34
A.D. Del Genio
individual storms suggests that the GCM is especially deficient in cloudiness along cold fronts,
perhaps because the model cannot resolve the tilted frontal structure and associated vertical
motions. But going to higher resolution (4° x 5° to 2° x 2.5°) does not improve the situation,
nor do higher order differencing schemes. Thus, an apparent problem in a different part of the
GCM affects the performance of the cloud parameterization, with no clear-cut solution.
A different situation exists in the tropics, where too little sunlight reaches the surface in the
GCM. This problem illustrates another common misperception in the climate community. It has
become popular to assume that model climate can be validated simply by ensuring correct TOA
broadband radiation fluxes or monochromatic radiances corresponding to particular satellite
instruments. Furthermore it is asserted that errors can be diagnosed simply by comparing
the individual shortwave and longwave contributions, without making use of information on
physical parameters because estimates of the latter are too dependent on retrieval algorithm to
be useful. This is based on the observation that high clouds have a distinct longwave signature,
while low clouds do not. But clouds reach unit longwave emittance at fairly low visible optical
depths (T ~ 4), while shortwave reflectance continues to increase with increasing optical depth.
Thus, for the optically thick cumulus anvil clouds that pervade the tropics, an error in cloud
cover will be manifested as an error in both the shortwave and longwave, but an error in optical
thickness will show up only in the shortwave. The latter is indistinguishable from an error in
low cloud cover or optical thickness using fluxes alone.
We can diagnose the tropical shortwave error in the GISS GCM only by combining satellite
and surface-based data sets (Fig. 2.2). First, comparison of the GCM with ERBE outgoing
longwave radiation indicates smaller tropical errors, both positive and negative, ruling out
anvil cloud cover errors. On the other hand, the GCM simulates much more low cloud cover in
these regions than is indicated by the surface-observed cloud climatology. Furthermore, ENSO
variations in tropical Pacific cloud forcing (Fig. 2.3) are consistent with ERBE (Ramanathan
and Collins, 1991), while the distribution of tropical high cloud optical thicknesses is consistent
with that deduced by ISCCP (Fig. 2.4).
The comparison of model and observed anvil cloud behavior illustrates two more points about
the validation of GCMs. The first is that variability reveals model performance in ways that
are impossible using information about the mean state alone. In the above example, ENSO
variability allows the anvil cloud effect to be isolated (because the primary tropical effect of
ENSO is a shift in the location of convection centers), while the mean state is a composite
of high and low cloud effects. Furthermore, a quantity and its derivative are not the same
thing, and validating a model's mean state gives no information about whether it will respond
realistically to climate perturbations.
Climate modelers are only now beginning to test their GCMs against observed variability.
Some types of variability are more useful than others in understanding climate change; what
is crucial is to determine the mechanism of variability in the current climate and whether it is
relevant to that which causes the climate feedback (cf. Del Genio, 1996a). For example, the
seasonal cycle is often said to be useless as a proxy for climate change because it is dominated
by seasonal shifts in the location of the Hadley cell, which are irrelevant to long-term climate
change. But poleward of ±30° latitude, baroclinic waves dominate the dynamics, and winter-tosummer weakening of the meridional temperature gradient, which affects baroclinic instability,
is a good proxy for the polar amplification of warming that occurs in decadal climate change
simulations. Thus, observations of seasonal water vapor variations (Del Genio et aI., 1994) in
midlatitudes may contain useful information about regional long-term water vapor feedback.
On the other hand, ENSO variations in the tropics mayor may not be a very good indicator
of long-term climate feedbacks: In ENSO, net equatorial warming strengthens the Hadley cell
(Pan and Oort, 1983) and thus dries the subtropics due to enhanced subsidence. In decadal
A.D. Del Genio
individual storms suggests that the GCM is especially deficient in cloudiness along cold fronts,
perhaps because the model cannot resolve the tilted frontal structure and associated vertical
motions. But going to higher resolution (4° x 5° to 2° x 2.5°) does not improve the situation,
nor do higher order differencing schemes. Thus, an apparent problem in a different part of the
GCM affects the performance of the cloud parameterization, with no clear-cut solution.
A different situation exists in the tropics, where too little sunlight reaches the surface in the
GCM. This problem illustrates another common misperception in the climate community. It has
become popular to assume that model climate can be validated simply by ensuring correct TOA
broadband radiation fluxes or monochromatic radiances corresponding to particular satellite
instruments. Furthermore it is asserted that errors can be diagnosed simply by comparing
the individual shortwave and longwave contributions, without making use of information on
physical parameters because estimates of the latter are too dependent on retrieval algorithm to
be useful. This is based on the observation that high clouds have a distinct longwave signature,
while low clouds do not. But clouds reach unit longwave emittance at fairly low visible optical
depths (T ~ 4), while shortwave reflectance continues to increase with increasing optical depth.
Thus, for the optically thick cumulus anvil clouds that pervade the tropics, an error in cloud
cover will be manifested as an error in both the shortwave and longwave, but an error in optical
thickness will show up only in the shortwave. The latter is indistinguishable from an error in
low cloud cover or optical thickness using fluxes alone.
We can diagnose the tropical shortwave error in the GISS GCM only by combining satellite
and surface-based data sets (Fig. 2.2). First, comparison of the GCM with ERBE outgoing
longwave radiation indicates smaller tropical errors, both positive and negative, ruling out
anvil cloud cover errors. On the other hand, the GCM simulates much more low cloud cover in
these regions than is indicated by the surface-observed cloud climatology. Furthermore, ENSO
variations in tropical Pacific cloud forcing (Fig. 2.3) are consistent with ERBE (Ramanathan
and Collins, 1991), while the distribution of tropical high cloud optical thicknesses is consistent
with that deduced by ISCCP (Fig. 2.4).
The comparison of model and observed anvil cloud behavior illustrates two more points about
the validation of GCMs. The first is that variability reveals model performance in ways that
are impossible using information about the mean state alone. In the above example, ENSO
variability allows the anvil cloud effect to be isolated (because the primary tropical effect of
ENSO is a shift in the location of convection centers), while the mean state is a composite
of high and low cloud effects. Furthermore, a quantity and its derivative are not the same
thing, and validating a model's mean state gives no information about whether it will respond
realistically to climate perturbations.
Climate modelers are only now beginning to test their GCMs against observed variability.
Some types of variability are more useful than others in understanding climate change; what
is crucial is to determine the mechanism of variability in the current climate and whether it is
relevant to that which causes the climate feedback (cf. Del Genio, 1996a). For example, the
seasonal cycle is often said to be useless as a proxy for climate change because it is dominated
by seasonal shifts in the location of the Hadley cell, which are irrelevant to long-term climate
change. But poleward of ±30° latitude, baroclinic waves dominate the dynamics, and winter-tosummer weakening of the meridional temperature gradient, which affects baroclinic instability,
is a good proxy for the polar amplification of warming that occurs in decadal climate change
simulations. Thus, observations of seasonal water vapor variations (Del Genio et aI., 1994) in
midlatitudes may contain useful information about regional long-term water vapor feedback.
On the other hand, ENSO variations in the tropics mayor may not be a very good indicator
of long-term climate feedbacks: In ENSO, net equatorial warming strengthens the Hadley cell
(Pan and Oort, 1983) and thus dries the subtropics due to enhanced subsidence. In decadal