Four-Dimensional Data Assimilation of Gale Data Using a Multivariate Analysis Scheme and a Mesoscale Model with Diabatic Initialization.

A method of assimilating 3-hourly sounding data is developed and successfully tested in this study. First, the successive corrections scheme of Bratseth (1986), which converges to optimum interpolation, is applied for the numerical analysis of data collected during the Genesis of Atlantic Lows Experiment (GALE). Univariate analyses of the mass and wind field are produced. The coupling of the mass and wind field is achieved by further iterations of the geopotential utilizing improving estimates of the geostrophic wind to extrapolate the geopotential to the grid points. The univariate wind analysis is then corrected for the new geostrophic wind. Next, diabatic forcing is incorporated into a vertical mode initialization scheme to provide more realistic initial conditions and to shorten the spinup time of the Naval Research Laboratory/North Carolina State University (NRL/NCSU) mesoscale model. Latent-heating profiles are computed from 'spun-up' model-generated and observed rainfall. The latent heating is distributed in the vertical according to the cumulus convective parameterization scheme (Kuo scheme) of the model. Compatibility between the specified heating during initialization and the heating during early model integration is retained by merging the model integrated rainfall and heating rates with those rates from the initialization. Finally, the multivariate, successive correction analysis scheme and the diabatic initialization procedure are combined with the NRL/NCSU model to form an intermittent data-assimilation system. Assimilations of the GALE data over a 2{1over2}-day period were performed with differing update cycles of 3, 6, and 12 h. Twelve-hour NMC hemispheric analyses served as the "no assimilation" control case for comparison. The assimilation of 3-hourly GALE data led to large decreases in background forecast rms errors and smaller decreases in analysis rms error. Better consistency in time was achieved between forecasts and analyses in the assimilation experiments. Rainfall prognoses from the assimilated states verified reasonably well with the observed rainfall and showed much more rapid spinup and better overall patterns than did the "no assimilation" precipitation forecasts.