|
Description:
|
Predictive vegetation mapping was employed to predict the distribution of vegetation
communities and physiognomies in the portion of the Scandinavian mountains in
Sweden . This was done to address three main research questions : (1 ) what
environmental variables are important in structuring vegetation patterns in the study
area ? (2 ) how well does a classification tree predict the composition of mountain
vegetation in the study area using the chosen environmental variables for the study ? and
(3 ) are vegetation patterns better predicted at higher levels of physiognomic
aggregation ? Using GIS , a spatial dataset was first developed consisting of sampled
points across the full geographic range of the study area . The sample contained existing
vegetation community data as the dependent variable and various environmental data as
the independent variables thought to control or correlate with vegetation distributions .
The environmental data were either obtained from existing digital datasets or derived
from Digital Elevation Models (DEMs ) . Utilizing classification tree methodology , three
model frameworks were developed in which vegetation was increasingly aggregated into
higher levels of physiognomic organization . The models were then pruned , and
accuracy statistics were obtained . Results indicated that accuracy improved with increasing aggregation of the dependent variable . The three model frameworks were
then applied to the Abisko portion of the study area in northwestern Sweden to produce
predictive maps which were compared to the current vegetation distribution .
Compositional patterns were critically analyzed in order to : (1 ) assess the ability of the
models to correctly classify general vegetation patterns at the three levels of
physiognomic classification , (2 ) address the extent to which three specific ecological
relationships thought to control vegetation distribution in this area were manifested by
the model , and (3 ) speculate as to possible sources of error and factors affecting
accuracy of the models . |