Estimating the characteristics of analysis error by using an OSSE

Unlike for data assimilation of real observations in which all information have unknown errors, OSSEs provide a framework in which truth is known. For an OSSE that mimics real nature well, the characteristics of real analysis error can thereby be estimated. That has been done in the framework of the GMAO OSSE.

The nature run is the T511L91 ECMWF year-long data set. The assimilation system is a recent version of the NCEP/GMAO GSI system applied to a 0.5 degree analysis grid. Results are examined for 4 times daily analysis during January of the NR period. The analysis error is simply determined as the difference between the analysis and truth evaluated on the NR grid and then truncated to the resolution of the analysis.

Several characteristics of the results are very interesting. The largest error variances are obtained over the central North Pacific Ocean. The error variances equal the variances of the truth at approximately wavenumbers 70 and 20 for rotational and divergent components of the wind, respectively. Although the formal resolution of the nature run is T511, the kinetic energy spectrum is likely much weaker than in reality for n>180. Other error characteristics will also be reported at the time of this presentation.