Geoscience Reference
In-Depth Information
two sets of fixed conditions to determine whether manage-
ment of human activities on the ground might be able to alter
sea ice-driven outcomes. In these “influence runs” we set
the states for all nodes over which humans might be able to
exert control (e.g., harvest, contaminants, oil, and gas devel-
opment) first to “same as now,” and then to “improved con-
ditions.” After doing so, projected probabilities of extinction
were lower at every time step (Plate 6). At and beyond mid
century, extinction was still the most probable outcome in
the PBDE and SIE. however, extinction did not become the
most probable outcome in the PBDE and SIE until mid cen-
tury. And in the SIE, with model runs based on GCMs retain-
ing the maximum sea ice, extinction, as the most probable
outcome, was avoided until year 75. Recall that extinction
was the most probable outcome in these ecoregions at year
25 in the original model runs. In contrast, results of these
influence runs suggested that on the ground management of
human activities could improve the fate of polar bears in the
AE and PBCE through the latter part of the century (Plate 6).
In summary, for our 50-year foreseeable future, it appeared
that management of localized human activities could benefit
polar bears in the PBCE and especially in the AE but was
likely to have little qualitative effect on the future of polar
bears in the PBDE and SIE if sea ice continues to decline as
projected.
To examine how much different, than projected, future
sea ice would need to be to cause a qualitative change in
our overall outcomes, we composed another influence run in
which we set the values for all non-ice inputs to uniform prior
probabilities. That is, we assumed complete uncertainty with
regard to future food availability, oil and gas activity, con-
taminants, disease, etc. Then, we ran the model to determine
how changes in the sea ice states alone, specified by our en-
semble of GCMs, would affect our outcomes. This exercise
illustrated that in order to obtain any qualitative change in
the probability of extinction in any of the ecoregions, sea ice
projections would need to leave more sea ice, at all future
time steps, than even the maximum-ice GCM projection we
used (Plate 6).
ateness of model structure. All of these can influence model
predictions. Uncertainty in our understanding of complex
ecosystems is virtually inevitable. We have, however, dealt
with this as well as possible by incorporating a broad sweep
of available information regarding polar bears and their en-
vironment. how to best represent our understanding of the
system in models, the second source of uncertainty, can be
structured in various ways. here, we captured and repre-
sented expert understanding of polar bear habitats and popu-
lations in a manner that can be reviewed, tested, verified,
calibrated, and amended as appropriate. We have attempted
to open the “black box” so to speak and to fully expose all
formulas and probabilities. We also used sensitivity testing
to understand the dynamics of BN model predictions [
John-
son and Gillingham
, 2004] (Appendix C). After BN models
of this type are modified through peer review or revised by
incorporating the knowledge from more than one expert into
the model parameterization, any variation in resulting mod-
els can represent the divergence (or convergence) of exper-
tise and judgment among multiple specialists.
Also included in the second category of uncertainty are
those associated with statistical estimation of parameters,
including measurement and random errors. The sea ice pa-
rameters we used in our polar bear models were derived
from GCM outputs that possess their own wide margins of
uncertainty [
DeWeaver
, 2007]. hence, the magnitude and
distribution of errors associated with our sea ice parameters
were unknown. To compensate for these unknowns, we ac-
commodated a broad range of sea ice uncertainties by ana-
lyzing the 10-member ensemble GCM mean, as well as the
minimum and maximum GCM ice forecasts. In the case of
polar bear population estimates, many are known so poorly
that the best we have are educated guesses. Pooling sub-
populations where numbers are merely guesses with those
where precise estimates are available, to gain a range-wide
perspective, prevents meaningful calculation and incorpo-
ration of specific error terms. We recognize that difficulty,
but because our projections are expressed in the context of
a comparison to present conditions, we largely avoid the
issue. That is, whatever the population size is now, the future
size is expressed relative to that and all errors are carried
forward.
The third category, uncertainty in predictions of species
abundance and distribution can be subject to errors because
of spatial autocorrelation, dispersal and movement of organ-
isms, and biotic and environmental interactions [
Guisan
et
al.
, 2006]. We addressed these error sources by deriving es-
timates of ice habitat area separately for each ecoregion from
the GCM models because sea ice formation, melt, and advec-
tion occur differently in each ecoregion. The BN population
stressor model accounted explicitly for potential movement
4. DISCUSSION
4.1. Uncertainty
Analyses in this paper contain four main categories of
uncertainty: (1) uncertainty in our understandings of the
biological, ecological, and climatological systems; (2) un-
certainty in the representation of those understandings in
models and statistical descriptions; (3) uncertainty in pre-
dictions of species abundance and distribution, and (4)
uncertainty in model credibility, acceptability, and appropri-
Search WWH ::
Custom Search