![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
|||||||
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
||
Model Comparison
We continue our comparison of global climate model simulations with those of
other models and with observed values. Figure
1 shows comparison of temperature
(top) and precipitation (bottom) for different global climate models in
different regions of the globe. Two sets of observations are listed on
this table to show that when averaging over large regions, particularly
where actual measurements are sparse, there is some lack of agreement. The
five regions represent the US Great Plains, Southeast Asia, the Sahel
(Africa just below the Sahara Desert), Southern Europe, and Australia. Notice again that no one model seems to
uniformly excel in representing temperature and precipitation. And also,
in some cases the low-resolution models give better results than
high-resolution models.
Figure 2 shows how one model, in this case the Goddard Institute for Space Studies (GISS) model, reproduces the global pattern of surface temperature difference between July and January. Observations are given in the upper panel and model results in the lower panel. Being a plot of difference, this map does not indicate how well the model simulates absolute temperature but rather the seasonal change from summer to winter. I have colored regions in the Northern Hemisphere with seasonal temperature differences larger than 25oC in green and regions larger than 40oC in red in both plots. The model evidently produces quite good seasonal temperature shifts as indicated by agreement between model and observations over large areas in North America and Russia. Figure 3 gives a similar plot for a different model (the NCAR model), with the model results in the upper panel and observations below. The results for this model are similar to those of the GISS model.
Figure 4, the accompanying two-panel map, gives the model results and observations for December-January-February (DJF) precipitation. This is from a fairly coarse resolution model (notice, for example, that the US does not include the Florida peninsula). If comparison is made for an isolated point on the map, the differences between model and observations will likely be large. However, the overall patterns are quite good and suggests that the model captures key large-scale features of global precipitation even if local values have large errors.
Incorporation of Physical Processes
The climate system, as has been discussed earlier, includes the
atmosphere, ocean, land surface, biosphere, and ice masses. Models of the
climate system must in some way consider the influences of each of these
components. For instance, surface winds drive ocean currents and create
waves that promote mixing in the upper layer of the ocean. This horizontal
movement and vertical mixing of ocean water transports heat poleward from
tropical and subtropical regions and downward from the surface. The heat
transported poleward is given up to the atmosphere at higher latitudes.
Failure to include ocean heat transport would create serious deficiencies
in atmospheric global climate models.
Ocean circulation modeling has proceeded in parallel with atmospheric modeling and is based on the same set of basic equations as the atmosphere (except the equation of state and density are different). The atmosphere exchanges heat, moisture, and trace gases (notably CO2 ) with the ocean, but the rates of change of temperature and speed of movement are much lower in the ocean.
Despite the difficult challenges posed by bringing together two complex numerical models, climate scientists have successfully coupled atmospheric and ocean circulation models. These ocean models may be fairly simple (simple heat diffusion poleward) or complex and use the full equations of motion as for the atmosphere.
Validity of Climate Models
According to Oreskes et
al (Science 263, 641-645), climate models are so large and describe so many
different processes that it is impossible to independently test and certify
that all components of the models are correctly representing physical
processes. However, present and past (paleoclimates, to be discussed
later) climates do offer several opportunities to validate model
performance. From
these comparisons, we have the following conclusions: Validation of global climate models
(Figure 5) (IPCC 1992 and personal
experience)
Models show considerable skill in reproducing large-scale maps of surface pressure, temperature, wind, and precipitation in both summer and winter. On regional scales (sub-continental) all models show significant departures from observations for both temperature and precipitation. Soil moisture comparisons are limited by lack of data, but results (where data are available) are in qualitative agreement. Snow cover is reasonably well simulated, except where temperatures are too high (e.g., at high latitudes in the Northern Hemisphere). Radiative fluxes at the top of the atmosphere are simulated well in some models. Daily and interannual variability is mixed, with most models giving good results in some locations but not good in others. Model response is good for slow changes in forcing such as El Nino, Mount Pinatubo, wet and dry periods in the Sahel, and select periods from the last 18,000 years. Ocean models reproduce large-scale features fairly well. Coupled atmosphere-ocean models do reasonably well in simulations of the last ice age.
