Accurate retrievals of precipitation rates and water paths require consistency between reality and the many physical assumptions underlying the retrieval method.  One test of this consistency involves comparisons of predicted and observed brightness temperatures over the same ensemble of storms.  Reasonable agreement has been reported between: (a) AMSU observations of brightness temperature histograms 50-190 GHz over 122 globally and seasonally distributed storms, and (b) 15-km predictions made using NCEP-initialized 4-6 hour MM5 forecasts.  The particular cloud-resolving form of MM5 used and the radiative transfer model are described in the attached preprint entitled "Comparison of AMSU millimeter-wave satellite observations, MM5/TBSCAT predicted radiances, and electromagnetic models for hydrometeors" by C. Surussavadee and D. H. Staelin (IEEE Trans. Geosci. and Remote Sensing, in press, 2006).

The Goddard explicit cloud physics model was used, which provides profiles for snow, graupel, cloud ice, and rain water.  Although Mie scattering was assumed, the density of the icy spheres was assumed to have a wavelength-dependent "ice factor" value F that was determined separately for snow and graupel using electromagnetic scattering computations (DDSCAT) for hexagonal plates and 6-pointed rosettes, respectively.  Reasonable agreement between observed and predicted brightness temperature histograms was also separately found for each of five classes of precipitating AMSU pixels (convective, stratiform, snowy, non-snowy, or non-glaciated), although the brightness temperature histograms for window channels over snow were degraded by the unmodeled surface emissivity spectrum of snow (an omission that can be remedied).  The validity of the DDSCAT computations was tested by varying F for each wavelength so as to minimize the discrepancy between the AMSU observations and the simulated radiances; the resulting two sets of F values agreed within ~0.1 or less, where 0 < F < 1.  We have separately noted that our F values derived using DDSCAT are a weak function of the size distribution of the icy hydrometeors for frequencies below ~190 GHz, but that size matters more at higher frequencies.

We have also documented additional tests of the robustness of these MM5 and radiative transfer comparisons with AMSU that involve varying several of the physical assumptions such as the altitudes and abundances of the snow and graupel predicted by MM5, the size distributions of the various ice habits, and the hydrometeor loss tangents, F values, back-scattering ratios, etc.

This work (Surussavadee, C., and D. H. Staelin, 2007: Millimeter-Wave Precipitation Retrievals and Observed-versus-Simulated Radiance Distributions: Sensitivity to Assumptions. J. Atmos. Sci., 64, 3808-3826) is discussed further here under the theme Retrieval Techniques.