UCSB-BRC Project Related
Publications Abstracts
Validating
Large-Area Land Cover Databases with Maplets
David M. Stoms
Point sampling
methods traditionally used for assessing accuracy of small-area thematic
maps are less suitable for large-area, lower resolution maps. Maplets,
or detailed maps of small areas, are proposed here as an alternative
source of independent reference data for accuracy assessment. As a
land cover census instead of a sparse sample of points, maplets can
provide information on the composition, heterogeneity, and accuracy
of individual map units, on boundary locational accuracy, and the
spatial pattern of errors in the large-area map. The techniques is
illustrated by a case study for a wildlife habitat map of southern
California, USA. Overall agreement between the maplet and the large-are
map was 63%. However, the total area of map units in which the label
agreed with the majority maplet class was 84%. This result suggests
that much of the "error" was actually generalization rather than misclassification.
|
