Havre 2012 pink pumice is evidence of a short-lived, deep-sea, magnetite nanolite-driven explosive eruption

The Havre 2012 deep-sea rhyolite eruption went unobserved and was initially recognised from a massive pumice raft at the sea surface. Havre pumices are predominantly white or grey, however pink pumice is common in the raft. In subaerial explosive eruptions, pink pumice is understood to result from high-temperature atmospheric iron-oxidation. The presence of pink pumice questions the effusive eruption model for the Havre raft. Here we report results from X-ray Absorption Near Edge Structure spectroscopy, magnetic measurements, TEM imaging and glass chemistry that collectively show the colour results from increasing amounts of magnetite nanolites in the raft pumice glass oxidizing to hematite. This suggests a short-lived but powerful explosive eruption phase penetrated the water column allowing hot pyroclasts to oxidise in air. Our results therefore challenge the known depth limits for explosive eruptions in the marine realm and suggest pink pumice can be an indicator of magnetite nanolite-driven explosive eruptions. Pink and white raft pumice from the 2012 Havre deep-sea eruption experienced prolonged high-temperature atmospheric oxidation of magnetite to hematite which suggests a short but powerful explosive phase, according to microanalytical investigation.

I n silicic subaerial explosive eruptions, magma originating from mid to upper-crustal chamber depths (~3-15 km) at high temperatures (~700-1000°C) can be ejected into the atmosphere as pyroclastic material, mostly pumice and ash in a matter of seconds to minutes [1][2][3] . Silicic pumice, which is formed by quenching of gas-rich highly vesiculated magma is typically white or grey, however, pink pumice is reported from some subaerial explosive eruptions [4][5][6][7] . Pink pumice has been experimentally produced by exposing pumice to high temperatures (>700°C) in oxygen-rich environments 5,8 . Natural pink pumice is therefore believed to be the result of exposure of hot pyroclasts to air during subaerial eruptions [4][5][6][8][9][10] . Due to the relatively short time pyroclasts are exposed to atmospheric oxygen conditions at high temperatures compared to oxygen diffusion rates in glass 11 , we hypothesize that the pink colourisation is likely due to oxidation of magnetite microlites or nanolites to hematite. Magnetite nanolites have recently been identified as having an important role in enhancing magma fragmentation and eruption explosivity as nanolite crystallization upon magma ascent triggers rapid increases in melt viscosity and bubble nucleation [12][13][14][15][16][17][18][19][20][21][22] . These studies have focused on subaerial eruptions but whether nanolites exist and can similarly promote increased melt viscosity and enhance eruption explosivity for deep sea eruptions remains unknown.
Large pumice rafts are typically associated with explosive eruptions [23][24][25][26][27] . The 18 July 2012 eruption of the Havre submarine volcano (Kermadec Arc, NZ) went unobserved but was later recognised from a massive pumice raft (1.2 km 3 ) through direct observation and satellite images two weeks after eruption 28,29 and was likewise initially thought to have recorded an explosive eruption 29 . However, the 2012 eruption was later concluded to be effusive [30][31][32][33][34] based on the discovery of many lava flows, domes, and ash produced during the eruption at the volcano summit 32,[35][36][37][38] and magma ascent models based on seafloor pumice melt properties 30 . The high hydrostatic pressures that would have existed at the summit depth of 900 m below sea level are thought to hinder magmatic volatile exsolution and therefore magmatic volatile-driven fragmentation 39,40 .
A new observation reported here is that in addition to the dominant white, crystal-poor pumice 41 some Havre raft pumice have a range of intensities of pink colouration (Fig. 1). The pink colour is pervasive through the pumice and demonstrably not just a surficial feature or a result of post-eruptive precipitation of hydrothermal minerals. Pink pumice has not been observed in sampled seafloor pumice deposits [30][31][32][33][34]42 . Notably, pink pumice has also been reported from pumice rafts of other unobserved submarine eruptions nearby at Healy (~1150 mbsl) and Raoul SW (~1200 mbsl) volcanoes in the Kermadec arc 43,44 .
