calendar Add meeting dates to your calendar.

 

Paper No. 3
Presentation Time: 8:00 AM-6:00 PM

FACTORS AFFECTING PRINCIPAL COMPONENT ANALYSIS (PCA) OF X-RAY ABSORPTION FINE STRUCTURE SPECTRAL DATASETS OF ARSENIC AND IRON COMPOUNDS


BROWN, Amy L.1, FOSTER, Andrea L.2, ALPERS, Charles N.3, HANSEL, Colleen4, LENTINI, Chris4 and KIM, Christopher S.5, (1)Department of Geological Sciences, University of Florida, 241 Williamson Hall, Gainesville, FL 32611, (2)U.S. Geological Survey, 345 Middlefield Rd., MS 901, Menlo Park, CA 94025, (3)U.S. Geological Survey, California Water Science Center, 6000 J St, Placer Hall, Sacramento, CA 95819, (4)School of Engineering and Applied Sciences, Harvard University, 40 Oxford Street, ESL, RM 305, Cambridge, MA 02138, (5)School of Earth and Environmental Sciences, Chapman University, One University Drive, Orange, CA 92866, amy.brown@ufl.edu

We are performing principal component analysis (PCA) on model compound test sets of arsenic (As) and iron (Fe) X-ray absorption fine structure (XAFS) spectra. The results are being used to assist in the interpretation of natural samples collected from Empire Mine Historic State Park (HSP), California, as part of a study on As bioavailability in mine waste from low-sulfide, quartz-hosted gold deposits.

Performing PCA on XAFS datasets prior to evaluation by linear combination least-squares fitting (LSF) provides: a model-independent way to view the variance in a spectral dataset, constraints on the number of unique spectra needed for LSF, and a quantitative process for selecting the appropriate model spectra to be used in LSF. Previous studies have highlighted key limitations or considerations for PCA, but the effect on PCA of factors such as element, number of data points, type of data (near-edge vs. extended XAFS), spectral noise, and low abundance species have not been described. Investigations into the magnitude of these effects are needed to establish the appropriate level of confidence to have in PCA on real datasets.

To date, we find that the correct number of components is identified in PCA of all 3-component test data sets. However, if 2 or more species in a dataset do not vary in relative abundance, the number of significant components identified by the procedure can be artificially low. The 3 significant components identified in XANES and EXAFS model compound data sets described an average of 99% of the total set variance, but the relative variance accounted for by each component varied considerably. In XANES data sets, components 1, 2 and 3 described an average of 90%, 6%, and 3% of set variance, respectively. For EXAFS, components 1, 2 and 3 described an average of 66%, 24%, and 9% of total set variance, respectively. Preliminary PCA on As and Fe EXAFS data sets collected on samples from the Empire Mine HSP indicate 2-3 components (species) for As and 3-4 for iron which cumulatively account for 69-89% of the set variance. Fe compounds identified as reasonable species for LSF of Fe EXAFS are ferrihydrite, lepidocrocite, nontronite, hematite, goethite, and Fe-smectite. Arsenopyrite, arsenian pyrite, As(V) sorbed on goethite, and As(V)sorbed on ferrihydrite are identified as reasonable species to use in LSF of As EXAFS.

Meeting Home page GSA Home Page