Relationships of Propagated Error in Fraction of Vegetation Cover Among the Retrieval Algorithms Based on a Linear Mixture Model

Fraction of vegetation cover (FVC) is often estimated by unmixing a linear mixture model (LMM). In the LMM-based algorithm, differences can be seen in assumptions and constraints imposed to the model such as spectral vegetation index, inducing variations in algorithm. As a result, robustness against noises in reflectance spectrum is somewhat different among those algorithms, depending on a target spectrum to be analyzed, endmember spectra used for unmixing, and choice of two-band VI assumed in the algorithms. Objective of this study is to propose an analytical technique for better algorithm selection under a twoendmember assumption. Robustness against noises in reflectance spectrum is considered as a criterion. This criterion is first derived analytically, and then demonstrated numerically. It is shown that our proposing technique based on the derived factor is an indicator to determine a better algorithm against noises for any target spectrum over the entire red-NIR reflectance subspace.