Accuracy Assessment with Complex Sampling Designs
Raymond L. Czaplewski
United States Forest Service Rocky Mountain Research Station Fort Collins, Colorado USA
Abstract: A reliable accuracy assessment of remotely sensed geospatial data requires a sufficiently large probability sample of expensive reference data. Complex sampling designs reduce cost or increase precision, especially with regional, continental and global projects. The General Restriction (GR) Estimator and the Recursive Restriction (RR) Estimator separate a complex sample survey into simple statistical components, each of which is sequentially combined into the final estimate. GR and RR produce a design-consistent Empirical Best Linear Unbiased Estimator (EBLUE) for any sample survey design, regardless of its complexity.
Keywords: Kalman filter; error matrix; GIS; geospatial database; MODIS; Landsat; LiDAR; photo-interpretation