Geographically weighted methods for examining the spatial variation in land cover accuracy

Geographically weighted methods for examining the spatial variation in land cover accuracy
Alexis Comber1, Peter Fisher1, Chris Brunsdon2, Abdulhakim Khmag1

1. Department of Geography, University of Leicester, Leicester, LE1 7RH, UK (,,
2. Department of Geography, University of Liverpool, Liverpool, L69 3BK, UK (

Abstract: The confusion matrix is used to describe land cover accuracy. It describes correspondence between alternative sources of land cover information and is a standard technique in remote sensing. BUT the confusion matrix is aspatial – it provides no information about the spatial distribution of accuracy. And, despite much work suggesting methods for describing the spatial variation of accuracy (Foody, 2002; 2005), these have not been adopted by the remote sensing community. This paper demonstrates how geographically weighted approaches can be used to analyse the spatial relationships between land cover data classified from remotely sensed data and data collected in the field, for both Boolean and Fuzzy classifications. These approaches each use a moving window or kernel to compute local accuracy measures, whose size is specified dynamically, and are ‘geographically weighted’ because the kernel allows for the fact that the more distant observations may be of less local relevance and their influence is weighted accordingly. Fuzzy and Boolean maps of the spatial distribution of accuracy were generated. This research demonstrates that data collected as part of a standard remote sensing validation exercise can be used to derive measures of accuracy that vary spatially and suggests that there is potential to move land cover validation from the aspatial to spatially explicit reporting of accuracy.

Keywords: confusion matrix, geographically weighted regression, spatial variation of accuracy, fuzzy difference

ComberAccuracy2012.pdf97.97 KB