Bachmann et al Page 1 Dec 2008

Appendix 1: The Raman method

1  Sample preparation and analyses

Characteristic pumices of the KPT from the four existing types (tube, frothy, crystal-poor, banded; Bouvet de Maisonneuve et al. 2008) were sampled in proximal and distal stratigraphic sections (Fig. 2 of main text). Pumices were crushed, sieved, and quartz crystals handpicked before being mounted in epoxy and polished in order to expose the largest possible surface of quartz grains. Those grain mounts were then photographed to locate the quartz grains and to find suitable melt inclusions for analysis. Each melt inclusion was also photographed and its depth from the grain mount’s surface measured.

All measurements were obtained using a Jobin Yvon Raman microspectrometer at the University of Geneva. An Olympus microscope is attached to the Raman spectrometer, and a 100x objective was used in most of the analyses. A 532-nm Nd-YAG laser was used to excite the sample. The applied laser output energy was ~500mW, and the energy measured on the sample’s surface was 15 mW. All measurements were done with a grating of 1800 to get high spectral resolution, a slit of 600 μm and a hole of 150 μm. Two spectral windows were measured, the first one from 50 to 1800 cm-1, and the second one from 2500 to 3950 cm-1. The total time of analyses was 3x60 seconds and each analysis was at least duplicated. For a melt inclusion intersected at the surface of grain mount, the analysis was done few microns deep in the inclusion. For the inclusion located below the surface of the grain mount, focus was aimed at the middle of the inclusion to make sure the excited volume was within the melt inclusions (and not partly in the host mineral).

2  Processing of the Raman Spectra

The treatment of the Raman spectra followed the method outlined by Zajacz et al (2005), which include several steps (Figure 1): (1) a first correction applied is to reduce the background effect (2) a frequency-temperature-intensity correction is used to obtain real spectral intensity. (3) The interval measured between 720 and 1260 cm-1 of the first peak is used as internal standard, allowing a normalization of the H2O peak (whose size can be influenced by various analytical parameters). (4) The 720-1260 cm-1 peak is deconvoluted and the area beneath of the H2O peak (3030-3740 cm-1) is calculated. (5) Finally, the ratio between those two areas (Si-O-Si peak and H2O peak) is multiplied by a calibration factor (obtained by using standards of known H2O content; see below) to give the H2O content of the quartz-hosted melt inclusion.

3  Calibration

Determination of H2O concentration by Raman spectroscopy requires both an external and internal standard (in order to normalize the peak at 3500cm-1). Several calibration procedures have been proposed (Chabiron et al. 2004; Zajacz et al. 2005; Di Muro et al. 2006). In this study, a calibration method similar to that of Zajacz et al. (2005) was used. We obtained calibration curves both by using: 1) synthetic experimental glasses of haplogranitic composition with known H2O content provided by Z. Zajacz, and 2) melt inclusion glasses from KPT whose H2O content were previously determined by SIMS.

To test the calibration obtained at the University of Geneva, we chose a set of melt inclusions in KPT quart crystals with excellent spectra and analyzed them both at the University of Geneva (with our calibration) and at ETHZ, using the calibration of Zajacz et al., (2005; using a Dilor LabRam II with 488nm laser). Best agreement between the two datasets (on the same melt inclusions) is reached when the calibration factor of 0.38 (using the H2O content of KPT melt inclusions determined by SIMS as standards) is used. We therefore applied this calibration factor despite the lower correlation index for the regression (R2=0.8; Figures 2 and 3)

To test the reproducibility of the Raman method, one large melt inclusion was analyzed eight times over the course of this study. All the analyses were done within a few months using the same calibration and procedure. The results show a slight variability between the different analyses (Figure 4), with a standard deviation of 0.21 wt% H2O.

4  Method limitation

4.1  Disturbed Raman spectra

The quality of the Raman spectra can vary significantly between melt inclusions for which there is no apparent difference under a petrographic microscope. Varying peak shapes were observed, which can lead to non-negligible variations on the results (e.g., H2O and Si-O-Si peaks are not well enough defined to be able to correct for background and obtain an accurate deconvolution). In some cases, adjusting some analytical parameters (adjustment of focus, increasing analysis time), lead to a significant improvement of peak shapes. However, most inclusions that yielded poor peak shapes could not be improved.

A common cause for disturbed Raman spectra is fluorescence (Chabiron et al. 2004; Zajacz et al. 2005; Behrens et al. 2006; Di Muro et al. 2006; Thomas et al. 2006). This effect disturbs the H2O analysis because it increases background significantly in the wavelength window of the H2O peak (in some cases, background completely covers the H2O peak; Figure 5).

