Geographically weighted correspondence matrices for local error reporting and change analyses: mapping the spatial distribution of errors and change

This letter describes and applies generic methods for generating local measures from the correspondence table. These were developed by integrating the functionality of two existing R packages: gwxtab and diffeR. They demonstrate how spatially explicit accuracy and error measures can be generated from local geographically weighted correspondence matrices, for example to compare classified and reference data (predicted and observed) for error analyses, and classes at times t1 and t2 for change analyses. The approaches in this letter extend earlier work that considered the measures derived from correspondence matrices in the context of generalized linear models and probability. Here the methods compute local, geographically weighted correspondence matrices, from which local statistics are directly calculated. In this case a selection of the overall and categorical difference measures proposed by Pontius and Milones (2011) and Pontius and Santacruz (2014), as well as spatially distributed estimates of kappa coefficients, User and Producer accuracies. The discussion reflects on the use of the correspondence matrix in remote sensing research, the philosophical underpinnings of local rather than global approaches for modelling landscape processes and the potential for policy and scientific benefits that local approaches support.