Spectroscopic data frequently contain trends that depend on several simultaneous or concurrent factors. SVD provides a method to decompose a data matrix into orthogonal basis spectra that can be interpreted later in terms of known components. SVD does not provide the interpretation itself, but it is the tool to separate various components in a data set and to permit a user to observe the various indpendent components.

A mathematical view of SVD is described in the following:

Matrix decomposition of a model matrix

A graphical view of SVD is provided in the following:

Components of a Gaussian spectral model

References