Figure 21. Probability density violin plots (Hintze & Nelson,
1998) for model runs. Violin plots show distribution of (a) weight
abundances (wt.%) and (b) grain sizes (µm) for each endmember.
Horizontal dashed lines show the values for the lowest RMSE fits for the
two sand bodies.
Inversions of Spectra to Chemical Compositions: Following the
approach outlined by Lapotre et al. (2017b), we inverted the
mineral-based results for our two endmember results to examine
compositional trends, simulating compositions that would have been
obtained by APXS (Table S3). Specifically, following VanBommel et al.
(2019a, 2019b), we did the inversions by first starting with APXS
spectra instead of oxides. This approach mitigates matrix effects within
XRF (e.g., APXS) data that can result in inaccuracies when performing
similar analyses via oxide mass balance. The simulated APXS spectra were
then converted to oxide abundances.
Following Lapotre et al. (2017a), we plot model results on a
SiO2 vs. MgO+FeO diagrams for all our CRISM spectral
retrieval results, along with compositions for the endmember spectra
used in the CRISM-based retrievals (Figure 22). Results highlight the
similarities between the two sand sheets in terms of the cluster of
results located in the approximately the same values of
SiO2 vs. MgO+FeO. Results also show the dominance of
pigeonite in controlling the differences between the two sand bodies.
Finally, it is noted that the best fit model results are not centered
with the bulk of the model results, again demonstrating that the models
need to be examined in a strict statistical sense.