Insights into the behaviour of S, F, and Cl at Santiaguito Volcano, Guatemala, from apatite and glass

The mineral apatite can incorporate all of the major magmatic volatile species into its structure. Where melt in- clusions are not available, magmaticapatite may therefore represent anopportunity to quantify volatile concentrations in the pre-eruptive melt. We analysed apatites and matrix glasses from andesites and dacites erupted from Santiaguito Volcano, Guatemala, between the 1920s and 2002. X-ray mapping shows complex zoning of sulphur in the apatite grains, but typically with sulphur-rich cores and sulphur-poor rims. Apatite microphenocrysts are enriched in F and depleted in Cl relative to inclusions. Matrix glasses are dacite to rhyolite and contain low F but up to 2400 ppm Cl. Overall, the data are consistent with progressive depletion of Cl in the most evolved melts due to crystallisation and degassing. In the absence of pristine melt inclusions, we used apatite, together with published partitioning data, to reconstruct the likely volatile contents of the pre-eruptive melt, and hence estimate long-term average gas emissions of SO 2 , HF and HCl for the ongoing eruption. The data indicate time-averaged SO 2 emissions of up to 157 tonnes/day, HCl of 74 – 1382 tonnes/day and up to 196 tonnes/day HF. Apatite may provide a useful measure of long-term volatile emissions at volcanoes where direct emissions measurements are unavailable, or for comparison with intermittent gas sampling methods. However, signi ﬁ cant uncertainty remains regarding volatile distribution coef ﬁ cients for apatite, and their variations with temperature and pressure. © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
The exsolution of dissolved magmatic volatiles into bubbles during magma ascent and eruption is one of the most important processes affecting the physical properties of any volcanic system. Whereas H 2 O and CO 2 are the most important volatiles by volume, S, F and Cl can have significant environmental consequences on a local to global scale, with relevance to atmospheric chemistry, human health, and ecology (e.g. Allen et al., 2000;Martin et al., 2009;Robock, 2000). Constraining the fluxes of these volatiles is an important means to assess the current and past impact of volcanic activity on the Earth's surface environment. In the absence of direct measurements of gas emissions, the volatile contents of melt inclusions, trapped in phenocrysts and isolated at depth, are routinely used to infer pre-eruptive melt volatile concentrations (e.g. Bouvier et al., 2008;Edmonds et al., 2001;Humphreys et al., 2008;Wallace, 2005). Comparison of these pre-eruptive volatile concentrations with those preserved in the matrix glass gives a petrologic estimate of volatiles degassed during volcanic eruptions (Devine et al., 1984;Thordarson et al., 1996).
However, in some magmas, melt inclusions may only be present in phases that are liable to leak or degas, or they may be present but too small for analysis, or have undergone devitrification or significant postentrapment modification. In such cases, an alternative method for assessment of pre-eruptive volatile contents is required. Here we explore and evaluate the potential use of apatite in place of melt inclusions, to infer pre-eruptive concentrations of S, F and Cl in the magmatic liquid at Santiaguito volcano, Guatemala, commenting on the advantages and limitations of the method. This work builds on previous studies, for example at Huaynaputina, Peru (Dietterich and de Silva, 2010) and Irazú volcano, Costa Rica (Boyce and Hervig, 2009). We use the data to infer pre-eruptive volatile concentrations in magmas erupted from Santiaguito volcano, and hence estimate the time-averaged gas emissions of this long-lived, but poorly monitored, volcanic dome complex.

