Robust Haze Reduction: an Integral Processing Component in Satellite-based Land Cover Mapping

Spatially varying haze is a common feature of archival Landsat scenes currently being used for large-area land cover mapping and can significantly affect product quality. Robust haze reduction that is image-based and involves minimal operator intervention, is therefore a necessary a-priori step to information extraction. The practical implementation of a suitable methodology, based on a Haze Optimized Transform (HOT), is described. The approach is being used in a program to map the Great Lakes watershed with archival Landsat Multi-Spectral Scanner (MSS) imagery. The impact of haze reduction is assessed using inter-scene classification consistency as a ‘surrogate’ measure of user classification accuracy. Consistency comparisons made between sets of raw and hazereduced scenes indicate that ‘rare’ class identification is most improved by this procedure.