Ground-Based Remote Sensing of Cloud Properties ...
353
done using the radar alone. However, much more powerful results can be achieved by using the
radar observations in combination with observations from other instruments. In this section, we
provide an overview of the various techniques being employed. Before discussing these in detail,
however, it is useful to consider the data needs from the perspective of the climate community.
As we have demonstrated above, millimeter-wave radar can provide a detailed and unique look
at cloud structure, including some aspects of cloud microphysics. It is important, however, that
we take the next step of translating this information into quantitative data on cloud properties.
Before discussing this process in detail, it is useful to consider from the perspective of the
climate community what the data needs are.
One of the major uncertainties in current climate models is the treatment of clouds as embodied
in the model parameterizations, and the impact of those clouds on the model radiation fields.
Cloud parameterizations vary in their input requirements and output variables but, as a general
rule, the output must specify cloud location (height and thickness) and amount of condensed
phase (liquid or ice). The latter is then converted in some fashion to an optical depth that can
be used as an input to the model radiative codes. More detailed parameterizations also produce
information about cloud particle size and the vertical distribution of the condensed phase as
outputs. Therefore, for a given set of parameterization inputs, verification of a parameterization
relies on validating the following basic predicted quantities: number of cloud layers, base and
top height of each layer, and either the amount of condensed phase or the particle drop size
distributions within each layer. Note that if the particle drop size distribution within a layer is
known, then the amount of condensed phase can be inferred. Because climate models at best
simulate the statistical distribution of these quantities, it is important that we also look at our
ability to provide a temporal distribution of these observed properties.
13.8.2 Models of cloud drop size distributions
Before describing several techniques proposed in the literature for retrieving water vapor path
WV P, liquid water content LWe and path LW P and ice water content [We and path [W P,
as well as the cloud particle number density distribution, it is useful to set forth the notation that
is commonly employed in descriptions of the retrievals. For cloud particles let n(D) represent
the number of particles per unit volume with a size between D and D + dD, where dD is a small
increment in the pertinent dimension of the particles. For spheres D represents the particle
diameter, whereas for nonspherical particles D may perhaps represent the maximum dimension
or the diameter of a spherical particle that has the same volume as the particle in question.
The total number of particles per unit volume becomes
fD m ••
Nt = iDm'n n(D)dD,
(13.75)
where we have adopted the notation of Flatau et al. (1989). For analytic convenience Dmin
and Dmax are generally set to 0 and 00, respectively; the error this appoximation introduces is
generally small.
The particle size density function n(D) is usually modelled either by a lognormal distribution
or the general modified gamma distribution. These two distributions are useful because they
fit the observed distributions of cloud particles and the moments of the distributions can be
defined in terms of known functions. The lognormal distribution is defined by
Nt
[(In(D/Dn))2]
nlog(D) = V21ruD exp -
.,fiu
'
(13.76a)
353
done using the radar alone. However, much more powerful results can be achieved by using the
radar observations in combination with observations from other instruments. In this section, we
provide an overview of the various techniques being employed. Before discussing these in detail,
however, it is useful to consider the data needs from the perspective of the climate community.
As we have demonstrated above, millimeter-wave radar can provide a detailed and unique look
at cloud structure, including some aspects of cloud microphysics. It is important, however, that
we take the next step of translating this information into quantitative data on cloud properties.
Before discussing this process in detail, it is useful to consider from the perspective of the
climate community what the data needs are.
One of the major uncertainties in current climate models is the treatment of clouds as embodied
in the model parameterizations, and the impact of those clouds on the model radiation fields.
Cloud parameterizations vary in their input requirements and output variables but, as a general
rule, the output must specify cloud location (height and thickness) and amount of condensed
phase (liquid or ice). The latter is then converted in some fashion to an optical depth that can
be used as an input to the model radiative codes. More detailed parameterizations also produce
information about cloud particle size and the vertical distribution of the condensed phase as
outputs. Therefore, for a given set of parameterization inputs, verification of a parameterization
relies on validating the following basic predicted quantities: number of cloud layers, base and
top height of each layer, and either the amount of condensed phase or the particle drop size
distributions within each layer. Note that if the particle drop size distribution within a layer is
known, then the amount of condensed phase can be inferred. Because climate models at best
simulate the statistical distribution of these quantities, it is important that we also look at our
ability to provide a temporal distribution of these observed properties.
13.8.2 Models of cloud drop size distributions
Before describing several techniques proposed in the literature for retrieving water vapor path
WV P, liquid water content LWe and path LW P and ice water content [We and path [W P,
as well as the cloud particle number density distribution, it is useful to set forth the notation that
is commonly employed in descriptions of the retrievals. For cloud particles let n(D) represent
the number of particles per unit volume with a size between D and D + dD, where dD is a small
increment in the pertinent dimension of the particles. For spheres D represents the particle
diameter, whereas for nonspherical particles D may perhaps represent the maximum dimension
or the diameter of a spherical particle that has the same volume as the particle in question.
The total number of particles per unit volume becomes
fD m ••
Nt = iDm'n n(D)dD,
(13.75)
where we have adopted the notation of Flatau et al. (1989). For analytic convenience Dmin
and Dmax are generally set to 0 and 00, respectively; the error this appoximation introduces is
generally small.
The particle size density function n(D) is usually modelled either by a lognormal distribution
or the general modified gamma distribution. These two distributions are useful because they
fit the observed distributions of cloud particles and the moments of the distributions can be
defined in terms of known functions. The lognormal distribution is defined by
Nt
[(In(D/Dn))2]
nlog(D) = V21ruD exp -
.,fiu
'
(13.76a)