ROBUST MULTIVARIATE-ANALYSIS APPLIED TO SEPARATION OF COMPONENTS IN ABSORPTION-SPECTRA

A new algorithm for the separation of standard absorption spectral components from an experimental composite spectrum is presented. It is in fact robust multivariate fitting. This algorithm assumes a double-sided exponential error distribution, is comparable to least squares in speed, and leads naturally to estimates of the component errors. Robust fitting is particularly applicable to absorption spectral measurements since the error distribution may be demonstrated to be non-Gaussian at high optical densities and further because a few grossly outlying data points have much less effect on the returned concentrations than the least squares case. © 1990.