Introducing residual errors in accuracy assessment for remotely sensed change detection

Accuracy assessment for map comparison is commonly found in urban planning research, especially for detecting error in remotely sensed imagery data. It is to compare two sources of spatial information. In analyzing such information quantitatively, the two datasets are summarized in a confusion matrix, which is represented in a form of percentage of predicted value against its actual data (ground truth). The common acceptable percentage is eighty percent and above. In this paper, we present a new way of accuracy assessment by introducing an additional value called residual error (or predicted error). The residual error is the percentage of error exists when two sources of major errors called mis-classification and mis-location are integrated. Such residual error is incorporated into the assessment so that the results are more accurate and comprehensive. As a case study, we calculate the residual errors of five independent image classifications from six different datasets. Therefore, the accuracy assessment is performed with more details that include not only the confusion matrix, but also the residual errors. In this way, the results of the change detection process can help in doing further analysis for urban growth and land development, particularly for town area.