Identification and quantification of single constituents in groundwater with Fourier-transform infrared spectroscopy and Positive Matrix Factorization

Fourier-transform infrared (FTIR) spectra of environmental samples usually comprise superposed pure component spectra of the individual constituents present. Thus, evaluation of band intensities and reconstruction of FTIR spectra by inverse techniques is often prone to ambiguity. Frequently applied multivariate curve resolution techniques (PCA, PCR, PLS) reduce the complexity in the observed spectra by isolating recurring components. However, due to mathematical constraints, such components usually do not resemble the physical spectra of the single constituents in mixed samples, which is, by contrast, principally possible with Positive Matrix Factorization (PMF). In the present study, we tested if PMF is appropriate to conduct quantitative FTIR spectroscopy on environmental samples. We quantified the concentration of humic acid-coated goethite (HA-Goe) in groundwater from two sites, which was sampled before and after HA-Goe was injected into the aquifers. PMF could successfully reconstruct the superposed FTIR spectra of the solids from freeze-dried groundwater, with explained variances of >0.99 and mean residuals of <0.01 (normalized on maximum absorbance). Due to normalization on an internal standard (potassium ferrocyanide), PMF allowed for the quantitative interpretation of compounds that were represented by dedicated PMF components (HA-Goe, kaolinite, montmorillonite, organic matter) even without a priori knowledge on the sample composition. The obtained HA-Goe concentrations agreed very well with those calculated from Fe concentrations determined with ICP-OES (correlation coefficients>0.96). Due to i) the use of objective criteria, which quantified the accuracy and uncertainty of each PMF solution, ii) repetitive PMF runs with randomized initial conditions and iii) physically meaningful PMF-components that resemble the FTIR spectra of single compounds in heterogeneously composed samples, the application of expert knowledge is replicable and less dependent on the operator than in other approaches of quantitative FTIR spectroscopy.