Introduction

The eruptive styles of active volcanoes are difficult to anticipate, but they are key for proper hazards assessment and crises response of impending eruptions. Eruptions can transition from effusive to explosive and vice versa between eruptions for example at Merapi (Costa et al. 2013; Preece et al. 2014), or within one eruptive period such as at Rabaul (Bernard et al. 2022), Redoubt (Bull and Buurman 2013), and more recently at Soufriere St. Vincent (Joseph et al., 2022). Such transitions in eruption style depend on several parameters, including melt volatile contents, melt compositions, magma rheology, and/or magma ascent rates, and these can rapidly change during transport in the conduit and ultimately determine whether magma “flows or blows” (Dingwell 1996; Rutherford 2008; Cassidy et al. 2018). Among the parameters that affect magma explosivity, magma ascent rates appear to exert critical control (Cassidy et al. 2018), but they are generally difficult to determine. Using geophysical data, the timing, and the location of seismicity that corresponds to the run-up to eruption have been used to obtain constraints on magma ascent rates (e.g. Passarelli and Brodsky 2012; Rasmussen et al. 2018; Roman and Cashman 2018), but most estimates are from experimental and petrological studies. These include quantitative analysis of crystal size distribution and microlites density (Couch et al. 2003; Rutherford 2003; Toramaru et al. 2008) and melt decompression experiments (Martel and Schmidt 2003). In addition, chemical zoning of volatile elements in minerals or melt embayments can be modelled using diffusion laws, from which timescales and magma ascent rates can be obtained (e.g. Myers et al. 2019; Li et al. 2020; Georgeais et al. 2021). Combination of micro-textures and geochemistry of the crystal cargo can yield timescales that are comparable to some volcano monitoring techniques (e.g. seismicitySaunders et al. 2012; Kilgour et al. 2014; Albert et al. 2015). These timescales of magma ascent can then be converted to ascent rates assuming magma storage depths are also known. Such depths below the volcano can be obtained from phase equilibria experiments (e.g. Andújar et al. 2013; Cadoux et al. 2014; Utami et al. 2022), or from mineral-melt geobarometry (Blundy and Cashman 2008; Putirka 2008). Alternatively, geophysical studies employing tomography can be used to calculate storage depth of many volcanoes (Puspito and Shimazaki 1995; Widiyantoro and van der Hilst 1996; Widiyantoro et al. 2018), although the uncertainty of magma storage depths of some of these studies yield larger ranges of potential magma ascent rates.

Among the various petrological tools available to constrain volatile budgets and magma ascent rates, apatite is an increasingly popular mineral (e.g. Stock et al. 2016; Humphreys et al. 2021; Bernard et al. 2022), as it can incorporate several volatile elements in its crystal structure [Ca5(PO4)3(Cl, F, OH)]. Apatite has been used to investigate a range of processes, including the pre-eruptive melt volatile concentrations (Scott et al. 2015); estimate volatile budgets and ascent rates (Li et al. 2021; Bernard et al. 2022); and reveal the relationship between volatile dynamics and an eruption’s explosivity (Stock et al. 2016); the record of metasomatism and fluid movement in the crust (Harlov 2015); track arc magma evolution (Nathwani et al. 2020); record carbonate assimilation into magma (Wang et al. 2014), or to investigate estimate lunar water contents (Boyce et al. 2014).

Here we use apatite to constrain magma ascent rates from three historical eruptions at Kelud (Kelut) volcano in East Java, Indonesia. Kelud has produced some of the most hazardous historical eruptions in Southeast Asia (Goode et al. 2018; Maeno et al. 2019; Utami et al. 2022). Hazards from explosive eruptions of Kelud are capable of wreaking havoc across the island of Java. For example, the 2014 eruption produced voluminous ash and ejecta that severely impacted several cities. Hazards included building collapse from tephra accumulation, destructive lahars, extensive building damage, and ash inhalation (Blake et al. 2015; Williams et al. 2020). On the other hand, dome emplacements such as in 2007 were less immediately hazardous but may have enabled accumulation of excess volatiles underneath the dome, and thus increase the explosivity potential of the next eruption (Utami et al. 2022). In this study we characterized the volatile chemistry of apatite crystals from the 1990, 2007, and 2014 explosive and effusive eruptions of Kelud. We also modelled the zoning of Cl, F, and OH in apatite crystals to obtain timescales of magma ascent and degassing. From this parameter we calculated magma ascent rates that, combined with magma viscosities, are used to infer the possible magmatic processes associated with each eruptive style.

