The recent study by Devine & Rutherford (2014), ‘Magma storage region processes of the Soufrière Hills Volcano, Montserrat’, published as Chapter 19 of the Geological Society of London Memoir, Volume 39, focused primarily on updating the record of magmatic temperatures recorded by Fe–Ti oxides in the andesite erupted from Soufrière Hills Volcano. However, a key result of the paper was that the compositions of some plagioclase-hosted melt inclusions indicate mixing of mafic magma components into the host andesite and trapping of the mixed melts within phenocrysts. This interpretation is consistent with our earlier work on the chemistry of melt inclusions (Humphreys et al. 2010) and of microlite crystal populations (Humphreys et al. 2009a, 2013). Humphreys et al. (2010) proposed this interpretation by showing that a subset of Soufrière Hills melt inclusions and matrix glasses has anomalous concentrations of K2O and/or TiO2, as have mafic inclusion matrix glasses (see Humphreys et al. 2010, fig. 2). However, in their Appendix, Devine & Rutherford (2014) suggest that enrichment in K2O can result only from decompression crystallization and not from magma mingling. They also assert that our melt inclusion dataset is fundamentally flawed and subject to faulty corrections for post-entrapment correction; and that our melt inclusions were actually matrix glasses. Finally, Devine & Rutherford (2014) doubt ‘whether or not one can look at melt-inclusion analyses and distinguish the effects of decompression crystallization … from the effects of mingling … with components derived from the injected mafic magma’.
Here we address the criticisms of the Devine & Rutherford (2014) Appendix, drawing on additional published data to support our arguments and explanations. We were unable to reproduce some of the key figures of Devine & Rutherford (2014), so we present revised, corrected versions. We demonstrate clearly the compositional effects of decompression crystallization and discuss carefully how this can be distinguished from mingling. Finally, we consider thoroughly, although ultimately we reject, the suggestion that enrichment of melt inclusions and matrix glasses in K2O can be derived only from slow crystallization. Instead we argue that the overall dataset is best explained as a result of mingling and hybridisation with mafic magmas that are compositionally variable through time.
Volatile contents of melt inclusions and matrix glasses
Devine & Rutherford (2014) firstly called into question our estimates of melt volatile contents for melt inclusions and matrix glasses, as published by Humphreys et al. (2010), and suggested that some of our melt inclusions might actually be matrix glasses. The volatile contents of the earliest erupted magmas at Soufrière Hills Volcano (from 1996) were reported to be up to c. 4.6 wt% H2O based on ‘volatiles by difference’ (VBD) from 100% analytical total by electron microprobe (Devine et al. 1998). This method has an uncertainty of c. 0.6 wt% H2O (Devine et al. 1995; Humphreys et al. 2006b). This result was supported by Fourier transform infrared (FTIR) analyses of six quartz-hosted melt inclusions (Barclay et al. 1998), with H2O contents of 3.52–5.05 wt% and <60 ppm CO2. We note that the Barclay et al. paper (on which both Devine and Rutherford were co-authors) also reports the volatile contents of two plagioclase-hosted melt inclusions, as 4.7 wt% and 0 wt% H2O. In comparison, our VBD estimates of melt H2O contents in pristine, plagioclase-hosted melt inclusions cover a similar range from 0 to 8.2 wt% (Humphreys et al. 2010). Obtaining anhydrous glass electron microprobe analysis data with analytical totals in the region of 100% is consistent with the propagated uncertainty of the electron probe analyses and the VBD method (Devine et al. 1995; Humphreys et al. 2006b); for our dataset this fully propagated uncertainty was typically 1.2 wt%. We ran a suite of hydrous glass secondary standards (Humphreys et al. 2006b) at the same time as the unknowns and these give an average absolute deviation of VBD relative to known H2O of c. 1 wt%. We agree that our highest VBD values are probably an overestimate, but broadly the high inferred H2O concentrations are supported by direct analysis of H2O in a subset of the same pumice-hosted inclusions by secondary ion mass spectrometry (SIMS), which gave 1.14–6.24 wt% H2O (Humphreys et al. 2009b; Fig. 1). Likewise, FTIR measurements by Mann et al. (2013) gave 1.2–6.7 wt% H2O in quartz- and plagioclase-hosted melt inclusions from pumices erupted in Vulcanian explosions during 1997 and 2003–2004. These values are consistent with our earlier measurements. We therefore do not consider our analyses to be a ‘divergence’ from the Devine & Rutherford (2014) dataset. There need be no expectation that the maximum volatile contents of multiple aliquots of magma erupted over 15 years should be identical; melt inclusion datasets from volcanoes globally commonly show variable H2O contents (e.g. Zimmer et al. 2010; Plank et al. 2013). However, the fact that the highest H2O contents were found in samples erupted in January 2007 by Humphreys et al. (2010), and also in samples erupted in August 1997 by Mann et al. (2013) suggests that maximum volatile contents have probably not changed substantially over the course of the eruption. Suites of melt inclusions typically show a wide range of H2O contents, so the very limited data presented in the original work (Devine et al. 1998) are entirely consistent with the more detailed, recent studies.
