558
A.D. Del Genio
the polarization of observed microwave brightness temperatures (cf. Petty and Katsaros, 1990);
(4) Multichannel statistical techniques (cf. Adler et aI., 1993); (5) Multichannel physicallybased radiative transfer algorithms (cf. Wu and Weinman, 1984; Smith and Mugnai, 1988;
Kummerowet a!., 1989; Liu and Curry, 1992).
A particular problem for the microwave is warm (coalescence-dominated) rain over land. Emission techniques do not work here because of the confusion of the land surface seen through the
rain, while ice scattering, which hides the land surface and permits microwave rain retrieval
for strong systems over land, is nonexistent. Algorithm intercomparisons indicate that shallow,
orographic warm rain is simply missed by typical microwave algorithms (cf. Adler et aI., 1993).
20.4.3 Visible/Infrared Sensor (VIRS)
By far the most widely available data on clouds consist of the visible and infrared window channel imagery from geostationary and polar orbiter operational satellites. The obvious appearance
of precipitating systems in such images as (visibly) bright and (thermally) cold regions suggests
that a quantitative measure of precipitation might be derived from such data, which are global
in coverage and several decades in length. The problem is that rainfall itself has little direct
signature in either the visible or infrared. Optical thickness is inversely proportional to particle
size, so large rain droplets contribute little to the visible reflectance of precipitating systems
relative to smaller cloud droplets. Infrared emission depends on the temperature of the cloud
top for an optically thick cloud, and in general higher intensity rain is associated with higher
altitude (and therefore colder) cloud tops. But moderate optical thickness non-raining cirrus
can appear just as cold as heavily precipitating cumulonimbus, and spatial/temporal variations
in atmospheric temperature and water vapor prevent the association of a unique rainfall rate
with a given infrared brightness temperature.
The standard infrared approach to rain estimation is the GOES precipitation index (GPI;
Arkin, 1979), which relates rain rate over a 2.5 0 x 2.5 0 area to the fraction of the area whose
IR brightness temperature is colder than 235 K via a simple linear regression. This algorithm,
based on GATE data for the tropical Atlantic, not surprisingly works well over that region
and is used to produce an operational rainfall product for the entire tropics. As a general
indicator of rainfall patterns and their seasonal or interannual variability, the GPI is a useful
index. But its quantitative accuracy is limited, it d0es not work well outside the tropics, where
non-convective rain is more common, it does not work well for shallow rain systems in general,
and it can give spurious information on the diurnal cycle of rain (cf. the discussion in Fu et
aI., 1990).
To address the problem of distinguishing precipitating cumulonimbus from equally cold nonprecipitating cirrus, Inoue (1987) developed the "split-window technique". This method relies
on the different wavelength dependences of the infrared emissivities of ice and liquid water. By
utilizing differences between the brightness temperatures at 11 and 12 J-Lm, Inoue separates the
cirrus (ice-only) cloud field from the cumulonimbus (ice and liquid) area, and bases the derived
rainfall on the latter. VIRS contains channels at 10.8 and 12 J-Lm to test this approach.
A different philosophy has been adopted recently by Rosenfeld (personal communication). A
common feature of current operational satellites is the existence of near infrared channels that
are sensitive to cloud droplet effective radius near cloud top (Han et a!', 1994). While such
channels also do not see the precipitating particles themselves, the possibility exists that the
droplet size distributions as a whole are shifted to larger sizes in precipitating systems. If
so, precipitation might be inferred indirectly in the near-IR as a function of cloud-top droplet
radii. This approach may provide a means of detecting the shallow rain systems over land that
A.D. Del Genio
the polarization of observed microwave brightness temperatures (cf. Petty and Katsaros, 1990);
(4) Multichannel statistical techniques (cf. Adler et aI., 1993); (5) Multichannel physicallybased radiative transfer algorithms (cf. Wu and Weinman, 1984; Smith and Mugnai, 1988;
Kummerowet a!., 1989; Liu and Curry, 1992).
A particular problem for the microwave is warm (coalescence-dominated) rain over land. Emission techniques do not work here because of the confusion of the land surface seen through the
rain, while ice scattering, which hides the land surface and permits microwave rain retrieval
for strong systems over land, is nonexistent. Algorithm intercomparisons indicate that shallow,
orographic warm rain is simply missed by typical microwave algorithms (cf. Adler et aI., 1993).
20.4.3 Visible/Infrared Sensor (VIRS)
By far the most widely available data on clouds consist of the visible and infrared window channel imagery from geostationary and polar orbiter operational satellites. The obvious appearance
of precipitating systems in such images as (visibly) bright and (thermally) cold regions suggests
that a quantitative measure of precipitation might be derived from such data, which are global
in coverage and several decades in length. The problem is that rainfall itself has little direct
signature in either the visible or infrared. Optical thickness is inversely proportional to particle
size, so large rain droplets contribute little to the visible reflectance of precipitating systems
relative to smaller cloud droplets. Infrared emission depends on the temperature of the cloud
top for an optically thick cloud, and in general higher intensity rain is associated with higher
altitude (and therefore colder) cloud tops. But moderate optical thickness non-raining cirrus
can appear just as cold as heavily precipitating cumulonimbus, and spatial/temporal variations
in atmospheric temperature and water vapor prevent the association of a unique rainfall rate
with a given infrared brightness temperature.
The standard infrared approach to rain estimation is the GOES precipitation index (GPI;
Arkin, 1979), which relates rain rate over a 2.5 0 x 2.5 0 area to the fraction of the area whose
IR brightness temperature is colder than 235 K via a simple linear regression. This algorithm,
based on GATE data for the tropical Atlantic, not surprisingly works well over that region
and is used to produce an operational rainfall product for the entire tropics. As a general
indicator of rainfall patterns and their seasonal or interannual variability, the GPI is a useful
index. But its quantitative accuracy is limited, it d0es not work well outside the tropics, where
non-convective rain is more common, it does not work well for shallow rain systems in general,
and it can give spurious information on the diurnal cycle of rain (cf. the discussion in Fu et
aI., 1990).
To address the problem of distinguishing precipitating cumulonimbus from equally cold nonprecipitating cirrus, Inoue (1987) developed the "split-window technique". This method relies
on the different wavelength dependences of the infrared emissivities of ice and liquid water. By
utilizing differences between the brightness temperatures at 11 and 12 J-Lm, Inoue separates the
cirrus (ice-only) cloud field from the cumulonimbus (ice and liquid) area, and bases the derived
rainfall on the latter. VIRS contains channels at 10.8 and 12 J-Lm to test this approach.
A different philosophy has been adopted recently by Rosenfeld (personal communication). A
common feature of current operational satellites is the existence of near infrared channels that
are sensitive to cloud droplet effective radius near cloud top (Han et a!', 1994). While such
channels also do not see the precipitating particles themselves, the possibility exists that the
droplet size distributions as a whole are shifted to larger sizes in precipitating systems. If
so, precipitation might be inferred indirectly in the near-IR as a function of cloud-top droplet
radii. This approach may provide a means of detecting the shallow rain systems over land that