Kelud volcano and its historical eruptions

Kelud (Kelut) is an active volcano in East Java, Indonesia classed as a basaltic andesite semi-plugged strato-cone (Whelley et al. 2015) that has produced > 30 explosive eruptions and effusive dome extrusions since 1000 CE, with a mean recurrence interval of ~ 23 years (e.g. Global Volcanism Program 2013; Kusumadinata and van Padang 1979; van Padang 1983). The volcano is monitored with a dedicated observatory (Global Volcanism Program 2013) by the Centre of Volcanology and Geological Hazards Mitigation of Indonesia (CVGHM, or PVMBG in Bahasa Indonesia). The explosive eruptions in 1990 and 2014 were short-lived, with unrest signals escalating within hours from the onset of unrest to eruption, making them challenging to forecast (Nakamichi et al. 2017). In contrast, the 2007 dome extruded effusively after a month of unrest, which included elevated seismicity and carbon dioxide emissions into the lake (Hidayati et al. 2009; Caudron et al. 2012). Unrest signals for all three eruptions consisted of heightened seismicity, deformation, and changes of the lake temperature above background levels (Lesage and Surono 1995; Vandemeulebrouck et al. 2000; Hidayati et al. 2019) (see Table S1). Both the 1990 and 2014 explosive events were preceded with an almost exponential increase in seismicity around 12 h before each eruption (Lesage and Surono 1995; Hidayati et al. 2019), but this trend was not observed before the 2007 dome. Several short-lived unrest episodes or crises occurred after the 1990 eruption and until the 2014 eruptions (Global Volcanism Program 1991) including 1996 and 2001, but they did not result in an eruption (Global Volcanism Program 1996; 2001).

Previous studies on timescales of magmatic processes at Kelud used the Mg-Fe interdiffusion in reversely zoned pyroxenes before the 1990 eruption (Utami et al. 2021), which they compared with timings of seismic and hydroacoustic unrest and run-up times (Lesage and Surono 1995; Vandemeulebrouck et al. 2000). Timescales of Mg-Fe interdiffusion in pyroxenes span weeks to months, and were attributed to degassing and fluid accumulation. These timescales also overlapped with the onset of seismic and hydro-acoustic unrest that began around three months before the eruption. The exponential run-up in seismicity that began ~ 12 h before the onset of the 1990 eruption gave magma ascent rates of ~ 0.2 m s−1 based on magma storage depths of 6–7 km (Lesage and Surono 1995; Utami et al. 2021, 2022). Other estimates of magma ascent rates for the more recent 2014 eruption are around 4–5 m s−1 (Cassidy et al. 2019), which were calculated using shallower storage depths of 2–3 km.

Methodology

We studied apatite crystals from one main-stage and one late-stage pumices of the 1990 deposits, one fragment from the 2007 lava dome, and two pumices from the sub-Plinian phase of the 2014 eruption. These are the same samples as in Utami et al. (2021) and Humaida (2013), where more details about sample locations and descriptions can be found.

Chemical Analyses of Apatite

We obtained backscattered electron (BSE) images of carbon-coated thin sections for each sample, and identified apatite based on their Electron Dispersive Spectra (EDS) using a JEOL-JSM 7800F scanning electron microscope (SEM) at the Earth Observatory of Singapore (EOS). We then mapped the apatite grains for Mg, Fe, Cl, F, and S in Wavelength Dispersive Spectra (WDS) mode on the JEOL JXA-8530F field emission electron microprobe (EPMA) at the Facility for Analysis Characterisation Testing and Simulation (FACTS-NTU), Nanyang Technological University, Singapore (NTU). Guided by these elemental maps, we selected target crystals for EPMA analyses.

Target apatite crystals were analyzed in spot mode broadly following the methodology outlined in Li et al. (2021), using a 15 kV beam and 10 nA current, but with 1–4 µm beam size. On peak counting times are 550 s (F – 60 s; Na – 10 s; Mg – 100; Fe – 40 s; Mn – 40 s; Ce – 60 s; Ca – 20 s; Cl – 60 s; P – 20 s; Si – 60 s; S – 80 s). Profiles of zoned apatite crystals from the 1990, 2007, and 2014 eruptions were analyzed using a 10 nA current and 1–4 µm beam size. The structural formula of each apatite was calculated using approach #1 of Ketcham (2015), which was adapted based on an estimate of the [OH] content. We used a threshold of < 5% deviation from the ideal cation ratio of 5/3 to discard non-stoichiometric apatite analyses and/or mixtures. The mole fraction of volatile elements in the M2 column anion sites are given as XCl, XF, and XOH, respectively. The OH anion was calculated by subtracting the atom per unit formula of Cl and F from the site total and normalizing the value according to:

$${X}_{OH}= {~}^{1}\!\left/ \!{~}_{2}\right.\times \left[2- \left(Cl +F\right)\right]$$
(1)