Geological background and petrology
Activity at the silicic lava dome complex of Santiaguito, Guatemala, began in 1922 and continues at the time of writing (2015). The dome sits on the shoulder of the much older Santa María volcanic edifice, which in 1902 was the site of a major bimodal explosive eruption, dominated by dacite pumice. Activity at the Santiaguito edifice is Glomerocrysts of plagioclase ± orthopyroxene ± olivine are common and contain large pools of interstitial glass (Fig. 1). These glomerocrysts preserve asymmetry at plagioclase-plagioclase-melt boundaries (Fig. 1d), due to the development of curved plagioclasemelt interfaces, rather than simple impingement textures with planar crystal surfaces. This suggests changes in the differential growth rates between different plagioclase crystallographic axes. These textures are similar to those observed in slowly cooling gabbroic cumulates (Holness et al., 2012) and, by analogy, suggests very slow growth. We therefore infer that these glomerocrysts may represent fragments of disrupted mush that would have gone on to form solid plutonic rocks at depth. Matrix and glomerocryst glass compositions range from 66 to~76 wt% SiO 2 and are similar to the compositions of melt inclusions (64.5-73.5 wt% SiO 2 , Singer et al., 2014).
Thermobarometry based on amphibole phenocryst compositions suggests magma crystallisation temperatures of~940-980°C (± 22°C) at moderately oxidising conditions in the region of NNO + 0.5 to NNO + 2 Singer et al., 2014), and this agrees with observed maximum surface eruption temperatures (850-950°C, Sahetapy-Engel et al., 2004). Fe-Ti oxide compositions from the 1902 eruption give temperatures of 860-885°C for the dacite and 925-1040°C for the andesite (Singer et al., 2014). Petrological and geochemical studies of Santiaguito show that the lavas have become more mafic with time since the eruption recommenced in 1922 (Escobar Wolf et al., 2010;Scott et al., 2013).
Apatite is present in all samples as microphenocrysts and/or as inclusions within phenocryst phases (typically clinopyroxene), indicating early apatite saturation in the melt (Fig. 2). Some crystals are fully included within the host mineral while others are partly open to the matrix (Fig. 2), permitting variable degrees of equilibration with the host melt. The common occurrence of apatites included in pyroxene may be related to synneusis. The inclusions are equant and thus clearly distinct from the acicular quench crystals commonly observed in plagioclase phenocrysts elsewhere (e.g. Bacon, 1986;Wyllie et al., 1962), which are thought to form as a result of growth from a melt boundary  (v). (b) Typical groundmass texture with abundant euhedral microlites of plagioclase, pyroxenes and oxides. (c) Matrix glass (arrowed) can be found as small patches and embayments near the margins of glomerocrysts. (d) Cumulate-type grain boundary textures are found in some plagioclase glomerocrysts. This is manifest as marked asymmetry of plagioclase-plagioclase junctions, resulting in small filaments of feldspar (arrowed; expected grain boundary marked with dashed line) joining adjacent grains of the glomerocryst. This is similar to that observed in gabbros (Holness et al., 2012) and suggests very slow cooling. Dark blebs are partially devitrified melt inclusions. Scale bar is 1 mm in all images except d (100 μm).
layer at the crystal-melt interface. Apatite microphenocrysts are texturally similar to those present as inclusions; their timing of crystallisation relative to the inclusions is unclear but we assume that the microphenocrysts were at least open to significant equilibration with the host melt. The apatites are relatively large, up to 150 μm in length, which is typically significantly larger than groundmass plagioclase, orthopyroxene and titanomagnetite microlites.

Magma supply and fractionation
There is clear evidence of open-system processes at Santa-María -Santiaguito, with a large range of magma compositions erupted, from basaltic andesite to dacite. The deep supply of magma is dominated by hybrid basaltic andesites, fractionating amphibole in the deep crust and assimilating crustal material to form more silicic compositions (Jicha et al., 2010;Singer et al., 2011Singer et al., , 2014. The shallow magmatic system is thought to comprise an elongate, perhaps chemically stratified magma storage region (Rose, 1972;Scott et al., 2013), in which magmas decompress, degas and crystallise. There is clear evidence for mixing of more mafic magmas with the dacites (Singer et al., 2011(Singer et al., , 2014 including reversely zoned plagioclase and the presence of mafic enclaves, as well as plutonic material.

Samples studied
Our dataset comes from analysis of 24 samples from Santiaguito, representing many of the dome and flow units of the complex, and dating from the 1920s to 2002, as reported in Scott et al. (2012, supplementary Scott (2012). We consider in detail the glass dataset of Scott et al. (2013, supplementary table D) together with some new glass analyses and a large dataset of apatite compositions.

Analytical methods
Mineral analyses were obtained by electron probe microanalysis on a four-spectrometer JEOL JXA-8600 electron microprobe in the Research Laboratory for Archaeology and the History of Art, University of Oxford. For apatite, long exposure to the electron beam results in sample damage in the form of volatile migration; this effect is strongly anisotropic and is most significant for halogen analyses conducted parallel to the c-axis of the crystal (e.g. Goldoff et al., 2012;Stock et al., 2015;Stormer et al., 1993). Selection of analytical beam conditions is a trade-off between the accuracy of halogen concentrations (needing a lower accelerating voltage and beam current to minimise electron beam-induced migration) and the precision of heavy and minor element analyses (e.g. Fe, Mn, requiring at least 15 kV accelerating voltage and higher beam currents, Stock et al., 2015). For most analyses, we used relatively short (30s) peak count times for all elements, a 15 kV accelerating voltage and a 15 nA defocused (5 μm) beam, with F, Cl and P analysed first. We found no discernible difference between analyses of grains with different crystallographic orientations, within the uncertainty of our analyses and the variance of the crystal population. We also analysed a subset of analyses using a 10 nA, 5 μm electron beam and these were consistent with the lower-F compositions of those analysed at 15 nA, albeit with slightly larger analytical uncertainties (see Table 1). For these analyses, 120 s peak counting times were used for F, Cl and S. Analyses with totals b 95 wt% were excluded, as were those that did not give good stoichiometric formulae. Wilberforce and Durango apatite, oriented both parallel and perpendicular to the electron beam, were used as secondary standards to check the accuracy of the analyses. These did show slightly higher F contents for crystals oriented with the c-axis parallel to the electron beam, as demonstrated previously (e.g. Stormer et al., 1993). The sulphur peak position was checked prior to analysis and S was calibrated using BaSO 4 . Analytical precision was typically better than 0.2 wt% for F, 0.13 wt% for Cl and 500 ppm for S, and is given in Table 1.
We also performed element mapping on eighteen apatites from five different samples, including microphenocrysts and inclusions in pyroxene, using a JEOL JXA-8800 electron microprobe at the University of Oxford with a 15 kV, 15 nA electron beam. These crystals did not subsequently undergo quantitative analysis. Mapping used WDS for S, Cl, and F, and simultaneous EDS for all other elements (Al, Ca, Fe, K, Na, P, Si, Ti). Resulting images were 256 by 256 pixels, with a count time of~45 microseconds per pixel.
For glasses, we used the existing matrix glass dataset of Scott et al. (2013) (see Table 2). We also analysed a small set of interstitial glasses from the glomerocrysts, using a 15 kV, 6 nA defocused (10 μm) beam, with alkalis analysed first to avoid electron-beam damage (e.g. Devine et al., 1995;Humphreys et al., 2006a). Peak counting times were 90s for F and S, 60s for Cl, 80s for Mg, 12 s for Na and 30s for all other elements. The sulphur peak position was checked prior to analysis and calibrated using BaSO 4 .

