Hydrothermal eruption dynamics reflecting vertical variations in host rock geology and geothermal alteration, Champagne Pool, Wai-o-tapu, New Zealand

Hydrothermal eruptions are characterised by violent explosions ejecting steam, water, mud, and rock. They pose a risk to tourism and the operation of power plants in geothermal areas around the world. Large events with a severe destructive threat are often intensified by the injection of magmatic fluids along faults and fractures within volcano-tectonic rifting environments, such as the Taupo Volcanic Zone. How these hydrothermal eruptions progress, how craters form and the scale of ejecta impacts, are all influenced by the local geology and reservoir hydrology. By analysing breccia lithology, undisturbed strata proximal to the explosion sites, and conducting tailored decompression experiments, we elucidate the eruption sequence that formed Champagne Pool, Wai-o-tapu, New Zealand. This iconic touristic site was formed by a violent hydrothermal eruption at ~ 700 years B.P. Samples from undisturbed drill cores and blocks ejected in the eruption were fragmented in shock-tube experiments under the moderate pressure/temperature conditions estimated for this system (3–4 MPa, 210–220 °C). Our results show that this was a two-phase eruption. It started with an initial narrow jetting of deep-sourced lithologies, ejecting fragments from at least a 110-m depth. This event was overtaken by a larger, broader, and dominantly shallower eruption driven by decompression of much more geothermal fluid within a soft and porous ignimbrite horizon. The second phase was triggered once the initial, deeper-sourced eruption broke through a strong silicified aquitard cap. The soft ignimbrite collapsed during the second-phase eruption into the crater, to repeatedly choke the explosions causing short-term pressure rises that triggered ongoing deeper-sourced eruptions. The eruption spread laterally also by exploiting a local fault. These results are relevant for hydrothermal eruption hazard scenarios in environments where strong vertical variations in rock strength and porosity occur.


Introduction
Steam-driven hydrothermal eruptions commonly occur in both volcanic and geothermal systems (Stix and Moor 2018). The most deadly eruptions in recent times have been associated with vent-hosted hydrothermal systems on volcanoes, e.g. 2014 Ontake eruption, Japan (Yamamoto 2014) and 2019 Whakaari/White Island, New Zealand . Sites of intermittent volcanism in humid environments are highly prone to steam-driven eruptions (Kilgour et al. 2019;Jolly et al. 2020). Hydrothermal eruptions are also very common in other settings because they only require hot and pressurised geothermal fluid circulation (Browne and Lawless 2001). In these cases, rapid fluid decompression could be triggered by tectonic fault rupture, landslides and sudden hydrological (e.g. lake level) changes. Within active volcanotectonic rifts, such as that of the central North Island of New Zealand (Rowland and Simmons 2012;Wilson and Rowland 2016), failed dyke intrusions and magmatic fluids rising along fracture systems may drive especially violent hydrothermal eruptions in areas distant from volcanoes (Hedenquist and Henley 1985;Nairn et al. 2005;Montanaro et al. 2020).
Driven by flashing of superheated groundwater or geothermal fluids at shallow depths (Mastin 1995;Thiéry and Mercury 2009), hydrothermal eruptions violently eject fragmented host rock, caustic fluids, steam, and fine ash as ballistics and wet debris jets. They may also generate steamrich density currents (surges) of poorly sorted rock debris Maeno et al. 2016;Kilgour et al. 2019). The wide diversity of rock types in geothermal areas causes a great variability in hydrothermal eruption dynamics (Browne and Lawless 2001;Montanaro et al. , 2020. By examining the characteristic lithologies and particle properties within hydrothermal eruption deposits, the dynamics and sources of eruptions can be estimated Mayer et al. 2015;Montanaro et al. , 2020Pittari et al. 2016;D'Elia et al. 2020). Valuable insights into the source rock properties come from analysing blocks within the hydrothermal breccias Heap et al. 2017), or from rare shallow (<400 m) nearby drill cores. Wells also provide direct information on the hydrothermal system state (pressure, temperature) and rock properties such as porosity, permeability, and alteration state (Dobson et al. 2003;Mielke et al. 2015;Milicich et al. 2016Milicich et al. , 2020Carlino et al. 2018).
Here, we examine the large hydrothermal eruption that formed Champagne Pool, which is one of the main New Zealand geothermal tourist attractions in the Wai-o-tapu geothermal field ( Fig. 1; Browne and Lawless 2001). We examined the Champagne Pool breccia and ejected particles (blocks and matrix) as well as samples from shallow drill cores adjacent to the explosion site. Adding laboratory decompression experiments, we explored the fragmentation and ejection behaviour of the variable substrate materials in order to reconstruct the evolution of the Champagne Pool hydrothermal eruption and determine the relationship between hydrothermal eruption properties and parent host rocks.

