Some Aspects of Uncertainty in Predicting Sea Ice Thinning - Arctic Sea Ice Decline: Observations, Projections, Mechanisms, and Implications - page 71

6. DISCUSSION rather than thickness change to illustrate this narrowing of

uncertainty in time. The distribution in thickness becomes
more sharply peaked after doubling CO 2 , especially when
the gain from ice-albedo feedback is included.
In the experiments described here, CO 2 was increased
from 355 to 710 ppm, so the albedo effect is evaluated for
a perturbation that transforms most of the perennial ice to
seasonal ice in the Arctic Ocean in CCSM3. This forcing is
roughly equivalent to the total anthropogenic forcing in the
first half of the 21st century of the SRES A1B scenario. Ex-
periments were run with CO 2 raised to just 550 ppm as well
(not shown), which gave nearly the same estimate for the
ice albedo feedback factor on ice thickness f in CCSM3. Be-
cause f depends little on the magnitude of the perturbation, I
expect f would not vary much during a transient integration
either.

I have estimated the uncertainty from two primary ther-
modynamic feedbacks: ice-albedo feedback and the growth-
thickness feedback. No doubt there are also feedbacks
between the ice and ocean that vary from model to model,
and these feedbacks may also depend on the mean state. Be-
cause I have not accounted for them, I have focused on ice
thickness north of 70°N, where I expect far less influence
from the ocean than in the subpolar seas. I have also not
tried to quantify the uncertainty in how models treat sea ice
dynamics and ice export. My estimates of the distribution
widths should be thought of as a lower limit.
For simplicity (and by analogy to the work of Roe and
Baker [2007]), the distributions here are meant to represent
the climate in equilibrium after doubling CO 2 . In the future
scenario shown in Figure 1, the trends of sea ice thinning in
the early 21st century span almost an order of magnitude. I
have assumed that the uncertainty in equilibrium ice thick-
ness change from doubling CO 2 would be similar.
Earlier I noted that the CMIP3 models with thicker ice
in the late 20th century thin at a faster rate in the early 21st
century (see Plate 1a). As a result the uncertainty in mean ice
thickness among CMIP3 models tends to decline in time over
the two centuries. Figure 6 recasts estimates of the probabil-
ity density functions from the section 5 in terms of thickness

7. CONCLUSIONS

The average sea ice thickness north of 70°N in CMIP3
models ranges from less than 1 m to more than 3 m in the late
20th century. The rate of sea ice thinning in the 21st century
in these models is a strong function of the late 20th century
thickness, such that models with above-average thickness
also thin faster than average. The average ice thickness north
of 70°N across the CMIP3 models is highly correlated with
the September ice extent and therefore strongly influences
marine ecosystems and early winter surface temperatures.
Because sea ice thickness change depends sensitively on the
mean state, error in a model's climatology gives rise to error
in future predictions of ice thinning and extent.
I have shown that uncertainty in the strength of ice albedo
feedback is probably not a major source of uncertainty for
ice thinning in future predictions. This result stems from the
fact that the ice-albedo feedback factor on ice thickness f is
rather small. I estimated f in a global climate model by hold-
ing the surface albedo of sea ice and ocean fixed while dou-
bling CO 2 . Ice-albedo feedback causes sea ice to thin about
26% more compared to a model run without ice-albedo feed-
back. A gain of 26% corresponds to a feedback factor of
only f = 0.21 ± 0.02, where the error here is an estimate of
uncertainty in this one model (which is bound to be much
smaller than the range of f across models).
Such a small value for f can only give rise to a fairly nar-
row estimate for thickness change provided the range of f
across models is the sole source of uncertainty. Even if the
uncertainty of f across models is as high as 100% (ranging
from 0 to 0.42), it causes little uncertainty in the ice thick-
ness change.
Instead, the uncertainty in the mean state has a much
larger influence on uncertainty in the thickness change. I

Figure 6. Distributions recast as a function of ice thickness rather
than change in ice thickness, showing that the distribution narrows
as the ice thins. The grey line is the initial assumption of a normal
distribution of ice thickness with - = - 1.8 m, s h = 0.77 m. The
other two lines are distributions after doubling CO 2 without ice-
albedo feedback ( f = 0 , dashed line) and with ice-albedo feedback
( f = 0.21 and s f = 0 , black line).