The Role of Natural Versus Forced Change in Future Rapid Summer Arctic Ice Loss - Arctic Sea Ice Decline: Observations, Projections, Mechanisms, and Implications - page 134

is quite thick (2-3 m), whereas for the central ice pack it is
only 1-2 m. This is perhaps not too surprising as the shelf re-
gions generally reach September ice-free conditions earlier
in the simulation. While the mean ice conditions averaged
over the ice loss regions vary considerably across the differ-
ent events, this could be unduly influenced by the different
regions where ice loss occurs. Thus, it is still possible that a
common spatially variable “critical ice thickness” is present
but that the region and time where this thickness is achieved
varies widely from run to run. If true, abrupt loss would be
possible if a considerable region of the ice pack reached the
appropriate conditions near the same time. The different
character of each abrupt ice loss event could then be associ-
ated with what region reaches these “critical” conditions.
Figure 2 shows the area of spring sea ice within 25 (and
50) cm of a common “critical ice thickness” for each en-
semble member. This “critical ice thickness” is defined by
the ensemble mean of the May thickness 5 years prior to
the time at which a transition to September ice-free condi-
tions results within each model grid cell (Plate 4, bottom
right). Some abrupt events are clearly preceded by anoma-
lously large regions of the ice cover reaching this “critical
ice thickness” (e.g., run 4, the second event in run 6, the
second event in run 7). However, a number of runs show
anomalously low areas of “critical ice thickness” just prior
to abrupt September ice loss (e.g., run 1, the first event in
run 7). Thus, it is not always the case that an anomalously
large region of the ice pack reaches such critical conditions
just prior to abrupt ice loss. This suggests that indeed there
is no common critical ice state present among the ensemble
members that leads to an abrupt ice loss. So, in addition to
an adequately thin ice cover, a forcing perturbation appears
necessary to initiate the events. This forcing perturbation is
unique to the individual ensemble member and results from
natural variability. It reinforces changes in external forc-
ing that are applied identically across the members. For this
process to occur, a thinned ice cover appears a necessary but
not sufficient condition.

tions, we will attribute this ensemble mean response as the
“forced” response that results from the changing external
forcing. As shown in Figure 3a, this forced response is siz-
able with a 1.1 million km 2 per decade rate of September
ice loss from 2000 to 2050. As a result, the ensemble mean
reaches near September ice-free conditions by 2050.
Subtracting the ensemble mean from the ice extent time
series for each individual realization leaves us with a char-
acterization of the “natural” variations. Figure 3b shows this
time series for the extreme ice loss event simulation in en-
semble run 1 (Plate 2a). Two things are apparent. First, there
is a relatively uniform distribution of positive and negative
anomalies about the ensemble mean. Second, the natural
variability in September ice extent increases during the early
21st century and then decreases again after 2050 at which
time little September ice cover remains. This is quantified in
Figure 3c, which shows the 20-year running standard devia-
tion of September ice extent over the 20th to 21st centuries
for run 1. Interestingly, while the variance increases, there is
little change in skew or kurtosis of the ice extent anomalies
as the ice thins (not shown).
The increase in ice extent variability with a thinning ice
cover is a consistent feature of the CCSM3 simulations
(Plate 5) and indeed of most climate models (not shown)
and makes good physical sense. As the ice cover thins, nat-
ural variations in atmosphere and/or ocean conditions that
modify the sea ice mass budgets translate more readily into
a change in ice area as large regions of the thin ice pack can
completely melt out. The thinning ice cover is in essence
more vulnerable to intrinsic climate perturbations. In turn,
the retreating ice cover causes an amplification of the change
because of the surface albedo feedback, resulting in still
larger ice extent change. Other feedbacks, associated with
changing longwave forcing, for example [ Gorodetskaya and
Tremblay , this volume], may also contribute.

3.2.2. Coarse-resolution results. One issue with the above
analysis is that the separation into natural and forced change
relies on a relatively small ensemble member size. To com-
plement these results, we here examine a large 29-member
ensemble of simulations runs from 2000 to approximately
2060 using the SReS A1B scenario forcing. These simula-
tions use a coarser-resolution (T42, approximately 2.75°) at-
mospheric component of the CCSM3 model coupled to the
identical ocean and sea ice components as used in the IPCC-
AR4 simulations. Aspects of the Arctic climate conditions in
control climate integrations with this model are discussed by
Holland et al . [2006b] and DeWeaver and Bitz [2006].
Because of the different atmospheric model resolution,
different mean late 20th century sea ice conditions result. In
particular, as shown in Plate 6a, the sea ice is considerably

3.2.1. Quantifying natural versus forced change. There
are eight realizations of 20th to 21st century integrations in
the suite of IPCC-AR4 CCSM3 scenario runs that are forced
with identical external forcing and only vary in their initial
conditions. While this is a relatively small number, a com-
parison between the different realizations can provide insight
into the role of natural versus forced change in driving rapid
sea ice loss. Figure 3a shows the ensemble mean September
sea ice extent. Because of the relatively small ensemble size,
this time series exhibits the signature of some individual
realizations. However, as averaging across these ensemble
members generally removes the uncorrelated natural varia-