Geological setting
Wai-o-tapu (Fig. 1) is a geothermal area covering 10.5 km 2 , as delimited by its low resistivity anomaly (Bibby et al. 1994). It is known for its high heat flow and localised precipitation of gold and silver and is one of the most photographed natural features in New Zealand, Champagne Pool (Hedenquist and Henley 1985). Two dacite domes, Maungaongaonga and Maungakakaramea, bound the geothermal area in the north, whereas the upthrown, footwall of the faulted Paeroa block (> 600 m of normal offset) lies to the west (Fig. 1). In the south, Wai-o-tapu is dissected by ring faults associated with the northern margin of the Reporoa caldera (Nairn et al. 1994) which, along with the NE-striking Ngapouri fault, host numerous hydrothermal eruption vents (Hedenquist and Henley 1985;Nairn et al. 2005).
The subsurface hydrology has been deduced from seven drill holes (WT 1-7, Fig. 1) penetrating to depths of 500 to 1100 m and spring chemistry (Hedenquist and Browne 1989;Hedenquist 1991;Wilson et al. 2010). Deep fluids ascend along temperature and pressure gradients that are close to boiling in the vicinity of WT-6 and Champagne Pool (Fig. 2a). The latter is one of the more than twenty hydrothermal eruption vents in the Wai-o-tapu area, which are commonly located where regional or caldera faults cross the geothermal field (Fig. 1). The highest chloride concentration occurs in waters at Champagne Pool, indicating that it currently represents a highly permeable flow path that taps the deepest known water in the system (Hedenquist 1991). The stratigraphy in the Wai-o-tapu area comprises almost flat-lying felsic pyroclastic deposits and locally reworked derivatives of these, along with andesite and dacite lava flows, and fine-grained lacustrine sediments in varying states of lithification. The variations in texture and permeability form a series of stacked aquifers, capped by aquicludes (Hedenquist and Henley 1985;Wilson et al. 2010;Wilson and Rowland 2016). Volcanic rocks penetrated by the Wai-o-tapu wells show highly variable degrees of alteration and permeability (Hedenquist and Browne 1989). Primary mafic minerals are often completely replaced and plagioclase is usually completely altered to a mineral assemblage, including variable amounts of albite, adularia, calcite, mica, and/or epidote. The groundmass is generally re-crystallised, silicified, and pyritised. Below 150 m (2 00°C on the boiling point-depth curve), the main alteration minerals in approximate order of decreasing abundance include quartz, white mica (mainly illite), albite, adularia, pyrite, calcite, epidote, wairakite, mordenite, and galena. Above this silicified zone at up to depths of 50 m, surficial acid sulphate waters cause irregularly distributed advanced argillic alteration. The mineralogy here consists of kaolin, alunite, cristobalite, native sulphur, and fine pyrite (Hedenquist and Henley 1985).

Champagne Pool eruption and local stratigraphy
Champagne Pool, together with several other hydrothermal eruption vents in Wai-o-tapu, formed around 700 years B.P. The origin of these eruptions are linked to intrusion and arrest of magma, or related magmatic fluids within faults prior to the large rhyolitic effusive and explosive A.D. 1315 Kaharoa Plinian eruption sequence at Tarawera volcano, located < 20 km to the northeast (Nairn et al. 2005). The Champagne Pool eruption occurred soon after the explosive phase of the Kaharoa eruption (Nairn et al. 2005). Dykes intruding across the region surrounding Mount Tarawera were channelled by existing fault and fracture networks to inject CO 2 (and or other  Fig. 1), including the main lithological units, alteration/silicification zones, fractures, and surficial features, as well as the projected location of drill hole no. 6 (modified from Hedenquist and hot gases) to "prime" the steam-driven eruptions (Nairn et al. 2005).

Field study
We documented the breccia deposit along the tourist tracks around Champagne Pool crater (Fig. 1b), to avoid damage to the vegetation within the Wai-o-tapu Thermal Wonderland nature reserve. We sampled the ash-and lapilli-rich breccia matrix, as well as blocks representative of the main lithologies, to be used for component and grain size analyses. Examination of macroscopic alteration features (e.g. colour differences), as well as the occurrence of alteration halos and silicified crusts and veins, indicated the relative degree of alteration within the breccia. We compared breccia-derived material with drill hole samples to infer eruptive source lithologies and excavation depth. Additional information of the field study are reported in the Online Resources.

Drill hole materials
We collected samples from the seven drill cores collected in 1956-1958 (WT 1-7) and stored within the Geothermal Collection at the University of Auckland. Material was sampled from the drill hole WT-6 located in close proximity to Champagne Pool ( Figs. 1 and 2). The cored samples (Figs. 2a and 3a) include densely welded tuff, purple-grey coloured, with lithoidal spherules and lithophysal cavities, derived from the lithophysal zone of the Waiotapu ignimbrite (WIS); densely welded tuff, purple-grey in colour with pumice-rich lenses derived from the Waiotapu ignimbrite (WI); coarse-grained, grey-coloured, pumice-rich tuff (CT-P); moderately dense, grey-coloured, coarse-grained tuff rich in crystals and lithics (CT-L); and fine-grained, dark-grey, crystal-rich tuff of the Rangitaiki ignimbrite (RI).

Breccia stratigraphy
The breccia erupted from the Champagne Pool has an oval surface distribution with its long axis directly towards the north (Fig. 1b). Immediately south of the crater (< 20 m), the deposit is 1.5 m thick (Fig. 1b), but it thins to zero, within 100-120 m. By contrast, the breccia is~4 m thick at 50-100 m north of the crater and is still~1 m thick at a 200-m distance northward. The breccia is characterised by two main beds showing distinctive matrix colour and matrix/block componentry.
The ≤ 1-m-thick lower brown-coloured bed is lenticular in shape and is distributed southward of the crater, although the lack of exposure means that we cannot exclude any deposit to the north. It is dominated by a brownish coarse ash-rich matrix and supports sub-rounded to angular lapilli to blocks up to 60 mm in diameter. The blocks in the lower bed comprise slightly to moderately altered fine-grained tuff, along with few blocks of silicified and brecciated material (e.g. B1-B2; Figs. 2, 3a, and 4).
The upper yellow-coloured breccia bed is more extensive and encircles Champagne Pool (Figs. 1, 2, and 4). It reaches a maximum thickness of 1 m south of the crater, and up to 4 m to the north. It comprises a fine ash-rich matrix, containing sub-rounded to angular lapilli-to-block-sized fragments up to 250 mm in diameter. The blocks comprise mostly moderately to highly altered fine-grained and accretionary lapilli-rich tuffs, silicified and brecciated clasts, and breccia tuffs (e.g. Y1-Y3 Fig. 2). Rare sintered clasts and tuffs with spherulitic texture occur in the upper half of the bed.
The two beds are separated by an irregular horizon marked by a layer of larger clasts and blocks, up to 1 m in diameter, forming low-(< 50°) to high-angle (> 50°) ballistic impact sags into the lower breccia, or impacting within the Taupo tephra (Figs. 2 and 4a-f). These blocks show the greatest evidence of alteration, comprising mainly silicified brecciated clasts and breccia tuffs.

