A robust quantification of H2O in silicate glasses through micro-Raman spectroscopy: insights on the compositional effect

Abstract

Micro-Raman spectroscopy is considered one of the most promising, rapid and economical methods to measure the water content in small volumes (a few μm3) of natural glasses (e.g. volcanic glass matrix and melt inclusions in minerals), which is a fundamental parameter to obtain information and constraints on the feeding systems of volcanoes. In this article we report a new method for calibrating micro-Raman spectroscopy for water determination based on the analysis of a large number of silicate glasses (101) synthesized over a wide compositional (basalt, basaltic andesite, andesite, haplogranite, shoshonite, latite and phonolite) and physico-chemical range (T = 1000–1275 °C, P = 50–500 MPa, fO2 = ±2.6 log units from the Ni-NiO buffer) with H2O contents between 0.2 and 7.6 ±0.1 wt.%. A robust and reproducible method for baseline subtraction and measurement of the water and silicate areas and their ratio (Rw/s) is proposed. Rw/s data show an average error of ±0.07 confirming the high potential of Raman spectroscopy of accurately determining the water content. Regression analysis allows the achievement of improved water equations as a function of Rw/s and the anhydrous composition of the glasses, accounting for relatively low H2O uncertainties (average error = 0.16–0.17wt.%; standard error of the estimate, σest, = 0.21–0.22wt.%) compared to previous calibrations based on smaller compositional ranges and sample numbers. These equations are reproducible and easily applicable to both experimental and natural glasses. Our analyses also indicate that the H and H/Σcations (hydrogen and sum of other cations, calculated on the 8-oxygen basis) vs. Rw/s relationships are affected by the amount of trivalent (Al3+ and Fe3+) and pentavalent (P5+) cations. This hypothesis requires additional investigations as, at the state of the art, the estimate of Fe3+/Fe2+ in experimental glasses is not sufficiently accurate.