Search

In estimation of fractional area covered by vegetation (fraction of vegetation cover, FVC) often involves choices of endmember spectra. The determination/estimation of those spectra is performed with uncertainties, hence becomes a source of error in FVC. This study aims at investigating those propagated errors in FVC targeting a retrieval algorithm with a constraint by spectral vegetation index. In this algorithm, two types of endmember species, vegetation and non-vegetation, are assumed. A band correlated noise was also assumed in both spectra. It was found that when the bias errors in the endmember spectra satisfy certain conditions, the propagated errors in FVC are cancelled out. This condition was derived analytically in terms of the bias-free (true) endmember spectra and a spectra of a target pixel. On the contrary the biases of the endmember spectra will be empathized when those biases satisfy a condition that is also derived analytically in this study. [ABSTRACT FROM PUBLISHER]