Breccia grain size and componentry
The matrix of the lower bed shows variable grading in particle size while the upper bed shows a general normal grading (Fig. 5a, Figure ESM1 and Table ESM1 in Online Resources). The lower bed shows normal-toreverse grading in the lower 50 cm (with fine ash, > 3 ϕ increasing from 0.91 to 1.68%), and reverse grading in the upper 50-80-cm interval (returning to 0.53% fine ash). In the transitional layer between the upper and lower breccia (80-90 cm), the median diameter (D50) decreases from − 2.77 to − 2.2, and fine ash increases from 0.53 to 1.1 wt.%. The upper breccia shows a normal/reverse/normal grading in the matrix from 90 to 140 cm. The content of fine ash reaches its peak of 7.6% at 90-100 cm, decreases to 0.6% at 100-120 cm, and 1.1% from 120 to 140 cm. Block-sized clasts throughout the upper bed (Fig. 4), do not reflect the matrix grain size variations.
Block lithologies vary between beds (Figs. 2b and 3a). The lower bed contains rhyolitic ash-rich tuff, dark-grey coloured, with abundant millimetre-seized pumices and rare accretionary lapilli (e.g. B1) and coarse-grained brecciated tuff, with fine-grained silicified brown ashy matrix containing lithic and pumice clasts (e.g. B2). Upper breccia blocks comprise silicified brecciated clasts and tuff breccia, yellow-coloured, showing a variety of textures including fine-ash tuff with rare fine lapilli (up to 5 mm), and coarse lapilli tuffs containing lithics, pumice, and crystals (e.g. Y1-1 to Y1-3 and Y2 Fig. 3a). Some blocks (e.g. Y3) are clearly derived from the OI unit, based on its in situ exposure near Champagne Pool. These consist of friable to weakly consolidated rhyolitic ash-rich tuff with abundant accretionary lapilli. Blocks similar to Y3, but with a siliceous sinter-patina, were common in the block-rich horizon between the lower and upper Champagne Pool breccias, but are rarely found above or below this zone.
The matrix of the two breccia units consists mainly of slightly altered to unaltered fragments of similar tuffs to the blocks, together with pumice, undefined lithics, loose quartz and P-wave velocities (v p ) as as a function of connected porosity of representative substrate lithologies involved in the CP eruption. In the upper graph, only v P of exploded samples were measured. The experimental error on these measurements is < 1% and is therefore within the symbol size crystals, chlorite, and altered, blue-stained grains ( Fig. 5b and Table ESM2 in Online Resources). The lower breccia matrix comprises mainly white tuff clasts (55-75%), and quartz crystals (20-40%), with < 5% of the remaining components. From the transitional layer upward, the upper breccia contains common yellow tuff fragments (> 20-70%), with a secondary abundance of quartz crystals (15-50%), white tuff (20-30%), and minor (< 5%) lithics, pumices, bluish-grey altered clasts, and other crystals. The uppermost part of the upper yellow breccia shows a contrasting zone comprising up to 70% of dominantly white tuff particles.  Fig.1). a Type section showing the two main breccia beds, along with the underlying Taupo tephra and Oruanui ignimbrite. b-f Detailed images illustrating several ballistic blocks (varying from silicified, brecciated, and accretionary lapilli-rich blocks) with prominent bomb sags at the contact between the lower and upper breccia bed, and at the contact between the upper bed and the Taupo tephra. g massive breccia facies of the upper bed rich in large blocks, north of the CP crater. h Contact between the upper bed deposit and the underlying units, ca.100 m south of CP

Laboratory study
We measured the bulk/matrix density, (connected) porosity, and P-wave velocities (v P ) of between four and fifteen cylindrical core samples per lithology type from breccias and drill core samples ( Fig. 3) (76 core samples in total). A sub-sample of 33 water-saturated samples were rapidly decompressed in shock-tube steam-flashing experiments (Table 1; Mayer et al. 2015Mayer et al. , 2016. Our experiments aimed to mimic the decompression of heated and pressurised fluids at similar conditions as the Wai-o-tapu field, with temperature and pressure conditions described by Hedenquist and Henley (1985) and illustrated in Fig. 2. Our experimental conditions were calibrated using: i) breccia componentry indicating a maximum excavation depth around 170 m (also noted by Hedenquist and Henley 1985); ii) drill hole pressure data (from WT-6) (Dench 1963); and iii) estimated temperature of equilibration for alteration minerals (Hedenquist and Henley 1985;Hedenquist and Browne 1989). Thus, experimental conditions were between~210 and 220°C, and of 3 to 4 MPa as confining pressures. These conditions, together with the proportion of pore water, were used to estimate the bulk explosive energies fuelling primary fragmentation and clast ejection . We recovered the fragmented and ejected material and analysed grain size distribution and componentry. Finally, we used high-speed footage of the expelled fragments to measure ejection velocities (Mayer et al. 2015), and pressure sensors above and below the samples to record fragmentation velocities (Scheu et al. 2008).
The main methodological approach for petrophysical property analyses, decompression experiments, and estimation of explosivity of water have already been detailed by us in (Mayer et al. (2015 and ; we thus do not restate these methodological details here. However, for further detail on the methodological approach applied here, see Online Resources 1, 2, and 3, respectively.

