| Crynodeb: | <p>Understanding asteroid composition is key to better understanding a variety of specific fields within planetary science. Thermal infrared (TIR; 5 − 25μm) spectroscopy is a valuable method for estimating modal mineralogy due to its non-destructive nature, and utility in laboratories on Earth and in spacecraft instrumentation to provide remote sensing observations. The current most widely used forward model (i.e., mathematical expression to describe phenomena), and algorithm (i.e., computational technique to estimate parameters of interest) are linear mixing and a linear least squares, respectively. Whilst this method for spectral interpretation is accurate for whole rock/coarse grained samples, and applications for which we have a wealth of contextual information (e.g., Mars), this combination of model and algorithm becomes unsuitable where fine grains (< 50μm) dominate, and when there is little additional information about the measurement target. The simplicity behind the linear mixing model and linear least squares algorithm raises the question whether a more complex model is required, or can linear mixing still be used effectively within more sophisticated algorithms?</p>
<p>As the linear mixing model has advantages over others, this project has investigated linear mixing within a selection of increasingly sophisticated spectral unmixing algorithms, whilst adopting a Bayesian approach. Issues of degeneracy and the use of implicit a priori information, that are intrinsic to current methods, have also been tackled. Results from this investigation (as presented within this thesis) showed that an alternative model is required; thus, also presented, is preliminary work towards a novel method of parameterising linear mixing via empirical studies of quantitative spectral morphology changes with decreasing grain size. Initial results show promise, but refinement through more in depth future spectral morphology studies is required.</p>
<p>This work has demonstrated the unsuitability of linear mixing with linear least squares for fine-grained mixture composition estimates, and begun laying foundations for improved spectral interpretation tools.</p>
|
|---|