Summary

The unique problem when running SOO RMSE (aka LMM_FAE) on the dataset is its failure to significantly distinguish the difference between npv, and soil.

To emphasize the difference between their spectra, choosing an interval of wavelengths where soils derivative spectra is zero, and scaling each spectra by 100,000, we ran this data frame through LMM_FAE. (Derivative, scaled, and interval of wavelength data frame).

The chosen wavelength interval which revealed the best results for when soil’s derivative is zero was 1500-1650 nm. The endmembers used for LMM_FAE were chosen arbitrarily, but consistent with other Science members default test endmember indexes.

Report:

Report_Interval_Scaling_Derivative_LMM_FAE.pdf

Code:

Interval_Scalling_Derivative_LMM_FAE.py

Results

This technique proved to increase accuracy of true abundances vs optimized abundances (y=x). And what was previously hindering this was LMM_FAE inaccurately distinguish between npv and soil.

Analysis/Conclusion

To investigate further:

1. Testing other intervals where one of the spectra’s derivative is zero
2. Altering scaling factor
3. Using optimized endmembers rather then arbitrary ones.