A Bayesian Network Modeling Approach to Forecasting the 21st Century Worldwide Status of Polar Bears - Arctic Sea Ice Decline: Observations, Projections, Mechanisms, and Implications - page 229

Because polar bears are tied to the sea ice for obtaining
food, major changes in the quantity and distribution of sea
ice must result in similar changes in polar bear distribution.
Therefore, the distributional effects of projected changes
are most apparent. Whereas it is fairly certain that polar
bears will not remain in areas where habitat absence is too
prolonged to make seasonal use practical, it is less certain
how many bears from areas of former habitat may be sus-
tained in areas with remaining habitat. It is not surprising,
therefore, that overall outcomes projected for polar bears
appeared to be driven more by distributional effects than nu-
merical effects. This is largely due to the parameterization
in the model. Some input variables such as hunting or direct
bear/human interactions might be expected to most imme-
diately affect bear numbers rather than distribution. history
has shown, however, that these things can be managed ef-
fectively to maintain sustainable populations when habitats
are adequate. Our model incorporated the manageability of
these human effects in the conditional probability tables.
In contrast, polar bears cannot be maintained where their
habitats are absent, and GCM projections suggest existing
habitat areas will be progressively declining. Regardless of
whether some concentration of numbers is possible in areas
with remaining habitat, and there is great uncertainty regard-
ing the relevance of this to the future, polar bears are not
likely to survive in any numbers in areas where their current
ice habitats no longer exist.
The most probable outcomes for factor A (habitat Threats)
of the proposal to list polar bears as a threatened species were
“major restriction” (Table 3). Numerical responses of polar
bears to future circumstances were forecast to be more mod-
est than changes in distribution. In all regions, reduced den-
sity was the most probable outcome for numerical response.
One way to interpret that outcome may be that where habitat
remains, polar bears will remain even if in reduced numbers.
This is consistent with our BN model results suggesting that
polar bears may persist in the AE through the end of the
21st century. Declines in distribution and number are likely
to be faster and more profound in the PBDE and the SIE
than elsewhere. Sea ice availability in both the PBDE and
SIE already is declining rapidly in these ecoregions [ Meier
et al. , 2007; Stirling and Parkinson , 2006]. The loss of sea
ice habitats in the PBDE is projected to continue, and pos-
sibly to accelerate [ Holland et al. , 2006; Durner et al. , 2008;
Stroeve et al. , 2007].
Plate 7 illustrates how distribution changes driven by
changes in the sea ice appeared to be the major factor lead-
ing to our dire predictions of the future for polar bears. For
projection purposes, we binned the number of additional
months during which the sea ice was projected to be ab-
sent from the continental shelf (node C) into four categories

which included the range from 1 month less than current
(−1) to ≥3 months longer than current. Similarly, we binned
the maximum distance the ice edge could move away from
the shelf (node N) into four categories including the range
from 200 km less than current (−200) to ≥800 km additional
distance. It is clear from node D (see Appendix B) that we
parameterized the model such that more distant ice retreat
and longer ice absence meant reduced availability of critical
foraging habitats as documented by Durner et al. [2008].
Such reduced availability has been shown to have negative
impacts on polar bears [ Regehr et al. , 2007a]. In the PBDE
as an example, the general circulation models that we used
to project future ice conditions, indicated values for nodes
C and N will range from 1.8 to 2.2 additional months of ice
absence and 234 km to 1359 km additional ice distance by
mid century. Similarly, foraging habitat quantity is projected
to decline between 16 and 32% by mid century. As Plate 7
illustrates even the smaller of these values for temporal and
spatial retreat of sea ice place factor A node F2 (see Appen-
dix B) into the category of major habitat restriction. That,
in turn, pushes the distribution response toward extirpated
which pushes the most probable overall population outcome
into the “extinct” category. The outcome percentages in this
example differ from the overall outcomes presented in Ta-
ble 2 because results shown in this example occurred with-
out changing any other inputs included in the full model.
hence, this result provides an example of how the projected
changes in sea ice alone influenced the dire projections of
our BN model. Outcomes in the PBDE are even more dire
when the GCMs that lose the most ice are used or when we
look farther into the future. In contrast, as Table 2 and Plate
4 illustrate, outcomes are less alarming in the PBCE and AE
because of the more modest changes projected for sea ice in
those regions. Sensitivity analyses described below confirm
this role of sea ice in driving the expected future for polar
bears.

4.3. Sensitivity Analyses

Sensitivity analyses offer an opportunity to interpret
model outcomes at every level. The overall population out-
come was most sensitive to change in habitat quantity (node
B) and temporal habitat availability (node C). The other
major habitat variable, change in distance between ice and
the continental shelf (node N) was the 6th most influential
factor on the overall population outcome, despite its being
relevant only to the polar basin ecoregions. Our BN model
recognized that sea ice characteristics, and how polar bears
respond to them, differed among the four ecoregions. In
the SIE, for example, all members of the subpopulation are
forced ashore when the ice melts entirely in summer. In the