Apatite compositions
Apatites from Santiaguito are typically fluorapatite with~0.6-1.5 wt% Cl (Table 1). Minor elements include~0.2-1.2 wt% FeO, 0.1-0.35 wt% MgO, 0.1-0.5 wt% MnO, up to~0.7 wt% SiO 2 , and up tõ 4000 ppm sulphur. There are no significant compositional differences between apatites in andesitic samples and those in the dacites, or between different phases of the eruption. The volatile contents of the   Santiaguito apatites are similar to those of some other subductionrelated systems for which data are available (Fig. 3). Apatite sulphur contents are similar in all textural associations. Those fully included in their host phenocrysts contain on average 1656 ppm S (1 σ 687 ppm; n = 52), while those that are open to the matrix contain 1367 ppm S (1 σ 602 ppm, n = 42) and microphenocrysts contain 1396 ppm S (1 σ 804 ppm; n = 41; Table 1). In other words, the mean S concentrations from the population of apatites in each textural category are separated by less than one standard deviation. However, the data suggest that there are detectable differences in halogen concentrations for apatite in different textural situations (Table 1; see also Supplementary Figure), with the mean values of each apatite category separated by more than 1 standard deviation. Microphenocrysts record higher mean fluorine contents (1.98 wt% F, 1 σ 0.45 wt%) and lower mean chlorine (1.00 wt% Cl, 1 σ 0.11 wt%) than inclusions within phenocrysts (1.50 wt% F, 1 σ 0.46 wt% and 1.19 wt% Cl, 1 σ 0.15 wt%). Median concentrations are slightly lower for the microphenocrysts and inclusions (Table 1), reflecting a spread of a minority of data points to high F contents. Inclusions that are partially open to the matrix tend to record slightly higher mean F (2.25 wt%, 1 σ 0.54 wt%) but similar mean Cl (1.01 wt%, 1 σ 0.18 wt%) to the microphenocrysts. The rims of individual microphenocrysts systematically record lower Cl and higher F than the cores. We estimated OH contents for apatite using stoichiometry and assuming a fully occupied Z site (e.g. Piccoli and Candela, 2002); average calculated OH contents are 0.82 pfu for the apatite inclusions, compared with 0.46 pfu for inclusions open to the matrix and 0.60 pfu for the microphenocrysts (see Table 1). However, propagated OH uncertainties are very high and the assumption of complete stoichiometry may be unrealistic, particularly if significant C is present (e.g. Suetsugu et al., 2000). Element mapping of individual crystals confirms that S, F, and Cl zoning is common even in very small apatites, but demonstrates that the form of the zoning may be quite variable. Apatites containing melt inclusions, or those entrapped adjacent to melt inclusions in the host phenocryst, commonly have patches enriched in F and Cl adjacent to the melt; S contents tend to be unaffected (Fig. 4a). Included apatites may contain sulphur-rich cores but typically do not show significant zoning of halogens. Inclusions open to the matrix also typically contain sulphur-rich cores and may show enrichment in fluorine towards the matrix, but with no equivalent systematic pattern in Cl contents (Fig. 4). Microphenocrysts may have sulphur-rich cores, and typically show F-rich rims (Fig. 4). Some grains show more complex sulphur zoning (Fig. 4).

