Calibrating Area Estimate Bias on Categorical Maps Using the Contingency Table

This paper presents and compares approaches of estimating true area on the ground and calibrating quantitative errors of area estimate on categorical maps from the contingency table. Results directly estimated from the contingency table and those from two calibration methods were compared on two maps of 10 different land cover classes with known errors between them. The estimated true area percentage from the contingency table and two calibration approaches showed obvious improvement when compared with uncalibrated values. However, there is no significant difference among the estimates from the contingency table and the two calibration methods. Although the inverse method led to mean estimates closer to the true values for all classes than other methods, comparing the individual area estimates for each class showed that the inverse method did not always produce the most accurate estimate. Homogeneous classes with high classification accuracy have a better chance of achieving more accurate estimates from calibration than heterogeneous classes. Compared with large classes, classes covering a small percentage of a map are more vulnerable to the quantitative error and more sensitive to sampling error.