Comparison of volatile contents estimated by secondary ion mass spectrometry (SIMS) and ‘by difference’ from electron microprobe totals, for melt inclusions from Humphreys et al. (2009b).
Our volatile data for pumice-hosted melt inclusions equate to entrapment pressures of at least 220 MPa, or more if substantial CO2 is present, for which there is good evidence from both gas emissions and melt inclusion data (Edmonds et al. 2014). Thus the melt inclusions define a range of pressures that is consistent with a vertically protracted magma reservoir of the type inferred from seismic tomography and from ground deformation (Elsworth et al. 2008; Voight et al. 2010; Elsworth et al. 2014). We interpreted the range of H2O contents to be the result of differential entrapment pressures of the melt inclusion suite during decompression crystallization, as previously proposed elsewhere including Shiveluch Volcano, Kamchatka (Humphreys et al. 2006a, 2008), Mount St Helens (Blundy & Cashman 2005), Stromboli, Italy (Metrich et al. 2001) and Jorullo, Mexico (Johnson et al. 2008) amongst others. This is consistent with the low volatile contents found for matrix glass (generally less than 0.5 wt% H2O; see below) and with the overall geochemical variations of the melt inclusion and matrix glass data. These data show increasing concentrations of incompatible elements (e.g. K, Ti, Mg) and decreasing concentrations of elements that are compatible in plagioclase (e.g. Ca, Na) with increasing SiO2. We accept that in part this may be related to our application of corrections for post-entrapment crystallization, although this does not affect the key elements of interest (e.g. K, Ti, Fe; see below for further discussion). However, the combined dataset for matrix glasses (which of course are uncorrected) and melt inclusions is consistent with progressive fractionation of plagioclase together with minor pyroxenes + oxides during decompression (see fig. 1 of Humphreys et al. 2010). This interpretation fits with observations of substantial groundmass crystallization of these same phases. The compositions are also generally consistent with the results of phase equilibria experiments of Couch et al. (2003), which resulted in crystallization of plagioclase and pyroxene (see below). We note that Mann et al. (2013) interpreted the variable H2O contents of melt inclusions to reflect hydrogen loss from the melt inclusions during magma stalling in the conduit rather than variable entrapment pressures. However, the co-variation of H2O and Cl measured by SIMS (e.g. Humphreys et al. 2009b) in some pumice-hosted melt inclusions would argue against this possibility, as Cl presumably cannot be lost easily by diffusion through the host phenocryst but is known to degas together with H2O during decompression (e.g. Villemant & Boudon 1999).
