Sampling Methodology, Sample Placement, and Other Important Factors in Assessing the Accuracy of Remotely Sensed Forest Maps

摘要

This paper examined two important factors regarding accuracy assessment that have not been well studied: (1) the effect of sampling methodology on estimates of accuracy, and (2) the exact placement of the samples. Several trials of simple random sampling, stratified random sampling, and systematic sampling were performed using totally random sample placement as well as samples chosen from homogeneous areas only. Error matrices were generated for each sampling method, with overall accuracy, normalised accuracy, and the KHAT statistic computed for each matrix. KHAT statistics were then tested for significant differences to statistically determine the effects of sampling methodology and placement. The results suggested that Kappa may be as efficient with systematic sampling and stratified random sampling as it is with simple random sampling, but also indicated that sample placement has more of an effect on estimates of accuracy than sampling methodology.