Edinburgh Research Explorer Integrated constraints on explosive eruption intensification at Santiaguito dome complex, Guatemala

explosion intensiﬁcation. Seismic and infrasound signals reveal progressively longer repose intervals between explosions and deeper fragmentation levels as the seismic energy of these events increased by up to four orders of magnitude. Evidence from geothermobarometry, bulk geochemistry and groundmass microlite textures reveal that the onset of large explosions was concordant with a relatively fast ascent of a deeper-sourced ( ∼ 17–24 km), higher temperature ( ∼ 960–1020 ◦ C) and relatively volatile-rich magma compared to the previous erupted lavas, which stalled at ∼ 2 km depth and mingled with the left-over mush that resided beneath the pre-2015 lava dome. We interpret that purging driven by the injection of this deep-sourced magma disrupted the long-term activity, driving a transition from low energy shallow shear-driven fragmentation, to high energy deeper overpressure-driven fragmentation that excavated signiﬁcant portions of the conduit and intensiﬁed local volcanic hazards. Our ﬁndings demonstrate the value of multi-parametric approaches for understanding volcanic processes and the triggers for enigmatic shifts in eruption style, with the detection of vicissitudes in both monitoring signals and petrological signatures of the eruptive products proving paramount.


Introduction
Subduction-related volcanic systems often exhibit transitions in eruption style over short (minutes to days) and long (months to years) timescales. Shifts in volcanic activity are commonly attributed to a switch from open-system to partially closed-system degassing (e.g., Fink et al., 1992), regulated by a combination of deep (e.g., magma injection, deep convection, mush instabilities and volatile fluctuations; Murphy et al., 2000;Williamson et al., 2010;Burgisser and Bergantz, 2011) and shallow (e.g., volatile ex- transitions in eruptive behaviour (e.g., Wadge et al., 2014;Stock et al., 2018).
Reconstructing the architecture of magma plumbing systems, commonly estimated via experimentally-calibrated geothermobarometers (e.g., Putirka, 2008) and petrological changes as magma ascends to the surface, can provide key insights to factors controlling eruption style. During magma ascent, decompression drives the exsolution of volatiles from the melt phase, which generates an effective undercooling ( T , defined as the difference in temperature between that of the melt and liquidus, Swanson, 1977) that induces microlite crystallisation; thus groundmass textures of eruptive products reflect their ascent history (e.g., Cashman and Blundy, 2000;Cassidy et al., 2015;Preece et al., 2016). Microlite crystallisation rapidly increases magma viscosity, inhibiting flow in the upper conduit leading to plug formation (e.g., Sparks, 1997). The outgassing efficiency through such a plug determines the development of pore pressure and explosive capacity in the magmatic column (e.g., Burgisser et al., 2010). If the rate of pore pressure increase induces strain rates that exceed the relaxation rate of magma it can cause rupture and generate an explosion (Dingwell, 1996); alternatively, if strain rates are lower than the rate of relaxation then the lava will erupt passively, either as a dome, a flow or spine (e.g., Melnik and Sparks, 2002).
Santiaguito-Santa María is an active dacitic-andesitic dome complex in south-west Guatemala ( Fig. 1a-b). The complex formed following the 1902 cataclysmic eruption of Santa María leaving a 0.5 km 3 explosion crater on its SW flank. After 20 years of quiescence (in 1922) extrusion began in the centre of the explosion crater (Fig. 1c) and eruptive activity has been continuous for the past century from four E-W trending vents (Rhodes et al., 2018): Caliente (active during 1922-39 and 1967-present), La Mitad (1939, El Monje (1949-58) andEl Brujo (1958-86). Activity over the last 50 years has remained at the eastern Caliente vent (Rose, 1972). In recent decades, eruptive activity has been characterised by slow extrusion of blocky lava (Harris et al., 2003) and regular, small-to-moderate gas-and-ash explosions (2-3 per hour) with plumes reaching 0.5-2.0 km above the vent De Angelis et al., 2016). However, from December 2014, explosion frequency gradually decreased from >30 per day to <10 per day by July 2015 while seismic energy of the explosions increased . This heightened explosive activity generated ash-rich plumes reaching 7 km above sea level (a.s.l.) and deeply excavated the crater by mid-2016, before returning to Santiaguito's regular explosive activity and renewed lava extrusion in late-2016 .
Petrological and geochemical assessments of extruded lava between 1922 and 2002 constrained a magma storage region residing between ∼12 and ∼24 km depth at temperatures ∼940-980 • C (Scott et al., 2012), with no definitive evidence of a shallower reservoir. The composition of Santiaguito lavas have shown a progressive decrease of ∼4 wt.% SiO 2 (∼66 to 62 wt.%; Rose, 1972;Harris et al., 2003;Avard and Whittington, 2012;Scott et al., 2013), leading to models of a vertically-stratified and thermally zoned storage region at mid-crustal levels grading from basaltic-andesite at depth to dacitic in the shallow region (Scott et al., 2013;Singer et al., 2014). Decreasing overall extrusion rates, taking place in 7-15 year cycles (nine documented; Harris et al., 2003;Rhodes et al., 2018), have led to a hypothesis that the magma source beneath Santiaguito is depressurising, and may completely exhaust between 2014 and 2024 (Harris et al., 2003). However, the eruption intensification in 2015-2016 raises questions on the current state of magma storage and transport, and thus associated hazards (Rhodes et al., 2018;Lamb et al., 2019).
Eruption intensifications at other systems have been attributed to deep magmatic process, particularly the injection of new magma (e.g., Murphy et al., 2000), fluctuations in volatile content (e.g., Williamson et al., 2010), changes in ascent style and rate (e.g., Cassidy et al., 2015) or cyclic pressure accumulation and release within a high-viscosity plug (e.g., Voight et al., 1999). Additionally, a change in magma properties in the source region (e.g., due to assimilation, differentiation or magma mixing) can lead to multiscale transitions between different eruptive regimes (e.g., Melnik and Sparks, 2002). Thus, identifying the cause of eruptive transitions requires an understanding of the magma plumbing system from source to surface. In this study, we integrate petrological and geochemical analyses of tephra deposits, generated from explosions in 2015-2016, with monitored seismic, infrasound and thermal signals to characterise a phase of explosion intensification at the Santiaguito dome complex. We examine pre-eruptive conduit processes, along with magma storage condition estimates, that build upon the current models for eruptive activity at Santiaguito and investigate potential trigger mechanisms driving explosions in 2015-2016.