We do not dispute that plagioclase-hosted melt inclusions in many Montserrat dome samples (including those dome samples studied by Humphreys et al. 2010) may have substantially or completely lost their water; they are ‘leaked’ melt inclusions and not matrix glasses. This is the reason that we excluded these samples from our study of H2O and Cl degassing (Humphreys et al. 2009b). By comparison with the well-documented similar effect in both natural and experimental olivines (Hauri 2002; Portnyagin et al. 2008; Gaetani et al. 2012) we infer that water loss may result from diffusion of H through the host phenocryst, as demonstrated experimentally by Johnson & Rossman (2013). The extent of diffusive loss probably depends on clast size (Lloyd et al. 2013) as well as the permeability of the dome, and probably occurs during prolonged storage at low pressures (perhaps in the lava dome itself) and at, or close to, magmatic temperatures. We would anticipate that H loss through the host phenocryst may explain anomalous trends in CO2–H2O space that are otherwise attributed to re-equilibration with highly CO2-rich vapours (e.g. Collins et al. 2009, for olivine-hosted inclusions). If this is the case, we would also expect these inclusions to have anomalously oxidizing compositions (e.g. Gaetani et al. 2012; Humphreys et al. in press).
Corrections for post-entrapment crystallization
In the supplementary dataset to our previous study (Humphreys et al. 2010) we provided petrographic information for each melt inclusion in our original dataset (including a description of shape, colour, presence or absence of bubble, textural association in the host phenocryst, area, equivalent radius and volume, area and volume of any post-entrapment crystallization if present, as well as the raw, PEC-adjusted and normalized anhydrous compositions). We now also present representative backscattered SEM images of typical inclusions in Figure 2. We also estimated the amount of post-entrapment crystallization (PEC) in each melt inclusion, and attempted to correct the melt inclusion compositions for the observed PEC. In our explanation of the correction procedure we used the term ‘pristine’ to refer to melt inclusions that were free from post-entrapment crystallization and thus did not need correction. For the other inclusions, our post-entrapment crystallization procedure followed that of Saito et al. (2005), as described in the supplementary information to Humphreys et al. (2010). The method was indeed developed for basaltic melt inclusions but the principle is not changed by the composition of the melt or host phenocryst. For plagioclase-hosted melt inclusions in particular, it is not always clear exactly what host composition is truly in equilibrium with the inclusion as the textures may be complex; precisely defining the volume and surface area of the inclusions may also be very difficult (Humphreys et al. 2008). For this reason, we used a constant plagioclase composition of An40 and made simplifying assumptions about the melt inclusion shape. This may have introduced some uncertainty or error into the corrected dataset, mainly for the key components of the host plagioclase (SiO2, Al2O3, CaO and Na2O; see discussion above), although we also provided the uncorrected compositions for reference. For completeness we show the effect of these corrections in the haplogranite ternary Ab–Or–Q (Fig. 3). However, we argue that for the minor elements of particular interest to this discussion (i.e. FeO, TiO2 and K2O) the PEC correction makes no significant difference.
Backscattered electron SEM images for typical melt inclusions. White arrows mark inclusions analysed. Scale bar represents 100 microns in all cases.
Haplogranite ternary projection showing melt inclusion and matrix glass datasets discussed in the text. Black arrow illustrates effect of 10% post-entrapment crystallization correction. Filled squares, Humphreys et al. (2010) melt inclusions, PEC-corrected. Open circles, Humphreys et al. (2010) matrix glasses. Black filled circles, Devine & Rutherford (2014) melt inclusions and matrix glasses. Black horizontal bars, Mann et al. (2013) melt inclusions. Crosses, Couch et al. (2003) experimental glasses.
Devine & Rutherford (2014) commented on the presence of normative Wo in some corrected melt inclusion compositions. All our compositions plotted were suitable for plotting in the haplogranite projection following Blundy & Cashman (2001), which requires glasses with <20 wt% normative anorthite. All our published glasses have ≤16 wt% normative anorthite but typical values are <<10 wt%. Normative wollastonite is calculated for a minority of compositions but this amounts to an average of 0.85 ± 0.54 wt% normative Wo. This amount of normative Wo is equivalent to a very minor shift in the position of a few glasses within the haplogranite ternary, away from the Qz'' apex (a shift smaller than the size of the symbols in Fig. 3). If anything, this may show that our PEC-corrected compositions have slightly overestimated CaO concentrations, but this does not affect the nature of our interpretations, which are largely based on minor element compositional variations. We would also add that many of the data of Devine & Rutherford (2014) show just as high dispersion in the haplogranite projection as ours (e.g. Devine & Rutherford 2014, figs 19A.6 and 19A.7) and that our data are not unique in having a small amount of normative wollastonite (see also Edmonds et al. 2002; Mann et al. 2013).
