Improving land change detection based on uncertain survey maps using fuzzy sets |
| |
Authors: | Stefan Leyk Niklaus E Zimmermann |
| |
Institution: | (1) Department of Geography, University of Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland;(2) Land Use Dynamics, Swiss Federal Research Institute WSL, Zuercherstrasse 111, CH-8903 Birmensdorf, Switzerland |
| |
Abstract: | In this paper we present a method for correcting inherent classification bias in historical survey maps with which subsequent
land cover change analysis can be improved. We linked generalized linear modelling techniques for spatial uncertainty prediction
to fuzzy set based operations. The predicted uncertainty information was used to compute fuzzy memberships of forest and non-forest
classes at each location. These memberships were used to reclassify the original map based on decision rules, which take into
consideration the differences in identification likelihood during the historical mapping. Since the forest area was underestimated
in the original mapping, the process allows to correct this bias by favouring forest, especially where uncertainty was high.
The analyses were performed in a cross-wise manner between two study areas in order to examine whether the bias correction
algorithm would still hold in an independent test area. Our approach resulted in a significant improvement of the original
map as indicated by an increase of the Normalized Mutual Information from 0.26 and 0.36 to 0.38 and 0.45 for the cross-wise
test against reference maps in Pontresina and St. Moritz, respectively. Consequently subsequent land cover change assessments
could be considerably improved by reducing the deviations from a reference change by almost 50 percent. We concluded that
the use of logistic regression techniques for uncertainty modelling based on topographic gradients and fuzzy set operations
are useful tools for predictively reducing uncertainty in maps and land cover change models. The procedure allows to get more
reliable area estimates of crisp classes and it improves the computation of the fuzzy areas of classes. The approach has limitations
when the original map shows high initial accuracy. |
| |
Keywords: | Predictive uncertainty modelling Fuzzy sets Land cover change analysis Classification bias Correction of survey maps Area estimation |
本文献已被 SpringerLink 等数据库收录! |
|