Here we show that the Havre pink pumice reflects hightemperature oxidation of magnetite nanolites and microlites and we interpret an explosive eruption column ejected pumice into the atmosphere at near magmatic temperatures (890 ± 27°C 41 ). Magnetic analysis, backscattered electron (BSE) imaging, transmission electron microscopy (TEM) images and X-ray fluorescence microscopy (XFM) elemental maps indicate magnetite nanolites are pervasive in the Havre white raft pumice and are oxidized to hematite in the pink raft pumice. Fe K-edge X-ray absorption near edge structure (XANES) spectroscopy shows that both magnetite and hematite are more abundant in pink pumice than in white raft pumice, and importantly that magnetite nanolites and hematite are absent from the seafloor pumice. We also show that the pink colour can be reproduced experimentally by equilibrating white Havre 2012 pumice at temperatures abovẽ 800°C in air within minutes, reinforcing the interpretation that the pink colouration is the result of pumice interacting with the atmosphere at near magmatic temperatures 5,8 . Our results suggest that magnetite nanolite nucleation played an important role in initiating an explosive phase of the Havre 2012 eruption and is essential for the formation of pink pumice by providing the mineral precursor that oxidises to hematite when exposed to air.

Results and discussion
Havre 2012 pumice characterization. The 18 July 2012 Havre eruption evacuated~1.3 km 3 (bulk volume) of rhyolitic pumice, where 93 % (~1.2 km 3 ) of the erupted pumice volume produced a pumice raft and the remaining 7% (0.1 km 3 ) formed a deposit of giant seafloor pumice surrounding the volcanic summit 29,32 . Near-vent deposits of sunken pumice cover >35 km 2 of the summit caldera and pumice clasts are distinctively large, ranging in size from~1 to 9 m in diameter 30,32,33 . The volumetrically dominant raft pumice and small-volume seafloor pumice are similar in appearance, general composition and texture, suggesting they had shared the same effusive eruptive origin 30 . Although recently it has been proposed submarine eruption models based on seafloor pumice deposits are not necessarily appropriate to estimate mass eruption rates of pumice rafts 41,45 .
Havre pumice raft material began washing ashore across the SW Pacific 28,41,46 from November 2012 (Supplementary Notes 1, Supplementary Fig. S1). While the majority of the pumice comprising the raft and seafloor deposits is white to banded white/grey, a substantial proportion of the stranded raft pumice collected from Tonga and along the Eastern coast of Australia is light brown-tan to red-pink 41 . (Fig. 1a-d; Supplementary Fig. S2; colours based on the 2018 Munsell colour chart). This colour commonly occurs in highly vesicular white pumice clasts that contain coalesced vesicles (up to 1 cm diameter) where pumice clast permeability is highest.
Pumice colour can reflect magma composition and mineralogy 47,48 . Havre pumice consists of vesiculated glass, phenocrysts and microlites. All Havre pumice types have a similar minerology 17,21 and comprise phenocrysts of plagioclase, orthopyroxene, and magnetite with trace clinopyroxene and rare quartz 32,41 with a subtly higher proportion of phenocrysts present in the seafloor pumice (compare Fig. 1d and Fig. 1e, f) 34 , ruling out elemental variations between pumice types as a cause for differences in colour.
Micron scale XFM images show discrete regions of high Fe abundance ( Fig. 1i-l) in the white and pink raft pumice, which we interpret as magnetite microlites. In pink raft pumice, high resolution BSE imaging shows that magnetite is typically observed as microlites and nanolites commonly lining vesicle walls in the pink pumice consistent with where colouration is observed and concentrated (Fig. 2, see also Supplementary Notes 5 and Supplementary Fig. S4). XFM analysis also reveals magnetite is present in the seafloor pumice as larger microlites (Fig. 1i). and microphenocrysts (Fig. 1e).
Nanolite Identification. Nanolites can drastically affect melt rheology, and their crystallization can result in an abrupt change in melt viscosity altering eruption behaviour from effusive to explosive eruptions 13,14,[16][17][18][19]22 . Nanolite identification and characterization is therefore becoming increasingly important in understanding eruption dynamics and hazard potential. To document nanolite minerology, size and relative abundances in the Havre 2012 white and pink raft and seafloor white pumice, we use a multifaceted approach of rock magnetics (i.e., magnetic hysteresis and frequency dependence of magnetic susceptibility), transmission electron microscopy (TEM) and complementary EDS analysis of TEM imaged areas.