Possible causes for triggering fluorescence in KPT melt inclusions are poorly understood. Fluorescence appeared in certain samples, but not in others. As no optically and chemically obvious differences between fluorescent and non-fluorescent samples could be detected (all have the same major element composition), a suite of 34 trace elements was analyzed by LA-ICP-MS (at the University of Lausanne; Table 3, and see section 4.4 for analytical details) in two populations of melt inclusions (one showing high fluorescence and the other not showing any) in order to determine whether trace element variations were the cause of this fluorescence effect. The inclusions selected had a diameter >40 microns and were all at the surface of the host crystal. The two populations of melt inclusion did not present any significant differences in the measured trace element contents (Table 3). Therefore, fluorescence in Raman spectra must be controlled by factors other than trace element chemistry.

4.2  Depth attenuation

Despite the fact that melt inclusions do not necessarily need to be exposed to the surface of the grain to obtain a usable Raman spectrum, Chabiron et al. (2004) have shown that there is an intensity reduction of the H2O peak when the melt inclusions are analyzed below the surface of the grain mount. Therefore, a depth correction to the determined H2O content needs to be applied. We followed the method of Thomas et al. (2006). Several melt inclusions (seven in our case) at different depths were chosen out of five quartz crystals coming from sample KPT05-3 (banded pumice from unit E) mounted in epoxy. The depth of each melt inclusions was precisely measured, prior to determining their H2O content by our Raman method. The crystals were then polished for ~10μm and their H2O content re-measured. A total of seven polishing steps were accomplished. Figure 6 shows how the H2O peak becomes dimmer as the inclusion becomes deeper (whereas the T-O peak remains relatively constant). As the ratio of those two areas is used to calculate the total H2O content of the inclusions, it is necessary to apply the depth correction. Using a linear fit to regress peak ratio vs. depth, a correction factor of 0.0046 was obtained.

Figures

Figure 1: Treatment of the Raman spectra obtained by Raman spectroscopy, showing (a) and (b) the original spectra with the background correction for Si-O-Si and H2O. (c) and (d) Spectra with the Frequency-Temperature-Intensity correction peak deconvolution peak area measurements.

Figure 2: Linear regression of the peak ratio determined by Raman vs. H2O determined by SIMS for 11 melt inclusions of the KPT and by experimental haplogranitic glasses (provided to us by Z. Zajacz, ETHZ Zurich). Calibration factor (slope) is 0.38 for the linear regression using SIMS data, and 0.41 for the regression using the experimental glasses.

Figure 3: Comparison of H2O content determined by Raman spectroscopy on 10 melt inclusions (in KPT samples) using the calibration of Zajacz et al. (2005) in Zurich and the both calibrations from Geneva (1: using experimental glass standards, calibration factor = 0.41, and 2: using SIMS data, calibration factor = 0.38). Best agreement with the Zurich data is obtained using calibration 2 (using SIMS data points)

Figure 4: Reproducibility test on a large KPT melt inclusion; eight analyses performed over the course of this study leads to a standard deviation of ~0.2 wt% H2O.

Figure 5: Example of two analyses showing a high fluorescence in the Raman analyses (from the same sample KPT05-2, an early-erupted crystal-poor pumice).

Figure 6: Raman spectra at different depths. Note how the signal improves (lower background, sharper peaks) as the melt inclusion becomes closer to the surface.

5  References

Behrens H, Roux J, Neuville DR, Siemann M (2006) Quantification of dissolved H2O in silicate glasses using confocal microRaman spectroscopy. Chemical Geology

Bouvet de Maisonneuve C, Bachmann O, Burgisser A (2008) Characterization of juvenile pyroclasts from the Kos Plateau Tuff (Aegean Arc): insights into the eruptive dynamics of a rhyolitic caldera-forming eruption. Bulletin of Volcanology DOI 10.1007/s00445-008-0250-x

Chabiron A, Pironon J, Massare D (2004) Characterization of water in synthetic rhyolitic glasses and natural melt inclusions by Raman spectroscopy. Contributions to Mineralogy and Petrology 146:485-492

Di Muro A, Giordano D, Villemant B, Montagnac G, Scaillet B, Romano C (2006) Influence of composition and thermal history of volcanic glasses on water content as determined by micro-Raman spectrometry. Applied Geochemistry 21(5):802-812

Thomas R, Kamenetsky VS, Davidson P (2006) Laser Raman spectroscopic measurments of water in unexposed glass inclusions. American MIneralogist 91:467-470

Zajacz Z, Halter W, Malfait WJ, Bachmann O, Bodnar RJ, Hirschmann MM, Mandeville CW, Morizet Y, Müntener O, Ulmer P, Webster JD (2005) A composition-independent quantitative determination of the water content in silicate glasses and silicate melt inclusions by confocal Raman spectroscopy. Contributions to Mineralogy and Petrology 150(6):631-642