A spatially stratified, multi-stage cluster sampling design for assessing accuracy of the Alaska (USA) National Land Cover Database (NLCD)

Abstract

Assessing the accuracy of a land-cover map is typically expensive, and at the planning stage it is often uncertain what final sample size will be affordable. The aim of this study is to develop an accuracy assessment sampling design that accommodates an ‘in progress’ change in target sample size without sacrificing other desirable design criteria. The sampling design constructed to assess the accuracy of the National Land Cover Database (NLCD) for Alaska achieves these desirable criteria. Spatial stratification provides the flexibility to accommodate a change in sample size and cluster sampling contributes to the cost-effectiveness of the design. We describe the advantages of these design features when the difficulty of accessing remote, large areas is a primary driver of the choice of a sampling design for accuracy assessment. Estimators for overall, user's, and producer's accuracies along with approximate standard errors are provided for the stratified, multi-stage cluster sampling design proposed.

Acknowledgements

We thank Collin Homer and Alexa McKerrow for reviewing an earlier version of the manuscript, and the three anonymous reviewers for their helpful comments. This manuscript has been subject to review and approved by the United States Geological Survey (USGS). The views and opinions expressed in this article do not necessarily represent USGS policy.