Seismic and acoustic infrasound
A geophysical network, consisting of 11 seismometers (six broadband and five short-period) and five acoustic infrasound microphones located around the dome complex, was deployed between November 2014 and May 2017 ( Fig. 1b; see Lamb et al., 2019). A waveform characterisation algorithm, that matches the seismic waveforms from each event with triggered signals (using infrasound waveform shape, amplitude and frequency), was used to detect individual explosions (Bueno et al., 2019). This approach provided a means for tracking the evolution of intensity and magnitude of explosions across the entire period of deployment, recording over 6000 explosion events. The energy released by each explosion was calculated from the corresponding seismic waveform, following Johnson and Aster (2005) as: where r is source-to-receiver distance, ρ earth is rock density, c earth is seismic velocity (2150 m s −1 ), A is attenuation, S is seismic site response, and U is the particle velocity. The integral is evaluated over the duration of the event.
Integrating seismic signals with infrasound observations can be used to gain insights into explosion source mechanisms. In particular, the time difference between the two signals can be used as a proxy for explosion depths in the conduit (e.g., Hagerty et al., 2000), thus seismic-infrasound time arrivals were also assessed across the 2014-2017 period.

Thermal radiance
Thermal infrared satellite data were obtained via the MOD-VOLC algorithm (http://modis .higp .hawaii .edu) to assess evidence for surficial temperature changes throughout the 2014-2017 period. MODVOLC uses the radiance emitted to estimate the radiant heat flux from the surface (in W) following the method of Wright et al. (2016). All data were collected from night-time due to high cloud coverage during day-time, thus any day-time explosions were not identified. It is important to note that not all thermal anomalies are explosion derived, but increased radiance may be due to, for example, increased extrusion of hotter lava at the surface.

Eruptive products and sampling
Ash and bombs generated from explosions in 2015-2016 were periodically sampled during multiple field campaigns over a 6month period between December 2015 and June 2016. Ash was also collected from a single explosion in January 2018. Ash (∼100-200 g) was sampled directly from the fallout of individual explosions using either high-sided ash collectors or laying clean canvas sheets on the ground prior to eruptive activity to avoid external contamination. Fresh airfall bombs (∼150 mm in diameter) were also collected from the base of Caliente following the largest explosions in 2016 (two from 7th February and one from 21st April). A total of 11 ash samples were collected at both proximal and distal locations (0.1-10.0 km away) from the active El Caliente vent ( Fig. 1; Table 1); proximal and distal ash showed no systematic difference in ash characteristics (e.g., componentry). For comparison with the 2015-2016 products, ash samples from small-to-moderate, regular explosions in January 1968, January 2003, November 2014 and January 2018 were also analysed in order to relate petrological and textural characteristics to the concurrent eruption dynamics.
The ash was first dried in a furnace at 70 • C for 6 hours to remove any residual moisture and then dry sieved into six grain size fractions (1 mm, 0.5 mm, 0.25 mm, 0.125 mm, 0.063 mm and <0.063 mm). Componentry was carried out to quantify the abundance of different clast types in the 0.25-0.5 mm size fraction using a binocular microscope, counting >300 grains per sample for statistical representation. Bombs were thin sectioned and ash particles were set in resin, polished and carbon coated for textural and chemical analysis.

Mineralogical and geochemical analysis
Mineralogy and phase abundances were quantified using QEM-SCAN, an automated SEM-EDS (Scanning Electron Microscopy-Energy Dispersive X-ray Spectroscopy) system, as described by Hornby et al. (2019a). Analyses were performed using 15 kV accelerating voltage and a ∼5 nA beam current, with the beam rastering along the sample at 2 μm intervals to generate a quantitative colour-coded phase distribution map for each ash and bomb sample. Bulk X-ray powder diffraction (XRD) analysis was performed on three ash samples (from 31st December 2015, 21st April 2016 and 19th June 2016) to independently verify the mineral assemblage identified in QEMSCAN. Analyses were performed on a Bruker D8 Advance powder diffractometer at the University of Leicester with Bragg-Brentano theta-theta geometry using Cu K α radiation and a LynxEye position sensitive detector (PSD). The start angle was 4.0 degrees, a step size of 0.010309 degrees and a step time of 0.5 seconds.
Bulk-rock major and trace element compositions were measured by X-ray Fluorescence (XRF) for all ash and bomb samples using a PANalytical Axios Advanced XRF spectrometer at the University of Leicester. Dry, raw ash samples (i.e., before sieving) and bombs were crushed to a fine powder in a planetary ball mill (agate) for 20 minutes. Vesicular and dense domains from the April 2016 bomb were carefully segregated and analysed separately. Major elements were measured on glass beads fused from ignited powders, and trace elements were measured using pressed powder pellets. Results are reported as component oxide weight percent and have been recalculated to include loss on ignition (LOI). Relative precisions and accuracies are better than 1-2% for all major elements and better than 5% for trace elements based on repeat analyses of international reference materials (BH-1 microgranodiorite, JR-1 rhyolite and BCR-1 basalt).
Chemical composition of crystals and interstitial glass were determined using a Cameca SX100 electron microprobe (EPMA) at Ludwig-Maximilians-Universät München. Crystal analyses were performed using a 15 kV accelerating voltage, 20 nA beam current, and a focused (∼1 μm) beam; all analyses on glass used a 5 nA beam current and a defocussed 10 μm beam. Calibration of the spectrometers was performed using a range of silicate and oxide standards (e.g., albite, wollastonite), and during data acquisition Na peaks were counted first to avoid alkali migration during the analyses. To ensure accuracy and precision, eight working standards Table 1 Bulk-rock chemical composition of major and trace elements for ash and bombs collected between 2015 and 2016, along with eruption date and collection location. Sample locations include El Caliente (El), VIP camp (VC), El Brujo (EB), La Mitad (LM), El Palmar (EP), Monte Bello Farm (MBF), Llano de Pinal (LdP) and OVSAN (see Fig. 1b). from the Smithsonian collection (incl. VG-568 rhyolitic glass and Kakanui hornblende) were measured at regular intervals throughout the analyses. All elements had standard deviations <0.5 and relative accuracies were better than 3% for major elements and 20% for minor elements. Multiple glass measurements were taken from the same sample to check consistency with totals and composition (outlying data with totals outside 97-101 wt.% were excluded from the results).