Petrophysical properties and textures
Samples from drill cores show narrow ranges of porosity which distinguish each lithological unit ( Fig. 3b and Table ESM3 in Online Resources). The deeper-derived WI and WIS samples have porosities between 20.5 and 23.4%; intermediate-depth CT-L and CT-P samples show porosities from 31.5 to 40.3%; whereas the shallow-seated RI samples yield porosities between 50.6 and 54.1%. The measured bulk/ matrix densities profile mirrors the P-wave velocity profile ( Fig. 3b and Table ESM3 in Online Resources), with the seismic velocities for the more porous RI (1.7 to 1.9 km/s) being much lower than those for the less porous and denser CT-P/L and WI/WIS (2.5 to 4.4 km/s). In general, these properties strongly correlate to rock textures, with lower porosity/high density and high v P values associated with densely welded and heterogeneous tuffs, and mid-high porosity/lower density and low v P values typical of pumice/lithic-and ash-rich mildly to weakly compacted tuffs.
Breccia block samples show variable ranges of porosity depending on their stratigraphic position, lithology, and texture ( Fig. 3b   Experimental conditions (rock connected porosity, mass of water, pressure, temperature), and estimated/measured parameters including explosive energy (E Expl ), fragmentation (v Frag ), and particle ejection (v Ejec ) velocities, amount of fine > 3 ϕ and median grain size (D50) Sample Porosity (%) Decompression experiment results show how the explosive energy is coupled with porosity, scattering due to the initial pressure/temperature experimental conditions, and natural variability/degree of saturation of the core samples (Fig. 6a). All samples with < 30% porosity did not fragment completely. Only 30-60% of each of these samples fragmented, independent of the initial pressure/temperature conditions.

Grain size distributions of experimentally fragmented samples
The median grain size and the amount of fine ash generated by experiments (Table 1) show higher proportions of fines for conditions with higher fragmentation energy (Fig. 6b, c). Grain size distributions from fragmented drill core samples are finer (Fig. 7a) as a function of increasing sample porosity. WI and WIS generally produced the coarsest grain size distributions (WIS1: D50 from − 1.76 to − 2.66 ϕ; WIS2: D50 from − 3.14 to − 3.20 ϕ) with little fine ash (< 4 wt%). We note that these samples did not completely fragment (< 1/3 of the whole sample). Experiments on CT-P and CT-L show very similar trends, with the pumice-rich CT-P producing a slightly lower median grain size than CT-L (0.75-0.88 ϕ vs. 0.33-0.42 ϕ respectively). Both CT-P and CT-L produced similar amounts of fine ash (5.5-9 vs. 5.4-7.3 wt%). The RI sample series produced the finest median grain size (from 1.75 and 2 ϕ) and the highest amount of fine ash (11.6 and 20.6 wt%).
Grain size distributions from breccia block samples span from very coarse to very fine (Fig. 7b-d), in a similar pattern to the drill core samples. The fine-grained B1 samples, are very similar to RI, and produced median grain sizes of 1.45-1.63 ϕ, and a large proportion of fine ash (9.2-18.4 wt%). Brecciated B2 samples only partially fragmented (< 2/3 or < 1/3 of the whole sample), and produced a coarse grain size distribution with median grain sizes from − 1.08 to − 1.25 ϕ, and low amounts of fine ash (1.7-5.1 wt%); the latter depending on the initial proportion of fine-grained matrix in the cored sample. For Y1 samples, the low-porosity samples (~23-28%) only partially fragmented (< 2/3 of the whole sample), resulting in coarse grain size distributions with a median between − 1.55 and − 0.7 ϕ. Mid-porosity Y1 samples (~31-41%), completely fragmented, generating fine median sizes Fig. 6 Explosive energies (E Expl ) (a), median grain size (b), and weight % of fine ash (3 ϕ) (c) of experimentally produced fragments as a function of samples porosities (in a iso-energetics are calculated based on the experimental conditions). Cases with porosity < 32% do not have sufficient energy to completely fragment the samples and produce few coarse clasts with a small quantity of fine ash. In cases with porosity > 32%, for all initial pressure-temperature conditions tested, the entire sample is fragmented and these experiments produce finer grain sizes and larger amount of fine ash. The scattering in the explosive energy reflects the natural variability in sample textures from − 1.1 to − 0.29 ϕ. Both sample suites produced small amounts of fine ash (2-3.9 wt%). Tuff breccia Y2 samples produced grain size distributions intermediate between CT-P/ L and RI, with a coarse median diameter (− 0.49 to 0.7 ϕ), and moderate amounts of fine ash (11.3 to 18.1 wt%). Y3 samples produced finer median grain sizes (2.23-2.40 ϕ) than B1 or RI and a significant proportion of fine ash (31.2-33.9 wt%).

