The Role of Natural Versus Forced Change in Future Rapid Summer Arctic Ice Loss - Arctic Sea Ice Decline: Observations, Projections, Mechanisms, and Implications - page 137

thicker in 2000 compared to the IPCC-AR4 integrations.
The ice is also distributed differently across the Arctic basin
than in the higher-resolution runs, with the thickest ice cover
present in the east Siberian Sea in contrast to observations.
This is consistent with the sea ice change in response to at-
mospheric resolution variations in CCSM3 control climate
integrations as discussed by DeWeaver and Bitz [2006].
The initially thicker ice cover modifies the projection of ice
extent. As shown in Plate 6b, the September ice extent loss
over the 21st century is delayed in the coarse-resolution runs
with similar conditions occurring about 20 years later than in
the IPCC-AR4 runs. Plate 6c shows the September ice extent
as a function of the May mean Arctic ice thickness. On the
basis of this measure, a more consistent picture emerges for
the two different sets of model integrations. This suggests
that the difference in initial ice thickness is a primary factor
responsible for the different projected ice extent loss.
The coarse-resolution ensemble members exhibit rapid
September ice loss events as defined above (not shown) but
generally do so later in the 21st century, consistent with their
initially thicker sea ice. The events typically start around
year 2050, but the earliest abrupt transition begins in 2021.
The event lengths are almost uniformly distributed between
3 and 7 years. They have typical trends in the smoothed time
series of −0.4 million km 2 per a −1 over the event length with
the trends varying from −0.6 million km 2 per a −1 to −0.3 mil-
lion km 2 a −1 for different events. This is similar to the identi-
fied trends in the IPCC-AR4 CCSM3 simulations. Of the 29
ensemble members, 17 (about 60%) exhibit an abrupt event.
The remaining 12 ensemble members have no periods that
meet the defined abrupt event criteria. However, the simula-
tions only run to approximately 2060 and still have about 3
million km 2 of ice cover in the 2060 ensemble mean. Addi-
tional abrupt events are thus possible if the simulations were
extended past 2060.
By subtracting the ensemble mean ice conditions shown
in Plate 6b from each individual ensemble member, we are
left with information about the intrinsic variations within
each simulation. As occurs in the IPCC-AR4 CCSM3 runs,
the natural variability in September ice extent (defined as
the deviation from the ensemble mean) increases with the
thinning ice pack (Plate 6d). Quantitatively, this agrees well
with the IPCC-AR4 simulations, suggesting that the small
ensemble size in the higher-resolution simulations does not
substantially bias those results. Previous work has shown
that anthropogenic climate change modifies other aspects of
natural variability, such as interannual air temperature and
precipitation fluctuations [e.g., Giorgi and Bi , 2005].

variability in September ice extent plays an important role
in the simulated abrupt ice loss events. There are numerous
factors that can contribute to natural variations in the sea
ice cover, including both dynamic and thermodynamic forc-
ing anomalies. As discussed by HBT, pulse-like increases
in ocean heat transport to the Arctic play an important role
in triggering the simulated ice loss events. These appear in
part to be driven by variations in the atmospheric forcing
that are suggestive of the North Atlantic Oscillation/Arctic
Oscillation [ Hurrell , 1995; Thompson and Wallace , 1998]
(Figure 4).
Other intrinsic variations with the modeled climate also
likely play a role in the simulated rapid ice loss. However, an
analysis of the surface atmosphere-ice heat exchange terms
shows little indication of extreme changes preceding the ice
loss events (not shown). In contrast, there are indications of
feedbacks associated with the changing ice conditions, and
many surface flux terms exhibit considerable changes during
the ice loss events. As discussed by HBT, the surface albedo
feedback and changing net surface shortwave radiation ap-
pears to be critically important. Also of interest, there are
changes in cloud conditions and cloud radiative forcing at
the surface that may play a role in these events. These feed-
backs will be examined further in future work.
Of course, natural variations also play an important role
in the simulated 20th century Arctic climate. For example,
pulse-like ocean heat transport anomalies are present within
the 20th century. These have an important effect on simu-
lated ice volume variations but do not translate into an ice
area change because of the relatively thick ice that is present.
This again highlights that a necessary, if not sufficient, con-
dition for the modeled abrupt ice loss is that the ice cover is
adequately thinned.

3.3. Possible Predictability of the Ice Loss Events

As discussed above, the abrupt ice loss events appear to
result from a growing intrinsic ice extent variability coupled
to considerable forced change. There is little indication of a
critical ice threshold that causes these events, and the cha-
otic nature of the natural variations makes the potential pre-
dictability of these events difficult. However, as changes in
ocean heat transport are implicated as a trigger for the abrupt
ice loss and the ocean varies on long timescales, it is possible
that knowing the prior ocean state may provide some predic-
tive skill for these events.
Here, we further examine the importance of the prior ice
and ocean conditions in the simulation of the events. We
focus on the long-lived abrupt ice loss present in ensem-
ble run 1 (Plate 2a) and use a set of sensitivity experiments
which are initialized at year 2020 with various ice and ocean

3.2.3. Processes contributing to natural variations. The
above analysis suggests that a large and growing intrinsic