Matrix glass compositions
The glass analyses presented include matrix glass, glass embayments at the margins of phenocrysts and glass trapped within glomerocrysts. The supplementary data from Scott et al. (2013) are given together with the new data in Table 2. Matrix glasses show a continuous range from 66-80 wt% SiO 2 and follow systematic major element variations. All the glasses show decreasing CaO, Na 2 O and Al 2 O 3 and increasing K 2 O with increasing SiO 2 . The MgO, CaO, FeO and TiO 2 contents of the matrix glasses decrease systematically with increasing bulk rock SiO 2 content, and the least evolved matrix glasses become more Si-rich (Fig. 5). The glasses plot along a systematic trend in the haplogranite ternary, and this has been interpreted as reflecting decompression crystallisation   fig. 11).
Glasses from most individual samples show a clear increase of K 2 O and TiO 2 with increasing SiO 2 , with glasses from the most evolved bulk rocks falling to lower concentrations at N75 wt% SiO 2 , but the overall picture is scattered (Fig. 5). A similar overall pattern is seen for FeO although the degree of scatter is higher and the downturn to lower FeO contents occurs at lower SiO 2 . FeO does not correlate with K 2 O or Al 2 O 3 , but correlates well with MgO (Fig. 5). The matrix glass compositions generally compare well with those of plagioclase-hosted melt inclusions from the 1902 dacite pumice and basaltic andesite scoria (Singer et al., 2014;Fig. 5). The glomerocryst interstitial glasses are similar to the matrix glass, but with slightly lower CaO, slightly higher K 2 O and TiO 2 , and markedly higher FeO (Fig. 5).
Analytical totals are high in all the glasses analysed (Table 2), which suggests low dissolved H 2 O concentrations. The matrix glasses contain up to 2400 ppm Cl, but the majority have much lower concentrations (average 835 ppm, 1 σ 352 ppm, Table 2; Fig. 6). Overall, variations of Cl with SiO 2 show a similar pattern to TiO 2 , with concentrations increasing with fractionation and then dropping towards lower Cl contents in the most evolved glasses (Fig. 6). F concentrations were consistently below detection limits (~0.35 wt% F). Only a few glasses had sulphur contents above detection limit (~135 ppm S); these contained 0.05-0.13 wt% SO 3 (200-520 ppm S). The interstitial glasses from the glomerocrysts typically contain higher Cl concentrations (1780 ± 440 ppm Cl, Fig. 6) but similar sulphur concentrations to the matrix glasses.
These data are more or less consistent with previously reported glass compositions in samples erupted from Santa María-Santiaguito, with the exception of apparently low H 2 O contents in our samples, inferred from the lack of significantly low analytical totals. Villemant et al. (2003)

Interpretation of glass compositional variations
Taken together, the matrix glass compositions exhibit chemical trends that indicate progressive fractionation driven by decompression and degassing; this is consistent with the progressive decrease in H 2 O content seen in plagioclase-hosted melt inclusions (Singer et al., 2014). The increase in MgO, FeO, TiO 2 and K 2 O with SiO 2 in individual samples suggests that fractionation is dominated by plagioclase ± pyroxene, consistent with the observed modal abundance of~75-80% plagioclase within the phenocryst assemblage (Scott et al., 2013). The most silica-rich glasses can only have formed at very low pressures, and the wide range of normative SiO 2 contents seen in the whole dataset is consistent with crystallisation of hydrous magma over a wide pressure range (Blundy and Cashman, 2001). The presence of amphibole phenocrysts in some of the more evolved rocks indicates crystallisation at~150 MPa or more (assuming H 2 O saturation; e.g. Browne and Gardner, 2006). Phase equilibria experiments (Andrews, 2014) suggest that the Santa Maria 1902 dacite was stored at pressures similar to 150-170 MPa (if H 2 O-saturated) at 850°C prior to eruption. Face-value application of the thermobarometer proposed by Ridolfi et al. (2010) indicates crystallisation depths N12 km (Scott et al., 2013), and there is also geochemical evidence of substantial fractionation of amphibole from the magma, seen as a decrease in Dy/Yb with increasing SiO 2 and La/Lu in the whole-rock dataset of Scott et al. (2013). Although pressures predicted by the Ridolfi et al. (2010) barometer are likely to be over-estimated (Andrews, 2014;Erdmann et al., 2014), these data together indicate polybaric crystallisation in the Santa Maria -Santiaguito system. The interpretation of decompression crystallisation is supported by the well-developed groundmass and amphibole breakdown textures ; the high H 2 O contents of melt inclusions (Singer et al., 2014) and successful phase equilibria   (Andrews, 2014); and positive correlations between FeO, TiO 2 and MgO, and CaO and Al 2 O 3 in the glasses, suggesting crystallisation of pyroxene, Fe-Ti oxides and plagioclase. The high FeO and TiO 2 contents of the glomerocryst glass embayments may support the interpretation that some of the erupted crystal load is derived from disaggregated plutonic material that represents the fractionation products of the magma, perhaps prior to saturation of Fe-Ti oxides. The glomerocryst and matrix glasses have very low H 2 O contents compared with melt inclusion compositions reported in previous studies (c.f. Balcone-Boissard et al., 2010). We interpret this as variable diffusional loss of H 2 O during magma ascent, degassing and crystallisation. However, the glomerocryst glasses retain high Cl and F concentrations, probably due to much slower (minimal) diffusion of halogens from these crystal clots during ascent. Overall, the glomerocryst glass compositions suggest that they represent partially re-equilibrated fragments of early-formed crystal mush.

