Categorical Coefficients of Agreement for Assessing Soft-Classified Maps at Multiple Resolutions

An important issue regarding map comparison involves examining agreement of pixels between two categorical maps. Information in the maps can become less precise spatially as resolutions of the maps change from fine to coarse. This paper consists of two major components addressing this issue. First, cross-tabulation matrices are produced for multiple resolutions using hard classification and three different soft pixel classification operators: Multiplication, Minimum, and Composite. Second, the crosstabulation matrices are analyzed through various statistical measures to produce the following categorical coefficients of agreement: user’s accuracy, producer’s accuracy, conditional kappa by row, and conditional kappa by column. These statistical measures are graphed to demonstrate their behavior over multiple resolutions. Land-cover maps of the same subject area for two different years are compared to illustrate the analysis. The area examined is a part of Worcester County, Massachusetts which has experienced about 10% change in land cover between 1971 and 1999. The results from the analysis show that over multiple resolutions, the Hard operator behaves chaotically, the Multiplication operator decreases agreement, the Minimum operator is difficult to interpret, while the Composite operator offers increasing agreement, is interpretable, and is recommended for a multiple resolution analysis.