Grain componentry of experimentally produced grains
The experimentally produced fragments from drill core and breccia block samples have similar componentry to the Champagne Pool breccia matrix (Fig. 5), but also include additional grey tuff, silicified material, and crystals of calcite and mica (Fig. 8, Figures ESM3, 4 and Table ESM4, 5 in Online Resources). The samples that only partially fragmented show similar components in similar proportions to the fully fragmented samples.
Within the fragmented material of brecciated clasts and tuff breccia (Fig. 8b), especially in fine-grained samples of B1, Y2, and Y3, white tuff is the dominant component (~30 to 76%), quartz crystals make up 16 to 49%. Most of the brecciated samples from B2 and Y1 are instead dominated by silicified material (~40 to 70%), with minor white tuff (10-36%) and quartz (5-22%). Only one finegrained sample of Y1 contained white tuff fragments (79%) and minor silicified clasts (10%). Yellow tuff is found in small amounts (~1-7%) by B2, Y1, and Y2 samples, besides one Y2 sample showing~21.7%. Pumice fragments are found in all exploded samples, especially from Y2, with quantities varying between 1 and 14%. Lithics fragments were generated from all samples and make up between < 1 to 10% of the particles. Chlorite is only found as a minor component (< 2%) within fragmented Y1 and Y2 samples. The proportions of the main components are similar between the experimentally produced fragments of drill cores and breccia blocks. The only exception is the almost total absence of silicified material in the produced fragments from drill core samples (Table 2, Fig. 8).

Ejection velocities
Ejection velocities (v Eject in Table 1) are computed from the displacements of individual particles tracked across five successive frames of high-speed footage. For a few experiments, the initial part of the plume was obscured by large vapour clouds so that only minimum velocities could be estimated. In general, the ejection velocities weakly correlate with explosive energy (Fig. 9a), with data scatter due to initial pressure/ temperature conditions and natural variability in sample strength (cf., . For the drill core sample series, the CT-P fragments show the lowest velocities (196 ± 19 m/s), with the highest velocities in CT-L (233 ± 15 m/s) and RI (215 ± 14 m/s) samples. Among the breccia block series, B1 samples eject the slowest fragments (179 ± 12 m/s), while higher velocities are reached by particles from Y1 (197 ± 15 m/s), Y2 (182 ± 16 m/s), and Y3 (218 ± 11 m/s).

Source lithologies of champagne Pool eruption
Drill core data from the nearby WT-6 core (Dench 1963;Hedenquist and Henley 1985) provide stratigraphic and lithologic control, and insight into local hydrothermal alteration. Noteworthy, among the drill core samples, the only lithology showing silicified material is the deeply seated Waiotapu ignimbrite (Fig. 8). This suggests that in the WT-6 area, fluids circulated and partially altered rocks only within deeper fractures. By contrast, upwelling of silicifying fluids that occurred CT-L) and Rangitaiki ignimbrite (RI). b Detailed componentry shown in percentage for brown tuffs (B1) and brecciated clasts (B2) of the lower bed, and for yellow brecciated material (Y1, Y2) and yellow tuffs (Y3) from the upper bed to shallower levels must have occurred via a pre-existing permeable pathway (fracture zone) beneath the present Champagne Pool (Fig. 2). Despite the differences in silicification, there are similarities in componentry (e.g. the abundance of tuff types and minerals) between the experimental results from drill core and breccia clast samples (Fig. 8, Table 2). Considering this, experimentally fragmented material remains a robust proxy to correlate the inherited characteristics of components with the original pre-eruptive lithologies (Geshi et al. 2011). These are used alongside petrophysical properties to fingerprint the host rock involved at different stages of the eruption (cf., Heap et al. 2019).
In the lapilli-to-block-sized fraction, the lower bed of the Champagne Pool breccia is rich in altered fine-grained ashrich tuffs (B1), and silicified brecciated clasts (B2), containing lithics and pumices (Fig. 3a). Their appearance, textures, petrophysical properties, and componentry (Fig. 3, Table 2) suggest that these lithologies originate from fragmented deepseated CT-P/L or RI units. The nature of the breccia block B2 also indicates that a fractured and silicified portion of these units was involved in the eruption. Experimentally produced fragments from B1 show finer grain sizes, but similar componentry to CT-P (Figs. 7 and 8), while partially fragmented B2 samples show coarser sizes and a componentry dominated by silicified material, which are absent in drill core samples.
The lapilli-to-block-sized fraction of the upper Champagne Pool breccia bed is dominated by weakly consolidated ashrich and accretionary lapilli-rich tuffs (Y3), and secondary silicified brecciated clasts (Y1) and tuff breccia (Y2) with variable textures, and containing lithics, pumices, and abundant crystals (Fig. 3, Table 2). The Y3 samples are derived from Oruanui ignimbrite. The fragmentation of such porous, low-density and accretionary lapilli-rich tuffs, results in fine grain sizes generation (Fig.7) and high fragmentation velocity (Fig. 9). Petrophysical properties of mid-porosity brecciated Y1 cores are similar to CT-P/L, while the low-porosity samples show similar features to WI (Fig. 3). In the experiments the first type of Y1 cores fragmented completely, generating grain sizes very similar to B2 and CT-P/L, while the second type fragmented only partially, producing size distributions similar to WI (Fig. 7). The componentry of fragmented particles from all Y1 cores are dominated by silicified clasts (Fig.  8) with minor white/yellow tuff and lithics. This is very similar to particles produced by fragmentation of silicified CT-P/ L and WI units. Breccia tuff Y2 samples have petrophysical properties that almost completely overlap those of CT-P/L (Fig. 3b). The experimentally produced fragment size distributions (Fig. 7) and componentry of Y2 include chlorite (Fig.  8) and are very similar to those produced by CT-P/L.
The matrix of the lower Champagne Pool breccia bed is dominated by white tuff and quartz, with minor pumice and lithics. Overall, this resembles a mixed componentry derived from CT-P and RI lithologies. The transition layer between the two breccia beds shows an increasing content of yellow tuff, which becomes dominant in the upper bed, together with white tuff and minor quartz, lithics, and pumice. The abundance of yellow tuff in the transition layer could originate from an input of deeper units, or more likely, a greater involvement of the argillic-altered portions of the Oruanui Ignimbrite, where kaolinite and alunite zones produce yellow-coloured alteration products (Fig. 2a; Hedenquist and Henley 1985;Hedenquist and Browne 1989). Caution is needed in comparing these components due to (i) subsurface mixing and sorting occurring during the debris jet emplacement  and (ii) the efficiency of steamflashing in fragmenting the host rock into ash-fine lapilli sizes . For instance, the matrix of both lower and upper bed breccia lacks silicified material, although this is abundant in the coarse lapilli and blocks. This might indicate that fragmentation processes were not efficient for the silicified lithologies (cf., Kueppers et al. 2006).
Decompression experiments showed that low porosity lithologies only partially fragment under the investigated conditions. This is consistent with the absence, or rarity, of WI/ WIS clasts and their components (e.g. the characteristic kfeldspar; Fig. 8), within the breccia blocks and matrix.