Magnetic hysteresis analysis provides information on the magnetite mineralogy and grain size characteristics 17,50 . Magnetic data for the white raft and seafloor pumice both indicate the presence of magnetite, whereas the pink raft pumice displays a wasp-waisted hysteresis suggesting a mixture of magnetite and hematite ( Fig. 3. c, f) based on the high coercivity of hematite controlling the closure of the hysteresis loop above 0.6T 51 . The raft and seafloor pumice display thin hysteresis loops with coercivities approaching zero consistent with a superparamagnetic (SP) signature (Fig. 3a-c) which represents the presence of magnetite nanolites <20 nm [52][53][54][55] . However, the hysteresis loops are not perfectly reversible, and the magnetic remanence (M r ) Fig. 1 Characteristics of the main pumice types from the 2012 Havre eruption. a-d Stereoscope images showing the representative colours and textures of the 4 pumice types analysed in this study; white and pink refer to pumice raft material. Note the grey and white pumice bands in the seafloor pumice. e-h XFM-Fe, Ti, Ca-RGB stacked elemental maps for the different pumice types. Boundary between white and grey pumice domains for the seafloor pumice indicated by white dashed line in (e). XFM elemental maps are colour-coded based on a RGB triangle where red, green and blue are end members correlating to high Fe, Ti and Ca, respectively. Where multiple elements are co-located, the pixel displays a colour that represents the relative proportion of each element's concentration. For example, Fe-Ti oxides will appear yellow, orthopyroxene crystals will appear pink and plagioclase crystals blue. i-l Corresponding XFM Fe elemental maps for the areas outlined by white boxes in (e-h). Fe elemental maps are set to a fire colour scale (µg*cm −2 ) where white pixels correlate to areas of high Fe-intensity, indicating Fe-Ti oxide microlites. Yellow pixels represent either orthopyroxene microlites or Fe-Ti nanolite glass mixtures. Scale bars in top row (a, e, i) applies to the entire column below (e.g., a, b, c, d are in the same scale as presented in (a). and coercivity (H c ) is not at zero (Fig. 3d-f), attesting to either a mixture of SP grains (i.e., magnetite nanolites <20 nm) and single domain (SD; magnetite nanolites <76 nm 56 ) and/or multi domain (MD; magnetite >76 nm 56 Supplementary Fig. S5). We note that a number of variables (e.g., mixing of magnetic phases, particle shape, mineral stoichiometry, internal stress, etc.) can influence the domain state interpretation in a Dunlop (2002) plot 58 . This places the high-field magnetic hysteresis data for all pumice types above the theoretical curves for MD + SP or SD + MD trends and data fall in the lower area of SD + SP mixes for SP particles <10 nm 57 ( Supplementary  Fig. S5). At first glance, the H cr /H c ratios (>4) for all pumice types are consistent with purely MD grains. However, the M r /M s ratios are too large (>0.05) in the pink raft pumice and the H cr /H c values are much too high (>10) in the white and pink raft pumice to represent purely MD grains. Although the white seafloor pumice displays a similar hysteresis shape as the white raft pumice and low M r /M s indicates a SP signature, the H cr /H c ratio is between 4 and 10, which is more typical of larger magnetite microphenocrysts with MD grain behaviour (>30 µm) 59 .
To further investigate SP signatures relating to magnetite nanolites <20 nm in the Havre pumice suite, the frequency dependence of magnetic susceptibility (χ fd ) was measured over one order of magnitude frequencies (465 to 4650 Hz). The maximum theoretical χ fd value for a sample with 100% SP magnetite is~16% 60 . In the white and pink raft pumice there is a χ fd of~2.2% and 4.2%, respectively, indicating that there is an abundance of magnetite nanolites <20 nm, mixed with a component of microlites (Fig. 2). In contrast, a χ fd = 0% for the white seafloor pumice indicates the absence of nanolites, attesting to purely microlite to microphenocryst sized magnetic carriers (Fig. 1e, i), despite the similar hysteresis shape to the white raft pumice. This provides further evidence that there is increasing abundance of magnetite nanolites from the white to pink raft pumice, with no magnetite nanolites present in the seafloor pumice.
