Spatial Assessment of Categorical Maps: A Proposed Framework

Meaningful accuracy assessment is a sine qua non condition for the use of increasingly precise and available thematic maps. Despite several caveats raised in the literature, the kappa coefficient is still extensively used as an accuracy assessment tool. However, it fails to take into account the spatial nature of remote sensing data, penalizes agreement on area estimation and relies upon hypotheses that are explicitly violated by classification methods such as Geographic Object-Based Image Analysis. Based on the fact that area estimation accuracy measurement is trivial, an innovative and comprehensive framework focusing on the spatial accuracy of categorical maps is built and discussed in this paper. The framework considers not only overlaps between the reference and the classification but also spatial vicinity and does not rely on hypotheses that are in contradiction with the assessment of spatial data. Two class-specific indicators are then designed and the ability of generating probability maps explained. The theoretical developments are followed by simulations illustrating specific cases.

DE ARAUJO MARINHO FILHO Eduardo;  FASBENDER Dominique;  DE KOK Roeland; 

2013-03-18

National Institute for Space Research (INPE)

JRC72481

978-85-17-00059-1,   

http://mtc-m18.sid.inpe.br/col/sid.inpe.br/mtc-m18/2012/05.14.14.40/doc/168.pdf,    https://publications.jrc.ec.europa.eu/repository/handle/JRC72481,