Eruption trigger and mechanism
The lower bed of the breccia includes deep-seated local lithologies such as CT-P and RI, as well as shallow OI clasts at the base of the unit (Fig. 3). The CT-P clasts indicate that the initial explosion began at least at a depth of~110-100 m (Fig. 2). In addition, some of the CT-P clasts were brecciated and silicified (Figs. 3 and 10). Similar brecciated and recemented clasts are found in other hydrothermal breccia deposits (e.g. Nairn and Wiradiradja 1980;Hedenquist and Henley 1985;Kennedy et al. 2020;Montanaro et al. 2020) and drill cores from geothermal fields (White et al. 1975;Krupp and Seward 1987;Dobson et al. 2003). Their textures were attributed to hydraulic fracturing (Phillips 1979;Bromley and Mongillo 1994), produced by over-pressurised fluids along fault/fractures that break rocks and precipitate silica along fractures. Such mechanisms generally predate and sometimes trigger explosive events (Nairn and Wiradiradja 1980). Silicification of fractures and pores of surrounding rocks can reduce matrix permeability of hydrothermal reservoir rocks of orders of magnitude, and enhance fracture formation . Yet, at the same time, silicification may form an effective aquitard or aquiclude horizon (Lowell et al. 1993;Dempsey et al. 2012). Any excess CO 2 or other magmatic gas rising from a basaltic dyke (> 1000 t/day; Nairn et al. 2005), would be channelled within fractures within silicified portions of the deeper units (Uysal et al. 2009;Rowland and Simmons 2012), but ultimately trapped by a cap rock (e.g. silicified and/or argillic-altered levels). Overpressurization thus leads to hydraulic fracturing and explosive failure (Browne and Lawless 2001;Morgan et al. 2009;Christenson et al. 2010;Rosi et al. 2018;Stix and Moor 2018;Heap et al. 2019). A pressurised aquifer likely breaches near a horizontal discontinuity, such as a textural break, or cap/seal layer (Stearns and McDonald 1949;Browne and Lawless 2001). This leads to a top-down rarefaction wave, in concert with a returning pressure wave that fragments and ejects clasts (Germanovich and Lowell 1995;Montanaro et al. 2020). This mechanism explains the opening stage of the Champagne Pool eruption, recorded by emplacement of the lower breccia bed (Fig. 10).
The limited distribution of the lower bed and the high-angle ballistic impact sags (Fig. 4a, b), suggest that the initial crater was located close to the current pool with deep-seated explosions forming a narrow, diatreme-like crater (Maeno et al. 2013;Taddeucci et al. 2013). Variation in the upper breccia bed componentry and its broader irregular distribution suggests multiple explosion locations, as well as unstable or directed jets, affected by the initially formed crater walls (Taddeucci et al. 2013;. In particular, the large abundance of OI material with low impact angle (Fig. 4a-f), and the large deposit thickness to the north, point to multiple shallow explosions at later stages of the eruption Table 2 Comparison of petrophysical properties, and median-fine grain size and componentry of naturally/experimentally fragmented material from drill cores vs. breccia blocks and matrix Drill cores Chlorite (1) Pumice (4.7) Fig. 9 Experimental results of ejection velocities (a) and fragmentation velocities (b) plotted as a function of bulk explosive energies (E Expl ) that migrated/excavated northward Kennedy et al. 2020;Montanaro et al. 2020). Geothermal features such as hot-spring conduits, hydrothermal eruption vents, and collapsed craters, are most commonly found close to faults (Curewitz and Karson 1997;Rowland and Simmons 2012;D'Elia et al. 2020). Structural features can also control vent migration during explosive events involving fluids rising through fractures (Ort and Carrasco-Núñez 2009). A local graben contains the collapsed craters and the Champagne Pool and creates a morphological buttress on the west and south sides of the pool. A SSW-NNEstriking fracture cuts through the crater site ( Fig. 1; Lloyd 1959;Curewitz and Karson 1997). These all indicate a fracture-controlled migration of the explosion loci, especially during its later eruption stages (Jordan et al. 2013;López-Rojas and Carrasco-Núñez 2015;Graettinger 2018).
Combining the structural and breccia features, we suggest that the initial deep-sourced jet occurred in the southern part of the crater area, and was followed by a decompression/flashing front propagating northward through the OI-hosted shallow reservoir (Fig. 10). Continuous flashing of fluids likely fuelled shallow explosions and subsequently enlarged the crater. A similar mechanism is described for the formation of craters at Lake Okaro and Whakaari/White Island Montanaro et al. 2020). Slumping of shallow unconsolidated deposits into the central vent area likely intermittently blocked it, inducing further deep explosions and sporadic debris jets transporting deeper clasts into the shallow ejection zone (cf., Sweeney and Valentine 2015). This, along with recycling of the lower breccia bed during crater widening, explains the deeply sourced fragments in the OI-dominated upper breccia bed. The presence of argillic-altered layers within the base of the OI deposit (Hedenquist and Henley 1985) could have also contributed to slumping/collapse during the northward excavation (Fig. 10). The breccia distribution pattern may also reflect strongly asymmetrical jets (cf., . Once the eruption ended, collapse and compaction of the material occurred in the northern basin (Fig. 10). This suggests that the final size of the crater produced by the eruption itself was approximately 100-120 m (Fig. 1b).