Microtextural analysis
Groundmass microlite textures in the juvenile ash and bomb samples were analysed using a Hitachi TM3000 scanning electron microscope (SEM) operating in backscattered electron (BSE) mode at 15 kV accelerating voltage and 10 mm working distance. The vesicular and dense domains within the April 2016 bomb were analysed separately for comparison. Plagioclase microlites (defined as crystals <100 μm 2 ) were manually outlined using BSE images of the groundmass within a ∼20,000 μm 2 area in Adobe Illustrator.
The digitised images were imported to ImageJ for crystal measurements using the best-fit ellipse tool (Schneider et al., 2012), including the number of crystals per area, total area % of crystals, individual and mean crystal size, and crystal dimensions (long and short axes). From this data, microlite area %, plagioclase microlite fraction (ϕ), and microlite number density (N A , mm −2 ) were calculated following the method of Hammer et al. (1999). Crystal shapes were determined using CSDSlice (Morgan and Jerram, 2006), which objectively converts 2D measurements into known 3D shapes to obtain a short (S), intermediate (I) and long (L) axes.
CSDCorrections software (Higgins, 2000) was used to provide 3D crystal size distributions (CSDs) by stereological conversion of the 2D crystal intersections measured in ImageJ. For all samples, the rock fabric was set to massive, crystal shape values (S:I:L) were derived from CSDslice and crystal roundness was estimated at 0.1 (based on a 0-1 scale where 0 is rectangular and 1 is an ellipsoid). Data binning was set at five per decade so that each bin was 1.6 times the size of the next smallest one. From the CSD data, volumetric plagioclase number densities were calculated us- where α is the slope of a straight segment of a CSD curve (in mm −1 ) and n o is the y-intercept (Blundy and Cashman, 2008). The CSDs can also be used to estimate microlite nucleation rates ( J ) and growth rates (G) if crystallisation times (τ ) are known (Marsh, 1988): α = −1/Gτ and J = n o G. As τ at Santiaguito are unknown, maximum and minimum crystallisation times were estimated from the time difference between the date of the explosion and the beginning of phase 1 (i.e., a τ of months) and repose times between explosions (i.e., a τ of minutes-hours), respectively. As the geophysical network did not extend further back than November 2014, a maximum crystallisation time for microlites in the November 2014 ash was estimated as the onset of lava extrusion in May 2014. On 9th May 2014, Santiaguito's eruptive activity was punctuated by the partial collapse of Caliente's eastern flank, which was shortly followed by the extrusion of a blocky lava flow that continued to descend the newly formed collapse scar until December 2014(Global Volcanism Program, 2015. Lava extrusion was interspersed with regular (>30 per day), small-to-moderate ashpoor explosions (<1 km above the vent) emerging from an infilled crater (Fig. 2b), consistent with the multi-decadal preceeding activity Sahetapy-Engel et al., 2008;Johnson et al., 2014;De Angelis et al., 2016;Lamb et al., 2019). Monitoring signals from seismic and infrasound stations for a number of selected explosion events in late-2014 show consistent arrival-time delays of up to 6 seconds ( Fig. 2h). Regular, weak thermal anomalies were recorded during this period (Fig. 2g).

2014-2017 eruption chronology and associated monitoring signals
From February to September 2015 (phase 1), explosions became progressively less frequent (<10 per day; Fig. 2g) forming darker, ash-enriched plumes and occasionally pyroclastic flows (Global Volcanism Program, 2015), while repose intervals and seismic energies increased synchronously; seismic-infrasound time delays remained consistent to those in pre-2015. Phase 1 was also marked by a sharp reduction in thermal anomalies, although when detected heat flux values remained low (Fig. 2g).
Phase 2 was marked by intense explosions from September 2015 through February 2016, which generated ballistics and pyroclastic flows along with dense ash plumes (up to 7 km a.s.l. with ash travelling laterally >20 km), interspersed by smaller explosions (Global Volcanism Program, 2016a). Repose intervals during this phase often increased to ∼12 hours and explosions became more energetic (Fig. 2g). Seismic-infrasound time arrival differences and the frequency of thermal anomalies remained similar to phase 1, although occasionally high heat fluxes were detected ( Fig. 2g-h).
Explosions in April and May 2016 (phase 3) were the largest reported since the early 1990s (Global Volcanism Program, 2016b). Seismic energies for these explosions were up to four orders of magnitude greater than explosions pre-2015 ( Fig. 2g), which generated ash-rich plumes several km above the vent and excavated the summit crater to ∼300 m wide and ∼175 m deep (Fig. 2d).
Metre-sized blocks and bombs were found up to 3 km from the Caliente vent. Similar, irregular explosive activity continued until October 2016 ( Fig. 2e), accompanied by infrequent pyroclastic flows. Satellite thermal anomalies remained infrequent, although large spikes in the thermal heat flux were recorded, with the largest in mid-April which coincided with explosions that excavated the crater. Comparison of the seismic-infrasound time arrival differences with phase 1 and 2 revealed overlapping delays (∼6 seconds); however, some explosions during this high-energy phase had delay times up to 17 seconds (Fig. 2h).
No large explosions were reported from October 2016 onwards (phase 4), and activity gradually returned to regular, small-tomoderate explosions. Contemporaneously, lava began extruding in the excavated Caliente summit crater (Fig. 2f), which was completely filled by March 2017 and began to overflow, generating block-and-ash flows (Global Volcanism Program, 2017). Interestingly, no thermal anomalies were observed during this time. By late-2017 ash plumes rarely exceeded 1 km above the vent (Global Volcanism Program, 2018).
Annual SO 2 emissions during 2014-2017 were obtained from Carn et al. (2017), measured by the Ozone Monitoring Instrument (OMI) on NASA's Aura satellite. In 2014, an SO 2 flux of 45 kt yr −1 (123 t d −1 ) was recorded from OMI satellite (Carn et al., 2017), consistent with field-based measurements from 2008-2009(Holland et al., 2011. This increased to 82 kt yr −1 (225 t d −1 ) and 85 kt yr −1 (233 t d −1 ) in 2015 and 2016, respectively, coinciding with phases 1, 2 and 3 when explosive activity intensified. Annual The satellite radiant heat flux data (in green and right vertical axis), as derived from MODVOLC, overlays the seismic data (g). The coloured arrows located above the seismic time sequence in (g) indicate eruption dates for the samples analysed: lava flow (yellow), ash (red), bombs (green). (h) Comparison of selected seismic and acoustic infrasound signals displaying larger delays in arrival times for explosion events in phase 3. SO 2 flux in 2017 increased significantly to 184 kt yr −1 (504 t d −1 ), corresponding to renewed lava extrusion in phase 4.

