PARENT SESSION
Oral Session #48: Vegetation Change and Response.
Presiding: D. Zobel
Tuesday, August 6. 1:00 PM to 4:45 PM. Mesquite Room, Radisson.
Comparing vegetation change maps: another alternative to the kappa statistic.
Drapek, Raymond*,1, Lenihan, James1, Neilson, Ronald1, Bachelet, Dominique2, 1 USDA Forest Service, Corvallis, OR2 Department of Bioengineering, Corvallis, OR
ABSTRACT- A problem that arises with modeling vegetation change over large spatial scales is coming up with quantitative methods for comparing maps. Common methods include tables of the fractions of map-cells changing, or the kappa statistic. Both of these treat vegetation differences as a simple hit or miss and do not take into consideration varying degrees of similarity between vegetation classes. In 1999, Sykes, Prentice, and Laarif proposed an attribute-based measure of dissimilarity to be used in comparing vegetation maps. We had problems implementing the method of Sykes et al. for a couple of reasons. 1) We found the attribute parameters difficult to apply to the broadly defined vegetation classes we wished to compare. 2) The dissimilarity values resulting from our best parameterization of the Sykes et al. method were at times counter-intuitive. For this reason, we came up with an alternative approach to produce dissimilarity values that we found more simple to implement and more intuitive. In our method, the dissimilarity value is based on a tree diagram we constructed to represent our subjective conceptualization of the relationships between the various vegetation classes. Though our method introduces a degree of subjectivity to the analysis, the simplicity of the tree diagram makes the logic of our dissimilarity values easy to understand and therefore makes it easier for anyone to evaluate our map comparisons.
KEY WORDS: kappa statistic, dissimilarity values, map comparison, vegetation map