Evaluation of Processing Speed and Accuracy of Various Radiative Transfer Techniques on Reflectance Near-Infrared Spectroscopy of Mixtures of Powders in Controlled Environment

This presentation is a summary of tests of various versions of spectral unmixing : linear spectral unmixing, an approximate version of the Hapke model, the Shkuratov model, and a combination of Hapke + linear unmixing. The tests were for computation speed and accuracy against laboratory spectra of powder mixtures of clinopyroxene, orthopyroxene, olivine and carbon black. This work is also relevant to another project of mine : mapping of olivine on Vesta, which I presented at DPS last week. The main results are: - On intimate powder mixtures of minerals, both Hapke and Shkuratov models provide abundance estimates 50% better than the linear spectral unmixing model, and they take 5 to 6 times longer to compute - On areal powder mixtures of minerals, the linear spectral unmixing model is very accurate, whereas the Hapke and Shkuratov model result in abundance estimates off by 40% and 30% respectively. - On Vesta's surface, linear spectral unmixing agrees with the Hapke model by 97.7% for olivine detections (not abundances). I haven't made the calculation for other minerals yet, but visually I can tell the differences are higher than for olivine. - On Vesta's surface, Hapke's model olivine abundance estimates are 40% higher than linear spectral unmixing results, and the computation time is ~10 times longer, even with the approximate version of the Hapke model. - I think it is possible to perform the approximate version of the Hapke model and the Shkuratov model to make maps, however maybe not at full resolution for all M3 data of all the areas of interest for this project : I have yet to try on M3 data. I only made tests on VIR data of Vesta. - Future testing is planned for full radiative transfer model accounting for coating of native iron and carbon black. Once I finish implementing this model, it is expected to be performed only on a selection of spectra, because it may take up to one hour to model just one...