Here we refer to the calculated OH as OHc. Relative errors from the EPMA analyses are 1–3% for XCl, 1–4% for XF, and the propagated errors are 1–33% for the calculated XOH. Some of the compositional profiles in the 1990 apatite crystals had major element oxide totals that were < 98 wt. %, perhaps due to thicker carbon coating (> 15 nm) present in some thin sections. We determined the orientation of the crystallographic axes of apatite (α and γ for a and c-axes, respectively) using the Electron Back-Scatter Diffraction (EBSD) mode in the SEM at EOS. These crystallographic orientations were used to calculate the angle between each EPMA profile and the crystallographic axes needed for diffusion coefficient correction (Costa et al. 2008), with an EBSD conversion code employing MTEX Texture Analyses Toolbox. The EBSD conversion code for apatite and other minerals can be obtained with the following link: https://github.com/jago2216/EBSD-Codes.

Modelling timescales and ascent rates from apatite

The Cl-F-OH diffusion in apatite was modelled as a multicomponent system and involves interaction between volatile species (Li et al. 2020), and we used a MATLAB code associated with the online ApTimer application that models multi-component Cl-F-OH diffusion in apatite (Li et al. 2020) (See Supplementary Material). The model also used a finite difference solution of the diffusion equation in one dimension, as it was appropriate given the length of the profiles and size of the crystals we investigated. Due to the large anisotropy of diffusion for many elements in apatite, we used the acute angles between the profiles and the crystallographic a- and c-axes (α and γ, respectively) to correct for Cl, F, and OH diffusion coefficients from Li et al. (2020). The anisotropy is such that the diffusion parallel to the c-axis is considerably faster than parallel to the a-axes (Li et al. 2020). Thus a correction factor was applied to the diffusion coefficients (Di) of three profiles that were parallel to the a-axes:

$${D}_{Cl || a}=166 \times {D}_{Cl || c}\text{, where }{D}_{Cl || a}={D}_{Cl || b}$$
(2)
$${D}_{F || a}=19 \times {D}_{F || c}\text{, where }{D}_{F || a}={D}_{F || b}$$
(3)
$${D}_{OH || a}=46 \times {D}_{OH || c}\text{, where }{D}_{OH || a}={D}_{OH || b}$$
(4)

These factors were determined by first calculating the diffusion coefficients at 950 °C for Cl, F, and OH parallel to the c-axis, and then dividing their values by diffusion coefficients parallel to the a-axis at the same temperature. This was necessary because Li et al. (2020) only determined the diffusion coefficients of these volatile elements parallel to a-axis at 950 °C. For further information, readers are encouraged to consult Li et al. (2020). Here we assume that the activation energy for these elements does not change with crystallographic orientation, which is the case for some minerals and elements (e.g. Fe–Mg in olivine; Dohmen and Chakraborty 2002). We mostly used the Cl profile to obtain best fit timescales as it was the most prominently zoned and precisely measured element, but in a few cases, we used the F profile instead. The initial boundary conditions were chosen by visual inspection of the 1-D profile shape and used to determine the most suitable core and rim compositions. A constant rim composition was used, and diffusion towards the core generated the observed concentration gradients (Li et al. 2020). This is a simplification of the diffusion process, as the composition of the apatite crystal rim likely changed in a complex manner during magma ascent due to loss of volatiles and microlite crystallization. Using compositions that change with time at the boundary would give comparatively shorter times (Costa et al. 2008). We used temperatures of 974–980 ± 39 °C for the three eruptions, as constrained through geothermobarometry and phase equilibria results reported in Utami et al. (2022). All inputs into the ApTimer application are listed in Table 1.

Table 1 Apatite diffusion coefficients and timescales modelled using ApTimer (Li et al. 2020), along with best fit magma ascent rate (vascent)
Full size table

For unzoned apatite from the 1990 and 2014 eruptions, we obtained the maximum time of diffusion using the Cl concentration on the condition that if there was a given diffusion length (see below), we would have been able to detect it in our elemental Cl maps and EPMA profiles. We thus neglected the multicomponent diffusion formulation and used a simplified diffusion equation for the penetration depth from the crystal rim towards the interior to calculate time (Costa et al. 2008):

$$x=\sqrt{4{D}_{Cl}t}$$
(5)

where x is the diffusion distance, which we chose to be 4 µm. DCl is the diffusion coefficient for Cl from Li et al. (2020); no correction factor necessary as several grains we investigated had profiles that were parallel to the c-axis. Magma ascent rates were calculated using the timescales and magma storage depths of 7 km, or storage pressures of 175 MPa derived from phase equilibria experiments and geothermobarometry (Utami et al. 2021, 2022), along with a mean crustal density of 2640 kg m−2 (Jeffery et al. 2013).