Eruption dynamics and energetic considerations
Here, we compare the Champagne Pool eruption to studies of natural explosion craters (Yokoo et al. 2002;Graettinger and Valentine 2017;D'Elia et al. 2020), and field-based explosion experiments (Goto et al. 2001;Ohba et al. 2002;Taddeucci et al. 2013;Graettinger et al. 2014;Sonder et al. 2015;Macorps et al. 2016). The experiments show that eruption dynamics and  (1), to overpressure geothermal fluids until aquitards were disrupted (2), and rapid fluid decompression began (3). The breaking of harder/low porosity host rocks at depth led to a first narrow/localised bottom-up explosions, eventually only breaking to the surface with low energy and emplaced a small breccia deposit (brown lower bed). The second phase, much more extensive, produced wider ejection of larger particles and a final thicker breccia (yellow upper bed). This occurred when weak/unconsolidated and highly porous units near the surface were disrupted. The transition between the two phases of eruption occurred when deeper eruptions broke through silicified and stronger aquitard layers. The soft substrate disrupted during the second phase collapsed into the crater centre to repeatedly choke and induce ongoing deeper-sourced eruptions while also spreading laterally along a local fault strike (3-4). Final collapse of the crater rim material lent the crater its final shape emplacement of ejecta depend strongly on the combination of explosion depth and energy, and the local topography. The relationship between energy and depth of explosion can be characterised by the scaled depth (physical depth divided by the cube root of energy; Goto et al. 2001). The scaled-depth condition during which the largest craters are produced is 0.004 m/J 1/3 . At scaled depths of~0.008 m/J 1/3 and greater, explosion energy is predominantly confined below ground, with no ejecta . However, if deepseated fragments are incrementally transported upward by explosions to optimal depths (Ross and White 2006;Sweeney and Valentine 2015), they may be eventually transported to the surface. If we assume that the largest crater produced during the initial Champagne Pool eruption was~60 m in diameter (the current pool dimension), the Goto et al. (2001) relationship would suggest an explosive energy of~10 12 J. This value is likely an order of magnitude too low, since it does not include the energy loss to fragmentation (Alatorre-Ibargüengoitia et al. 2010;. Neither does it include the effect of explosivity-enhancing gases (e.g. CO 2 ; Thiéry et al. 2010;Hurwitz et al. 2016) or steamflashing of heated groundwater (Mastin 1995;Thiéry and Mercury 2009;. Thus, an explosive energy of~10 13 J is more likely, corresponding to an optimal scaled depth between 86 and 93 m, and a confinement depth between 172 and 187 m . The lower breccia bed is a massive deposit with abundant blocks sourced from different stratigraphic levels, including clasts of the 100-110-m-deep CT-P, which is well within the scaled conditions suggested above. The upper breccia is thicker, wider spread, and associated with numerous larger blocks within proximal and medial facies. These clasts are dominated by material excavated from the OI unit (~80-70 m). The texture, componentry, and explosion depth estimated for both breccias imply a deposition from ballistic curtains, with extremely rapid sedimentation from a spreading and collapsing jet ( Fig. 10; Graettinger and Valentine 2017).
Theoretical considerations and experimental results reveal an increasing explosive energy with porosity, suggesting that reservoirs hosted within the shallow RI and OI (< 100 m) had a higher explosive potential (17 to~21 MJ/m 3 ) than deepseated (> 100 m) reservoirs (~8 to~16 MJ/m 3 ) within CT-P/L and WI/WIS (Fig. 6a). The exploded host rock lithologies also modulated the fragmentation and ejection behaviour . In this, alteration-driven changes in host rock properties must be considered. Secondary mineralisation can reduce matrix permeability, especially formation of expanded hydrated minerals (clays) and silica precipitation within pores and veins (Dobson et al. 2003;Heap et al. 2019Heap et al. , 2020. Porosity reduction results reduce the content of hot and pressurised fluids, ultimately reducing the fragmentation efficiency . In our study, weaker explosivity is indicated by the coarser median grain size and lower weight % of fine ash (3 ϕ) in experiments (Fig. 6b, c). In addition, silica-strengthening of lithologies affected the fragmentation speed, which does not show the expected positive correlation with explosive energies . For the estimated energies, rocks with porosity < 32% are only partially fragmented (Fig.  6). Among the rocks with higher porosities, the relatively more cohesive/compacted lithologies (which are also generally deeper in this case) show comparatively low production of fines (~1-18 wt% vs.~12-33.9 wt%), as well as lower ejection (196 ± 16 m/s vs. 216 ± 13 m/s) and fragmentation velocities (158 ± 6 m/s vs. 166 ± 9 m/s) than the shallow, weakly silicified units (Figs. 6c and 8a, b). The higher fragmentation velocity and the greater fines generation of the weakly consolidated OI compared with the firmer RI further highlights the effect of rock strength on eruption style (Haug et al. 2013;. Experimental results and field interpretation indicate that the deeper lithologies require higher energy to be fragmented and ejected, whereas fragmenting/ disaggregating the OI and enlarging a crater requires less energy (Morgan et al. 2009;Alatorre-Ibargüengoitia et al. 2010;Galland et al. 2014;Montanaro et al. , 2020. The soft OI lithology, combined with shallow explosions, produced large volumes of ejecta , that were mostly dispersed to the north (Figs. 1b and 10).
The lateral extent of a ballistic curtain deposit also depends upon the explosion energy and on its scaled depth . Although exposure is limited, both the lower and upper breccia bed have high aspect ratios, with thick deposits near the crater, and rapid thinning away from it. The lower bed contains fewer and smaller blocks than the upper bed, the latter including > 50-cm-diameter blocks launched over 200 m from the crater. Thus, if the two beds were sourced from explosions below the main Champagne Pool crater, the upper breccia must have resulted from higher energy conversion into kinetic energy. Experimental data show that deepsourced lithologies (e.g. CT-P/L) produce low ejection velocities, compared with OI and RI. Also, the Champagne Pool eruption likely produced a dense particle-rich jet, so that the friction within the mixture, and between the erupting material and the conduit walls, reduces the kinetic energy (Mastin 1995). This friction loss may have been most important during deep-sourced explosions (Fig. 10).

