A Comparison of Sampling Schemes Used in Generating Error Matrices for Assessing the Accuracy of Maps Generated from Remotely Sensed Data


Sample means and variances, obtained through computer simulation, were compared with the corresponding population means and variances for five sampling schemes typically used in assessing the accuracy of land-cover maps derived from remotely sensed data. The five sampling schemes were simple random sampling, stratified random sampling, cluster sampling, systematic sampling, and stratified systematic unaligned sampling. Three data sets of varying spatial complexity, including an agricultural area, a range area, and a forest area, were investigated. The patterns of error in each data set, as measured by spatial autocorrelation analysis, greatly influenced the appropriate sampling scheme to be used for assessing the map accuracy. The results indicate that simple random sampling always provides adequate estimates of the population parameters, provided the sample size is sufficient. For the less spatially complex agriculture and range areas, systematic sampling and stratified systematic unaligned sampling greatly overestimated the population parameters and, therefore, should be used only with extreme caution. Cluster sampling worked reasonably well. However, clusters should not be taken of size greater than 25 pixels and preferably 10 pixels.


Natural Resources and the Environment

Publication Date


Journal Title

Photogrammetric Engineering and Remote Sensing


American Society for Photogrammetry and Remote Sensing and Remote Sensing

Document Type



©1988 American Society for Photogrammetry and Remote Sensing