Creating Temporal Hyperspectral Regional Endmember Bundles (THREBs): Automatic imaged based EM extraction and library reduction to derive regional fractional vegetation cover

Abstract

With the increasing availability of hyperspectral images (HSI) through missions like PRISMA, DESIS, and EnMAP, as well as the upcoming CHIME mission, the opportunities for large scale analyses are now being realized. Simultaneously, the establishment of operational processing chains is becoming increasingly important. With that in mind, the Earth Observation Centre at DLR has developed an fCover processor which extracts and classifies endmembers from HSI and, following a spectral unmixing step, generates soil and vegetation cover maps. To generate homogeneous fCover maps over large areas it is necessary to overcome challenges like generalized EM libraries, incoherent unmixing, and unintentional use of false EMs that can occur in single scene processing. The methodology presented here alters the established fCover processor by exploring the potential of processing multiple scenes, thereby creating temporal hyperspectral regional endmember bundles (THREBs) and generating coherent region-wide fractional vegetation cover maps in an automated fashion.