48
A.D. Del Genio
temperature errors in the tropical and polar upper troposphere. SST climatologies generally
differ from each other by 0.5-1.0°C, but there are important differences in interannual variation
among various data sets as a result of different weighting of in situ and satellite observations.
-General circulation.
Conventional climatologies exist (cf. Oort, 1983), but information on synoptic time scales is
restricted to analysis products provided by ECMWF, NMC, etc. (Raschke, 1996, chapter 1).
These are fairly accurate in northern midlatitudes, where observations strongly influence the
analysis, but problems exist in the tropics, where the parent GCM's cumulus parameterization
is more likely to determine the nature of the flow, and in the southern midlatitudes, where
virtually no observations constrain the analysis. Thus, such analyses must be used judiciously
as validation tools.
-Sea ice.
Estimates of sea ice extent vary significantly among different data sets; the ten-year record
compiled by AMIP to force GCMs contains a sharp discontinuity early in the 1980's which is
probably not real. Other important properties of sea ice, such as thickness and areal coverage
of leads, which dramatically affect the polar surface energy balance, are even more uncertain.
This subject has not received the attention it deserves, given the extreme sensitivity of GCMs
in general to sea ice variations.
b) Physical process data sets.
The validation data described above can only tell a modeler whether a particular parameter is
being simulated correctly or not in the GCM. To correct flawed or incomplete parameterizations,
information is required on the physical processes that control various climate parameters.
Some processes are not understood fundamentally. A good example is cloud top entrainment
instability, which was originally touted as the mechanism that controls the breakup of marine
stratocumulus decks (Randall, 1980). Recent observational and theoretical evidence suggests
that this mechanism occurs if at all under more restrictive instability conditions than originally
thought, and mayor may not be important relative to other processes that can detach stratus
from the surface moisture source (Kuo and Schubert, 1988; Wang and Albrecht, 1994).
Other processes are understood on the cloud scale but not on the GCM grid scale. The most
obvious example is the relationship between cloud cover and relative humidity. On the cloud
scale, condensation occurs very close to 100% relative humidity, at least for liquid clouds. But
an area the size of a GCM grid box is generally not saturated, yet some fraction may be occupied
by clouds because of subgrid-scale variations in temperature and moisture. Some GCMs assume
a correlation between cloud cover and relative humidity on the grid scale; observations suggest
that this is true statistically but with a considerable degree of scatter (Soden and Bretherton,
1993; Walcek, 1994). Other GCMs assume a subgrid-scale distribution of some thermodynamic
quantity to diagnose cloud cover, but what controls this distribution is not known, and arbitrary
assumptions can sometimes have significant unintended feedback effects (Miller and Del Genio,
1994).
Process studies can involve either actual physical mechanisms that control a variable or diagnosis of the component parameters that determine an integrated measure of radiative effect.
Process understanding is best accomplished by combining satellite data, in situ measurements,
theory, and model sensitivity studies. An instructive ongoing case in point is the question of the
factors controlling optical thickness feedback, a major aspect of disagreement among climate
models. Basic thermodynamics indicates that the adiabatic liquid water content of a lifted
cloud should increase with temperature (Betts and Harshvardhan, 1987), and aircraft data for
low and midlevel clouds generally seem to verify this (Feigelson, 1978). This implies a negative
optical thickness feedback if other factors are held fixed (Somerville and Remer, 1984).
A.D. Del Genio
temperature errors in the tropical and polar upper troposphere. SST climatologies generally
differ from each other by 0.5-1.0°C, but there are important differences in interannual variation
among various data sets as a result of different weighting of in situ and satellite observations.
-General circulation.
Conventional climatologies exist (cf. Oort, 1983), but information on synoptic time scales is
restricted to analysis products provided by ECMWF, NMC, etc. (Raschke, 1996, chapter 1).
These are fairly accurate in northern midlatitudes, where observations strongly influence the
analysis, but problems exist in the tropics, where the parent GCM's cumulus parameterization
is more likely to determine the nature of the flow, and in the southern midlatitudes, where
virtually no observations constrain the analysis. Thus, such analyses must be used judiciously
as validation tools.
-Sea ice.
Estimates of sea ice extent vary significantly among different data sets; the ten-year record
compiled by AMIP to force GCMs contains a sharp discontinuity early in the 1980's which is
probably not real. Other important properties of sea ice, such as thickness and areal coverage
of leads, which dramatically affect the polar surface energy balance, are even more uncertain.
This subject has not received the attention it deserves, given the extreme sensitivity of GCMs
in general to sea ice variations.
b) Physical process data sets.
The validation data described above can only tell a modeler whether a particular parameter is
being simulated correctly or not in the GCM. To correct flawed or incomplete parameterizations,
information is required on the physical processes that control various climate parameters.
Some processes are not understood fundamentally. A good example is cloud top entrainment
instability, which was originally touted as the mechanism that controls the breakup of marine
stratocumulus decks (Randall, 1980). Recent observational and theoretical evidence suggests
that this mechanism occurs if at all under more restrictive instability conditions than originally
thought, and mayor may not be important relative to other processes that can detach stratus
from the surface moisture source (Kuo and Schubert, 1988; Wang and Albrecht, 1994).
Other processes are understood on the cloud scale but not on the GCM grid scale. The most
obvious example is the relationship between cloud cover and relative humidity. On the cloud
scale, condensation occurs very close to 100% relative humidity, at least for liquid clouds. But
an area the size of a GCM grid box is generally not saturated, yet some fraction may be occupied
by clouds because of subgrid-scale variations in temperature and moisture. Some GCMs assume
a correlation between cloud cover and relative humidity on the grid scale; observations suggest
that this is true statistically but with a considerable degree of scatter (Soden and Bretherton,
1993; Walcek, 1994). Other GCMs assume a subgrid-scale distribution of some thermodynamic
quantity to diagnose cloud cover, but what controls this distribution is not known, and arbitrary
assumptions can sometimes have significant unintended feedback effects (Miller and Del Genio,
1994).
Process studies can involve either actual physical mechanisms that control a variable or diagnosis of the component parameters that determine an integrated measure of radiative effect.
Process understanding is best accomplished by combining satellite data, in situ measurements,
theory, and model sensitivity studies. An instructive ongoing case in point is the question of the
factors controlling optical thickness feedback, a major aspect of disagreement among climate
models. Basic thermodynamics indicates that the adiabatic liquid water content of a lifted
cloud should increase with temperature (Betts and Harshvardhan, 1987), and aircraft data for
low and midlevel clouds generally seem to verify this (Feigelson, 1978). This implies a negative
optical thickness feedback if other factors are held fixed (Somerville and Remer, 1984).