TRMM observations Directory UMM :Data Elmu:jurnal:A:Atmospheric Research:Vol57.Issue1.2001:

Table 2 Cloud model simulations Ž . Model Ex. reference Cloud system Resolution Time steps min Ž . km Goddard Cumulus Ensemble Tao and Simpson squall line 1.0 120, 150, 180, 210 Ž . Model, GCE-1 1993 TOGA-COARE Goddard Cumulus Ensemble Tao and Simpson squall line 3.0 180, 240, 300, 360 Ž . Model, GCE-3 1993 TOGA-COARE Ž . Large Eddy Model CETPr Lafore et al. 1998 squall line 1.25 300, 360, 420, 480 Meteo-France, Meso-NH TOGA-COARE and include cloud microphysics for five different particle types. For further information Ž . on mesoscale model intercomparison of this case, refer to Redelsperger et al. 2000 . Table 2 gives a summary of the employed model simulations.

4. TRMM observations

Fig. 4 presents an example of bright band occurrence in a storm system observed Ž . with the TRMM Precipitation Radar PR off the Northern coast of Australia. Spatial resolutions are ; 4 km in the horizontal and 0.25 km in the vertical planes. Melting particles account for 5–10 dBZ increases of radar reflectivity as predicted by the model Ž . simulations Fig. 4a . The bright band is observed at altitudes between 4.0 and 4.5 km, according to the vertical velocity and temperature distributions. Fig. 4d indicates a high frequency of occurrence of bright bands compared to other rain types. However, due to the cross-track scanning illumination geometry of the PR, the bright band detection deteriorates towards the scan edges. Ž . A comparison of observed and simulated brightness temperature TB distributions was carried out to investigate the representativity of the described model approach. For the above introduced cloud model, four simulation time steps were chosen covering the mature and decaying stages of the squall line. The melting layer was introduced and Ž . radiative transfer simulations were carried out, applying the model of Bauer et al. 1998 Ž . which includes the imaging specifications of the TRMM Microwave Imager TMI . TBs at 10.7, 19.35, 37.0, and 85.5 GHz and both polarizations were generated, assuming a Ž constant zenith angle of u s 52.88 and varying azimuth angles. The first channel 10.7 . GHz was simulated with an upgraded spatial resolution according to a deconvolution Ž . technique later applied to the TMI measurements Bauer and Bennartz, 1998 for the sake of an increased dynamic range of TBs. Ž . The standard output of the PR level 2 2A25 product provided by NASA was used in Ž . combination with TMI level 1 1B11 output. PR and TMI observations were colocated so that only PR observations were taken when they covered at least 80 of the TMI Ž . 10.7 GHz footprint at a resolution of 44 km = 27 km . The rain system classification included in the 2A25 files was used to distinguish between stratiform clouds and stratiform clouds for which a bright band was detected. Due to the bright band detection Ž . Ž . Fig. 4. TRMM PR reflectivity vertical cross-section a , estimated rain rate at 2 km altitude b , altitude of Ž . Ž . bright band c , and rain system identifier d for tropical storm off the coast of Darwin, Australia, on December 9, 1998. problem near the scan edges, only the inner 16 TMI footprints were analyzed. Forty-eight orbits from various seasons were chosen. Table 3 shows the orbit numbers and interesting rain cloud features represented in the measurements. The case from Fig. 2 was also included. The effect of melting layers in observations and simulations was isolated as follows. 4.1. ObserÕations Ž . TBs were accepted if at least 80 of the TMI 10.7 GHz field of view FOV was covered with valid PR measurements and if at least 50 of the FOV was classified as rain-contaminated. Of these, observations were classified as stratiform, including a melting layer when the PR classification gave a minimum coverage by bright bands of Ž . 50 sample size n s 6916 . For stratiform clouds with no bright band, a maximum Ž . coverage by bright bands of 10 was permitted sample size n s 2012 . This represents a compromise between discrimination of effects and large enough samples. A very important aspect was the isolation of bright band effects from effects of systematic differences introduced into the samples from other sources: Fig. 5 shows histograms of near-surface rain liquid water contents, W , and remaining convective PR cloud fractions, C , from both datasets. c Those cases, including a bright band, contained slightly higher rainfall amounts and significantly less convection. To avoid biases in the following analysis, the original datasets were corrected by scaling with a common distribution obtained from the minimum frequency per liquid water and fractional convection interval. The distribution of C over TB generally shows a wider distribution with less total amounts in the c presence of bright bands, i.e., wider spread stratiform clouds with less occurrence of embedded convection. For stratiform clouds without bright bands, much less homoge- neous conditions are present, introducing a convection bias at narrow TB ranges—pre- ferably at the lower end of the hydrometeor emission contribution to the total signal at the respective frequency. Table 3 Selected cases for database evaluation Case Date Orbit no. System contained Lat.rion. of center 1 02r10r98 1171 tropical cyclone, Indian Ocean 158Sr608E 2 02r11r98 1186 tropical cyclone, Indian Ocean 208Sr608E 3 02r16r98 1273 deep convection, central Pacific 58Sr1458W 4 08r26r98 4283 hurricane ‘Bonnie’, Atlantic Ocean 328Nr788W 5 08r26r98 4285 hurricane ‘Bonnie’, Atlantic Ocean 328Nr788W 6 08r27r98 4299 stratiform, Pacific Ocean 188Nr1258W 7 08r27r98 4304 stratiform, Pacific Ocean 188Nr1258W 8 09r19r98 4656 hurricane ‘George’, Atlantic Ocean 148Nr508W 9 09r19r98 4666 hurricane ‘George’, Atlantic Ocean 158Nr548W 10 12r09r98 5931 tropical cyclone, off Darwin 128Sr1288E 11 12r09r98 5938 tropical cyclone, off Darwin 128Sr1288E 12 01r21r99 6620 scattered convection, Indian Ocean 58Nr808E 13–48 04r22r98– 2288– frontal systems with midlatitude origin 04r24r98 2323 Fig. 5. Frequency distribution of near-surface rain liquid water contents from stratiform, bright band, and Ž . Ž . Ž . common observation datasets a . b As in a for fractional TMI EFOV coverage by convective clouds. 