Abstract
In this research study, non-linear spectral mixing models have been developed and employed to achieve unmixing the proportions of components accurately and retrieving the physical parameters of the mixture. The first focus is on a comparative study of linear and non-linear spectral mixing models. A carefully-controlled experiment was conducted in the laboratory. The aim is to test both models to unmix ternary powdered-chalk mixtures by using the directional reflectance data. The results demonstrated the superiority of the non-linear model over the linear model. However, there was at least one case when the linear model produced more accurate results than the non-linear model. As a consequence, a hypothesis was made that the directional reflectance data obtained from certain measurement geometries may not contain useful information for deriving surface parameters. In order to investigate this, an error analysis was employed to observe the sensitivity to error of a physical parameter, which is needed in the non-linear unmixing, when estimated from directional reflectance data at certain measurement geometries. This theoretical investigation was tested against reflectance data of mineral mixtures obtained from a laboratory experiment. The results showed that the unmixing could be improved when the angular measurements were carefully chosen. Information contained in each surface measurement can be useful or damaging depending on the measurement geometry and the brightness of the surface itself. The next focus is on the utilisation of non-linear spectral mixing model to retrieve the biophysical properties of vegetation canopies by means of a canopy reflectance modelling. A two-layer model of the bidirectional reflectance of homogeneous vegetation canopies was proposed in this study. The anisotropic scatterings of both the vegetation canopy and the background were taken into account. This new model was validated against simulated and field-measurement data. The results showed that this model can be used to model the bidirectional reflectance and to retrieve the optical properties of canopy elements (leaves) and background of a homogeneous canopy. Finally, a simple non-linear spectral mixing model was developed. The second order interaction between vegetation and soil was taken into account. Results from the experiments showed that the vegetation cover and leaf area index of moderate density canopies can be retrieved by using this model.