Magma viscosity calculations

To quantitatively compare the magma physical properties between the different eruptions, we first calculated the melt viscosity using the formulation of Giordano et al. (2008). For this, we used the calculated matrix glass composition calculated from Utami et al. (2022) for a range of pre-eruptive temperatures (average T = 980 ± 39 °C) and the same pre-eruptive water content in the melt as our starting point (4 wt. %), which was chosen based on previous studies on the same samples (Utami et al. 2021, 2022). To incorporate the effect of crystals, we calculated the effective viscosity for all three eruptions using the Einstein-Roscoe equation (e.gEinstein 1906; Roscoe 1952; Gonnermann and Manga 2013):

$${\eta }_{eff}= {\eta }_{melt} \times {\left(1- \frac{\phi }{\omega }\right)}^{-2.5}$$
(6)

where ηeff is the effective viscosity, ηmelt is the melt viscosity, ∅ is the volume fraction of crystals, and ω is the threshold volume fraction of crystals required for to transition to a rigid system, for which we used the value of 60 vol.%. The phenocryst content of the 1990 and 2014 eruption deposits range from 28 to 47 vol. % (Utami et al. 2022), which allowed us to use the Einstein-Roscoe equation. The phenocryst content of the 2007 dome are up to 70 vol. %, which precludes accurate estimation of the effective viscosity for the dome rocks with Eq. 6. We thus compared several viscosity models capable of calculating effective viscosities for up to 80 vol. % (Caricchi et al. 2007; Costa et al. 2009) and followed a conceptually accurate model (Faroughi and Huber 2015; Moitra and Gonnermann 2015; Moitra et al. 2018), which is explained in greater detail below.

Some viscosity models such as the Moitra and Gonnermann (2015), required detailed information on crystal shape parameters, packing fraction and aspect ratios for both microlites and phenocrysts. However, we made a simplification by concentrating on crystal size and shape parameters of phenocrysts. The mineral assemblages of all samples are dominated by prismatic plagioclase crystals, with some prismatic pyroxene and crystal clots, as previously documented in Utami et al. (2021, 2022). The relatively high crystal content of the 1990 (31–47 vol. %) and 2007 (67–71 vol. %) eeruptions made the maximum packing fraction (ϕm) the most important factor. The aspect ratios of phenocrysts and crystal clots were determined with Feret’s diameter of BSE images using ImageJ (Schneider et al. 2012) for the 2007 dome (n = 80), the 1990 (n  = 100) and 2014 (n  = 78) eruptions. Based on this results, we used parameters corresponding to a suspension of large spheres (S) for the 1990 and 2014 eruptions, and a mixture of large and small spheres (Ss) for the 2007 dome. The parameters we used are reported in Table 3 and nmin was estimated from Fig. 10 of Moitra and Gonnermann (2015). To be able to compare the effective viscosities of all three eruptions, we also calculated the effective viscosities for the 1990 and 2014 eruptions with the same viscosity models as the 2007 dome.

These viscosity models also require the ascent rate and conduit diameter, which are a priori unknown. However, the mass discharge rate (MDR) for the three eruptions were calculated by Maeno et al. (2019), and these can be related to the exit velocity (ve) in m s−1, magma density at discharge (\({\rho }_{discharge}\)) in kg m−3, and conduit cross-sectional area (\({A}_{conduit}\)) in m2 through the following equation (See Table 2 for a summary of calculated parameters):

Table 2 Estimated pre-eruptive conduit geometry of the 1990, 2007, 2014 eruptions
Full size table
$$MDR= {\rho }_{discharge} \times {v}_{e}\times {A}_{conduit}$$
(7)

The magma exit velocity is in principle different and higher than the magma ascent velocity we have calculated from the apatite diffusion models, and the magma density at pre-eruptive conditions is also different and higher than that at exit, when magma can be full of gas bubbles due to volatile exsolution. However, for the 2007 dome eruption the magma ascent velocity and the density we calculate at depth are very similar to the values at exit, and we have used them to obtain a range of conduit diameters and thus calculated the viscosity of the 2007 magma. For the 1990 and 2014 eruptions we used the Einstein-Roscoe equation to calculate their respective viscosities. Moreover, we also used the MDR values and measured density (1400 kgm−3) for the 1990 and 2014 eruptions reported by Maeno et al (2019), which we report in Table 3 and Fig. 6. These values, together with the ascent rates from the apatite, were used to calculate the magma viscosity of the 1990 and 2014 magmas using various viscosity models that would better account for the crystal content and strain rate (Caricchi et al. 2007; Costa et al. 2009; Faroughi and Huber 2015; Moitra and Gonnermann 2015; Moitra et al. 2018). For the 2007 dome, the strain rate was needed for the Caricchi et al. (2007) and the Costa et al. (2009) viscosity models, which we calculated by dividing the magma ascent rate with the conduit radius (Moitra et al. 2018) (Table 3).