TEM imaging is utilized here to characterize the size of nanolites bound in the pumice glass. In the white and pink raft pumice we find an abundance of nanolites which are <10 nm in size (Fig. 4a, b). Nanolites are often observed in amalgamations in scanning transmission electron (STEM) images (Fig. 4c, d) consistent with high concentrations of Fe speckles in the corresponding energy dispersive spectroscopy (EDS) analysis areas (Fig. 4e, f). In comparison, no nanolites could be detected in the white seafloor pumice glass consistent with a lack of frequency dependence of magnetic susceptibility.  where black spaces correspond to vesicles or void spaces. Note the view of the preserved concave vesicle wall (outlined by green box) that contains abundant magnetite nanolites to microlites. Blue circles highlight magnetite microlites occurring along vesicle walls. The large yellow crystal in the centre of the image is an orthopyroxene microlite. c An enlarged area of the XFM Fe elemental map of the pink raft pumice (Fig. 1i) showing a vesicle lined with numerous pixels with high Fe concentrations (white) interpreted to be magnetite microlites/nanolites. XFM elemental maps resolution is limited to 5 µm.
recorded at the XFM beamline of the Australian Synchrotron 61 to determine the proportion and distribution of magnetite and hematite nanolites/microlites within the glass framework of a suite of Havre 2012 pumice ( Fig. 3; Supplementary Fig. S7). The results, together with Fe elemental maps are presented in Fig. 5 and show that magnetite is present in the seafloor pumice while both magnetite and hematite are present in all samples of the white and pink raft pumice, with a progressive increase in the amount of magnetite and hematite from white raft pumice to pink raft pumice. Hematite is absent in the seafloor pumice. Comparison of Fe-elemental maps (Fig. 3a-f) and magnetite and hematite distributions (Fig. 5g-l) reveals that higher magnetite and hematite proportions correspond to regions of higher Fe concentration suggesting that bright regions in the Fe maps ( Fig. 5a-f, Fig. 2c) are likely to be magnetite or hematite-altered nanolites and microlites. Close inspection of the Fe maps ( Fig. 5a-f) shows that these nanolites and microlites are concentrated around the vesicle rims (Fig. 2).
The Fe-XANES spectra indicate that the proportion of Fe as Feoxides (magnetite + hematite) in the white seafloor and some white raft pumices are small and nearly identical, with the seafloor pumice containing magnetite and the white raft pumice containing magnetite and small amounts of hematite ( Fig. 5g-i, o). In contrast, pink raft pumice has considerably higher proportions of  Table S3) and frequency dependence of magnetic susceptibility (χ fd ) are utilized to determine that the white and pink raft pumice hysteresis loops represent a mixture of superparamagnetic (SP) < 10 nm and MD magnetite grains, and that seafloor white pumice contains purely MD magnetite. In addition, c, f Havre pink raft pumice displays a classic wasp-waisted hysteresis signature closing past 0.6T suggesting that both magnetite and hematite are present 51 (Fig. 5j-l, o), in agreement with MND analysis that showed substantially more magnetite in the pink raft pumice compared to the seafloor pumice. A strong positive correlation exists between the percentage of both magnetite and hematite and the intensity of reddening in the raft pumice (Fig. 5h-l, o).
To simulate possible eruptive conditions and air interactions that might cause magnetite oxidation and pink colouration, we undertook atmospheric heating experiments (Supplementary Notes 8). White raft pumice was heated to temperatures between 675 and 900°C for 2.5-25 min under atmospheric conditions ( Supplementary Fig. S8). No colour change occurred at 675°C, while the orange-pink to pink-red colour observed in natural Havre raft pumice was produced near magmatic temperatures 41 of 800 and 900°C within 2.5-10 min (Supplementary Fig. S8c, d,  S9a-d). The speckled pink to red staining ( Supplementary Fig. 8d) implies the colour change is due to magnetite microlites and nanolites oxidizing to hematite 8,18,62 . The experimental samples with more abundant larger vesicles that display a greater degree of interconnectivity developed more intense and penetrative colouration ( Supplementary Fig. S10).
Tait et al. 8 hypothesized subaerial pink pumice forms from high-temperature oxidation of Fe stored either within the pumice glass or magnetite microlites. Our analyses indicate that hightemperature oxidation of magnetite nanolites and microlites to hematite is the primary mechanism for the development of pink colouration. Furthermore, the results suggest a substantial proportion of magnetite microlites and nanolites need to be available to oxidize to hematite for a visible pink colouration to develop. Some white raft pumice does not have sufficient hematite to become visibly pink. We hypothesize that this is due to insufficient time spent in the air at high temperatures to convert the available magnetite to hematite. The required time is likely dependent on the amount of precursor magnetite available in the pumice glass and clast permeability to allow oxygen penetration and interaction with the nanolites.
