Geoscience Reference
In-Depth Information
can benefit from the experience derived from other fields of science (e.g.,
RS
).
Methods for summarizing the accuracy of raster and vector maps using point
samples and error matrices are now widely used in
GIS
and are beginning to
find their way into ecological applications as well. However, standard tech-
niques have not had universal acceptance for a number of reasons. For exam-
ple, a number of alternative sampling designs have been proposed for analyz-
ing the accuracy of imagery. The choice of sampling design is often subject to
the particular problems associated with the area to be ground-truthed. How-
ever, a number of general trends are obvious from the
GIS
and
RS
literature.
Random and stratified random sampling are acknowledged to maximize pre-
cision and accuracy (though at a higher cost than cluster sampling or system-
atic sampling). It should be noted that in highly heterogeneous landscapes
(e.g., native forests, especially tropical forests), stratification is often too costly
to consider. Cluster sampling offers reduced sampling costs, but in order to be
effective it depends on low intracluster variance. Systematic sampling schemes
may lead to a bias in parameter estimation if periodic errors align with the
sampling frame (e.g., as a result of image banding or linear topographic fea-
tures, as in the Allegheny Mountains of Pennsylvania).
GIS data layers contain numerous errors. These pose a number of problems
as errors accumulate during the process of analysis and model building.
Although modeling the accumulation of error during
GIS
overlay analysis is still
in its infancy, some methods for measuring error accumulation during
GIS
analysis have been discussed.
Any procedure to reduce mapping error in individual layers in a
GIS
will
improve the mapping accuracy of an overlay generated from the
GIS
. Until bet-
ter error-modeling techniques are developed for
GIS
s, descriptive statistics
should ideally be calculated for each layer in a
GIS
, as well as for each layer pro-
duced by
GIS
modeling. The descriptive statistics should include overall map-
ping accuracies as well as class mapping accuracies. An alternative way of defin-
ing the performance of a
GIS
model, thus assigning a level of reliability to its
results, is sensitivity analysis that identifies crucial parameters. These parame-
ters are those that, within their range of variation, determine the highest vari-
ation in the model output.
Conclusions
j
A common opinion among epistemologists is that we are facing a break
between the development of advanced technologies and our needs and abilities
