Regional stability of an ERS/JERS-1 classifier

The potential of combined ERS/JERS-1 SAR images for land cover classification was demonstrated for the Raco test site (Michigan) in recent papers and articles. The authors´ goal is to develop a classification algorithm which is stable in terms of applicability in different geographical regions. Unlike optical remote sensing techniques, radar remote sensing can provide calibrated data where the image signal is solely determined by the physical (structural) and electrical properties of the targets on the Earth´s surface and near subsurface. Hence, a classifier based on radar signatures of object classes should be applicable on new calibrated images without the need to train the classifier again. This article discusses the design and applicability of a classification algorithm, which is based on calibrated radar signatures measured from ERS-1 (C-band, vv polarized) and JERS-1 (L-band, hh polarized) SAR image data. The applicability is compared in two different test sites, Raco, Michigan and the Cedar Creek LTER site, Minnesota. It was found, that classes separate very well, when certain boundary conditions like comparable seasonality or soil moisture conditions are observed