4.2. Simulations TBs from the simulations were taken for which the root mean square difference between simulations including and excluding a melting layer over all frequencies exceeded 2 K. This was carried out for three different models but the same simulation experiment in order to investigate the impact of different model physics and resolutions Ž on the magnitude of the melting layer effect sample size n s 2392, 6452, 1197 for . GCE-1, GCE-3, and Meso-NH, respectively . Ž . Fig. 6 compares the normalized polarization differences NPD of each profile for the melting vs. no melting case at 10.7, 19.35, and 85.5 GHz and all databases. NPD is defined as: TB y TB Ž . v h cloud NPD s 21 Ž . TB y TB Ž . v h clear Ž . Ž . Ž . Fig. 6. Simulated NPDs melting vs. no melting at 10.7, 19.35, and 37.0 GHz from GCE-1 a , GCE-3 b , Ž . and Meso-NH c model experiments. Ž . with v,h denoting the vertically and horizontally polarized TBs. For clear scenes, NPD s 1 while for totally opaque atmospheres NPD s 0. Since NPD is equivalent to the decrease in transmittance, t , the offset to the line represents the decrease in t originating from the melting layer. All model simulations behave similarly with respect to the melting effect even though GCE-1 represents a less opaque situation. The decrease in t increases slightly with frequency as does the shielding of melting layer Ž . effects by the upper cloud layers, Bauer et al., 2000 so that the maximum effect is observed where t f 0.5 with Dt f 0.05. Since only from the simulations a profile-by-profile intercomparison is possible, accumulated distributions of TBs were compared to identify systematic differences between stratiform profiles containing significant amounts of melting particles and those without melting layer. The results between melting and no melting classes as well as observations and simulations are presented in Figs. 7–9. The thinrthick lines represent simulationsrobservations, respectively. One obvious difference between mo- del simulations and observations is the range of TBs and their distribution over this Ž range. While the model simulations always represent an isolated cloud system over a . certain time of its evolution , the observations cover a larger variety of system and cloud stages as well as surface conditions. This causes the observed TB-range to be wider with a more shallow slope compared to a more confined TB-variation in the simulations. The different models and resolutions already suggest that even when the same initialization was implemented, there is a strong dependence of TB-statistics on cloud model. The coarser GCE-3 produces stronger radiometric effects because the Ž . averaging over the TMI footprint size includes less grid cells see Fig. 8 . Thus, larger TBs occur at lower frequencies and lower TBs at higher frequencies. An obvious fea- ture in the Meso-NH simulation is the larger number of high TBs at 10.7 GHz and low TBs at 85.5 GHz, suggesting more precipitating water and ice than in the GCE simulations. The melting layer effect is consistently represented in all simulations with amounts between 0 and 8 K over the whole TB range at lower frequencies. Theses values represent the shift in the TB-distributions, thus local differences can be much higher Ž . Olson et al., 1999 . At 85.5 GHz, the opacity of the cloud layers above the melting Ž . layer reduces the TB-increase significantly Figs. 7d–9d . The TB-distributions from the observations compare very well with those from GCE-1 at frequencies above 10.7 GHz, keeping in mind the wider TB-range due to the larger variety of rain systems. Even at Ž 85.5 GHz, where the strongest weakness of current cloud models i.e., the simula tion of . precipitating ice would be expressed, the observed reflects the simulated TB- distribu- tion. Larger differences occur for the experiments at higher frequencies while Meso-NH seems to compare best at 10.7 and 19.35 GHz. The observations indi cate a similar magnitude of melting layer effects on upwelling radiances as the simula tions. Only at 37.0 and 85.5 GHz the stratiform clouds show higher TBs than those with a bright band. This can be explained by the ice concentrations above the melting layer. On the footprint size of the TMI, no melting layer effect can be expected at 85.5 GHz as indicated by the simulations. However, the presence of bright bands requires consi derable amounts of large frozen particles above the freezing level. This will cause stronger scattering of microwave radiation, thus, a TB-depression compared to similar cloud amounts with less ice. This is not represented in the simulations because the inclu sion of a melting layer was held against identical cloud profiles without melting layer only. Ž . Fig. 7. Accumulated frequency distributions of TBs from GCE-1 simulations thin lines and TRMM Ž . Ž . Ž . observations thick lines classified into stratiform solid lines and stratiform with melting layer dashed lines Ž . Ž . Ž . Ž . at 10.7 a , 19.35 b , 37.0 c , and 85.5 GHz d . Fig. 8. As in Fig. 7 but for GCE-3. Fig. 9. As in Fig. 7 but for Meso-NH.

5. Summary and discussion