Comparing estimators of gross change derived from complete coverage mapping versus statistical sampling of remotely sensed data

Abstract

Area of gross change in land cover can be derived from a complete coverage land cover change map of a region of interest or estimated from a statistical sample of the region. Sampling may produce significant cost savings and more timely results because change is determined over a smaller total area than required by complete coverage mapping. Mean square error (MSE) defined in the context of a survey sampling measurement model is used to compare gross change estimators obtained from the two approaches. Measurement error bias attributable to error in classifying land cover change may occur with either the sampling or complete coverage mapping approach. An additional contribution to MSE attributable to sampling variability exists for the sampling-based estimator, but not the complete coverage estimator. If this sampling variability is small, the classification error bias of the sampling approach need not be reduced very far relative to the classification error bias of complete coverage to achieve similar MSE. Data from several published change accuracy error matrices are used to provide MSE comparisons for specific applications.