Figure 19A.11 of Devine & Rutherford (2014) shows the anomalous composition of the bulk groundmass from Soufrière Hills magma, as recognized by several researchers previously. Although Devine & Rutherford (2014) state that this may be due to analytical artefacts caused by inaccurate raster analysis of inhomogeneous groundmass, there are alternative explanations. One is that the groundmass is contaminated by disaggregation of mafic enclaves and transfer of small crystals into the andesite groundmass, including clinopyroxene and Ca-rich plagioclase (Humphreys et al. 2009a, 2013). This is consistent with the interpretation that the anomalous glass compositions may be related to hybridization (see below; Humphreys et al. 2010; Devine & Rutherford 2014). It has also been suggested that the bulk groundmass composition could be affected by incorporation of substantial additional SiO2 through precipitation of cristobalite in vesicles (Horwell et al. 2013).
K2O enrichment due to slow decompression crystallization or mingling?
The substantive criticism of Devine & Rutherford (2014) about our earlier work is that the variably high K2O contents seen in a subset of our matrix glasses and melt inclusions may be related to slow crystallization of the andesite during ascent and decompression, rather than to hybridization with mafic enclave components as we suggested (Humphreys et al. 2010). Occasional high-K glasses can also be observed in several previous studies of Soufrière Hills Volcano eruptive products and include matrix glasses and melt inclusions in both plagioclase and hornblende (e.g. Edmonds et al. 2002; Harford et al. 2003; and in particular Buckley et al. 2006). Anomalous high-K melt inclusions have also been observed at Colima Volcano, Mexico, where they have been ascribed to recording heterogeneity in the melt caused by the breakdown of amph–bt cumulate nodules (Reubi & Blundy 2008). In our earlier work we considered this process, alongside the breakdown of amphibole during heating or decompression (Buckley et al. 2006), but concluded that this could not fully explain either the compositions or the textural associations of the glasses. We also considered the possibility of entrapment of anomalous boundary layer melts into melt inclusions (Baker 1991) but rejected this on the basis that matrix glasses were also affected.
The suggestion of Devine & Rutherford (2014) that K2O variability in matrix glasses and melt inclusions could be due to variations in magma ascent rate (and thus extent of decompression crystallization) is interesting. This was based on experiments conducted by Hammer & Rutherford (2002), which produced high-K, low-Na glasses through decompression crystallization of plagioclase. In their experiments, isothermal decompression of Pinatubo dacite resulted in groundmass crystallization of plagioclase + quartz, with other phases in minor abundance. K2O is essentially incompatible in these phases, so K2O contents of the experimental matrix glasses increase with increasing groundmass crystallinity, accompanied by decreasing CaO, which is compatible in plagioclase. In those experiments, increasing the dwell time at the final pressure actually resulted in a greater spread of K2O contents, rather than a uniform increase. However, the same trend of increasing K2O and decreasing CaO is also observed, more clearly, in the experimental studies of Couch et al. (2003), Martel & Schmidt (2003) and Brugger & Hammer (2010). The experiments of Martel & Schmidt (2003) are particularly appropriate because their starting material was chosen to have the composition of the most evolved (i.e. highest SiO2 on an anhydrous basis) plagioclase-hosted melt inclusions from Devine et al. (1998), equivalent to melts in equilibrium with phenocryst rims during decompression crystallization (Martel & Schmidt 2003). In all sets of experiments, the very high-K compositions are achieved only at very low final pressures (e.g. <25 MPa, Hammer & Rutherford 2002; 15 MPa, Martel & Schmidt 2003; <20 MPa, Brugger & Hammer 2010), presumably as the melt approaches saturation in K-feldspar. The strong decrease in CaO that accompanies K2O enrichment has important implications for plagioclase–melt thermometry (Humphreys et al. 2014).
