Since the launch of MERIS on board of ENVISAT long term activities using vicarious calibration approaches have been set in place to monitor potential drifts in the calibration of the radiance products of MERIS. In this paper, a stable and well monitored reference calibration site named Railroad Valley Playa (Nevada, USA) is used to derive the calibration uncertainties of the MERIS FR TOA radiance over time. Subsequently, a linear interpolation of these uncertainties is performed for a set of images covering the whole of the Netherlands (which is used as a validation site). After this, the images over the Netherlands were corrected on the basis of the previously interpolated uncertainties and classified in 9 land use classes using linear spectral unmixing and matched filtering techniques.
The classification endmembers were derived from an image-based land use map of the Netherlands (LGN4) after determining the most homogeneous areas for each land use type by means of a standard purity index and a moving window filter to minimize possible adjacency effects.
Finally, the impact of the calibration accuracy over the land use classification is assessed by comparing classification results both for corrected and uncorrected images. We conclude that the classification performance may significantly be increased, when taking into account long-term vicarious calibration results.