INTERNAL: (Review) Spectral Unmixing Results
Anonym
The "Spectral Unmixing Results" section of Chap 9: Spectral Tools is not well worded and can cause some confusion.
"Higher abundances (and higher errors for the RMS error image) are represented by brighter pixels..."
This statement suggests that higher abundances and higher RMS error values are directly correlated. Unfortunately, this is incorrect. Higher abundances are not associated with higher RMS error.
The series of gray-scale images are often refered to as abundance images or fraction images. In these images, floating point values indicate the fraction of the pixel that is filled by each particular endmember material. So, a value of 1.0 would indicate that the entire pixel is filled by that material, and a value of 0 would indicate that there is none of that material in the pixel.
RMS error images use the results of the abundance images to determine the overall error of all of the endmember abundance values for each pixel. RMS error images should appear as noise. If you see areas with distinct recognizable features in the RMS error image, that indicates that there are probably problems with the endmembers, or perhaps some are missing from your analysis.
The RMS error is calculated by:
RMS = sqrt(SUM[(data - lib_model)^2] / # of bands)
where lib_model is library scaled to the same range as the data.