Devine & Rutherford (2014) rightly point out that we do not find both high-K and low-K matrix glasses in the same sample. However, we do find both high-K and low-K melt inclusions in the same sample, and in the same crystal. Although our most evolved matrix glass compositions are similar to the very low-pressure, high-K experimental glasses (Figs 4 and 5), it seems clear that slow, low-pressure decompression crystallization cannot explain the enrichment of those compositions in TiO2 (Fig. 4). Specifically, the experiments show that the very high K2O contents can be reached only after extensive crystallization at low pressure, resulting in extremely low CaO and MgO or FeO concentrations; this is not consistent with the spread to higher CaO and FeO at high K2O contents. We therefore conclude that the most likely explanation for the anomalous, high-K melt compositions is still that of hybridization with mafic-derived melt components.
Variation of K2O and TiO2 concentrations in melt inclusions and matrix glasses from Soufrière Hills Volcano, equivalent to figure 2 of Humphreys et al. (2010). Red (= dark grey) filled squares, Humphreys et al. (2010) melt inclusions. Open circles, Humphreys et al. (2010) matrix glasses. Filled diamonds, Humphreys et al. (2010) mafic enclave glasses. Red (= dark grey) filled circles, Devine & Rutherford (2014) mafic enclave glasses. Black filled circles, Devine & Rutherford (2014) matrix glass and melt inclusions. Black horizontal bars, Mann et al. (2013) melt inclusions. Open squares, Mann et al. (2013) mafic enclave glasses. Crosses, Plail et al. (2014) Type A mafic enclave glasses. Asterisks, Plail et al. (2014) Type B mafic enclave glass. Black filled squares, matrix glasses from Harford et al. (2003). Fields with grey shading or dashed outlines show the compositional space of experimental glasses from Martel & Schmidt (2003) and Couch et al. (2003), and the matrix glass and melt inclusion data of Buckley et al. (2006).
Variation of (a) CaO v. K2O and (b) FeO v. K2O concentration in melt inclusions and matrix glasses from Soufrière Hills Volcano. Symbols as in Figure 4. Fields are as in Figure 4; black arrow and numbers denote the effect of decreasing pressure in the Martel & Schmidt (2003) experiments, with approximate equilibration pressures in MPa. Extremely low-pressure crystallization is required to produce residual matrix glasses with strongly enriched K2O contents.
The compositional overlap of high-K melt inclusions and matrix glasses with mafic enclave glasses led us to propose that the anomalous, high-K glasses were derived from disaggregation of mafic material and hybridization with the andesite (Humphreys et al. 2010). This would be consistent with similar observations made on the compositions of microlite crystal populations (Humphreys et al. 2009a, 2013). In our original study, the key observation was that both K2O and TiO2 anomalies are observed in melt inclusions and matrix glasses, and that they were apparently decoupled (see fig. 2 of Humphreys et al. 2010). This decoupling led us to propose our model of diffusive fractionation, based on the facts that (1) K and Ti are enriched in mafic enclave glasses relative to residual liquids in the andesite, generating a chemical gradient, and (2) the diffusivity of K in melt is significantly faster than that of Ti (e.g. Bindeman & Davis 1999; Richter et al. 2003) raising the possibility for fractionation. In the main body of their paper, Devine & Rutherford (2014) supported our interpretation that the variance of glass chemistry was caused by disaggregation and mixing with mafic-derived components, although they based this on FeO contents in addition to TiO2. Their issue with our interpretation seems to arise from the fact that their mafic enclave glasses were not enriched in K2O, whereas ours were (Figs 4⇑–6). We have discussed the possible alternative for K enrichment, that of low-pressure crystallization, above. If this were the controlling influence on mafic glass composition, comparison with experimental studies would imply that the mafic enclaves studied by Devine & Rutherford (2014) were quenched at very high pressure (c. 200 MPa) whereas those studied by Humphreys et al. (2010) quenched over a range of pressures, but below c. 35 MPa; this seems implausible.
