modeling
PEM
wetlands, while Figure 2 shows the conditional
importance for
PFO
/
PSS
wetlands. Conditional variable impor-
tance varied between the ecological subregions and also when
comparing the
PEM
and
PFO
/
PSS
models, although some com-
mon variables were consistently shown to be of importance
in all models, especially the variable calculated employing
the distance from waterbodies weighted by slope using the
Cost Distance tool. This variable was the most important for
prediction within each ecological subregion for
PEM
wetlands,
except for the Western Coal Fields, and was also generally
found important for predicting
PFO
/
PSS
wetlands. Roughness,
which to our knowledge has not been previously investigated
for mapping wetlands, was found to be of value. Also, dissec-
tion, which is similar to the previously explored topographic
position index, was found to be especially valuable. Surpris-
ingly, variables that have previously been used for wetland
mapping tasks, surface curvature (Knight
et al
., 2013) and
CTMI
(Knight
et al
., 2013; Rampi
et al
., 2014), were gener-
ally found to have little importance in the
PEM
and
PFO
/
PSS
models in that the
OOB
error barely increased when they were
excluded from the model.
We attribute the differences in conditional variable im-
portance between ecological subregions to the variability in
topographic signature of wetlands in differing landscapes
Figure 2: Conditional predictor variable importance for modeling PFO/PSS wetlands as estimated using OOB mean decrease in accuracy
for individual models for each ecological subregion (A through E) and a model for the entire state (F).
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2016
443
