490 
A three-dimensional time-dependent cloud mesoscale model, named University of 
Wisconsin - Regional Atmospheric Modeling System (UW-RAMS), has been used for generating cloud 
structures (the primary cloud dataset), explicitly describing the detailed vertical distribution of four species of 
hydrometeors: cloud drops, rain drops, graupel particles, and ice particles (Smith et al„ 1992). The number of 
cloud layers has been automatically reduced to at most seven homogeneous layers in order to simplify the 
radiative transfer calculations (Basili et al„ 1992b). Then the primary cloud dataset has been extended by 
means of a Monte Carlo statistical procedure, based on the use of the correlation matrix of the hydrometeor 
equivalent water contents (Basili et al., 1992a). In this way. a dataset of five thousand cloud structures has 
been statistically generated retaining the physical and statistical features of the primary cloud model. 
In order to associate upwelling brightness temperatures to each structure of the statistically- 
generated cloud dataset, a radiative transfer scheme based on the discrete-ordinate method has been used to 
calculate multi-frequency unpolarized Tb's emerging from a multi-layer medium (Basili et al., 1991). Within 
each layer, temperature is supposed to be linearly dependent on the height, and the gaseous absorption is 
determined by means of the Liebe model (Liebe, 1985). The surface background has been modeled as a 
Lambertian source. The hydrometeors have been supposed spherical and characterized by size-distributions 
according to the UW-RAMS model. As a result, a cloud-radiation dataset consisting of five thousand cloud 
structures and the associated T B 's has been generated and considered as a random sample of space-borne 
microwave radiometer observations of precipitation. 
3 - SPACE-BORNE MICROWAVE SIGNATURE OF PRECIPITATION 
The Special Sensor Microwave / Imager (SSM/I) was launched for the first time on June 19, 1987 on board the 
Defense Meteorological Satellite Program (DMSP) Block 5D-2 Spacecraft F8 (Hollinger et al„ 1989). 
Subsequently, the second SSM/I was launched on December 1, 1990 on board the F10 spacecraft, and the third 
SSM/I, on November 28, 1991 on the Fll spacecraft. While the first SSM/I functioned properly for a couple 
of years and the third one is still providing the full set of data, in contrast, the second one has not been used so 
far because the F10 spacecraft is flying on a wrong orbit. The SSM/I satellite radiometers observe the 
microwave emission from the Earth at four frequencies (19.35 GHz, 22.235 GHz, 37.0 GHz, and 85.5 GHz) 
and provide information on a variety of environmental parameters, including atmospheric water, wind speed, 
and sea ice. The SSM/I orbit is circular, near-polar, and sun-synchronous with an altitude of 860 km and 
inclination of 98.8°. The orbital period is 102 minutes, and the local time of the ascending equatorial node is 
6:12 for the F8 spacecraft, and 17:04 for Fll. This orbit provides complete coverage of the Earth, except for 
two small circular sectors of 2.4° centered on the North and South Poles. Dual polarization measurements are 
taken at 19.35, 37.0, and 85.5 GHz, and only vertical polarization is observed at the 22.235 GHz water vapor 
channel. Earth observations are taken during a 102.4° rotation of the conical sc anning system and correspond 
to a swath width of 1394 km on the Earth surface. The spatial resolution of the images depends upon the 
frequency; specifically, the 3-dB foot print sizes (along-track by cross-track) are 69x43 km, 50x40 km, 37x29 
km, and 15x13 km for the 19,22, 37, and 85 GHz channels, respectively. 
3.1. SSM/I observation of an intense storm 
Figure 1 shows the vertically-polarized T B 's at 85.5 GHz measured by SSM/I on September 27, 1992 during 
the DMSP Fll satellite ascending pass over the Italian peninsula at 16:15 UTC. Three areas of low Tb's, 
corresponding to whiter pixels, are present along the cold front within the convective cloud system. The first 
one includes part of the Toscana region (central Italy) and the coast of the Tyrrenian Sea close to Rome; the 
second one, covers a relatively small area of South-Eastern France; the third one, that will be analyzed in this 
paper, extends from the coast of Liguria, near Genova, to the North-West borders of Italy. The whiter areas 
within the cloud system are associated to the most intense cells, in which large ice particles scatter the 
upwelling radiation, emitted from the lower cloud and rain layers. In general, scattering from large drops and 
ice particles is the physical mechanism responsible for the overall appearance of the storm as a cold feature 
over a warmer continental background. As a result, precipitation signature may be very similar to that of other 
natural emission sources (like snow cover, sea, lakes, etc.), thus generating a possible confusion when 
operating an automatic image classification. 
It must be pointed out that a technique for improving the spatial ground-resolution of the 
lower-frequency channels (19.35, 22.235 and 37.0 GHz) has been applied to the calibrated T B ’s (Farrar and 
Smith, 1991). This method, based on the bidimensional Backus-Gilbert filter, enhances the lower-frequency