Haplogranite ternary showing variation in mafic enclave glass compositions. Symbols as in Figure 4.
An alternative explanation is offered by compositions reported by more recent work (Mann 2010; Plail, 2014; Plail et al. 2014). Plail et al. (2014) identified two texturally and geochemically distinct types of mafic enclaves in Phase 5 of the eruption (rocks erupted during 2009–2010), as well as a third type that was a hybrid of the first two. Type A enclaves are glassy and vesicular, with more mafic compositions, and the framework-forming phase is high-Al amphibole. Type B enclaves are more evolved (less mafic, with higher SiO2) in composition, with higher crystallinity and lower vesicularity; the framework-forming phase is plagioclase and high-Al amphibole is rare to absent (Plail et al. 2014). Type A enclaves are thought to form by rapid thermal equilibration, crystallization and vesiculation of the enclave magma during injection into the andesite, whereas type B enclaves were inferred to have resulted from significant hybridization of enclave magma with the andesite, associated with slower cooling and crystallization (Plail et al. 2014). The two types of mafic enclaves have very different residual matrix glass compositions, even for enclaves erupted in the same andesite magma (Plail et al. 2014). Type A enclaves are characterized by variably high FeO, TiO2 and MgO, whereas type B enclaves have compositions that are more similar to the host andesite matrix glass (Figs 4 and 5). Importantly, the rapidly quenched, primitive Type A enclaves show strong enrichment in K2O, whereas the more slowly crystallized Type B are only slightly enriched (Figs 4 and (5). Figures 4⇑–6 show clearly the disparity between Type A and Type B mafic enclave glasses.
Devine & Rutherford (2014) suggested that the difference between their mafic enclave analyses and ours (see Devine & Rutherford 2014, fig. 19A.10) occurs because the mafic enclaves in our study had undergone slower crystallization in the conduit, resulting in a spread towards K2O enrichment. In fact, the comparison with additional data shows that the Devine & Rutherford (2014) mafic glass compositions are unusual in having particularly high TiO2 and FeO concentrations, yet no K2O enrichment (Figs 4 and 5). The contrast in Type A and Type B enclave glasses also suggests that variations in decompression rate are unlikely to be the cause for the observed compositional differences, as the most K2O-enriched (Type A) enclave glasses are derived from enclaves that quenched rapidly (Plail et al. 2014), not those that crystallized slowly. Based on minor element compositional variations (Figs 4 and 5), we therefore suggest that it is more likely that the Soufrière Hills magma system is regularly fluxed by mafic melts of variable composition. This should be investigated further using trace elements and isotopic compositions, to investigate possible heterogeneity in the magma source regions and thus melt generation processes.
Conclusions
We have thoroughly reviewed the melt inclusion dataset presented by Humphreys et al. (2010) and showed the results to be robust. High volatile contents inferred ‘by difference’ from electron microprobe analysis are supported by direct measurements in multiple independent studies. Plagioclase-hosted melt inclusions in many dome lavas have low H2O contents, which we attribute to diffusive loss through the host mineral. The anomalous K2O enrichment found in a subset of our melt inclusions and matrix glasses is supported by several other independent studies, and can be best explained as a result of mingling and hybridization with mafic magma, as originally proposed. Our corrections for post-entrapment crystallization did not introduce significant artefacts into the dataset that would contradict these conclusions. Additional mafic glass data from the recent literature show that, instead, intruding mafic magmas at Soufrière Hills were probably variable in composition; we therefore recommend further investigation using trace elements and isotopes.
Acknowledgements and Funding
This work benefited from helpful journal reviews by P. Ruprecht, M.-J. Brounce and K. Cashman. M.C.S.H. was supported by a Royal Society University Research Fellowship.
- © 2015 The Author(s)
References
Discussion on ‘Magma storage region processes of the Soufrière Hills Volcano, Montserrat’, Geological Society, London, Memoirs, 39, 361-381
