Weighted misclassification rate: a new measure of classification error designed for landscape pattern index

Thematic maps classified from remotely sensed data are usually used to derive landscape pattern index (LPI). As the classification error can propagate into the LPI calculation, correctly estimating the LPI error caused by the classification error is necessary in order to obtain a reliable landscape analysis. However, the widely accepted classification accuracy indices are not suitable for quantitatively describing the LPI error because the spatial distribution information of the classification errors is not considered. In this article, we developed a weighted misclassification rate (WMR) by assigning a certain weight to each misclassified pixel according to the spatial distribution information of the pixel. The results reveal that WMR is highly correlated with the error of number of patches (NP), total edge (TE), contagion (CONTAG), double-logged fractal dimension (DLFD) and aggregation index (AI). Moreover, the results suggest that the proposed WMR, as a supplement to the overall accuracy (OA) and the kappa coefficient (κ), is potentially a practicable measure of classification error for landscape analysis.