Behaviour of the halogens
With SiO 2 or K 2 O as an index of differentiation, Cl contents of the glasses increase during fractionation from~0.03 wt% to N 0.15 wt% Cl, then decrease after~75 wt% SiO 2 . This indicates incompatible behaviour of Cl until the later stages of crystallisation, when it undergoes exsolution into a fluid phase, consistent with the conclusions of Villemant et al. (2003) based on correlation between H 2 O and Cl contents of glass. Fluorine concentrations of the glasses were below detection limits. However, we can use the volatile contents of apatite to give more information on the evolution of volatiles during magma ascent and crystallisation.
The F/Cl ratio of apatite is dependent on the ratio of halogen fugacities, f HF /f HCl , as well as temperature and pressure (Piccoli and Candela, 1994). Therefore, in the absence of additional information, the cause of the observed compositional change is difficult to determine without ambiguity. The observation of higher F and lower Cl contents in the apatite microphenocrysts relative to inclusions could be produced by decreasing temperature, or by increasing pressure (Doherty et al., 2014;Piccoli and Candela, 2002). Recent work also indicates that Cl partitioning between apatite and melt in the presence of a fluid phase is dependent primarily on melt halogen (Cl) concentration, with a subsidiary dependence on pressure (Cl concentrations in apatite increasing with decreasing pressure; Doherty et al., 2014;Webster et al., 2009). We cannot rule out that decreasing temperature during crystallisation played a role in the changing apatite compositions. However, a number of factors suggest that the changing apatite chemistry is related to compositional variations in the coexisting melt and/or fluid. Firstly, the correlation between H 2 O and Cl in glasses (Villemant et al., 2003), indicates that the magma reached fluid saturation and exsolution resulted in decreasing Cl concentrations in the melt. Loss of Cl from the melt by exsolution of a fluid phase, together with incompatible, nonvolatile behaviour of F (i.e. F does not migrate into the free fluid phase but increases in concentration in the melt) would be consistent with increasing f HF /f HCl during crystallisation and degassing. This would require loss of Cl to occur between crystallisation of the pyroxene or plagioclase phenocrysts (that host the apatite inclusions) and crystallisation of the apatite microphenocrysts.
The Cl depletion of microphenocryst rims relative to microphenocryst cores could also reflect primary variations in melt composition during apatite crystallisation, or re-equilibration of the microphenocrysts with the partially degassed matrix melt. The variability in halogen contents of apatites open to the matrix, and the small-scale zoning observed when inclusions are trapped adjacent to glass, indicates relatively rapid halogen diffusion in apatite. However, Cl zoning in the microphenocrysts is typically more diffuse than that seen in the partially enclosed crystals, and occurs on similar scales in all crystallographic orientations, whereas halogen diffusivities are strongly anisotropic (Brenan, 1994). This suggests that the microphenocrysts do in fact record primary growth zoning and not partial re-equilibration.

Behaviour of sulphur
The population of apatite analyses records a wide spread of sulphur concentrations and there is no statistically significant difference in S content between microphenocrysts and inclusions. However, the X-ray mapping shows that apatite microphenocryst cores are enriched in sulphur relative to the rims, with minor but detectable fluctuations between core and rim in some grains (see Fig. 4). Although earlyformed sulphide inclusions are occasionally found in the cores of magnetite and pyroxene phenocrysts, no sulphides are found in the groundmass and this is consistent with the relatively high measured oxygen fugacity of + 1 b ΔNNO b +2 (Andrews, 2014;Singer et al., 2014). This suggests that sulphur is not compatible in any late-crystallising phase in these magmas, and therefore that the decrease in sulphur contents in apatite rims must indicate either a decrease in sulphur concentration in the melt resulting from degassing, or a decrease in the partition coefficient D S ap-m due to a change in conditions in the magma. The S partition coefficient for apatite depends on oxygen fugacity (Peng et al., 1997), melt sulphur content and temperature Holtz, 2004, 2005). At Santiaguito, we have no evidence for strong changes in magma oxygen fugacity, which is relatively high (estimates range from NNO + 0.5 to NNO + 2, Andrews, 2014;Scott et al., 2012;Singer et al., 2014). Temperature variations may have been important during magma fractionation and ascent, given that temperature estimates for the dacites are 860-885°C but 925-1040°C for the basaltic andesite (Singer et al., 2014). However, D S ap-m increases with decreasing temperature (Parat and Holtz, 2004, see later), so cooling during fractionation would have the effect of increasing the S content of apatite in equilibrium with the melt, resulting in reverse zoning instead of the observed normal zoning. We therefore conclude that for the most part, the volatile contents of apatite at Santiaguito are related to changes in melt volatile concentration and degassing.

Quantification of pre-eruptive volatile contents
We used published apatite-melt partition coefficients to estimate pre-eruptive melt volatile concentrations from the analysed apatite compositions. Apatite inclusions in the phenocryst phases are essentially protected from the external melt environment and should therefore retain a reliable record of their original volatile contents, as long as they are not in contact with any melt pockets (see above; Fig. 4). To determine the volatile concentrations in the melt prior to ascent and degassing, we take representative compositions of apatite inclusions and the cores of microphenocrysts (Table 3). Volatile concentrations in the melt after decompression and immediately prior to extrusion are derived from matrix glass compositions.

