GSA Annual Meeting in Seattle, Washington, USA - 2017

Paper No. 172-10
Presentation Time: 9:00 AM-6:30 PM

CHEMICAL ZONING HAS INSIGNIFICANT EFFECT ON PREDICTION OF HOST LITHOLOGY OF TOURMALINE USING MULTIVARIATE ANALYSIS OF LASER-INDUCED BREAKDOWN SPECTROSCOPY (LIBS) SPECTRA


MOUNT, Cole1, MCMILLAN, Nancy J.2, DUTROW, Barbara L.3 and HENRY, Darrell J.3, (1)Geological Sciences, New Mexico State University, PO Box 30001, MSC 3AB, Las Cruces, NM 88003, (2)Geological Sciences, New Mexico State University, Box 30001 MSC 3AB, Las Cruces, NM 88003, (3)Department of Geology and Geophysics, Louisiana State University, Baton Rouge, LA 70803, mcmount@nmsu.edu

Tourmaline incorporates a wide range of elements during formation, recording the composition of the host lithology. Tourmaline growth often occurs in dynamically changing chemical environments, commonly resulting in chemical and isotopic zoning. Because tourmaline is chemically and mechanically resistant and has low diffusion rates, its original composition and zoning are preserved. Thus, tourmaline is an ideal mineral for use in determining sediment provenance. The zoning of tourmaline raises the question of whether the host lithology can be accurately identified from all zones in a single crystal. This study approaches the question by modeling LIBS spectra with the multivariate technique Partial Least Squares Regression (PLSR). LIBS spectra contain information on the concentration of every element including isotopic, chemical and structural information: thus, each spectrum is a unique and detailed signature of the material.

In this study, 60 spectra from tourmaline from six different host lithologies (calcareous metamorphic rocks, hydrothermal deposits, Li-poor pegmatites, Li-rich pegmatites, pelitic metamorphic rocks, silicic igneous rocks) were used to build a matching algorithm. The matching algorithm is a sequence of five PLSR models, each of which classifies the spectra into two groups: the lithology being defined and the group of all other lithologies. After a lithology was defined, all associated spectra were removed from subsequent models. Models were validated with 15 spectra not used for model calibration. Success rates are the percent of correctly classified spectra. Overall, 92.3% of the test set spectra were correctly classified; success rates ranged from 83.3% to 100% for individual models. Thirty-four spectra of zones in 12 tourmalines, all from Li-rich pegmatites without detrital cores, were used to test the hypothesis that the lithologic host of all zones can be correctly classified. Thirty-three of the spectra were correctly classified; the host of one spectrum was classified as Li-poor pegmatite. This study indicates that multivariate analysis of LIBS spectra can “see through” tourmaline zoning and identify each zone as the correct lithology.