Conclusions
Host rock properties of geothermal systems strongly affect cratering processes, eruption style, and related hazards during steam-driven hydrothermal eruptions (Mastin 1995;Browne and Lawless 2001;Haug et al. 2013;Montanaro et al. , 2020Heap et al. 2020;Kennedy et al. 2020). Analysis of the Champagne Pool breccia and drill core materials, along with decompression experiments, allowed us to fingerprint the eruption source depth and fluid state during all phases of the eruption. Stratigraphic boundaries, along with variable degrees of alteration affecting porosity, and host rock strength played a key role in eruption dynamics. Heat and gas pressure from magma below was focused into pre-existing faults and fractures, to overpressure geothermal fluids until aquitards were disrupted and rapid fluid decompression began. The geometry of the deposits with harder, low porosity units at depth led to the first narrow/localised explosions, initially only breaking to the surface with low ejection energy. The second phase was laterally much more extensive and, overall, more violent at the surface, with wider ejection of larger particles and formation of thicker deposits. This occurred when shallow weak/unconsolidated and highly porous units were disrupted. The transition between the two eruption phases occurred when deeper eruptions broke through a silicified, strong aquitard layer. The contrasting rock properties remain reflected in the champagne-glass-like geometry of Champagne Pool.
Investigations of similar eruptions under different geological conditions should be undertaken to widen our understanding of how local host lithologies and associated hydrological conditions affect priming/triggering mechanisms, eruption dynamics, crater size, and deposition mechanisms. From this study, we suggest that if predominantly soft, unconsolidated, and highly porous strata are disrupted during hydrothermal eruptions, broad and irregular craters will form. Crater shapes may reflect also concentrated decompression and fluid expansion along rift or fracture orientations. Eruptions in soft substrates may be highly irregular, with repeated blocking and explosive clearance of central deeper vents by repeated slumping of crater walls. By contrast, hydrothermal eruptions within harder, less porous material, may cause narrower, deeper craters, with a greater tendency to focus fluids expansion into smaller, localised areas. Consequently, these may also produce more localised deposits from lower-intensity debris jetting and a more rapid caseation of eruption as pressure equilibrates because crater walls are stable.
Substrate/host rock settings play a key role in hazard intensity metrics of hydrothermal eruptions, which are fundamental for impact assessment and planning. It is apparent that both parent geology and alteration-induced lithological changes (e.g. silicification) are important for generating a depthvariability of host rock properties (e.g. porosity and rock strength) that strongly influence the dynamics and spatial impact of hydrothermal eruptions. By understanding the local geology, geothermal alteration properties, and hydrology of active geothermal areas in active tectonic settings, first-order eruption scenarios can be proposed for different geological and geometric settings across susceptible areas, taking into account pre-existing faults and other structural features that focus fluid flow from below.
Author contributions A.G. and C.M. led the project and wrote the manuscript draft. Fieldwork and sample collection were carried out by A.G., C.M., and S.C.. A.G. and C.M. performed the laboratory measurements of porosity, density, and P-wave velocities with help from B.S.. A.G. and C.M. also performed the decompression experiments, grain size, and video analyses. A.G. and C.M. performed the thermodynamics calculations. All authors contributed to the writing of the manuscript.
Funding Open Access funding enabled and organized by Projekt DEAL.

Compliance with ethical standards
Competing interests The authors declare that they have no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.