Ash componentry and bomb textures
Ash erupted from explosion events during phases 2 and 3 of the 2014-2017 eruptive period comprised eight clast types: dense brown (DB), transparent (TC), dense grey, white vesicular, dull lithic, hydrothermally altered, and single felsic and mafic crystals. DB clasts have a brown, glossy surface texture that is glass-rich, has angular edges and low vesicularity (Fig. 3a). BSE images of DB clasts contain a varying abundance of mafic microlites in the groundmass (Fig. 3b). TC clasts are light grey to transparent with a glossy surface texture (i.e., glass-rich) and initially appear poorlyvesicular, yet bubble walls are often observed at the edges of TC fragments (Fig. 3c). Unlike the DB clasts, the TC clasts consistently contain a lower abundance of mafic microlites in the groundmass (Fig. 3d). As DB and TC clasts are glass-rich, we refer to these as the juvenile components. The dense grey, dull lithic and white vesicular clasts reveal no fresh glass, which has degraded to SiO 2 phases (Fig. A.1), and thus are regarded as non-juvenile.
Quantification of the temporal changes in the ash components are displayed in Fig. 3g. Ash from explosions during phase 2 were dominated by DB clasts (∼60-70%), which switched to TC dominated (up to 60%) during the highest energy explosions in phase 3 (April 2016). Ash from explosion events during May and June 2016 of phase 3 contain a higher amount of non-juvenile dense grey clasts (up to 30%), which coincides with a reduction in the amount of TC clasts. Ash from a small explosion in January 2018 has a similar componentry to ash from the largest explosions in phase 3, although contains more hydrothermally altered fragments.
The two bombs from 7th February 2016 are porphyritic (phenocrysts up to 2 mm) and have a low vesicularity (<5 vol.%), and display groundmass textures similar to the TC clasts found in the ash. The bomb from 21st April 2016 has a similar porphyritic texture, although noticeably contains discrete interfingers of highly vesicular (20-25 vol.%) and poorly vesicular (<5 vol.%) regions, termed herein as vesicular and dense domains respectively. The vesicular domains contain vesicles up to 500 μm in diameter, consistent with the bubble walls observed at the edges of the TC clasts (Fig. 3c). In thin section, the dense domains display a darkercoloured groundmass (Fig. 3e) with microtextures more similar to the DB clasts ( Fig. 3b and f), while the vesicular domains display lighter-coloured groundmass (Fig. 3e) with similar microtextures to the TC clasts ( Fig. 3d and f).

Mineralogy
All erupted samples contain phenocrysts of plagioclase, clinopyroxene, orthopyroxene, amphibole and titanomagnetite. The groundmass contains microlites of plagioclase, pyroxene, titanomagnetite, and minor apatite, situated within an interstitial glass. Silica-rich phases are present in the groundmass of the non-juvenile ash clasts and are locally found in clusters in the dense domains of the April 2016 bomb. QEMSCAN mineral maps were collected on raw ash samples (Fig. 4a-f) and bombs (Fig. A.2). The mineralogy is broadly consistent across all samples (Fig. 4g), although subtle differences are observed between the ratio in Si-rich phases (Si polymorph) and glass that are consistent with the ash componentry. For example, the increase in a silica-rich phase in ash samples in May and June 2016 is in agreement with the increased abundance of dense grey lithic clasts, and the higher glass content of the bombs indicates their juvenile nature. XRD further confirms an increase in silica polymorphs in ash from late-2015 to mid-2016, identified as tridymite (Fig. 4h).

Bulk-rock compositional differences
All erupted products are andesites (61.3-62.7 wt.% SiO 2 ; Table 1) that abide to Santiaguito's historical mafic trend (Fig. 5a), making them the least evolved magmas since its inception. However, subtle changes exist across the 2014-2017 eruptive period (Fig. 5b). The 2014 lava flow is the most evolved with 62.7 wt.% SiO 2 and lowest combined FeO and MgO content (6.3 wt.%). Ash and bombs erupted during phase 2 record a decrease in SiO 2 (by 1.0-1.5 wt.%) and increase in combined FeO and MgO (by 1.2-2.3 wt.%). During the high energy explosions in phase 3 the bulk composition remained similarly more mafic, although became systematically more evolved for explosions in May and June 2016, closely approaching the composition of the 2014 lava. The vesicular domains within the April 2016 bombs are compositionally more mafic (61.5 wt.% SiO 2 ) than the denser domains (62.1 wt.% SiO 2 ); however, this is close to analytical uncertainty (0.5 wt.% SiO 2 ) thus should be assessed with caution.

Amphibole phenocrysts
Amphibole crystals in the eruptive products from 2015 to 2016 are sparse (≤2 area %) and are found as clusters or individual phenocrysts. They typically display breakdown rims of anhydrous mineral phases, including plagioclase, orthopyroxene, clinopyroxene and titanomagnetite. Rim textures vary from thin (<50 μm), fine grained symplectitic laths consisting of anhydrous crystals to thick (up to 300 μm), coarse-grained crystals in the rim (Fig. 5c). In some cases, amphibole crystals are completely broken down resulting in pseudomorphs. Amphibole are dominantly magnesiohastingsite (Mg-Hst), although a tschermakite (Tsch) crystal was present in ash from May 2016, using the classification of Leake et al. (1997). Their remnant core compositions of different pressuresensitive (e.g., Al) and temperature-sensitive (e.g., Si and Ti) elements can be used to estimate the conditions in which they crystallised (Ridolfi et al., 2010). By employing the empirical amphibole geothermobarometer of Ridolfi et al. (2010), the amphiboles from explosions in phase 3 were estimated to crystallise at temperatures between ∼970 and ∼1020 • C (±22 • C) and pressures from ∼450 to ∼620 MPa (±50 MPa), which represents conditions at the higher temperature and pressure end of the previously constrained storage region (∼940-980 • C and ∼330-620 MPa; Fig. 5d; Scott et al., 2012). The Ridolfi model also provides formulations for a hygrometer (primarily influenced by Al content), with estimated dissolved magma H 2 O contents of between 6 and 8 wt.%. Using the chemometric equations of Zhang et al. (2017) from major element multiple regression analysis, it is possible to reconstruct the melt composition that was in equilibrium with the amphiboles during crystallisation. Mg-Hst amphiboles from the andesitic products gave equilibrium melt compositions of 55-61 (± 3) wt.% SiO 2 (avg. 57 wt.% SiO 2 , n = 9) and 2.4-4.2 (±1) wt.% MgO, thus crystallising in a basaltic-andesite magma, while the single Tsch amphibole re- vealed crystallisation within a more evolved dacitic magma (68 ± 3 wt.% SiO 2 ).

Plagioclase phenocrysts
Plagioclase phenocryst core compositions from Santiaguito lavas between 1922 and 2002 record a temporal evolution in An content ( Fig. 5f; Scott et al., 2012), which reveals a bimodal distribution (peaks at An 40-50 and An 60 ) diminishing towards a unimodal distribution (single An 50-55 peak) through time. A weak An 85-90 peak was recorded in all of Santiaguito's pre-2002 lavas (<10% of the phenocryst population). Plagioclase phenocryst core compositions from the explosive deposits in 2015-2016 ( Fig. 5e and f) were bimodal with peaks at An 45-55 (42%) and an additional strong peak at An 85-95 (30%). Plagioclase phenocryst rim compositions record a single, unimodal distribution with a peak at An 50-55 , while microlites reveal a lower An content (An 22-57 ).

Interstitial glass chemistry
The interstitial groundmass glass for the ash and bombs pre-  (2000), which uses An-corrected normative Qz-Ab-Or, provides an approximate assessment of glass equilibrium state (Fig. 6a). Liquid lines of descent show progressively higher glass equilibrium pressures through time, with the bomb in April 2016 and ash in January 2018 revealing a higher degree of disequilibrium between the melt and microlites at lower pressures.
Magmatic temperatures and water contents of the erupted products were estimated using the experimentally calibrated plagioclase-melt geothermometer and hygrometer of Putirka (2008). For each clast type, the average glass chemistry was used as the melt composition along with the average chemistry of coexisting plagioclase microlites. These estimates show the TC clasts were hotter (up to 970 ± 23 • C) and more H 2 O-rich (up to 2.4 ± 1 wt.%) than the DB clasts (up to 900 ± 23 • C and 1.2 ± 1 wt.% H 2 O), consistent with the vesicular texture of the bombs and pore shaped geometry of the TC clasts. The bombs from February 2016 reveal similar temperature and H 2 O estimates to DB clasts (up to 910 ± 23 • C and 1.2 ± 1 wt.%), while the bomb from April 2016 is more similar to TC clasts (up to 940 ± 23 • C and 1.7 ± 1 wt.%) with negligible difference between the vesicular and dense domains.

Plagioclase microlite textures
The groundmass of the ash and bombs across the 2015-2016 eruptive transition consists of 25-40% plagioclase, 1-8% mafic crystals (combined pyroxene and titanomagnetite), 55-70% interstitial glass and <1% porosity (Fig. 3b and d). Textural characterisation Table 2 Quantitative textural analysis of groundmass plagioclase microlites for different juvenile clast types in the ash (transparent, TC, and dense brown, DB) and bombs from explosions in 2015-2016, compared to ash from 1968, 2003, 2014 and 2018 (n denotes number of crystals analysed). Crystal aspect ratios (short: intermediate: long axes) were calculated using the 3D stereological correction program CSDSlice (Morgan and Jerram, 2006). Growth rates and nucleation rates were calculated from equations of Marsh (1988). Ascent rates were calculated using the MND water exsolution rate meter of Toramaru et al. (2008 (Morgan and Jerram, 2006). Shaded areas in (b) show textural comparisons with other well characterised systems including 1991 Pinatubo eruption (Hammer et al., 1999), 1997 vulcanian explosions at Soufrière Hills Volcano (Clarke et al., 2007), 2006 Merapi dome eruption (Preece et al., 2013) and the dense dome (DD) material and white pumice (WP) of the 2010 effusive-explosive transition at Merapi (Preece et al., 2016). The latter has also been compared to (c) mean crystal size and (d) aspect ratios of Santiaguito samples.
of groundmass microlites can provide a quantitative assessment of the style of late-stage crystallisation. We focus exclusively on quan-tification of plagioclase microlites in the DB and TC clasts, as well as the bombs (Fig. 3a-f; Fig. A.3; Table 2).
Both DB and TC clasts have similar overlapping microlite number densities (N A ) (8725-14804 mm −2 and 8252-12425 mm −2 , respectively), although on average TC clasts lie towards lower N A values. Plagioclase microlite fraction (ϕ) for both clast types fall within a similar range (0.22-0.38) and show negligible correlation with N A (Fig. 6b). However, N A for both DB and TC clasts correlate with crystal size, with smaller crystal sizes having a higher N A (Fig. 6c). Although both clast types show similar ranges for mean crystal size, microlites in TC clasts on average are larger (28 μm 2 ) than those in DB clasts (21 μm 2 ). 3D crystal morphologies (determined using CSDSlice) reveal a distinct divide between the clast types (Fig. 6d). Plagioclase microlite aspect ratios (S/L) for DB clasts were typically less than 0.2 (i.e., elongate/acicular), while microlite aspect ratios in TC clasts were consistently higher (0.2-0.7; i.e., equant/rectangular prism).
3D crystal size distributions (CSDs) correct for the errors obtained by using 2D slices, which can provide a more quantitative assessment of the style of final magma ascent (see Fig. A.4 for all CSD curves). Using the slope (α) and y-intercept (n o ) of the steepest portion of the CSD curves (segment 1 in Fig. A.4 representing late-crystallisation), volumetric number densities (N v ), microlite growth rates (G) and nucleation rates ( J ) can be calculated (Marsh, 1988;Blundy and Cashman, 2008) and temporally assessed ( Fig. 7a-d). N v for ash from pre-2015 and phase 2 were all < 2 × 10 6 mm −3 , which increase significantly for explosions in phase 3 (up to 4 × 10 6 mm −3 ) and the return to < 2 × 10 6 mm −3 in 2018 (Fig. 7a). The temporal evolution in S/L values also show an increase for 2015-2016 explosions (phase 2 and 3), which remain high in 2018 (Fig. 7b). G values (Fig. 7c) and J values (Fig. 7d) show large ranges owing to the uncertainty of crystallisation timescales (τ ), thus should be taken with caution. However, by using repose times as the minimum τ , and the time difference between the explosion and the start of phase 1 as the maximum τ (assuming the onset of phase 1 was related to new magma entering the system), an average G and J can be calculated which provides an approximate realistic estimate. Average growth rates for plagioclase microlites in 2014 are ∼1.8 × 10 −6 mm s −1 , which drop to ∼4.8 × 10 −8 mm s −1 in 2015-2016, and increase again to 3.5 × 10 −7 mm s −1 in 2018. A similar trend is calculated for the average J (∼180 mm −3 s −1 in 2014, ∼22 mm −3 s −1 in 2015-2016, and ∼130 mm −3 s −1 in 2018), yet no clear differences in G and J were calculated for DB and TC clasts.

Ascent rates
In order to assess whether a variation in magma ascent rate took place during microlite crystallisation, ascent rates were estimated using the MND water exsolution rate meter of Toramaru et al. (2008). Firstly, decompression rates (dP w /dt) were calculated as: where c is a function of the water content, b is a constant (40 for plagioclase), N is the microlite number volume and a is a calculation combining both glass SiO 2 (C si ) and water content (C w ). A sensitivity analysis of these calculations shows that an error of 1% for N, C w and C si can result in differences in ascent rate of up to 1, 3 and 50%, respectively. The maximum water contents of 1.2 wt.% for DB clasts and 2.4 wt.% for TC clasts, calculated from Putirka (2008), were used as inputs. An ascent rate (V n ) was then calculated from the following: (3) Fig. 7. Temporal evolution of 3D plagioclase microlite textural characteristics for ash and bomb samples over the eruptive period. (a) Volumetric number densities (N v ), (b) short/long axes (S/L) determined using CSDSlice (Morgan and Jerram, 2006), (c) average growth rates (G) and (d) nucleation rates ( J ), and (e) calculated ascent rates using the MND water exsolution rate meter (Toramaru et al., 2008). Error bars in (c) and (d) represent minimum and maximum rates. Symbols are the same as those in Fig. 6. where ρ is the density (2500 kg m −3 ), g is gravity and dP /dz is the decompression rate at a given water content at a given depth (i.e., onset of microlite crystallisation). Estimated ascent rates (Fig. 7e) for the DB clasts are between 1-6 × 10 −5 m s −1 (Fig. 7e), while the TC clasts from 2015-2016 and 2018, along with the April 2016 bomb, ascended 1-2 orders of magnitude faster (1 × 10 −4 -1 × 10 −3 m s −1 ).

Architecture of Santiaguito-Santa María magma storage region
The origin and evolution of magma feeding persistent eruptive activity at Santiaguito-Santa María has been the focus of a number of studies (e.g., Jicha et al., 2010;Scott et al., 2013;Singer et al., 2014;Cisneros de León et al., 2019). Matching chemical signatures of the 1902 Santa María dacite and earliest Santiaguito lavas supports the reactivation of a shared plumbing system (Scott et al., 2013). Santa María volcanism is believed to have initiated by the accumulation of mantle-derived basalt at the base of a ∼40 km thick crust (Carr et al., 2003), which evolved by assimilation and fractional crystallisation and led to the eruption of hybrid basalticandesite magma (Jicha et al., 2010;Singer et al., 2014). During a repose period of ∼25 ka, the conduit sealed causing magma to stall in the mid-crust (∼12-24 km; Scott et al., 2013), possibly as multiple reservoirs, allowing the basaltic-andesite to evolve forming a vertically stratified and thermally zoned storage region (Scott et al., 2013), with dacite (∼870 • C; 65 wt.% SiO 2 ; Singer et al., 2014) at the top and basaltic-andesite (∼1020 • C; 54 wt.% SiO 2 ; Singer et al., 2014) below. It is interpreted that deep mafic magma injection into the shallower dacite reservoir triggered the 1902 eruption, and in 1922 led to the progressive extrusion of the chemically stratified mid-crustal storage zone resulting in the birth of the Santiaguito dome complex (Scott et al., 2013).
Geothermobarometric (Ridolfi et al., 2010) and chemometric (Zhang et al., 2017) constraints of the Mg-Hst amphibole crystals in the ash and bombs that erupted explosively in 2015-2016 indicate they crystallised from basaltic-andesite melt sourced from the hottest (∼970-1020 • C) and deepest (∼450-620 MPa) part of the pre-constrained storage region (Scott et al., 2012). This is also supported by the minor shift (∼1.5 wt.% SiO 2 ) to a more mafic bulkrock composition (Fig. 5b), which is consistent with the chemically stratified storage model (Scott et al., 2013). However, the single Tsch amphibole found in May 2016 suggested crystallisation within a cooler dacitic magma. Cisneros de León et al. (2019) analysed zircons in the 1902 dacite and ash from a moderate sized explosion in August 2016, finding evidence for dacitic remnants from the pre-1902 reservoir becoming entrained in the magma during mafic recharge events, which may explain the presence of the Tsch amphibole. Based on estimated melt compositions and H 2 O contents (6-8 wt.%), it suggests the basaltic-andesite was waterundersaturated at the mid-crustal depths (Baker and Alletti, 2012).
Compositions of plagioclase phenocrysts, in particular growth zonations from core to rim (Fig. 5e), can provide useful insights into deeper magmatic processes as they are influenced by the composition of the surrounding melt, along with pressure, temperature and volatile contents. The bimodal plagioclase core compositions pre-1940 (An 40-50 and An 60 ; Fig. 5f) have been related to magma mixing of basaltic-andesite and dacite, becoming unimodal (An 50-55 ) due to a lower prevalence of magma mixing post-1940, while the small, consistent An 85-95 peak has been associated with remobilised crystals from a mush zone (Scott et al., 2013). The significant increase in these high An cores in the 2015-2016 eruptive products (Fig. 5f) suggests increased remobilisation and/or convection of a relatively thick mush zone at ∼17-24 km depth (e.g., Burgisser and Bergantz, 2011). This mush zone likely contained crystals formed from a primitive melt, and became decoupled and entrained following the injection of magma into the storage region. Based on the near-homogenous rim compositions for all plagioclase phenocrysts ( Fig. 5f and 8a), it is apparent that they followed a similar ascent path through the shallow system.

