Among the most serious sources of uncertainty in current general circulatio
n models (GCMs) is the treatment of clouds and cloud-radiation interactions
. We have used a single-column model (SCM) diagnostically to evaluate param
eterizations against observations from the Atmospheric Radiation Measuremen
t (ARM) Program. We find that schemes with explicit cloud water budgets and
interactive radiative properties are potentially capable of matching obser
vational data closely. In our SCM, using an interactive cloud droplet radiu
s decreases the cloud optical thickness and cloud infrared emittance of hig
h clouds, which acts to increase the downwelling surface shortwave flux and
the outgoing longwave radiation. However, it is difficult to evaluate the
realism of the vertical distribution of model-produced cloud extinction, cl
oud emittance, cloud liquid water content and effective cloud droplet radiu
s until high-quality observations of these quantities become more widely av
ailable. We also find that in the SCM, cloud parameterizations often undere
stimate the observed cloud amount, and that ARM observations indicate the p
resence of clouds while the corresponding maximum relative humidity is less
than 80%. This implies that the underlying concept of a critical gridpoint
relative humidity of about 80% for cloud formation, as used in many GCM cl
oud parameterizations, may need to be reexamined. (C) 1999 Elsevier Science
Ltd. All rights reserved.