Decomposition of mineral absorption bands using nonlinear least squares curve fitting: Application to Martian meteorites and CRISM data

This study advances curve-fitting modeling of absorption bands of reflectance spectra and applies this new model to spectra of Martian meteorites ALH 84001 and EETA 79001 and data from the Compact Reconnaissance Imaging Spectrometer for Mars (CRISM). This study also details a recently introduced automated parameter initialization technique. We assess the performance of this automated procedure by comparing it to the currently available initialization method and perform a sensitivity analysis of the fit results to variation in initial guesses. We explore the issues related to the removal of the continuum, offer guidelines for continuum removal when modeling the absorptions and explore different continuum-removal techniques. We further evaluate the suitability of curve fitting techniques using Gaussians/Modified Gaussians to decompose spectra into individual end-member bands. We show that nonlinear least squares techniques such as the Levenberg–Marquardt algorithm achieve comparable results to the MGM model (Sunshine and Pieters, 1993 and Sunshine et al., 1990) for meteorite spectra. Finally we use Gaussian modeling to fit CRISM spectra of pyroxene and olivine-rich terrains on Mars. Analysis of CRISM spectra of two regions show that the pyroxene-dominated rock spectra measured at Juventae Chasma were modeled well with low Ca pyroxene, while the pyroxene-rich spectra acquired at Libya Montes required both low-Ca and high-Ca pyroxene for a good fit.

Research highlights
► We advanced curve-fitting modeling of absorption bands of reflectance spectra.
► We detailed a recently introduced automated parameter initialization technique.
► We considered issues related to continuum removal, band saturation and stability of solutions.
► We applied Levenberg–Marquardt curve-fitting to meteorites and total least squares to CRISM data.
► We developed new summary maps based on confidence levels of model parameters.