Species mapping

Since health estimates developed from hyperspectral data have been species
specific in our particular projects, we must first determine the species composition
of each image pixel.

Two approaches are taken to develop species maps - depending on the spatial resolution.
Both methods rely on the careful development of an endmember library in which the endmember
spectra represent the species of interest.  In our work, we develop these spectra from the
imagery itself, by careful geolocation of each tree canopy and extraction of those spectra 
representing healthy examples of each species.

Spectral Unmixing
For coarse resolution imagery (10m+), each pixel may contain a number of tree canopies, and 
the species may vary within the pixel. In this situation, we use mixture tuned matched filtering
to determine the abundance of each species within each pixel.

Just a note to us re the way mtmf works:
	Mapping the pixel-by-pixel relative abundance of vegetation
	was carried out
	using mixture-tuned matched-Wltering (MTMF) (Kruse
	et al. 2003). MTMF expands on a proven signal processing
	methodology (matched-Wltering; MF) that maximizes the
	response to known endmembers while suppressing the
	response of the composite background. MTMF adds to MF
	an infeasibility score (IS) that measures the likelihood that
	a specific pixel is a feasible mixture of the target endmember
	and the background distribution defned by the full
	scene covariance. The MF output from MTMF ranges from
	0.0 to 1.0, with 1.0 representing a perfect match and values
	<1.0 representing fractional abundances of identifed endmembers.

Insert fig showing tree canopies and large pixels

Spectral Angle Mapper
For fine-scale spatial resolution imagery (1-2m), each individual pixel is likely to cover part of a
single canopy.   In this instance, the similarity between pixel spectra and those in the endmember
library is used to assign a single species class to each pixel.
   
Insert fig showing tree canopies and samll pixels