Multivariate statistical approach to estimate mixing proportions for unknown end members

Journal of Hydrology
By: , and 

Links

Abstract

A multivariate statistical method is presented, which includes principal components analysis (PCA) and an end-member mixing model to estimate unknown end-member hydrochemical compositions and the relative mixing proportions of those end members in mixed waters. PCA, together with the Hotelling T2 statistic and a conceptual model of groundwater flow and mixing, was used in selecting samples that best approximate end members, which then were used as initial values in optimization of the end-member mixing model. This method was tested on controlled datasets (i.e., true values of estimates were known a priori) and found effective in estimating these end members and mixing proportions. The controlled datasets included synthetically generated hydrochemical data, synthetically generated mixing proportions, and laboratory analyses of sample mixtures, which were used in an evaluation of the effectiveness of this method for potential use in actual hydrological settings. For three different scenarios tested, correlation coefficients (R2) for linear regression between the estimated and known values ranged from 0.968 to 0.993 for mixing proportions and from 0.839 to 0.998 for end-member compositions. The method also was applied to field data from a study of end-member mixing in groundwater as a field example and partial method validation.
Publication type Article
Publication Subtype Journal Article
Title Multivariate statistical approach to estimate mixing proportions for unknown end members
Series title Journal of Hydrology
DOI 10.1016/j.jhydrol.2012.06.037
Volume 460-461
Year Published 2012
Language English
Publisher Elsevier
Contributing office(s) South Dakota Water Science Center, Dakota Water Science Center
Description 12 p.
First page 65
Last page 76
Google Analytic Metrics Metrics page
Additional publication details