Comparison of Moisture Variables in the Vertical Interpolations of a 4-D Data Assimilation System

摘要

In this study, we demonstrate that the choice of moisture variable and vertical interpolation algorithm in 4-D data assimilations of moisture is crucial to preserving a reasonable moisture distribution in humidity analyses. To support this conclusion, we test the use linear, logarithmic, power law, and exponential vertical interpolation in pressure (P) for each of the moisture variables of relative humidity (RH), dewpoint (T sub d), and specific humidity (Q). We test these interpolations in the transformations of humidity analyses from mandatory pressure surfaces (P) to forecast model sigma surfaces (sigma), and vice versa, using three January and three July 1978 FGGE III-A analyses. The synthesized analysis, obtained by applying the P yielding T and T yielding P interpolations in turn to the input FGGE III-A analysis, is verified against the input analysis. When the standard choice of logarithmic interpolation is employed, the verification errors show that using Q as the moisture variable yields much larger RMS errors than using either RH or T sub d. The large RMS error obtained when using Q results from a large positive moist bias in the Q-based synthesized analyses in the middle troposphere. The cause of this bias is the pronounced convexity of typical vertical profiles of Q when plotted against Ln(P). Cubic spline interpolation of Q in Ln(P) provided negligible improvement. Because we show Q on average closely satisfies a power law in P, power law interpolation of Q proves to be quite satisfactory. In interpolations of RH or T sub d, both logarithmic and power law interpolation yielded good results. (Author)