Mining below the Sub-Pixel Scale: Past Results and New Directions In Thermal Infrared Data Analysis of the Earth and Mars

Thermal infrared (TIR) spectroscopy and remote sensing have benefited greatly in the past two decades from advances in both instrumentation and data analysis techniques. Past Mars-orbiting and landed payloads have had TIR instruments, each of which produced datasets that have forever changed our perspective of surface processes operating on the red planet. Similarly, in the past decade Earth-orbiting instruments have enabled the generation of global emissivity maps, the analyses of petrologic/textural data of continents, and the study of surface thermal fluxes from active volcanoes to urban landscapes. An over-arching analysis technique in all these studies is linear spectral deconvolution, which uses a set of library or image end-members to derive sub-pixel areal percentages of those end-members. The model has been extensively tested in the laboratory, and on both Mars (e.g., TES, THEMIS, mini-TES) and Earth (e.g., MODIS, ASTER) datasets. The assumption that spectral end-members in the TIR mix in relative proportion to their areal percentages is valid due primarily to the high absorption coefficient of most rock-forming minerals. However, the limitations of such an approach are driven by the initial assumptions of the model, the spatial and spectral resolution of the data, as well as the breadth of the available spectral library. Previous studies have examined the validity of this model with complicating effects such as decreasing particle size, heterogeneous thermal structure, cavity emission, incomplete/incorrect atmospheric corrections, poor signal to noise, and iterative fitting of end-members from large spectral libraries. For many of these complications, linear deconvolution has been shown to remain valid, whereas the others may require a different set of initial modifiers in order to apply the model accurately. This paper will explore the scope and evolution of the linear deconvolution approach in the TIR from early results to the future possibilities using more advanced analysis techniques.