Downward-propagating eruption following vent unloading implies no direct magmatic trigger for the 2018 lateral collapse of Anak Krakatau

8 National Oceanography Centre, Waterfront Campus, University of Southampton, European Way, Southampton, SO14 3ZH, UK 9 Institute for Risk and Disaster Reduction, University College London, London, WC1E 6BT, UK 10 Department of Ocean Engineering, University of Rhode Island, Narragansett, R.I 02882, USA 11 Volcano Research and Monitoring Division, CVGHM Geological Agency of Indonesia, Jl. Diponegoro No. 57, Bandung, 40228, Indonesia


Introduction
Landslides generated by volcanic flank failure are significant hazards that can cause destructive tsunamis in island settings (e.g., Rosi  observations can constrain) and/or immediately followed by intense explosive activity (e.g., Perttu et al., 2020). The collapse cut the active conduit beneath sea level, resulting in Surtseyan eruptions that produced extensive ash deposits, rapidly buried the collapse scar, fed convective atmospheric plumes reaching 16-18 km (Prata et al., 2020), and involved a higher magma flux than anything recorded in recent decades (Gouhier and Paris, 2019). Here, we seek to determine the role of magmatic activity in the 2018 Anak Krakatau collapse, specifically addressing whether the intense accompanying volcanism was a driver or a consequence of edifice failure, by reconstructing magma ascent conditions spanning the syn-and post-collapse period. This is important not only to understand this particular event, but more generally to identify causes of edifice failure at both active and inactive volcanoes, to develop approaches for monitoring edifice stability, and to understand the relationship between surface mass redistribution and magma ascent behaviour (cf. Petrone et al., 2009; Watt, 2019).

2018 eruption and collapse observations
Observations spanning the collapse period provide important constraints for understanding eruptive behaviour, ash dispersal, and characteristics of the post-collapse tephra-stratigraphy. All times stated in this section are Western Indonesian Time (WIB; UTC + 7 hours).
The 2018 eruption began in late June (PVMBG, 2018). Discharge rates, calculated from MODIS data, indicate that the magma flux peaked in September, gradually waning after October. Intensified eruptions on 22 December produced another peak in activity, but discharge rates had reached comparable or higher levels on ten previous occasions between June and December . Fishers, familiar with the area and near the island during the collapse, reported that the activity on 22 December had increased but was not unusual (Perttu et al., 2020). Infrasound signals suggest intense activity started at ~13:00 on  (Perttu et al., 2020). The same authors report that fishers decided the island was too dangerous to return to after 19:00 and observed lightning in the Strombolian plume at 20:00; and that two highfrequency seismic signals between 19:50 and 20:00 are consistent with small-scale slope failures. In the hour before the collapse, Darwin Volcano Ash Advisory Centre registered a final pre-collapse plume at 20:10 (Perttu et al., 2020). A cessation in audible explosions in the 30 minutes before the collapse broadly coincides with an infrasound signal pause of a few minutes , suggesting a pre-collapse break in surface activity (Perttu et al., 2020).

Figure 1
U10 and U23-2 stratigraphic sections and grain size data. a The physical characteristics of the U10 sequence on northern Panjang (image, sedimentary log and grain size distributions (sieved and laser diffraction data)). b Inset map showing location of Anak Krakatau and sample localities (main sites of KRA-233, U23-2 and U10 = peach circles; additional sites of NP1 and NP2 = dark blue pentagons). c The physical characteristics of the U23-2 ash from southern Sertung (sedimentary log and grain size distribution (sieved and laser diffraction data)). White triangles mark samples analysed texturally and black diamonds mark samples analysed geochemically (XRF and/or EPMA).  et al., 2020). This signal has a sharp onset and peaks at ~21:00, forming an ice-rich but ash-poor cloud advected to the SW (Prata et al., 2020) with ash emission lasting ~40 minutes (described as a blast-like explosion by Gouhier and Paris, 2019). Gouhier and Paris (2019) derive a higher mass-eruption rate for this specific plume (9 × 10 5 kg/s) than for the subsequent sustained phase (5 × 10 5 kg/s). Perttu et al. Current observations, outlined above, demonstrate that the pre-collapse eruptive activity was intense but not unusual. There is no evidence of activity strongly accelerating in the hours before collapse; behaviour on 22 December represented a renewal of vigorous eruptions, but output peaked three months beforehand. A powerful explosive eruption accompanied the collapse, and was distinct from the sustained activity that followed within a few hours, with infrasound signals, satellite observations and aerial photographs suggesting an immediate switch from Strombolian to Surtseyan behaviour as water infiltrated the vent.

