Image Change Detection (ICD) methods are widely adopted to update large area land use/cover products. Uncertainty problems, however, are well known in such techniques, and a transparent assessment is necessary. In this study, a framework was proposed for evaluating binary land change utilizing remote sensing images. First, two widely adopted ICD methods were used to establish change maps. Second, binary decisions on Change (C) and Non-Change (NC) classes were reached through thresholding on change maps. Then, results were evaluated using two sampling designs: random sampling and stratified sampling. Analysis of results suggests that (1) for random sampling, with an increasing threshold on change variables, the overall accuracy increases and shows a large variance, which is highly correlated with the C omission error; and (2) comparatively, for stratified sampling, in which two strata (i.e., C and NC) were set, the overall accuracy shows a smaller variance and is highly associated with the NC commission error. The significant trends in accuracy assessments indicate the trade-offs between the C and NC classification errors in a binary decision and can present superficial or perfunctory accuracy evaluation in certain circumstances that the causes of error sources and uncertainty problems in ICD are not fully understood.
This is an open access article distributed under the Creative Commons Attribution License
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited