Calibration-based models for correction of area estimates derived from coarse resolution land-cover data

Calibration-based area correction models provide improved estimates of cover-type proportions measured at coarse scales. Three separate regions (Plumas and Stanislaus National Forests and Lake Tahoe Basin) in the northern Sierra Nevada serve as test sites: the first for model calibration and the others for extrapolation and validation of the area correction methods. The use of one (or a few) large sites to calibrate models for extrapolative application over large areas represents a global test-site sampling strategy that most easily can be employed for the development of global land cover and land-cover change products. The inverse estimator, calibrated on scale-specific interclass transition matrices, produces the best overall results at the coarser scales for both validation sites. Good results probably occur because the conditional probabilities in the transition matrices implicitly carry information about the spatial organization of the landscape. This method performs poorly at finer scales, or in cases where the calibration and validation sites have dissimilar spatial organization, or class confusions. The slope estimator, calibrated on the coefficients of scale-specific proportion transition lines, produces improved estimates over uncorrected values at all scales. The slope estimator appears to generalize most successfully because it characterizes the basic tendency of small classes to diminish and large classes to increase in size as the landscape is represented at increasingly coarser scales. The positive results achieved using the inverse estimator and the slope estimator indicate potential for using such a posteriori calibration methods to improve coarse resolution land-cover area estimates over large regions.