Tephra deposits
Ash samples (Supplementary Table 1) were collected from five localities on Panjang and Sertung, islands respectively east and west of Anak Krakatau. Access difficulties limited sampling to one site on Sertung, U23-2 (6.38 km SW of the vent; Fig.1). This was collected in healthy forest, on level ground, at ~100 m altitude and beyond the tsunami inundation limit. The dark volcanic ash sample occurred at the surface as a 1-cm thick structureless layer mixed with leaf litter above an organic soil, consistent with a fall deposit. U23-2 is aligned with the south-westward dispersal of the high-level, ice-rich plume described above, but not with the E-advected low-level plume. Visual and satellite observations show no evidence of post-collapse vegetation damage on Sertung. In contrast, significant ash deposition stripped leaves and branches from the dense forest on Panjang (Fig.S1), consistent with observations of ash-laden plumes drifting over the island for several days after the collapse. On Panjang, surface pits at four sites exposed a well-bedded ash stratigraphy consistent with predominant fall deposition, exceeding 20 cm in places (excluding remobilised surface deposits). At site U10, a flat, open area ~50 m from the shore on north Panjang, ~5 m above high-tide level and 4.09 km from the vent, these deposits directly overlie a pumicerich sand layer mixed with sparse marine shells, deposited on top of organic soil (Fig.1a). A comparable ash stratigraphy was observed further inland (Fig.S2), but with the pumice layer absent.
Samples from U23-2 and U10 were selected for further textural and geochemical analysis. All analytical methods are described in Appendix 1. For comparison, ash (KRA-233) from May 1997 (an earlier Strombolian eruption) was also analysed, collected from a fall deposit on Sertung (Fig.1b), ~3.80 km from the vent.

Physical overview of eruptive products
The physical features (i.e., grain size, componentry, exterior grain surfaces) of the 1997 ash (KRA-233) provide an insight into the products of pre-collapse (Strombolian) magma ascent and fragmentation conditions, which can be contrasted with the December 2018 samples ( Fig.1a and c) to evaluate changes in eruptive behaviour. KRA-233 is poorly sorted (1.2 f) with a unimodal grain-size distribution peak of 2-3 f and a fine ash content of 13% (>4 f or <63 μm; Wohletz, 1983). The sample is dominated by juvenile ash grains (84%) that are predominantly glassy, black and angular ( Fig. 2a  U23-2 is poorly sorted ash (1.2 f), with a unimodal grain size of 2-3 f, and comprises highly angular, juvenile grains (96%) that appear fresh, glassy and glossy. The unimodal grain size characteristics and homogeneous physical appearance of U23-2 ash suggest it is the product of a single depositional event, rather than an amalgamated deposit from pre-or post-collapse activity (i.e. an upwind equivalent of the Panjang Surtseyan deposits). The highly angular ash fragments (Fig.2c) and relatively narrow grain-size range of U23-2 contrast with the U10 samples ( Fig.1a and c), as do several characteristics discussed in later sections.
The U10 sequence can be subdivided into eight distinct ash units (some comprising multiple layers; U10-3 to U10-10), with a total thickness of 21 cm (excluding >6 cm of structureless, reworked surficial ash), overlying a pumiceous sand (U10-2). Beneath this, a black soil (U10-1), rich in ash and rootlets, is inferred to derive from pre-2018 Anak Krakatau activity. U10-2 is a poorly sorted (1.4 f) and structureless layer, defined by erosional contacts with a maximum thickness of 11 cm. Sub-angular/sub-rounded cream pumice fragments dominate the layer (52%), assumed to originate from local coastal exposures of the 1883 Krakatau ignimbrite (cf. Madden-Nadeau et al., 2021), alongside other volcanic clasts and minor (<4%) marine biogenic material (gastropod shells, sponge spicules). Based on these characteristics and an absence of this layer at more elevated sites further inland (i.e., NP1 and NP2; Fig.S2), U10-2 is interpreted as a tsunami deposit resulting from the 2018 landslide (Fig.2d). Above this, the ash units are characterised by generally poor sorting (0.9-1.6 f) and unimodal grain-size peaks at 2-3 f with a fine ash content ranging from ~8 to 33%. U10-3 (sampled twice; 3A capturing the bulk layer, and -3B the uppermost part to avoid contamination from U10-2) is a thin (0.5 cm) and slightly indurated purple ash with an oxidised yellow-brown crust, containing the highest proportion of altered and lithic grains (19%; Fig.2e) of any studied samples. U10-3 is also the only U10 unit that can be well-correlated, based on its distinctive colour, with other 2018 ash exposures on N Panjang (i.e., NP2-1; Fig.S2). Overlying this, U10-4 is an inverse graded, planar bedded brown ash overlain by a thin, very fine ash (3.5 cm in total), dominated by juvenile grains (95%) but with a generally duller and less angular appearance than U23-2. Grain characteristics are similar throughout the overlying sequence. U10-5 (2.8 cm) and U10-6 (2.5 cm) are very fine (modal peak, 3-4 f) brown ash beds, distinct from the rest of the sequence in displaying weak cross-stratification  that becomes more developed near the top of the layers; these may reflect deposition from base surges rather than fallout. The U10-7 -U10-10 layers are again characterised by parallel planar bedded structures ( Fig.1a). U10-7 (black medium ash; 2.5 cm) and U10-8 (a similar deposit; ~1 cm) show slight normal grading, and U10-9 (2.7-5.5 cm in total; up to 10 individual beds on a few-mm scale) and U10-10 (a 2 mm very fine basal brown ash overlain by three normally graded beds; 3.2 cm) display alternate fine and very fine ash beds. Based on its multi-bedded characteristics and aerial observations of NE/E-directed ash-rich plumes (PVMBG, 2018; Prata et al., 2020), we infer that the U10-3 -U10-10 sequence represents deposition from pulsatory post-collapse Surtseyan activity between 22 December and early January 2019, though we cannot correlate exact dates with individual layers.