Sulphur
We estimated D S ap-m by first calculating the apatite saturation temperature (i.e., the temperature at which apatite appears on the liquidus), following Piccoli and Candela (1994) and Dietterich and De Silva (2010): where AST is the apatite saturation temperature; C AST SiO2 and C AST P2O5 are the weight fractions of SiO 2 and P 2 O 5 in the melt at the point of apatite saturation, and X is the fractional crystallinity of the magma at the point of apatite saturation. In the Santiaguito magmas there are abundant apatite inclusions in phenocryst phases including plagioclase, pyroxene and Fe-Ti oxides, which suggests that apatite saturation occurred relatively early. This is also supported by the lack of significant P 2 O 5 enrichment in the bulk rock compositions (generally b 0.25 wt% P 2 O 5 , see Scott et al., 2013). We therefore assume that C AST SiO2 and C AST P2O5 are equivalent to representative bulk rock compositions (~62-65 wt% SiO 2 and 0.21-0.23 wt% P 2 O 5 , Scott et al., 2013). We also infer from the abundance of apatites included in phenocrysts that X may be rather low, and must certainly be less than~0.3 (the proportion of phenocrysts observed in the samples on eruption, Scott et al., 2012Scott et al., , 2013. This range of parameter values gives a range of calculated AST = 897-987°C (with 'preferred' values in the range 897-963°C, based on the observation of abundant apatite inclusions within phenocrysts including both plagioclase and pyroxene, leading to the assumption that a reasonable upper limit for the magma crystallinity at apatite saturation is X = 0.15, i.e., half of that on eruption). These estimates are at the upper end of (or slightly higher than) temperature estimates for the more evolved magmas (e.g. amphibole-plagioclase geothermometry, 840-950°C, Scott et al., 2012; two-oxide temperatures, 860-885°C, Andrews, 2014;Singer et al., 2014), and at the  (Humphreys et al., 2006b), Huaynaputina, Peru (Dietterich and de Silva, 2010), Monte Vulture, Italy (Liu and Comodi, 1993) and Yerington batholith (Streck and Dilles, 1998). 'Other arc volcanoes' (black field) are as reported in Webster et al. (2009), and comprise Krakatau, Indonesia;Pinatubo, Philippines;Mt St Helens, Washington;Santorini, Greece;Lascar, Chile;and Bishop Tuff, California. Insets highlight in grey the sections shown in the main figures.
We then used the empirical relationship obtained by Peng et al. (1997) for the El Chichón trachyandesite to determine a partition coefficient for sulphur: which gives D S ap-m of 2.4-6.4 for the 'preferred' apatite saturation temperatures (897-963°C). These experimental data were acquired at more oxidising conditions (equivalent to the MnO-Mn 3 O 4 to MH buffers) than the Santiaguito magma, and at more appropriate fO 2 conditions D S ap-m would be slightly reduced, by perhaps a factor of 2 (Peng et al., 1997). In contrast, the strong increase of D S ap-m with decreasing temperature (Peng et al., 1997) (Table 2). This suggests that the apatite compositions are a reasonable reflection of coexisting melt sulphur compositions, at least if temperatures are in the lower part of our range (leading to higher partition coefficients). Many of the apatites show systematic zoning, commonly with sulphur-poor rims. It seems unlikely that this is a result of changing temperature during crystallisation, as this would require significant heating (to 940-1100°C) to crystallise apatite with the lowest observed rim S concentrations (~400 ppm) without any change in melt concentration. It is possible that interaction with mafic magma at depth in the volcanic system could cause such a heating effect; however it seems more likely that the apatites may record syneruptive sulphur loss related to degassing.

Fluorine and chlorine
To determine Cl and F concentrations in the melt we used empirical partition coefficients determined for hydrous silicic melts (65-71 wt% SiO 2 ) in equilibrium with melt and a fluid phase at 200 MPa, 900-924°C and NNO to NNO + 2.1 (Webster et al., 2009). These parameters are a good match for the estimated magma storage conditions at Santiaguito, although true ternary F-Cl-OH exchange coefficients would be more strictly appropriate than apparent partition coefficients. The data of Webster et al. (2009) show that X F ap increases with increasing F and decreasing Cl concentration in the melt; X F /X Cl (ap) increases linearly with X F /X Cl (m), with X F /X Cl (ap) ranging from 0.26-14.9 (Webster et al., 2009). Their experiments were run at higher Cl contents than the Santiaguito glass, so we only used data from the less Cl-rich experiments that resulted in apatite with ≤ 2 wt% Cl.  (Table 3). This range is consistent with the low measured glass F compositions (lower than the detection limit for at~0.35 wt% F) and suggests that there has not been significant degassing of F during magma ascent and crystallisation. In contrast, melt inclusions have 700-1600 ppm Cl (Balcone-Boissard et al., 2010), which is substantially lower than the concentrations predicted from our apatite compositions; our matrix glasses analyses show even lower Cl contents (Table 2). This result is similar to that of previous studies (e.g. Boyce and Hervig, 2009;Webster et al., 2009), which also found anomalously high apatitebased estimates for pre-eruptive melt Cl (but not F) concentrations when compared to melt inclusions. It has been suggested previously that a discrepancy between melt inclusion Cl contents and those calculated from apatite could be due to exsolution of a low-density aqueous vapour from a higher density single-phase fluid coexisting with the magma during ascent (Webster et al., 2009). Subsequent segregation of the low density vapour would result in increasing salinity of the remaining saline fluid and reequilibration of apatite to more Cl-rich compositions (Webster et al., 2009). There is no direct evidence of the presence of a high density saline fluid at Santiaguito, although the melt Cl contents predicted from apatite may be approaching the concentration at which a dense fluid could become stable (Signorelli and Carroll, 2001;Webster, 2004). However, the process described by Webster et al. (2009) should result in partial equilibration of the larger apatite grains to leave Cl-rich rims, whereas Cl-poor rims are observed. It is unlikely that the discrepancy between predicted and observed melt Cl concentrations is the result of apatite growing at volatile-undersaturated conditions, because previous melt inclusion studies demonstrate Cl-H 2 O loss during magma ascent (Singer et al., 2014;Villemant et al., 2003). We can also rule out early crystallisation in a higher temperature, less fractionated melt, as this would result in a lower D Cl ap-m (Webster et al., 2009) and hence higher melt Cl contents for a given apatite composition. This leaves the most obvious explanation for the low Cl contents of the matrix glasses being that the matrix has substantially degassed, resulting in loss of Cl into the vapour phase during ascent. This is supported by the covariation of Cl and H 2 O in melt inclusions and residual matrix glasses of Plinian clasts (Villemant et al., 2003).

