Sampling method and sample placement: How do they affect the accuracy of remotely sensed maps?

The accuracy of remotely sensed forest stand maps is traditionally assessed by comparing a sample of the map data with actual ground conditions. Samples most often comprise clusters of pixels within homogeneous areas; thereby avoiding problems associated with accurately mapping edges (e.g., transition areas between two forest types). Consequently, they may well overestimate accuracy, but the degree of overestimation is unknown. This paper examines two important factors regarding accuracy assessment that are not well studied: the effect on estimates Of accuracy of (1) the sampling method and (2) the exact placement of the samples. Overall accuracy, normalized accuracy, and the KHAT statistic are computed from error matrices generated from simple random sampling, stratified random sampling, and systematic sampling using totally random sample placement and samples chosen from homogeneous areas only. The results indicate that Kappa appears to be as appropriate to use with systematic sampling and stratified random sampling as it is with simple random sampling, but suggests that sample placement may have more of an effect on estimates of accuracy than sampling method alone.