Exterior grain surfaces
Scanning electron microscope (SEM) secondary-electron images of mounted ash grains were used to examine micro-scale features and potential differences in fragmentation modes between samples. Many grains from all samples display brittle features including stepped (Fig.2g) and conchoidal fractures, as well as river-line patterns that indicate fracturing under mixed-mode stresses (Hull, 1999). Additionally, many grains in U23-2 and U10 have secondary minerals (e.g., cubic NaCl) and/or finer particles adhered to their surfaces; in some cases, these have annealed together, creating irregular moss-like grains (       compositional range in KRA-233 and U23-2 is similar (An48-68). In contrast, U10 microlites extend to An79 ( Fig.6a), with subsidiary peaks at An68 and An75. For all samples, microphenocryst and phenocryst core compositions are more primitive (i.e. more anorthitic; ranges of An62-91 and An45-89, respectively) than corresponding rims (An48-68 and An51-79, respectively). Rims also show more primitive compositions progressively higher up the U10 stratigraphy (U23-2 is also among the least primitive, but does not extend to anorthite contents as low as KRA-233); core compositions exhibit the same trend, albeit less strongly  are slightly more primitive than in rocks erupted in the 1970s (Camus et al.,1987) and KRA 233 (Fo56-68, microlites; Fo62-73, cores and rims) (Fig.6b).

Microtextural observations
2D microlite analysis was conducted on dark vesicular, microlite-rich grains ( Fig.7; Fig.S7) with a glassy lustre; these are inferred to represent juvenile material (D'Oriano et al., 2014), with their crystallisation fabrics assumed to record primary ascent conditions. All 2D textural data are summarised in Table 1.
Plagioclase areal number density (NA), which defines the number of groundmass feldspars per unit area (mm -2 ), is lowest in the 1997 KRA-233 sample (11,194 mm -2 ). NA in U23-2 is also relatively low (  ash). The overall homogeneity of U23-2 in both its deposit-and grain-scale characteristics suggests that it cannot be an upwind equivalent of U10, and was derived from a discrete event. We thus interpret U23-2 as a deposit from the initial explosive pulse that accompanied the lateral collapse at ~20: 55

