Spectral un-mixing of natural surfaces scenarios

Natural geological surfaces may have various mineralogical compositions and physical properties. Surface reflectance spectra acquired by field spectrometers result generally of contributions from several end-members (various soil/rock types, different sizes, etc.). Most of the algorithms commonly available these days to access sub-pixel information deal with linear (or spatial) un-mixing. However, natural environments represent complex situations, combining linear and non-linear mixtures at various scales. At this stage, empirical investigations are best suited to try and better understand the result of mixtures on spectral reflectance of natural surfaces. Surface spectral reflectance was acquired with an ASD FieldSpec spectrometer, covering the 0.4-2.5 μm range with a spectral resolution of 1 nm. A linear un-mixing algorithm was applied to several cases, then end-member abundances were estimated and compared to real abundances. Results showed non-ability of linear un-mixing to predict correct abundances of some end-members. This derives from non-linear contribution of some components of the mixture.