Diagnosis of the accuracy of land cover classification using bootstrap resampling

Since the 1970s, there has been a long history of assessing the accuracy of remotely sensed land cover types. Various accuracy assessment methods have been proposed, including the method of bootstrap resampling. However, the potential of the bootstrap resampling method to diagnose the accuracy of remote sensing monitoring of land cover has not been fully explored. Here, we used the land cover map classified in the Gannan Tibetan Autonomous Region as an example and investigated the impact of two parameters in the bootstrap resampling method on the assessment of land cover classification accuracy: the number of validation points and the number of bootstrap resampling times. Varying these two parameters can help diagnose the accuracy of remote sensing classifications of land cover. The main conclusions are as follows. First, the number of validation points required to achieve a stabilized accuracy estimate varies among land cover types. It indicates not only the level of homogeneity of the land cover type but also the level of difficulty associated with classifying it. Second, it is not necessary to have a very large number of bootstrap resampling times to estimate the classification accuracy - 50 was sufficient. In summary, we recommend that various numbers of validation points be employed in bootstrap to objectively and comprehensively assess the overall, producer's, and user's accuracies of remote sensing classifications of land cover.