improvements that have been obtained so far can be attributed to the change in methodology or to various technical aspects of the implementation.
The central question we address here is: how much improvement can we expect from assimilating radiances rather than retrievals, all other things being equal? We compare the two approaches in a simplified theoretical framework. Direct radiance analysis is optimal in this idealized context, while the traditional method of assimilating retrievals is suboptimal because it ignores the cross-covariances between background errors and retrieval errors. We show that interactive retrieval analysis (where the same background used for assimilation is also used in the retrieval step) is equivalent to direct assimilation of radiances with suboptimal analysis weights.
We illustrate and extend these theoretical arguments with several one-dimensional analysis experiments, where we estimate vertical atmospheric profiles using simulated data from temperature sounding channels of both the High-resolution InfraRed Sounder 2 (HIRS2) and the future Atmospheric InfraRed Sounder (AIRS). In the case of non-interactive retrievals the results depend very much on the quality of the background information used for the retrieval step. In all cases, the impact of the choice of analysis method is dwarfed by the effect of changing some of the experimental parameters that control the simulated error characteristics of the data and the retrieval background.
[1]
J. Eyre.
On systematic and their climatological mean values
,
1987
.
[2]
J. Joiner,et al.
Efficient methods to assimilate remotely sensed data based on information content
,
1998
.
[3]
Dick Dee,et al.
Maximum-Likelihood Estimation of Forecast and Observation Error Covariance Parameters. Part I: Methodology
,
1999
.
[4]
J. R. Eyre,et al.
Assimilation of TOVS radiance information through one-dimensional variational analysis
,
1993
.
[5]
Philippe Courtier,et al.
The ECMWF implementation of three-dimensional variational assimilation ( 3 D-Var ) . 111 : Experimental results
,
2006
.
[6]
John Derber,et al.
The Use of TOVS Cloud-Cleared Radiances in the NCEP SSI Analysis System
,
1998
.
[7]
Eldo E. Ávila,et al.
The effect of the cloud‐droplet spectrum on electrical‐charge transfer during individual ice‐ice collisions
,
1999
.
[8]
Dennis C. Reuter,et al.
An accurate radiative transfer model for use in the direct physical inversion of HIRS2 and MSU Temperature Sounding Data
,
1983
.
[9]
C. Rodgers.
Characterization and Error Analysis of Profiles Retrieved From Remote Sounding Measurements
,
1990
.
[10]
Hartmut H. Aumann,et al.
Atmospheric Infrared Sounder on the Earth Observing System
,
1994
.
[11]
R. Todling,et al.
Data Assimilation in the Presence of Forecast Bias: The GEOS Moisture Analysis
,
2000
.
[12]
Arlindo da Silva,et al.
Data assimilation in the presence of forecast bias
,
1998
.
[13]
Andrew C. Lorenc,et al.
Analysis methods for numerical weather prediction
,
1986
.
[14]
J. R. Eyre,et al.
On systematic errors in satellite sounding products and their climatological mean values
,
1987
.
[15]
C. Rodgers,et al.
Retrieval of atmospheric temperature and composition from remote measurements of thermal radiation
,
1976
.