Geoscience Reference
In-Depth Information
use of a statistical relationship between the large-scale
values calculated for the grid cell in order to determine
the effect of the parameterized process.
In order to gain confidence in the performance
of models in predicting future atmospheric states, it
is important to evaluate how well such models perform
in representing present-day climate statistics. The
Atmospheric Model Intercomparison Programme
(AMIP) is designed to do this by comparing models
from various centres around the world using common
procedures and standardized data (on sea-surface
temperatures, for example), as well as by providing
extensive documentation on the model design and the
details of model parameterizations. In this way common
deficiencies can be detected and perhaps attributed to
a single process and then addressed in future model
versions. Figure 8.3 compares simulated zonally aver-
aged surface temperature for January and July for all
AMIP participants with the observed climatological
mean. The general features are well represented qual-
itatively, although there can be large deviation between
individual models. The evaluation of models requires
analysis of their ability to reproduce interannual vari-
ability and synoptic-scale variability as well as mean
conditions. A comparison project for AOGCMs similar
to AMIP is now underway called the Coupled Model
Intercomparison Project (CMIP). Plate G illustrates
the 500-mb heights for northern winter and summer,
as observed (top) and as simulated by the National
Center for Atmospheric Research (NCAR) Community
Climate Model (CCM) 3, and the considerable differ-
ences between them in high latitudes.
Recent models incorporate improved spatial reso-
lution and fuller treatment of some previously neglected
physical processes. However, both changes may create
additional problems as a result of the need to treat
accurately complex interactions such as those between
the land surface (soil moisture, canopy structure, etc.)
and the atmospheric boundary layer, or interactions
between clouds, radiative exchanges and precipitation
mechanisms. For example, fine-scale spatial resolution
is necessary in the explicit treatment of cloud and rain
bands associated with frontal zones in mid-latitude
cyclones. Such processes require detailed and accurate
representation of moisture exchanges (evaporation,
condensation), cloud microphysics and radiation (and
the interactions between these processes) which are all
represented as averaged processes when simulated
at larger spatial scales.
Coupled model hierarchy
A
Atmospheric GCM
Swamp
SST from surface energy balance
B
Atmospheric GCM
Mixed layer
(slab)
SST from surface energy balance,
heat storage
Atmospheric GCM
C
SST from surface
energy balance,
heat storage,
advection,
diffusion
Ocean
GCM
Figure 8.2
Schematic illustration of the three types of coupling
of an atmospheric GCM to the ocean: (A) swamp ocean (B) mixed
layer, slab ocean (C) ocean GCM.
Source
: From Meehl (1992). Copyright © Cambridge University
Press.
concern is 'model drift' (a definite tendency for the
model climate to warm or cool with time) due to
accumulating errors from the various component
models. These tendencies are often constrained by using
observed climatology at certain high-latitude or deep
ocean boundaries, or by adjusting the net fluxes of heat
and fresh water at each grid point on an annual basis in
order to maintain a stable climate, but such arbitrary
procedures are the subject of controversy, especially for
climate change studies.
Many important weather and climate processes
occur on a scale which is too small for the typical GCM
to simulate with a grid of several degrees on a side.
Examples of this would be the radiative effects or
latent heating due to cloud formation or the transfer
of water vapour to the atmosphere by a single tree.
Both processes greatly affect our climate and must
be represented for a realistic climate simulation.
Parameterizations
are methods designed to take into
account the average effect of cloud or vegetation process
on an entire grid cell. Parameterizations generally make
