Assessment of biophysical vegetation properties through spectral decomposition techniques

This article demonstrates the use of spectral decomposition for analyzing the spectral response of different semiarid vegetation species found throughout Mediterranean Europe. Using this technique, it is possible to decompose a spectral data set into a smaller number of significant factors that represent the key variables affecting vegetation spectral response. The results presented here show how spectral decomposition can be used to determine the intrinsic number and identity of the significant factors affecting the multispectral response. For the dataset investigated here, which comprises field spectral recorded over 1130 wavelengths, using a GER single field-of-view IRIS (SIRIS) spectroradiometer, it was found that a combination of just four factors was responsible for the majority of spectral variance. Interpretation of these factors was carried out by graphical analysis stepwise regeneration of the original spectra, and correlation with biophysical data. Considering the identity of these factors, it was found that the second most significant factor (factor 2) was strongly related to the proportion of directly irradiated green leaves within the field-of-view of the spectradiometer. In addition, it was found that the fourth most significant factor (factor 4) provided a good summary of the spectral response of the different samples in the region of strong chlorophyll absorption. This demonstrates the possibility of using spectral decomposition techniques, particularly in environments dominated by spectrally similar vegetation classes, to model the mixed spectral population as mixtures of fundamental biophysical parameters rather than as mixtures of the classes themselves.