Estimating Erosion Surface Features by Linear Mixture Modeling

Spectral mixture modeling was performed for identification and mapping of land degradation features related to soil erosion processes in the Sacaba Valley, Bolivia. The model allowed use of up to five surface components to characterize the selected area, since six bands (1, 2, 3, 4, 5, and 7) of the Landsat TM sensor were used as inputs. Among the various methods commonly used to determine end-members from the satellite image, three were selected: a) identification of one “pure” pixel representing a particular surface component from false color composites; b) average of “pure” pixels to characterize a particular end-member; and c) a method based on principal components. The best characterization of end-members was achieved by using average pure pixel reflectance. The median of the abundance images showed that, in 95% of the cases, the individual pixel compositions were explained by the selected surface components. The research has demonstrated that regional patterns of soil surface erosion features can be reliably mapped using linear spectral mixture analysis. Extrapolation of this approach to other regions where soil degradation features are correlated with spectrally distinguishable surface characteristics is feasible, provided that an optimization of the unmixing model as a function of local or regional surface component types is completed.