Uncertainty characterization in remotely sensed land cover information

Abstract

Accuracy assessment has become increasingly recognized as an integral component in thematic classification of remotely sensed imagery, for which descriptors such as percentage of correctly classified pixels (PCC) and Kappa coefficients of agreement have been devised for statistical inference about significance of classification accuracy. However, such spatially averaged measures about accuracy neither offer hints about spatial variation in misclassification, nor are they useful for quantifying error margins in land cover derivative products, such as land cover change. Such limitations originate from the deficiency that spatial dependency is not properly accommodated in the conventional methods for classification accuracy assessment and error analysis. Geostatistics provides a good framework for uncertainty characterization in land cover mapping and change detection. This paper seeks to extend and consolidate geostatistical approaches to accuracy assessment and error modeling in land cover and land cover change. Methods for creating spatially explicit maps of misclassification and mis-detection of change will be developed on the basis of classified samples and, possibly, covariates, such as spectrally derived class memberships. It is anticipated that systematic research into uncertainty characterization will contribute to long-term development of large-area land cover and land cover change data, as important components in the comprehensive array of biophysical, environmental, climate, and socio-economic databases.