Miyakoda, K., 1986: Assessment of results from different analysis schemes. In International Conference on the Results of the Global Weather Experiment and Their Implications for the World Weather Watch, GARP Publications Series No. 26, (Volume 1), WMO/TD No. 107, 217-253.

Abstract: Associated with GARP activities, two kinds of data assimilation methods have been developed, i.e., the intermittent and the continuous schemes. These schemes include two focal points, i.e., the spatial interpolation technique and the initialization procedure. A great deal of progress has been made on the former aspect in terms of the optimum interpolation analysis. In particular, the multivariate optimum interpolation has been widely used. For the intermittent scheme of data assimilation, the multivariate version is instrumental for incorporating various sources of observed data. For the second aspect, the normal mode initialization has been devised, which turned out to be a breakthrough for obtaining the dynamical balance between the mass fields and winds. Yet the decision to use this procedure depends upon the philosophy of the analysis. The observed data contain a fair amount of dynamically unbalanced components both real and spurious, and it is difficult if not impossible, to distinguish between them. In this situation, it is optional to apply the dynamical balance to the variables in the resulting analysis.
The tropical analysis presents a unique problem. Part of the reasons for this is the sparcity of data, but the real reason is the dominance of cumulus convection and gravity modes. Cumulus convection is frequently associated with Kelvin waves. Therefore, if a large portion of these waves is suppressed in the analysis, a sufficient amount of condensation cannot be expected in the subsequent forecast. Cumulus convection is affected appreciably by the sea surface temperature in the tropics. For these reasons, the tropical analyses are influenced by the quality of the GCM's convection scheme and by the specified sea surface temperature.
In connection with the FGGE data set, a number of comparisons have been made, for example, Kung, Chen, Baumhefner, Rosen et al., Lau, and Hollingsworth et al. In this paper, it is attempted to review these comparisons. Assessment of data assimilation schemes is not straightforward. Evaluation consists of many items, i.e., the degree of data fitting, the accurate representation of inidividual storm structure, the adequate inclusion of general circualtion features, and the impact on the subsequent forecasts. Depending on which of these aspects is emphasized determines the type of analysis. Data assimilation schemes, in various centers, have evolved considerably in the last 5 years. The results of recent analyses for January 1980, 81, 82, and 83 at GFDL, NMC, and ECMWF have been applied to monthly forecast studies for 8 January cases from 1977 to 1983.
Even for seasonal forecasts, data assimilation must be thought of as an essential component. It will be necessary to use an air-sea-land coupled GCM that will innovatively incorporate observed sea surface temperatures and satellite data. However, in considering forecasts on time scales of a season one cannot ignore the effects of the model's own climate drift.