The Role of Natural Versus Forced Change in Future Rapid Summer Arctic Ice Loss - Arctic Sea Ice Decline: Observations, Projections, Mechanisms, and Implications - page 131

To further evaluate the influence of natural and forced
variations we also examine a large 29-member ensemble of
CCSM3 21st century model integrations. These runs employ
a coarser-resolution atmospheric component with T42 (ap-
proximately 2.75°) resolution. The late 20th century Arctic
sea ice state that these runs are initialized from is quite dif-
ferent because of the different resolution of the atmospheric
model (e.g., see DeWeaver and Bitz [2006] for a discussion
of resolution and Arctic sea ice in CCSM3 control runs). Of
interest here, the early 21st century Arctic sea ice is consid-
erably thicker in these simulations than in their T85 coun-
terparts. This modifies the projected change in ice cover as
discussed below. These simulations are also run using the
middle-range SReS A1B scenario.

In any given ensemble member, one does not expect a direct
match between a particular year and the corresponding year
in the observational record because of the effect of natural
variability on the temporal evolution of ice extent. However,
we can gauge the consistency of the model simulations and
the observed record by comparing statistics of the time series
among multiple ensemble members with the observations. An
analysis of the September ice extent trends from observations
and model simulations (Table 1) confirms that at the Arctic
basin scale, the simulated trends are consistent with observa-
tions. The observed September ice extent trend from 1979 to
2007 falls within the range of trends from the different en-
semble members. The ensemble members differ only in their
initial conditions, and as such, the different trends that are
present among the ensemble members result from the intrin-
sic (“natural”) model variability. Thus, the simulated trends
agree with the observations within these intrinsic variations.
As discussed by Stroeve et al . [2007], this is in contrast to
many other IPCC-AR4 models, which generally simulate
smaller trends than observed over the satellite record.
The 1979-2007 interannual standard deviation in Septem-
ber ice extent within the model ensemble members is also
generally consistent with observations (Table 1), as the ob-
served value falls within the range of values from different
ensemble members. For the detrended 1979-2007 Septem-
ber time series, the observed standard deviation is slightly
higher than that simulated in any of the ensemble members.
This is due to the very anomalous conditions of 2007, and the
simulated values for different ensemble members bracket the
observations if only the 1979-2006 time period is considered
(Table 1). In general, the IPCC-AR4 CCSM3 integrations
appear to simulate realistic year-to-year and long-term varia-
tions in the September ice cover compared to observations.

3. ReSULTS

3.1. Simulated Arctic Sea Ice

The ensemble mean IPCC-AR4 CCSM3 sea ice thickness
averaged from 1980 to 1999 is shown in Plate 1. In accord
with observations [e.g., Bourke and Garrett , 1987; Laxon et
al ., 2003], the thickest ice is present along the Canadian Arc-
tic Archipelago and north of Greenland. However, the model
also obtains relatively thick ice throughout the east Siberian
Sea in contrast to observations. This is a common problem in
many coupled climate models and is likely associated with
biases in the wind forcing [e.g., Bitz et al ., 2002; DeWeaver
and Bitz , 2006]. Satellite altimetry measurements suggest an
October-March mean ice thickness of 2.73 m for the region
south of 81.5N, excluding ice thinner than 1 m [ Laxon et al .,
2003]. Using a consistent method of averaging, the CCSM3
ensemble mean obtains a value of 2.6 m in good agreement
with observations.
The IPCC-AR4 CCSM3 has a reasonable winter ice edge
in the Greenland and Bering seas compared to observations,
but the winter ice is too extensive in the Labrador Sea and
the Sea of Okhotsk (Plate 1). Overall, the CCSM3 integra-
tions obtain a winter ice cover that is from 1 to 1.8 million
km 2 larger than observed (Figure 1). A recent study [ Jochum
et al ., 2008] indicates that substantial improvements in these
regions are realized in CCSM3 simulations with modified
ocean viscosity. From June to December, the simulated to-
tal Arctic ice extent, defined as the area with greater then
15% ice concentration, is in excellent agreement with the
observations and the difference between the two is typically
smaller than the interannual standard deviation. One small
discrepancy from observations during the summer months
is that the simulated ice edge is displaced poleward within
the Kara Sea.

3.2. Projected Changes in Ice Cover

Plate 2 shows the projected changes in the September Arc-
tic ice extent within the CCSM3 ensemble members. Abrupt
events are indicated by the grey shading. The events are
identified following HBT as times when the derivative of the
5-year running mean smoothed September ice extent time
series exceeds a loss of 0.5 million km 2 a −1 . The event length
is determined by the time around the transition for which
the smoothed time series derivative is larger than 0.15 mil-
lion km 2 a −1 . Only time periods that fit this definition and are
longer than 2 years in length are considered “abrupt events.”
Although subjective, this definition clearly identifies periods
of rapid retreat in the September ice cover. These vary from
4 to 10 years in length, and the trends over these events are
typically 3-4 times larger than comparable length trends in
the smoothed observational time series from 1979 to 2005. If