The 'petrologic method' using apatite
The pre-eruptive dissolved volatile concentrations predicted from apatite can be used to estimate the flux of SO 2 , HF, and HCl from Santiaguito, in the same way that the 'petrological method' is commonly used with melt inclusions (e.g. Dietterich and de Silva, 2010;Thordarson et al., 1996;Wallace, 2005). First, the amount of volatiles degassed is constrained using the difference between apatite-based estimates of pre-eruptive melt volatile concentrations (0.29-1.19 wt% Cl, 0.03-0.17 wt% F, 218-676 ppm S, Table 3) and matrix glass volatile concentrations. We use the upper estimates of   (Ebmeier et al., 2012;Harris et al., 2003;Scott et al., 2013); using a typical dacite magma density of 2500 kg m −3 gives a total erupted mass of 2.75 × 10 12 -5 × 10 12 kg. Assuming a mean phenocryst content of 30% , this equates to 1.9-3.5 × 10 12 kg melt. Thus the total mass of volatiles emitted since 1922 is up to 2.8 × 10 9 kg S, 2.4 × 10 9 -4.5 × 10 10 kg Cl and up to 6.3 × 10 9 kg F (Table 3).
This suggests time-averaged SO 2 emissions of up to 157 tonnes/ day (Table 3), which is similar to previous estimates based on sporadic field measurements (between 20 and 960 tonnes/day, see Table 4, Andres et al., 1993;Holland et al., 2011;Rodriguez et al., 2004). The same method suggests time-averaged estimates of 74-1380 tonnes/day HCl and up to 196 tonnes/day HF (Table 3). These results give gas species ratios of HF/SO 2~1 .25, and HCl/SO 2~0 .5-8.8. Because the melt Cl concentrations calculated from apatite are rather high compared with melt inclusions (see earlier), we consider that the lower HCl flux values are probably more reliable. There are no published field-based estimates of Santiaguito's halogen emissions, so we are unable to compare this with independent constraints on HCl flux from the volcano. It is not trivial to compare long-term petrological estimates with spot measurements of gas emissions at any individual volcano, primarily because gas fluxes may be highly variable in time, depending on the level of volcanic activity. For example, the HCl flux (and consequently the HCl/SO 2 ratio) at arc volcanoes is typically related to direct magma extrusion and falls to very low levels during periods of non-extrusion (e.g. Edmonds et al., 2001). Spot field measurements for SO 2 at Santiaguito are highly variable in time and also appear to depend on whether there is active extrusion at the lava dome Holland et al., 2011;Rodriguez et al., 2004).
A well-monitored volcanic system that also shows long-term domebuilding activity is Soufrière Hills Volcano, Montserrat. Cl is degassed from the andesite magma during extrusion but sulphur is mostly supplied by deeper degassing of unerupted mafic magma (e.g. Edmonds et al., 2001Edmonds et al., , 2010. Soufrière Hills Volcano has emitted approximately 4.0 ± 0.6 × 10 9 kg sulphur during the course of the prolonged 1995-2011 dome-forming eruption, including both SO 2 and H 2 S (Edmonds et al., 2014). Similarly to Santiaguito, SO 2 emission rates have been highly variable during the eruption (e.g. 42 to N1900 tonnes/day during 1996-1997, Young et al., 1998, with long-term time-averaged SO 2 emission rates~600 tonnes/day SO 2 Edmonds et al., 2014), approximately twice the upper estimate for the SO 2 flux emitted at Santiaguito (see Table 4). At Soufrière Hills Volcano, the HCl/SO 2 ratio is b 0.3 during pauses, N1 (up to~10) during active extrusion Edmonds et al., 2014), with HCl emission rates of N400 tonnes/day during dome-building and b80 tonnes/ day during pauses in extrusion (Edmonds et al., 2002). The inferred HCl flux at Santiaguito is therefore in line with that observed during dome growth at Soufrière Hills. However, the HCl/SO 2 ratios at Santiaguito extend to higher values than Soufrière Hills Volcano. One explanation for this is the apparently substantially lower SO 2 fluxes at Santiaguito. This may reflect differences in the details of the deep plumbing system (a substantial proportion of the SO 2 supply at Soufrière Hills is contributed by unerupted mafic magma, whereas the long-term petrologic estimates for Santiaguito consider only sulphur degassed from the magma that is erupted; alternatively mafic magma  Fig. 5), increasing and then decreasing after~75 wt% SiO 2 . Error bars shows ±1 sigma analytical uncertainty on the analyses; horizontal line represents the detection limit (~300 ppm Cl). Two grey ovals represent the range in composition of matrix glasses and melt inclusions from the 1902 Plinian eruption (dark grey) and later dome rocks (light grey) as measured by electron microprobe (Balcone-Boissard et al., 2010;Villemant et al., 2003).  Table 2). ¶ Average detected matrix glass concentrations used for Cl, Table 2.
supply rates at Santiaguito may differ from those at Soufrière Hills). The mismatch between apatite and melt inclusion Cl contents is another source of uncertainty here.