Evidence for a short-lived explosive phase of the Havre 2012 eruption. We have shown that pink pumice forms when hot, magnetite microlite and nanolite bearing pumice is exposed to air allowing magnetite to oxidise to hematite in sufficient quantities to be macroscopically visible (Fig. 2a). Magnetite and hematite are present in all raft pumice with increasing amounts of both minerals from the white raft to pink raft pumice (Fig. 5). Neither pink colouration nor hematite are present in the Havre seafloor pumice investigated here, consistent with previous studies that concluded that the Havre seafloor pumice did not interact with the atmosphere 30,31,34 . The results of our heating experiments (Fig. S8) support the existing evidence that for hightemperature Fe-oxidation to occur, clasts need to remain at high temperatures (>700°C) for several minutes in the air 8,62,63 .
The effusive phases of the Havre 2012 eruption produced large pumice clasts which could buoyantly rise to the ocean surface from the 900 mbsl; a 9 m sized clast with 90 % vesicularity could ascend in as little as 63 s 30 . Clasts with decreasing size and vesicularity will have increasing ascent times to the ocean surface; clasts <1 m become waterlogged and sink before reaching the surface 30 . This ascent time could allow the core of a large clast to retain some of its original magmatic entropy 64 . As pumice floats in, rather than on the water 26 , pumice that breaks up at the surface due to quench fragmentation will be immediately quenched by seawater 64 . Explosions of the large pumice clast at the ocean surface would therefore be required to eject fragments of initially shielded hot interiors into the air. A key requirement to generate pink pumice is hot pumice to remain airborne for several minutes, which is equivalent to reaching approximately 2 km above sea level 8,65 . Ocean surface explosivity has only been observed to ballistically eject fragments to heights of a few hundred meters 66,67 which does not allow sufficient time to become visually oxidized and form pink pumice. Therefore, any pumice clasts buoyantly rising through the ocean column and exposed to the atmosphere by secondary fragmentation and breakup at the ocean surface cannot be oxidized to form the pink raft or white raft pumice. Furthermore, such a mechanism would only result in a small proportion of the volumetrically dominant raft pumice becoming oxidized. All analysed white and pink raft pumice has been oxidized and contains hematite, indicating the majority of the volumetrically dominant pumice raft is comprised of oxidized material. Importantly then, the mechanism for delivering hot pumice to the surface needs to be capable of delivering the bulk of the pumice raft volume at near magmatic temperatures to the atmosphere where it can remain for several minutes before being deposited on the ocean surface. We propose that an unrecognized explosive phase of the Havre 2012 eruption is required to achieve the conditions necessary to oxidise magnetite to hematite in all raft pumice and generate the pink pumice. We envision a short-lived explosive eruption column breaching the ocean surface that would allow hot pumice to directly interact with the oxygen-rich lower atmosphere (Fig. 6), become air-filled and initiate high-temperature oxidation of magnetite nanolites to microlites in the pumice glass (Fig. 5).