It can be concluded that models have enough skill in simulating known climate features that they can be useful tools in trying to estimate the climatic impacts that are likely to occur from the changes in atmospheric chemistry that we discussed earlier in the semester. Recall that we concluded the section on radiative forcing with four scenarios of possible radiative forcing. There are two possible ways the global models can be used to estimate future climatic impacts. The most accurate procedure is to simulate the climate, day-by-day, year-by-year using the increases in radiative forcing previously discussed from the time of the Industrial Revolution until the year 2100 or some other point in the future. Experiments of this computing magnitude are now being done by a limited number of research laboratories having very large computers. Results of these simulations, called transient climate simulations because greenhouse gas concentrations are changing (increasing), will be described in Unit 2-4.
Double CO2 Models
An alternative way to estimate effects of future enhanced
greenhouse gases is to simulate a shorter period (perhaps 20 to 30 years)
with a constant amount of greenhouse gas (usually the equivalent of about
600 ppmv of carbon dioxide, which is approximately twice the pre-Industrial
Revolution level). Concentrations of this magnitude are expected to exist
in about the year 2050. These simulations, called simulations of an
equilibrium 2xCO2 climate, are much shorter to run on supercomputers and have
been done for the last 15 years by several research groups. The factor
not
accurately represented in equilibrium models is the difference in time scales of
components of the climate system. Oceans, soil moisture, and ice masses
are much slower to respond than the atmosphere, and the real climate system
is not expected to ever achieve an equilibrium state.
Coupled Models
Some examples of results from equilibrium simulations of a doubled CO2
climate will now be described. Figure 6 gives results for
coupled models that use a "mixed layer" (shallow ocean of constant
temperature and salinity in the vertical direction). The right-hand-most
column gives the global mean temperature change and precipitation change
projected by the simulation of a doubling of atmospheric CO2 . Note that the
temperatures projected for a doubling of CO2 are all positive (warming
rather than cooling) and range from 1.9o C to 5.2oC
with a mean value of 3.7oC. Similarly, all models report precipitation increases
ranging from 3 to 15% with a mean of 8.7%. Figure 7 gives
some of the most recent results from longer runs (see length of run given in right-hand column).
Changes in temperature and precipitation from a CO2 doubling are comparable
to the previous table.
Maps showing the global distribution of changes in temperature, precipitation, and soil moisture for a doubling of atmospheric CO2 are given in the following series of colored charts. These are from high-resolution models from the Geophysical Fluid Dynamics Laboratory and the United Kingdom Meteorological Office, respectively.
Temperature
The most notable features in the first graphs for
December-January-February (DJF) temperature change (Figure 8) are the dominance of
warming at high latitudes, particularly in the Northern Hemisphere. Both
models agree that the Northern Hemisphere polar region warms substantially
(more than 8oC over wide areas) but the exact boundaries of the areas are
different. Also, the GFHI simulation shows more intense warming at the
North Pole and less difference in warming between land and water areas at
high latitudes. The models disagree on the fate of warming over
Antarctica. This is not surprising in view of the wide model disagreement
over the pressure and circulation patterns over this part of the globe, as
previously discussed. Both models report only weak warming in tropical
areas.
Simulations of warming due to doubling of CO2 for June-July-August (JJA) are given in Figure 9. Compared to the previous graph, the GFHI simulation shifts the maximum warming to the Antarctic ice margin in keeping with the shift to winter in the Southern Hemisphere. The UKHI model calculates less warming, although the site of maximum warming is similar to that of the GFHI map. Tropical regions again show little warming.
Precipitation
Patterns of changes in precipitation over the globe show much less
spatial coherence, as shown in Figure 10 for DJF for the same
two models. Reasonable agreement is shown between models in Africa and mid
and northern latitudes in the Northern Hemisphere. This agreement may be
misleading, however, because comparison with a third model (Canadian
Climate Centre model, not shown) does not confirm the large-scale patterns
shown in the accompanying graphs. Maps for JJA (Figure 11) also
have some areas of agreement and large areas of disagreement.
Soil Moisture
Soil moisture is a climate parameter of considerable scientific and
practical (agricultural) interest that is not routinely measured over large
areas of the earth. This impedes attempts to make comparisons of
calculated values with actual measurements. Figure 12 shows DJF
soil moisture from the same two models. The tendency for agreement of the
two models on DJF precipitation in the mid and high latitudes of the
Northern Hemisphere leads to a tendency for agreement on soil moisture as
well. Values for JJA (Figure 13) show no
consistent agreement.
