A quantitative assessment of multiple scattering in plant-soil mixtures and the implications on nonlinear spectral unmixing models

Bilinear Model (BM) is one of widely used nonlinear spectral unmixing methods, which are developed to deal with nonlinearity resulted from the multiple scattering within mixed pixels such as plant-soil mixtures. In the BM, products of endmember spectra are used to represent multiple scattering effect, and this approximation needs a validation. This study applies a Monte Carlo ray-tracing (MCRT) model to test this approximation by exploring the correlations between multiple scattering reflectances and endmember products. The correlations are found linear, proving the rationality of this approximation. Besides, the correlations between multiple scattering coefficients in the BM and vegetation characters are analyzed. Scattering coefficients are found to have quadratic relationships with vegetation coverage and linear relationships with crown height. The correlations can be used as constraints when solving the BM to retrieve the abundance of each component, to relieve the collinearity problem which impacts the solution precision.