Impacts of assimilation of satellite and rawinsonde observations on the energetics of Southern Hemisphere baroclinic waves in the NCEP/NCAR reanalysis

Much of our current understanding of baroclinic wave structure and energetics are based on examination of gridded atmospheric analyses. Previous studies have found that the NCEP/NCAR and ECMWF reanalyses show some significant differences in depicting Southern Hemisphere (SH) baroclinic waves. Since the reanalysis is the model forecast modified by the observations after an assimilation cycle, it is useful to examine the impacts of the assimilation of observations (in the SH, mainly satellite and rawinsonde data), and the results based on the NCEP/NCAR reanalysis are presented in this study. Note that for the NCEP/NCAR reanalysis, not only are the gridded analyses available, but also the first guess (6-hr forecasts), the actual observations assimilated by the reanalysis, as well as forecasts (out to 8 days) made using the reanalysis grids as initial conditions. All of these have been examined to detail the impact of observations.

A one-year NOSAT analysis for 1979, in which all satellite data are removed during the assimilation process, is used to directly compare with the regular (SAT) reanalysis. It is found that the assimilation of satellite data generally decreases eddy kinetic energy (EKE) in the SH, with the greatest impact occurring over the oceans. This impact of satellite data on eddy amplitude is opposite to that found for ERA-40 in previous studies. It is further found that this decrease can be traced back to a low bias in retrieved satellite temperature (SATEMP) variance, which leads to a reduction in eddy available potential energy (EAPE) and a reduction of baroclinic generation. This turns out to have important impacts in the forecasts, as SH eddy energy (probably already biased low) decreases significantly during the first two days in forecasts made using the SAT analyses as initial conditions.

The NCEP/NCAR reanalysis as well as the corresponding first guesses are also analyzed. The EKE analysis increment, which is the difference between EKE based on the analysis and that based on the first guess, is a good measure to indicate the impacts of observations assimilated. Geographically, positive EKE analysis increments are found around the SH rawinsonde stations in the NOSAT analysis in 1979, indicating that the assimilation of rawinsonde data increases EKE significantly from the first guess. This also suggests that the NCEP/NCAR first guess is probably biased low, which is confirmed by comparing reanalysis with rawinsonde observations at several SH stations. The spatial pattern of EKE analysis increment is also examined for the SAT reanalysis in 1979, and positive analysis increments with even larger values are found around those rawinsonde stations compared to the NOSAT pattern, suggesting that with satellite data assimilated, the analysis and hence the first guess is reduced (see discussions above), therefore the rawinsonde observations have to further increase the EKE amplitude from the first guess. The patterns of EKE increment from the pre-satellite (1958-1977) and satellite (1979-1999) eras show high degrees of similarities to the NOSAT and SAT reanalysis patterns, lending further support to these findings.

Finally, the results from comparing the SAT reanalysis and the NOSAT analysis is applied to evaluate the impact of the introduction of satellite data on the trend of SH baroclinic wave activity. It can also be used to explain the difference between the eddy amplitudes in the NCEP/NCAR and ERA-40 reanalyses.