Model Validations
The following series of images gives examples of what several
global climate models project for the US under an equilibrium doubled CO2. These results are
from very early versions of the respective models, but nevertheless demonstrate several issues in model
validation. Recent model results show improved accuracy.
In each case we also will examine the model validation as compared with the
present climate. Models used in this comparison are the Oregon State
University (OSU) 2-level model, the NASA GISS low-resolution model, and the
GFDL model. The comparison will consist of a series of 4-panel displays of
regional patterns of temperature and precipitation over North America. In
each 4-panel display, the panel in the upper left corner will be the
observed values that serve as the reference pattern for comparison with
model results.
Figure 14 gives the comparison of model validations on mean January temperatures. Maps for each of the three models give the difference in temperature between the model produced value and the observed temperature for that local part of the domain. If a model gave a perfect simulation of the present climate, its map would be a field consisting of all zeros. If the numbers are positive, the model overpredicts the temperature (gives values that are too warm), and negative numbers give a simulated climate that is colder than observed. The GISS model gives values that tend to be positive, suggesting that simulated January temperatures are warmer than observed. Most values are within 2 degrees C of observed, although a few are markedly larger. The largest errors tend to be positive and occur at high latitudes.
The GFDL model is a higher-resolution model that gives a more even balance of positive and negative values but also has large positive numbers in the northeastern part of North America similar to the GISS model. The OSU model also has large positive and negative values but agrees with the other models in overpredicting January temperatures in eastern Canada.
Figure 15 gives a similar plot of July temperatures. In this case, the GISS values are more negative and some of the largest negative values are in Northeastern Canada. The GFDL model is biased too warm but with lower magnitude than for January. The OSU results show large negative values off the east coast of the US and in the mountainous western US. It can be concluded that each model has its own bias that differs from region to region and month to month.
Global climate models also have been used to simulate paleoclimates (climates of the past). In one example where a climate model is used to simulate climate change as Earth emerged from the last ice age, the distribution of spruce pollen observed to have been emitted over North America during this transition agrees very well with the locations where spruce trees would have grown for the climates simulated by the model.
Climate Change
Individual model biases must be taken into account if
they are to be used for projecting future climates. This can be done by
comparing the values for the future climate with the results the model
produces for the present climate, rather than the observed values for the
present climate. This strategy assumes that if the model is biased in a
certain region in a certain month, then it may likely be biased similarly in
the results for the future climate. Subtracting the model simulations of
the present climate (1x CO2) from the 2x CO2 simulation reduces, if not
eliminates, the model bias.
Surface Temperatures
Figure 16 shows the surface-temperature change predicted by
the models for January for an equilibrium climate with a doubled CO2
atmosphere. Since a temperature difference is being plotted, a value of
zero indicates the model calculates no change in surface temperature with a
doubled CO2 atmosphere. The GISS model gives very large values of warming
for the enhanced greenhouse climate, with generally higher values at high
latitudes. The GFDL model produces results that are less severe and more
uniform over the domain. The OSU results also are less severe and do not
have such large values at high latitudes. A common feature of all models
is that they all agree that the result of increased CO2 is to produce a
warming that averages something on the order of about 4o C.
The July calculations given in Figure 17 are similar to the previous results in that all models produce a pattern of warming. There is less evidence of severe warming at high latitudes in the July results, and there is somewhat better agreement among the models in the values produced. The GFDL model tends to produce higher temperature changes than the others.
Precipitation
Model capabilities for producing precipitation are revealed in Figures
18 and 19. Ratios of calculated to observed values are used for
precipitation because of the wide difference in precipitation for different
parts of the domain. As for temperature, the values in the upper left
panel give the observed values for each grid point. The first 4-panel
graph gives comparisons of 1x CO2 results with the observed climate.
Perfect simulations of the present climate would give 1.0 at each point,
since ratios are being plotted. Large percentage errors are produced by
the models for precipitation. A value of 2.0 indicates the model produces
twice the observed precipitation (100% error). Errors of 50 to 100% are
common, and much larger values are observed at many grid points. Such
results do not instill confidence in global climate model capability for
simulating precipitation. The graph for the ratio of 2x CO2 precipitation
to the models' 1x CO2 values gives values of 1.0 or less at most grid
points. Because of the large biases of the previous graph, there is little
justification for using these precipitation results in assessing impacts of
climate change.
The next unit discusses transient climate simulations, which give more realistic simulations of present climate and, hopefully, also future climates.