Magma ascent
Textural characterisation of groundmass microlites can provide a quantitative assessment of the style of crystallisation, which may thus be used to constrain pre-eruptive magmatic processes and conditions as magma ascends prior to eruption (Cashman and Blundy, 2000). Plagioclase microlite textures reflect the degree of undercooling ( T ) and thus ascent rate (e.g., Hammer and Rutherford, 2002). Theoretically, a larger T favours a nucleationdominated regime with the formation of many small microlites, whereas a lower T would result in a growth-dominated regime and the growth of fewer but larger crystals (Mollo and Hammer, 2017). A subtle difference in the DB and TC clasts is observed (Fig. 6c), with TC clasts generally having fewer and larger microlites suggesting these crystallised at lower T under a more growth-dominated regime. However, some overlap of microlite number densities (N A ), mean crystal sizes and microlite crystal fraction (ϕ) for the different ash fragments renders such an interpretation inconclusive (Fig. 6b-c). The differences in the crystal morphology (S/L) for the two clast types (Fig. 6d and 7b) can also be related to the style of crystallisation. Experimental studies have shown that crystals change from equant/tabular to acicular with an increasing T (Hammer and Rutherford, 2002;Couch et al., 2003). Microlites in TC clasts have consistently higher S/L values than the DB clasts, reinforcing that these clasts experienced a lower T .
Applying these results to the experimentally calibrated model linking crystal morphology with final crystallisation pressures (Hammer and Rutherford, 2002), DB clasts suggest continued crystallisation to ∼10 MPa in the conduit (∼400 m depth) as previously estimated for pre-2002 lavas (Scott et al., 2012), while TC clasts suggest final pressures of ∼50 MPa (∼2 km depth), possibly due to stagnation prior to eruption. This is supported by the observation that TC microlites show more pronounced growth zones ( Fig. 3b and d). Furthermore, the broad range in plagioclase microlite An contents (An 22-57 ; Fig. 5f) suggests microlites in both clast types formed at a range of temperatures, pressures and water-contents indicative of multi-step decompression. A comparable transition in eruption style was observed in 2010 at Merapi volcano (Indonesia), with the eruptive products showing similar relative differences in microlite textures to those observed during Preece et al., 2016), attributed to an influx of new magma. Preece et al. (2016) showed that removal of a dense dome material (due to overpressures within a plug) with smaller acicular microlites due to high T (akin to the DB clasts) gave way to a fresher white pumice consisting of larger, more equant microlites due to stalling ∼2 km depth in the conduit that favoured lower T (akin to the TC clasts).
The estimated temperature range during microlite crystallisation (900-970 • C; Putirka, 2008) is consistent with temperatures obtained by thermal infrared and spectral measurements of Caliente's crater surface (850-950 • C; Sahetapy-Engel et al., 2004) and fumarole surface temperatures (843 • C, Stoiber and Rose, 1969), thus suggesting limited cooling in the conduit (Scott et al., 2012). The higher estimated H 2 O contents and temperatures of the TC magma and April 2016 bomb (from Putirka, 2008) suggests limited degassing during final ascent, which is supported by the higher equilibrium pressures of the interstitial glass (Fig. 6a) and faster ascent rates relative to the H 2 O-poor DB clasts (Fig. 7e). Several lines of evidence thus indicate that the TC clasts and vesicular bomb were derived from a new magma source that influenced final ascent style, differentiation of the groundmass microlite textures and glass composition. Consequently, the mingling of the vesicular domains and dense domains in April 2016 likely took place at a shallow level in the conduit syn-or post-groundmass crystallisation.  Scott et al. (2013). Coloured-gradation reflects the degree of magma evolution from dacitic to basaltic-andesite. The storage region was previously constrained by amphibole geothermobarometry to reside at 12-24 km depth (Scott et al., 2012). Slow, continuous magma ascent from the deep to shallow system would have taken place, with crystallisation and volatile exsolution leading to the formation of blocky lava flows and generation of a plug. Low energy, shallow shear and friction-driven fragmentation likely resulted in small (<1 km high) ash-poor plumes. (b) Injection of new magma into the shallow magmatic system from depth, which passed through an extensive mush zone. Magma likely ascended as a volatile-rich pulse that experienced a relatively fast ascent towards the surface, stalling beneath the dome causing a lower degree of undercooling and facilitating mingling with pre-2015 magma. Overpressures from rising magma and gases from beneath led to the gradual excavation of the plugged vent causing greater delays in the arrival times in seismic and acoustic signals (as fragmentation depth increased), resulting in higher energy, ash-rich explosions and pyroclastic flows.
During magma mingling at shallow levels, it is likely that the eruptive products entrained material that had coupled to the conduit margin wall-rock. The systemically more evolved composition of ash from the large explosions in phase 3 (Fig. 5b), which coincides with an increased fraction of dense grey lithic clasts (Fig. 3g) that comprise increased silica polymorphs (Fig. 4g and h), suggests entrainment of the more evolved conduit material. Similar silicarich groundmass textures to those observed in the lithic fragments (Fig. A.1) have been characterised in detail for Santiaguito pre-2002 lavas (Scott et al., 2012), and were observed in many other dome lavas (e.g. Mount St. Helens, Pallister et al., 2008;Soufrière Hills volcano, Horwell et al., 2013). The formation of these textures has been attributed to glass breakdown at high temperatures and low pressures within the dome, and that the extent of such breakdown may be used as a relative estimate for extrusion rate (Scott et al., 2012). Thus, it suggests that complete glass breakdown in the lithic clasts was a result of longer residence (>2-3 hours) at high temperatures (∼900 • C) within the dome before being excavated during the explosions in 2016 (Scott et al., 2012;Horwell et al., 2013), which may be a result of the longer repose intervals between explosions during phases 2 and 3. Interestingly, if the increase in crystallised silica is responsible for the bulk SiO 2 increase in phase 3, it may suggest the bulk transport of Si into the dome either by magmatic gases or meteoric waters (Horwell et al., 2013).