Apatite: its potential for tracking volcanic degassing
There is considerable potential for using apatite to infer magmatic volatile contents and time-averaged gas emissions for S, F, and Cl, as an alternative to melt inclusion-based methods (e.g. Huaynaputina, Peru, Dietterich and de Silva, 2010). This may prove particularly useful when direct emissions measurements are unavailable, for historic and prehistoric eruptions, and for comparison with intermittent gas sampling methods, which are typically highly variable in time. However, there still remain significant problems associated with using apatite to infer magmatic volatile concentrations, not least in estimating the point at which apatite started to crystallise. One of the most significant problems is uncertainty over the presence and composition of any fluid(s) coexisting with the melt, coupled with a lack of constraints on the ultimate fate of a brine phase, if present. This uncertainty still exists for melt inclusion studies, but for apatite the problem is trickier because of the strong dependence of apatite halogen contents on fluid composition (Webster et al., 2009). More focus is required on demonstrating possible fluid immiscibility, including documenting the presence of multiple fluid bubbles in melt inclusions, as well as the composition of fluid inclusions. Melt inclusions are useful to constrain whether there has been volatile saturation in the melt.
Additional problems arise when there is also substantial variability in the compositions of apatite. Apatite inclusions and microphenocrysts presented here show considerable compositional variability, with 1 relative standard deviation~12-15% for Cl, 25-30% for F and 40-50% for S. For the most part this variability appears to be real, although improved precision (e.g. by use of ion microprobe techniques) would be helpful, as would direct analysis of OH for accurate determination of Cl/OH and F/OH ratios. Significant variability between grains means that it is difficult to demonstrate that apatites represent equilibrium compositions, as well as to determine which values are truly representative of pre-eruptive magmatic conditions, and at what conditions. Syn-eruptive diffusive equilibration of microphenocrysts with degassed matrix glass can, in principle, be distinguished from crystallisation effects by considering the lengthscale and anisotropy of compositional gradients. Accurate and precise knowledge of magmatic conditions (fO 2 , pressure and temperature) is required for sensible choice of partition coefficients, and thermodynamic calculations may help in this regard. Finally, application of apatite as a tracer of magmatic volatiles would be enhanced by knowledge of partitioning characteristics of all the volatiles (including OH and C), as well as direct determination of these elements in both apatite and melt.

Conclusions
The eruption of Santiaguito volcano, Guatemala, is highly active and amongst the longest-lived of its kind in the world. However, due to its location, climate and the surrounding terrain there are few constraints on gas emissions from this volcano. Apatite in Santiaguito lavas retains evidence of volatile zoning, recording loss of sulphur and chlorine between early entrapment of inclusions and crystallisation of microphenocryst rims. This is likely related to degassing of Cl and S from the magma together with aqueous vapour. Pre-eruptive melt volatile concentrations were determined from the apatite compositions using published partition coefficients. These were used, together with matrix glass compositions, to derive time-averaged estimates of SO 2 , HF, and HCl fluxes from Santiaguito. These results indicate time-averaged fluxes of up to 157 tonnes/day SO 2 , up to 196 tonnes/day HF, and 74-1380 tonnes/day HCl. Estimated ratios are HF/SO 2~1 .25, and HCl/SO 2 0.5-8.8. These fluxes are in line with estimates from other arc volcanoes; however the Table 4 Estimated volatile flux from Santiaguito, based on apatite analyses (this study), with directly measured flux from other volcanoes for comparison.
Halogen/ SO 2 ratios for Santiaguito were calculated from the petrological data as described in the text.  N400-6000 Allen et al. (2000), Edmonds et al. (2001Edmonds et al. ( , 2002, Young et al. (1998) Sakurajima, Japan Dome-building 0.009-0.015 0.14-0.33 Mori and Notsu (2003) uncertainties are large and additional work is needed to constrain volatile exchange coefficients between apatite and melt ± fluid(s), including in volatile-undersaturated systems, as well as direct analysis of OH in both apatite and melt. Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.lithos.2015.07.004.