Spatially Variable Thematic Accuracy: Beyond the Confusion Matrix

Abstract

An essential aspect of the increasing sophistication of ecological models is the use of spatially explicit inputs and outputs. Thus, the challenge of documenting the uncertainty of model parameters must expand to include the distribution of error across the surface of maps, satellite images, and other ecological data that are keyed by geographic location. It has become more common to report the overall accuracy of map data sets. Support for such accuracy statements is seen in the descriptive attributes that are defined in file format conventions (e.g., the spatial data transfer standard, SDTS; FGDC 1998). These attributes include documentation of the root mean square error for positional accuracy and error rates associated with the delineation of specific map features. The probability of mapping errors, however, is generally not consistent across the surface of a map data set (Congalton 1988a; Steele et al. 1998), and standard methods have not been adopted for presenting the spatial distribution of error in thematic maps. The confusion matrix is the most commonly accepted method for assessing the accuracy of thematic maps, but it is entirely devoid of spatial context. This chapter addresses shortfalls in various approaches to predicting the distribution of error in thematic maps derived from image data.