Havre rhyolite magma was primed for volatile saturation and vesiculation on eruption 41 . Magnetite crystallization was likely promoted by degassing and/or undercooling in response to initial decompression 17,20,22,68,69 (Fig. 6b). Continued decompression would explain the observed range in size of magnetite nanolites (Fig. 4) and microlites, where initial nanolites would have grown in size 70 to form magnetite microlites in the raft pumice ( Fig. 1k-l, Fig. 2). Once formed, the magnetite nanolites and microlites would collectively have resulted in a near instantaneous increase in melt viscosity, causing gas bubbles to become isolated in the stiffened melt 18 . Nanolites act as sites for enhanced exsolution of volatiles indicated by coalesced vesicles lined with nanolites and microlites in the raft pumice (Fig. 2) [12][13][14]17 . Through this process, trapped gas would have rapidly expanded on further decompression thereby initiating brittle fragmentation and a violent explosive eruption. The magnetite crystals then provide ample sites for oxidation once exposed to the atmosphere following rapid transfer in the penetrative explosive eruption column. In this model, pink pumice would correspond to the rapidly ascending core of the eruption column ( Fig. 6b; dashed  lines). A steam and gas-rich plume would provide extra protection from water interaction and cooling of the eruption column core, allowing the central parts of the submarine eruption column to potentially behave similarly to a subaerial  Supplementary Fig. S8. o Percentage of Fe as hematite (red) and magnetite (yellow) in each Fe-XANES analysis area. Percentages are calculated by averaging the portion of magnetite and hematite in every pixel of an Fe-XANES area determined by fitting every pixel as shown in (m). The error was calculated as the mean of the normalized sum of the residuals from the linear combination fitting. Samples: a, g white seafloor pumice (same as in Fig.1a,  e), b, c, h, i two different white raft pumice samples (b, c are the same white pumice in Fig. 1a, e), d-f, j-l) 3 pink raft pumice samples with different intensities of colouration: d, j tan-brown, e, k orange-pink (Fig. 1c, g), f, l red-pink pumice (Fig. 1d, h). eruption 34,71,72 . Differences in the proportion of magnetite microlites in the pumice raft is likely due to variable decompression rates in the volcanic conduit, resulting in lower magnetite microlite proportions in the white raft pumice. These variable decompression histories would place white raft pumice as material that is marginal in relation to the core of the explosive eruption column.
It is clear that the Havre eruption had many effusive eruptive phases to produce the multiple summit lava flows and domes 31,32,35 . Our results and analysis, however, suggest that the pink raft pumice is a visual marker recording a violent but short-lived explosive phase to the eruption, which generated an eruption column that breached the ocean surface. This eruption model puts in to question the relationship between the giant seafloor pumice and the pumice raft 13 . The larger magnetite microlite to microphenocryst size and lack of obvious magnetite nanolites (Fig. 1e) 33,34 suggest the seafloor pumice experienced more protracted syn-eruptive crystallization during slower ascent, and therefore, different trajectories on eruption. The lack of hematite in the seafloor pumice (Fig. 5g, o) precludes any hightemperature atmospheric oxidation. If the seafloor pumice is temporally related to the explosive eruptive phase that produced the pumice raft then we suggest that the seafloor pumice represents magma from the margins of the conduit and eruptive column 34 . Erupted material along the column margins would interact extensively with seawater enhancing cooling and promoting sinking 30,37,38,71,73 . Consequently, pumice in submarine explosive eruptions can either be air-cooled and therefore buoyant and float to form rafts 26,27 , or water-cooled and sink to the seafloor near the vent 30,42 . At Havre, this duality is produced by the respective locations of pumice material in the eruption column and resultant trajectories that lead to air cooling of the raft and water cooling of the seafloor pumice.
Our interpretation that the pumice raft of the 2012 Havre eruption originated from a magmatic volatile-driven explosive eruption challenges theoretical models 39,40 of the limiting depth of magmatic volatile-driven fragmentation in the marine realm. Others have shown the importance of magnetite nanolites that form upon decompression in changing the rheology of the melt and providing conditions needed for catastrophic explosive eruptions [17][18][19] . Future studies related to nanolite nucleation would benefit by constraining the rates of nanolite nucleation, the quantity of nanolites needed to shift viscosity over fragmentation thresholds and the rate of volatile exsolution in response to nanolite nucleation for varying melt compositions. Recent modelling of gas jet thrusts in submarine eruptions suggest there is some depth limitation (<1000 m) for an eruption to breach the surface 74 . However, such models could be expanded on to incorporate the effect of nanolite crystallization and the Fig. 6 2012 Havre explosive eruption model. a NASA Aqua Modis (Moderate-resolution imaging Spectroradiometer) satellite true colour image~21 h after eruption onset (01:26 UTC July 18,2012) showing the pumice raft (I) that had covered an area of~400 km 2 . 