New generalized statistical methods to measure agreement between two maps at multiple-resolutions, where each cell in each map has a multinomial distribution among any number of categories, are presented. This methodology quantifies agreement between any two categorical maps, where either map uses fuzzy or crisp classification. The method measures the agreement at various resolutions by aggregating neighboring cells into an increasingly coarse grid. At each resolution, the method partitions the overall agreement into correct due to chance, correct due to quantity, correct due to location, error due to location, and error due to quantity. In addition, the method computes six statistics that are useful to interpret the differences between maps, and shows how these statistics change with resolution. This technique is particularly useful for characterizing land-cover change and for validating land-cover change models. For illustration, this paper applies these theoretical concepts to the validation of a land-use change model for Costa Rica.
Photogrammetric Engineering and Remote Sensing
land cover maps, cartography, classification, deforestation, errors, forests, models, methodology, statistics, remote sensing, land use, vegetation, Costa Rica, Central America
Pontius, R. Gil, "Statistical methods to partition effects of quantity and location during comparison of categorical maps at multiple resolutions" (2002). Geography. 790.