Unloading effects on microlite textures
Edifice destruction causes an instantaneous pressure reduction in the underlying magmatic system (Pinel and Jaupart, 2005), which may be manifested in microlite textures that reveal decompression conditions during ascent (e.g., Preece (Lofgren, 1980). The textural similarities of U23-2 and KRA-233 indicate that both samples record steady-state conditions characteristic of the Strombolian feeder system that existed until the point of collapse, with relatively low ascent and decompression rates. The absence of textural disequilibrium suggests that U23-2 does not record a collapse-driven pressure perturbation, and we thus infer it derived from the surficial portion of the conduit, and that its fragmentation was an instantaneous, decompression-driven blast-like response to edifice failure (e.g. Alidibirov and Dingwell, 1996). This notion of 'conduit clearing' corresponds with the timing and short-lived nature of the initial explosive pulse However, these samples display decreased NA and feldspar crystallinity, and a slight increase in mean microlite size, consistent with a gradual re-stabilisation of the feeder system with reduced magma ascent velocities (with U10-6 representing the transition towards these conditions). While the absolute ascent and decompression rate values are subject to uncertainties (e.g., water content), the overall pattern from U23-2 and through the U10 sequence is systematic, reflecting disruption of ascent processes across a shorttimescale.
Considering the mixed textural characteristics of U10-3, it implies that unloading disrupted this magma batch while stalling. Compositions of U10-4 and U10-6 then suggest temporarily reduced crystallisation (low K2O, high MgO), consistent with the faster and deeper decompression revealed by microlite textures.
Glasses higher up the stratigraphy (U10-8 and U10-10; Phase C) have slightly higher K2O; reconciled with lower ascent rates, this suggests longer crystallisation times as the system progressively stabilised.
Plagioclase compositions are particularly sensitive to changes in temperature, water pressure (PH2O), and H2O content. U10 groundmass compositions (Phase B + C) extend to notably higher values (An67-79) than U23-2 (Phase A), with the more calcic populations suggesting relatively higher PH2O and temperatures ( Fig.5a) (Couch et al., 2003). The bimodal U10 populations imply tapping and mixing of rapidly ascending magma from across a broad depth range (see Appendix 4 and Supplementary Table 3 for barometry estimates), although the main microlite population is comparable (An50-63) across all samples, indicating crystallisation consistently extended to shallow levels (<80 MPa; Fig.8). Plagioclase phenocryst rims and cores also exhibit a slight increase in An-content higher up the U10 stratigraphy (Fig.6a), consistent with a hotter and deeper origin.

Fragmentation and magma-water interaction
Ash morphological and surface characteristics in U23-2 and the U10 samples indicate variable fragmentation modes caused primarily by two brittle mechanisms. The vitric nature, fracture patterns and relatively high angularity of U23-2 ( Fig. 2d and e) are consistent with sudden decompression, driven by a downward propagating decompression wave, producing brittle fracture and relatively denser textures (Alidibirov and Dingwell, 1996). In U10, blocky and sub-angular morphologies, alongside stepped or riverline fractures on grain surfaces, also signify a dominant brittle fragmentation process. However, the overall grain-size distributions compared to U23-2 (Fig.1a). The collapse also uncovered the subsurface hydrothermal system on the SW flank, indicated by the orange seawater plumes evident in post-collapse satellite images (Fig.S1b).

Reconstruction of syn-and post-collapse eruptive activity
Anak Krakatau's collapse, at ~20:55, immediately initiated decompression-driven brittle magma fragmentation in the shallow conduit. This magma had been feeding Strombolian eruptions and had ascended under conditions characteristic of the preceding months (Fig.9a). Unloading of the conduit elicited a highly explosive, short-lived eruption (Phase A), with rapid plume ascent reaching ~16 km The eruption of hotter, deeper magmas (Phase B), rapidly ascending through the decompressed feeder system (evident from a shift towards rapid nucleation of smaller microlites), followed the initial explosion.
Primary brittle fragmentation was driven by vesicle overpressure, as suggested by the increase in ascent rates and vesicularity. U10-3 defines the onset of extensive seawater interaction, with a vent widening stage leading to sustained Surtseyan activity, producing cock's tail jets and ash-laden, low-altitude plumes 2021), we suggest that the system returned to equilibrium pressurisation conditions within a 1-2 week period as activity waned.

