The COLA pilot reanalysis project, which was begun at the University of Maryland, was finished at the Center for Ocean-Land-Atmosphere Studies. The last nine months of assimilation (of 18 months total) were completed between July 1993 and January 1994. Post-processing and analysis of results has continued to the present time. Preliminary results were published in COLA Technical Report No. 5 (Paolino et al., 1994) and submitted to the Bulletin of the American Meteorological Society.
The COLA reanalysis was performed using an intermittent, multivariate optimum interpolation based on the Global Data Assimilation System (GDAS) of NMC (Dey and Morone, 1985), and including the complex quality control procedure (OIQC) as described by Woollen (1991). The assimilation merges observational data with a six hour forecast (the "first guess") produced by the COLA GCM (Kinter, et al., 1988), which included the SiB land surface model (Sellers, et al., 1986; Sato, et al., 1989; Xue et al., 1991). The analysis/first guess cycle was performed four times each day, at 0000, 0600, 1200, and 1800 UTC, for the period April 25 1982 - November 30, 1983. Five data streams were input to the analysis: station upper air soundings (radiosondes/rawinsondes/pilot balloons), aircraft wind reports, satellite cloud tracked winds, ship surface pressure and winds, and TIROS operational vertical soundings (TOVS). Observed monthly mean global sea surface temperatures and Northern Hemisphere snow cover were used to initialize each six hour forecast. Soil wetness observations were considered too sparse and unreliable, so the model predicted soil wetness was used to initialize each following six hour forecast.
To measure the representativeness of the COLA retrospective analysis in comparison to the actual state of the atmosphere, root mean square errors (RMSE) were computed between the analyses and the station upper air soundings. The same RMSE were computed for the contemporaneous ECMWF analyses. We reduced any bias in the selection of stations for the computation of the RMSE by considering all stations, regardless of any decisions made by the quality control procedures (including OIQC) of the COLA analysis. RMSE statistics were computed for each three month season, JJA 1982 - SON 1983, using all stations with at least 60 soundings / season (40 soundings between 20ºS and 20ºN). Observations failing a hydrostatic consistency check or a gross error check were discarded.
Table 1 displays the RMSE statistics, accumulated over 30º latitude bands, for the COLA and
ECMWF zonal wind analyses of June 1982 - November 1983. COLA errors tend to be lower
than those for ECMWF almost everywhere north of 60ºS. The COLA zonal wind RMSE is
particularly lower in the tropics at 500 mb and above. Similar statistics for the geopotential
height (not shown) show the RMSE for the COLA analyses are smaller than those for the
ECMWF analyses north of 30ºS. The RMSE for the COLA analyses south of 30ºS show very
large RMSE in the austral winter when compared to the ECMWF analyses. These errors were
probably due to excessive rejection of observations, perhaps due to a systematic error in the first
guess model. The divergent wind in the COLA analyses (not shown) was superior to those of
the earlier operational analyses. Further diagnostic evaluation of the COLA reanalysis has
continued and is described in more detail in the chapter entitled, "Climate Data Analysis."
Personnel: Doty, Kinter, Paolino, Shukla, Straus, Q.Yang
Summary prepared by: Paolino
last update: 5 November 1994