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 234

A declining habitat base, corresponding with FWS listing
factor A (habitat threats), is the overriding factor in projec-
tions of declining numbers and distribution of polar bears.
Other factors which correspond with FWS listing factors
B, C, and E, and which could result in additional popula-
tion stress on polar bears, are likely to exacerbate effects of
habitat loss. To qualitatively alter outcomes projected by our
models, it appears future sea ice would have to be far more
extensive than is projected by even the more conservative
of the general circulation models we used. Because recently
observed declines in sea ice extent continue to outpace most
GCM projections [ Stroeve et al. , 2007], more extensive sea
ice in the future seems unlikely unless greenhouse gas emis-
sions are reduced. This study and the others on which it was
based establish that the future security of polar bears over
much of their present range is threatened in an ecological
context. In May 2008, Secretary of Interior Dirk kemp-
thorne, upon review of this and other available information,
decided this ecological threat required policy actions and
declared polar bears a threatened species under the defini-
tion of the Endangered Species Act [ U.S. Fish and Wildlife
Service , 2008].

gist (S. Amstrup) with guidance from an ecologist modeler
(B. Marcot). The general underlying influence diagram for
the BN model is shown in Plate 2, and the full model is in
Plate 3. A BN model consists of a series of variables rep-
resented as “nodes” (boxes in Plate 3) that interact through
links (arrows in Plate 3). Nodes that have no incoming ar-
rows are “input nodes” (the yellow boxes in Plate 3, e.g.,
node T Parasites and Disease). Nodes with both incoming
and outgoing arrows are summary nodes (or latent variables,
e.g., node L2 Vital Rates). In our model, we also specified
four of the summary nodes as listing factors used by the U.S.
Fish and Wildlife Service (S. Morey, personal communica-
tion, 2007). Nodes with incoming arrows but no outgoing
arrows are output nodes (node D1 Overall Population Out-
come). The model (Plate 3) was run by specifying condi-
tions of all input nodes for each combination of ecoregion
(node M), time period (node Q), and climate data source or
GCM run (node R). In running the model, specifying ecore-
gion automatically adjusts two of its summary nodes (D and
U). The input data set (Tables D1a and D1b) also specifies
time period, GCM run, and all other inputs. Nodes Q and R
appear in the model unconnected as visual placeholders for
displaying the basis of each model run.
Each node in this model consists of a short node name
(e.g., node D1), a longer node title (e.g., Overall Population
Outcome), a set of states (e.g., larger, same as now, smaller,
rare, and extinct), and an underlying probability table. The
probability tables consist of unconditional (or prior) prob-
abilities in the input nodes, or conditional probabilities in all
other nodes, the latter representing probabilities of each state
as a function of (conditional upon) the states of all nodes that
directly influence it.
Table A1 presents a complete list of all nodes in the model
with their short code letters, titles, description, possible
states, and the group (Node Set, in Netica® parlance) to
which it belongs (input nodes, output node, summary node,
or summary listing factor node).

APPENDIX A: DOCUMENTATION OF ThE
BAyESIAN NETWORk POlAR BEAR
POPUlATION STRESSOR MODEl

Appendix A documents the structure of the Bayesian
network (BN) population stressor model. We used the BN
modeling shell Netica® (Norsys, Inc.) to create a model that
represents potential influences on distribution response, nu-
merical response, and overall population response of polar
bears under multiple stressors, which include anthropogenic
stressors, natural disturbances, and other key environmental
correlates to polar bear population amount and distribution.
The BN population stressor model was created to repre-
sent the knowledge and judgment of one polar bear biolo-

Table A1. Complete list of All Nodes in the Bayesian Network Population Stressor Model
Node
Name Node Title Node Description States

Input Nodes

T Parasites and

Disease

As the climate warms, regions of the Arctic are hospitable to parasites and disease
agents which formerly did not survive there. Polar bears have always been free of
most disease and parasite agents. Trichinella is one notable exception, but even rabies,
common in the Arctic has had no significance to polar bears. Changes in other species
disease vulnerability suggest that similar changes could occur in polar bears so that
they could move from a position where parasites and disease are not influential on a
population level to where they are influential.

influential
not