Determining spatial location of sub pixels in hyperspectral data for mangrove species identification

Hyperspectral data finds wide applicability in species level mapping of mangrove forest cover in pure and mixed stands. The high spectral resolution of hyperspectral data enhances discriminatory power of target objects (mangrove species) on one hand whereas on the other hand its low spatial resolution leads to challenging problems of mixed pixels. Though Linear Spectral Unmixing (LSU) has been successful in mangrove discrimination at sub pixel level, it only provides information about the fractional abundance of endmembers within each mixed pixel. The spatial location of the sub-pixels remains unknown, as the unmixing technique does not perform any enhancement of spatial resolution. Hence, even though the abundance of target sub pixels is known their distribution within the area is unknown. In this paper, we have attempted a technique for correctly locating, from a spatial point of view the fractional abundances of the sub pixels i.e. mixture of mangrove species within the mixed pixel. In the present study, the Hyperion sensor captures images at a spatial resolution of 30m within which there is high probability of presence of mixed mangrove species besides homogeneous patches. It is therefore necessary not only to identify the diverse mangrove species within this spatial resolution but also their geographic locations which will greatly help in obtaining classified maps at finer spatial resolution. This takes into account the information of pure end members extracted by automated target generation algorithm (NFINDR in this case) and applies them to determine the fractional abundance of individual sub pixels in pure and mixed pixels. The sub pixel locations are then determined using Simulated Annealing Algorithm. The study has been successful in finding out the spatial distribution of mixed mangrove species within a pixel in the Henry Island of Sunderban. The accuracy has been validated from field visits made in the study area.