Explosion mechanisms: linking petrological and monitoring observations
Early interpretations of Santiaguito's regular, weak explosive activity suggested magma-water interaction may be responsible (i.e., phreatomagmatic) owing to a combination of heavy rainfall and the active vent residing in the centre of the 1902 crater (Rose, 1972). However, more recent observations have led to different eruption mechanisms being proposed. Specifically, geophysical monitoring of the dome using tiltmeters revealed regular (every ∼26 minutes) inflation and deflation cycles throughout 2012-2014 , with explosions occurring during the most rapid inflation-deflation cycles. These cycles were attributed to gas accumulation from a pressure source, either beneath a plug or within a shallow reservoir, during periods of relative sealing of permeable pathways followed by its sudden release, causing rapid decompression and deflation of the surface with a corresponding gas flux of ∼10 1 kg s −1 . Comparison of inflation timescales to rock deformation experiments on Santiaguito lavas has shown that gas may accumulate in dilatant ten-sile fractures formed during brittle-ductile deformation (Hornby et al., 2019b). Combined seismic, acoustic and thermal monitoring of small-to-moderate explosions provided depth constraints for the source of fragmentation, occurring 100-620 metres beneath the vent (Sahetapy-Engel et al., 2008). The dominance of DB clasts in ash generated during similar events pre-2015 suggests fragmentation of an H 2 O-poor magma plug that resided at a shallow depth within the conduit (Fig. 8a). The low energy of these explosions (Fig. 2g) and relatively short delays in the time arrivals between the seismic and acoustic infrasound signals observed here support a shallow explosion source, likely representing fragmentation of the 2014 lava that filled the conduit. The capping of opensystem volcanic conduits with a high viscosity plug is a common phenomenon observed at many silicic volcanoes that show persistent explosive activity (e.g., Volcán de Colima, Lavallée et al., 2013;Sakurajima, Miwa et al., 2009). Conduit flow modelling suggests shear-induced fragmentation is a likely process in such cases (Gonnermann and Manga, 2003), owing to pulsatory ascent and subsequent localised failure near the conduit margins (Neuberg et al., 2006). At Santiaguito, this model is supported by the gas-rich, ash-poor nature of the pre-2015 eruption plumes (De Angelis et al., 2016), along with gas-and-ash emerging from arcuate fractures on the dome surface (Johnson et al., 2008) and frictional melts plus thermally-driven vesicles recorded in ash fragments (Lavallée et al., 2015). Furthermore, SO 2 gas fluxing during explosions in 2008 and 2009 have provided further evidence for localised shear fracturing, in which monitored SO 2 degassing remained continuous throughout repose intervals, suggesting no significant sealing of the permeable network for pressures to accumulate (Holland et al., 2011). Quantification of the relative fractions of constituent phases at ash particle boundaries for ash generated in similar sized explosions in 2012 led to the conclusion that faulting and abrasion mechanisms likely contribute to ash generation (Hornby et al., 2019a). We thus infer that the trigger mechanism for explosions up until phase 1 (Fig. 2g) were driven primarily by the sudden release of pressurised gas stored within fractures due to rapid shearing of magma resulting in fragmentation at shallow levels within the conduit (Fig. 8a). Fracture infilling (e.g., Castro et al., 2012) or healing (e.g., Tuffen et al., 2003) likely periodically impeded gas flow through the conduit leading to the accumulation of pressures in the dome, causing rapid inflation of the dome and explosion .
The explosion mechanism for the small-to-moderate gas-andash explosions is distinctly different to explosions in 2016, with anomalously high energies that were able to excavate the crater ( Fig. 2b and d). The gradual increase in the repose intervals and energy of explosions from phase 1 to phase 2 likely reflects a change in magma ascent, evolution and eruption mechanism influenced by the injection of a higher temperature magma modifying the pressure source in the conduit and the outgassing efficiency (Fig. 8b). An increase in the SO 2 flux from 2014 (∼45 kt yr −1 ) to 2015-2016 (∼85 kt yr −1 ) from OMI satellite observations (Carn et al., 2017) supports the petrological evidence of a fresh magma supply. During this time, the development and destruction of outgassing pathways, such as fracture networks and tuffisite veins, would play a non-trivial role (e.g., Tuffen et al., 2003;Castro et al., 2012). The largest explosions recorded in phase 3 (April 2016) coincided with the high abundance of TC clasts in the ash and mingled bombs, contemporaneous with the excavation of the crater. The longer delay in the seismic and acoustic time arrivals for some explosion events during this period suggests a potential deeper fragmentation source, which is consistent with the higher final crystallisation pressures (∼2 km depth) and more equant shaped (high S/L values) microlites of the TC clasts. The explosion intensification in 2015-2016 thus suggests a switch from shallow, low energy shear fragmentation to deeper, higher energy overpressure fragmentation as a consequence of the injection of new magma into the shallow system (Fig. 8b).

Conclusion
The shift in eruption style at Santiaguito from early-2015 to late-2016 has been explored by combining petrological, geochemical, geophysical and thermal observations. In this study, we show a unique dataset that documents this escalation in explosivity using a multi-parametric approach. Multiple lines of evidence suggest the 2-year phase in explosive eruption intensification was influenced by the injection of a higher temperature, relatively volatile-rich magma subsequently causing a change in the explosion mechanism from shear-triggered to decompression-triggered eruptions. Monitored seismic and acoustic signals indicate that this caused fewer, more erratic, larger explosions, generated from generally deeper fragmentation levels. These larger explosions disrupted the dome, excavated the crater and exposed the hot interior of the dome as observed by satellite-based thermal imagery. Although many questions regarding the processes taking place during 2015-2016 at Santiaguito remain, we highlight the importance of long-term multi-parametric monitoring techniques coupled with detailed investigation of the erupted products. Such efforts will assist in improving our understanding of pre-eruptive volcanic processes, the monitored signals these processes produce and shifts in eruption style at intermediate-silicic volcanic systems.

Funding sources
This work was supported by the European Research Council Starting Grant on Strain Localisation in Magma (SLiM, 306488). SDA, YL, and AR acknowledge funding from the Natural Environment Research Council (Grant NE/P007708/1). JEK acknowledges an Early Career Fellowship of the Leverhulme Trust (ECF-2016-325). A. H. acknowledges funding from H2020 Marie Skłodowska-Curie Actions (AVAST, 753900). PJG acknowledges the support from NERC Centre for the Observation and Modelling of Earthquakes, Volcanoes and Tectonics (COMET, GA/13/M/031, http://comet .nerc .ac .uk).

Data availability
The geophysical data and seismic catalogue supporting the conclusions of this manuscript will be made available upon request by any qualified researcher without undue reservation.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.