386
experts (0.73), followed by system stochasticity (0.12), and then the imperfect
knowledge component (0.01).
These results suggest that when a single or small group of similar experts pro-
vide the input to models, the results could deviate greatly from the true outcomes
of vegetation management actions if the ecosystem dynamics are not well under-
stood. However, even though the model results showed high variance, it is impor-
tant to note that they are useful because they represent the current understanding of
ecosystem dynamics and illustrate hypothetical depictions of landscape change.
The large among-expert variance also makes selecting among management actions
challenging because there is low confidence in model results. This variance could
be reduced (i.e., masked) by subsampling similar individuals from available experts
or aggregating opinions through consensus, both of which are very commonly
done in rangeland management steering committees, at the risk of biasing results
and to the detriment of management decisions. If experts are forced to form con-
sensus, this uncertainty cannot be identified and management actions may be sus-
ceptible to unexpected results. Ascertaining which parameters are most variable
among experts and collecting monitoring data to determine true rates and effects of
transitions is therefore needed, and may be necessary for STSM applications to
Bromus invaders.
There is no evidence provided in the example for the minimum number of
experts needed to ensure adequate confidence in model predictions. Minimum
expert sample sizes as high as 50 have been recommended, though typically less
Fig. 13.4 The average percent of cells in the desired vegetation state (low-density mature) over
time, for all models combined and for each expert. Error bars are stacked and represent the stan-
dard deviations for the variance due to imperfect knowledge (90 % confidence; inner bars) and
system stochasticity (from end of inner bars to outer bars). Results adapted from Czembor et al.
( 2011 )
L. Provencher et al.