28,29 , a gas/steam vapour plume (II) extending~50 km downwind, and suspended ash/tephra at shallow water depths (III) surrounding the pumice raft. b Eruption model showing an explosive eruption column triggered by exsolution of volatiles at depth, leading to subsurface magmatic volatile-driven fragmentation. Black arrows represent trajectories of pumice types examined in this study assuming the raft and giant seafloor pumice (GP) were erupted contemporaneously. The explosive eruption column punctures the overlying ocean water column allowing pumice to enter the atmosphere at near magmatic temperature 33 where it can then undergo oxidation. The white and pink raft pumice are erupted through the centre of the eruption column, where decompression rates were highest promoting syn-eruptive growth of magnetite nanolite and microlite crystals. While all pumice ejected into the atmosphere to form the pumice raft likely undergoes some Fe-oxidation, physical characteristics (i.e., clast permeability 5 , temperature & magnetite abundance) and residence time in atmosphere are limiting factors on Fe-oxidation occurring and pink colouration developing. Substantial non-explosive to explosive water-magma interaction along column margins (refs. 38,71,73 ) is also expected, quickly cooling and waterlogging pumice and quenching giant seafloor pumice precluding any high temperature oxidation of magnetite to hematite. Eruption model not to horizontal scale. We acknowledge the use of imagery from the NASA Worldview application (https://worldview.earthdata.nasa.gov/), part of the NASA Earth Observing System Data and Information System (EOSDIS). consequent effects on spontaneous enhanced melt viscosity and volatile exsolution has on explosivity at varying pressures and therefore water depths. This highlights the need for further investigation into nanolite occurrence and their influence on submarine eruption behaviour, particularly as it applies to eruptions which form large and hazardous pumice rafts. We have shown that high-temperature Fe-oxidation of magnetite nanolites to hematite results in the formation of pink pumice upon air interaction. As it is rare to observe these deep-sea eruptions, re-examining pink pumice reported from other deepsea submarine eruptions 43,44 , such as Healy (~1150 mbsl) and SW Raoul (~1200 mbsl) would be useful tests of the eruptive model proposed here. We conclude that pink pumice in pumice rafts is a valuable visual key for the detection of unrecognized atmosphere-penetrating explosive eruptions from submarine volcanoes.

Methods
Sample collection and preparation. Samples were collected from stranded pumice raft material deposited mainly on the eastern shores of Australia as outlined in Supplementary Materials (Fig. S1). Pumice samples for most micro analysis (i.e., BSE, XFM, Fe-XANES) were embedded in epoxy, sectioned, glued to a 1 mm thick fused pure quartz glass slide and then polished to 30 µm thickness.
Electron probe microanalysis (EPMA). Glass composition is quantified by electron probe microanalysis (EPMA) following the methods reported in Knafelc et al. 41 (Supplementary Table S1) and is further discussed in Supplementary materials. Analyses by EPMA were conducted on samples embedded in 1-inch epoxy resin mounts to assure coloured domains for analysis correlate to either white or pink pumice.
X-ray fluorescence microscopy (XFM). The prepared samples were placed in the 2.5 μm FWHM focus of a 18.5 keV X-ray beam from a double-crystal Si (111) monochromator and focused with a Kirkpatrick-Baez mirror pair at the XFM beamline of the Australian Synchrotron 75 . Large areas (5 mm × 5 mm) were mapped with 1 ms dwell per 5 µm pixel with excited fluorescent photons collected with a 384 element Maia detector system 76 placed in its usual backscatter position. Spectra were analysed using the Dynamic Analysis method as implemented in GeoPIXE 77 and quantified through measurements of an Fe foil (Micromatter, Canada). Individual elemental maps were exported as quantitative 32-bit tiff files for further analysis (Supplementary Notes 4; Supplementary Fig. S3).
Backscattered electron microscopy. BSE maps of 500 µm × 500 µm areas of slides from which Fe-XANES spectra were recorded (Fig. 3h, l) were acquired using a TESCAN MIRA3 scanning electron microscope (SEM). BSE maps were produced using an accelerating voltage of 15 kV and a beam current of 720 pA (beam intensity setting 13, spot size 11 nm) at a working distance of 15 mm, and stitching images with a FOV = 256 μm (Fig. S4). Microlite number densities (MND) were calculated using grey scale analysis in ImageJ software 78 of BSE images with FOVs of 276.8 μm (×1000 magnification) and 553.6 μm (×500 magnification) and details are further outlined in the Supplementary materials (Supplementary Table S2).
