A New Measure of Classification Error: Designed for Landscape Pattern Index

Classified thematic maps based on remote sensing data are usually used to derive Landscape Pattern Index (LPI). However, the classification error can be propagated into the LPIs calculation, while it is usually ignored in the previous literatures. Correctly estimating the LPI error is vital for reliable landscape analysis; however, the widely accepted accuracy assessment method without considering spatial information is not suitable for indicating LPI error. In this paper, we developed a new accuracy assessment index by giving a certain weight to each error pixel according to its isolated degree. The result shows that the new index is a good predictor for the error of Number of Patches (NP), Total Edge (TE) and Aggregation Index (AI). The effect of sample size was also studied, showing that the correlation between new index and LPI error become higher and more stable when the sample size increases. The results suggest that the proposed new index is potential to be a practicable measure of classification error for landscape analysis as supplement of overall accuracy and Kappa coefficient. * Corresponding author. This is useful to know for communication with the appropriate person in cases with more than one author.