The potential of spectral mixture analysis to improve the estimation accuracy of tropical forest biomass

A main limitation of pixel-based vegetation indices or reflectance values for estimating above-ground biomass is that they do not consider the mixed spectral components on the earth's surface covered by a pixel. In this research, we decomposed mixed reflectance in each pixel before developing models to achieve higher accuracy in above-ground biomass estimation. Spectral mixture analysis was applied to decompose the mixed spectral components of Landsat-7 ETM+ imagery into fractional images. Afterwards, regression models were developed by integrating training data and fraction images. The results showed that the spectral mixture analysis improved the accuracy of biomass estimation of Dipterocarp forests. When applied to the independent validation data set, the model based on the vegetation fraction reduced 5–16% the root mean square error compared to the models using a single band 4 or 5, multiple bands 4, 5, 7 and all non-thermal bands of Landsat ETM+.