Accuracy assessment curves for satellite-based change detection

The concept of "accuracy assessment curves" as they relate to image-based change detection is developed. These curves are a graphical representation of the relationship between typical accuracy assessment figures and a binary change detection threshold level. After an introduction to accuracy assessment curves, an example is presented. The relationship between typical accuracy assessment parameters and the location of the binary change threshold level is first shown with a series of error matrices. This relationship is then expanded across a range of thresholds and shown as accuracy assessment curves. These curves show how accuracy figures are dependent on the change metric's threshold level. This dependency indicates some limitations of binary change maps. These limitations provide the impetus for a continuous change product. We describe how a continuous Probability-of-change (poc) image, derived from statistical models, can be incorporated with the accuracy assessment curves to produce a more meaningful and informative change detection product.