Rock magnetics. Magnetic hysteresis measurements were conducted following the method of Knafelc et al. 79 . Here, a set of raft (i.e., white, and pink) and seafloor pumice samples were analysed five times and averaged to reduce the noise of the resulting hysteresis loops. The samples were prepared into~1 mm thick cylinder wafers to reduce the influence of phenocrysts (3-5% modal abundance) on magnetic glass analysis. Normalization to the highest magnetization and a 70% high field slope correction were applied to each hysteresis loop. Frequency dependence of magnetic susceptibility (χ fd ) was measured using a Bartington MS2B sensor over two orders of magnitude frequency (465-4650 Hz). Magnetic susceptibility was measured 10 times at each frequency for the raft pumice samples and 20 times for a white seafloor pumice sample.
Transmission electron microscopy (TEM). Transmission electron microscopy (TEM) images were acquired using a JEOL 2100 operated at 200 kV mounted with a TVIPS F416 4K CCD camera (Fig. 4). EDS analysis of TEM image areas was performed with an Oxford X-Max high-sensitivity silicon drift X-ray detector. Powdered samples were treated with a magnet to remove loose magnetic material prior to mounting. Treated powder was sonicated for 5 min in 100% EtOH and drop cast onto 300 mesh lacey carbon copper grids and left to air dry. Once dry, samples were plasma cleaned using an Evactron Plasma Cleaner prior to imaging.
Fe K-Edge X-ray absorption near edge structure (Fe-XANES). Regions of interest (500 μm × 500 μm) were selected from the XFM elemental maps for the acquisition of Fe K-Edge X-ray absorption near edge structure (Fe-XANES) spectra. The fluorescence intensity was recorded at 220 non-uniformly spaced incident energies from 6963 eV to 7463 eV with 2 ms dwell per 4 μm pixel for all samples except Fig. 3i, j, which had a dwell of 0.8 ms per 4 μm pixel using the experimental geometry described above for XFM analysis. Intensity maps for each incident X-ray energy were generated and exported as described above for elemental maps. The intensity maps at each energy were binned to achieve a constant per pixel dwell of 50 ms per pixel to increase the signal to noise ratio yielding a final pixel size of 20 μm for all regions except Fig. 3i, j, which achieved a final pixel size of 50 μm. The energy was calibrated to an Fe foil first derivative peak of 7112.0 eV in transmission mode, before the spectra at each binned pixel were background subtracted and normalised to an edge step of one by subtracting a linear fit to regions below the edge (7021.5 eV to 7081.5 eV) and subtracting a second linear fit to regions above the edge (E > 7200 eV). The spectra binned to a single pixel and normalised as described previously are shown in Supplementary Fig. S6 and show differences in both the peak and pre-edge shapes, together with minor differences in the edge energy. A spatially resolved linear combination analysis (Matlab® lsqnonneg) was applied to pixels where the Fe concentration was above the noise level to identify regions containing hematite and magnetite (Fig. 3m, n; Supplementary Fig. S7) from regions dominated by glass using hematite and magnetite reference spectra 80 and a glass rich rhyolitic pumice spectrum from an area of white raft pumice (Supplementary Table S1) void of visible microlites and nanolites. Pixels where the linear combination analysis yielded poor results were excluded from the analysis. Less than 2% of the pixels were excluded. The magnetite and hematite maps were normalized to the amount of Fe present in each pixel and the mean of the normalized magnetite and hematite presented in Fig. 3o. The error was calculated as the mean of the normalized sum of the residuals from the linear combination fitting. Area presented for Fe-XANES analysis were exported as quantitative 32-bit tiff files for further analysis and presentation.
High-temperature oxidation experiments. High-temperature oxidation experiments were conducted on separate pieces of Havre 2012 white raft pumice clasts between 675 and 900°C for a series of times ranging from 2.5 to 25 min in atmospheric conditions. Most experiments were conducted on white raft pumice reported here, but seafloor pumice was also used. Samples were cleaned thoroughly using a desalination method 41 before the experiments. Each sample was placed into a pre-heated Carbolite 1 bar furnace set to temperature. An inevitable rush of atmospheric air enters the furnace that cools the furnace 20-30°C initially, but the furnace heats back to the specified temperature in 1-2 min. Samples were removed from the furnace and cooled in air after the specified time. Air cooling effectively cools the samples below 600°C in seconds to tens of seconds. Samples used in the experiments ranged in size from 5 to 10 cm and there was no clear indication that the reaction is prohibited or enhanced by the size of the sample.

Data availability
The data used to produce the results of this article can be accessed through Cloudstor at the following link (https://cloudstor.aarnet.edu.au/plus/s/wVeA0yCE3vetH6x).