Implications for determining future collapse events and collapse impacts
We have observed no evidence for unusual magma ascent patterns preceding the collapse, and the patterns observed in the post-collapse tephra-stratigraphy can be explained as a magmatic response to a collapse-driven pressure perturbation of the feeder system. This implies that no distinctive pre-collapse magmatic signature (i.e., volcano seismicity, inflation or degassing) would have been apparent as a signal indicative of incipient collapse. However, progressive susceptibility of the SW flank to failure was evident from longer-term deformation and growth patterns. Lateral deformation of the SW flank was identified over ten years before  Pc1 represents an average precollapse conduit pressure. b Phase A: lateral collapse and unloading causes downward propagating decompression within the surficial conduit, and the pressure perturbation induces an intense explosion. U23-2 magma experiences limited seawater interaction, and the dispersal of U23-2 tephra in the convective, icerich plume heads towards the SW. c Phase B: destabilisation, decompression and deeper tapping of the conduit facilitates fast ascent of U10-3 (Surtseyan vent widening) along with U10-4 and U10-6 (sustained Surtseyan activity). Pc2 < Pc1 represents Phase B with conditions of a highly depressurised conduit (Pc2) relative to the pre-collapse conduit pressure (Pc1). d Phase C: system gradually re-stabilised with ascent characterised by lower velocities and decompression rates (U10-8 to U10-10). Pc3 > Pc2 represents Phase C conduit pressure re-stabilising following Phase B eruptions and partial edifice regrowth.
(e.g., Reid, 2004). Together, all these factors pre-conditioned the SW flank for its eventual collapse. For future monitoring of edifice stability at Anak Krakatau or elsewhere, an approach integrating short-and long-term edifice growth patterns with flank deformation monitoring (e.g., Gonzalez-Santana and Wauthier, 2020), and an improved understanding of edifice material properties (cf. Heap and Violay, 2021), may hold the best prospects for refining forecasts of collapse timing.
The Anak Krakatau collapse also reveals the impact of surface-unloading driven disruption on a shallow magmatic system. Although its volume was smaller than other historical collapses (e.g., Ritter Island; However, rapid edifice rebuilding may hinder opportunities to investigate such processes, by concealing failure scars and the stratigraphic record of collapse-associated volcanism.

Conclusions
Our physical, microtextural and geochemical analysis of syn-and post-collapse deposits shows no evidence that intrinsic magmatic changes preceded the lateral collapse of Anak Krakatau. Instead, the intense, accompanying volcanism is interpreted as a response to collapse-driven depressurisation of the magma system and can be divided into three main phases. Phase A involved a syn-collapse eruption triggered by decompression of the shallow conduit, generating a powerful explosive pulse and depositing ash to the SW. Textures in this ash record pre-collapse ascent conditions, excluding a direct magmatic trigger for edifice failure, suggesting the collapse resulted from longer-lived structural and gravitational instabilities arising from edifice development.
Phase B reflects successive post-collapse tapping of deeper, hotter magma batches from the depressurised conduit, with extensive degassing and accelerating ascent rates. Gradual re-stabilisation of conduit conditions occurred in Phase C, as rapid edifice regrowth led to waning activity.
The 2018 collapse highlights that lateral collapses are not necessarily directly triggered by immediate shifts in magmatic behaviour. Therefore, effective volcanic monitoring and forecasting of such events may need to focus on identifying areas with increased susceptibility to failure, as signalled by changing edifice growth patterns and flank deformation; this will be particularly relevant for the future growth of Anak  (Cashman, 1992).  (5) glass shards. Further subdivisions of vesicular and dense grains were made based on the degree of groundmass present (Fig.S7 and Fig.S8). Vesicular grains are moderately to highly vesicular with irregular vesicles, and can contain numerous small microlites (>50%) or few microlites (<50%) within a glassy matrix. Vesicle morphologies within grains vary depending on crystallinity. Microlitepoor grains typically exhibit smaller, spherical, and isolated vesicles, whereas microlite-rich grains contain irregular and coalesced vesicles. Dense grains are blocky and poorly vesicular with a variable microcrystalline texture (similar to vesicular grains mentioned above) in a glassy matrix.
Free crystals are either whole or fragments of feldspar and pyroxene crystals. Holocrystalline grains are fully crystallised, which may represent recycled grains or parts eroded from the conduit.
Glass shards are broken fragments of the glassy matrix, marked by highly concave outlines (c.f.

Liu et al., 2015a).
Between different grain size fractions, the most significant trends are associated with the vesicular grains (including glass shards) and dense grains (including free crystals and holocrystalline grains). The proportion of vesicular grains increases with larger grain size fractions, whereas the proportion of dense grains decreases with larger grain size fractions. Both trends are broadly consistent with the ash morphology results (see Section 4.3). U10-4 also displays a higher proportion of microlite-poor ash grains, which corresponds with its low K2O contents from glass data.

Appendix 4: Thermobarometry estimates
To constrain pre-eruptive magmatic temperatures and crystallisation pressures, we used the plagioclase-and orthopyroxene-melt thermobarometers of . Temperature and pressure estimates were both iteratively calculated in the Python3 tool Thermobar (v.0.0.9; Wieser et al., 2021) using phenocryst rims and average matrix glass compositions (Supplementary Table   3).