Table 3 Input parameters and results of melt and effective viscosity calculations for 1990, 2007, and 2014 eruptions
Full size table

Results

Apatite crystals from the 1990 main-stage pumices are found in the groundmass (Fig. 1a), and as inclusions in pyroxenes and partially reacted amphibole phenocrysts (Fig. 1b, 1c). These apatite crystals are typically subhedral with rounded edges to euhedral, with sizes of around 100–500 µm. We find apatite in similar textural positions in the late-stage pumice (Fig. 1d, 1e), except for the apatite found in the quartz-bearing enclaves, which could have previously been groundmass apatite (Fig. 1d, 1f). Some apatite crystals are found in close textural association with magnetite and ilmenite or are found attached to phenocrysts. The apatite of the 2014 eruption apatite has similar textural relations and sizes as that of the 1990 main-stage pumices (Fig. 1j, 1l), but it is also found inside melt inclusions (Fig. 1k). In contrast, apatite crystals in the 2007 dome, particularly in the groundmass, are significantly larger (< 1000 µm) than either of the 1990 and 2014 eruption deposits (Fig. 1g, 1i, 1h). Apatite in the 2007 dome rocks is also subhedral to euhedral and found as inclusions in melt embayments in glomerocrysts (Fig. 1g) or wedged between phenocrysts (Fig. 1h).

Fig. 1
figure 1

BSE/SEI images of apatite crystals in various textural environments in the 1990, 2007, and 2014 rocks. Abbreviations are as follows: Ap = apatite; Gm = groundmass; Amph = amphibole (pseudo = pseudomorph); Cpx = clinopyroxene; Opx = orthopyroxene; Mag = magnetite; Ilm = ilmenite; MI = melt inclusion. In 1990 main- and late-stage pumices, apatite is found in: (a) the groundmass; (b) as Amph-hosted inclusion; (c) Cpx-hosted inclusion; (d) Opx-hosted inclusion; (e) inclusions in Amph pseudomorphs; (f) as inclusions in quartz-bearing enclaves; (g) glomerocrysts; (h) embedded between phenocrysts; (i) groundmass phase; (j) Cpx-hosted inclusions; (k) melt inclusion daughter phase; and (l) Amph-hosted inclusions. White scale bars are 10 μm, whereas red scale bars are 100 μm

Full size image

Apatite from the 2007 eruption overlap in composition with 1990 and 2014 crystals, but feature texturally distinct populations (Figs. 2 and 3). For example, single 2007 dome apatite crystals mostly overlap with those from the other eruptions in F (1.4–1.7 wt. %) and Cl (0.6–1.0 wt. %) with only two grains recording lower Cl (0.6–0.8 wt. %). The 2007 dome apatite crystals in contact with the melt, such as those found in the groundmass or wedged in between phenocrysts, commonly show reverse apatite zoning with rims are generally slightly higher in Cl than in cores (Fig. 2b). This is confirmed in X-ray elemental maps, where we see that Cl increases from core to rim (Fig. 2c, e). Apatite not in contact with the melt is unzoned (Fig. 2d).

Fig. 2
figure 2

Ternary plots of volatile compositions (in mol %) of point analysis and profiles of apatite and elemental maps from the 2007 dome. Note that the crystals show significant chemical zoning. Ternary plot of volatile compositions (in mol. %) of 2007 apatite point analyses, with white symbols are the 2007 apatite and black symbols are apatite from other eruptions. (b) Ternary plot showing cores and rims compositions of apatite profiles, where rims are relatively more enriched in Cl compared to cores. Elemental maps show spatial variations in Cl and F concentrations: (c) apatite wedged between two phenocrysts; (d) Cpx-hosted apatite, and (e) groundmass apatite

Full size image
Fig. 3
figure 3

Ternary plots of volatile compositions (in mol %) of point analysis and profiles of apatite and elemental maps from the 1990 and 2014 eruptions. Note that the crystals show limited to no chemical zoning of these elements. (a) Ternary plot of volatile compositions (in mol. %) of 1990 apatite point analyses, with white symbols are the 1990 apatite and black symbols are apatite from other eruptions. Circles are main-stage apatite, and squares are late-stage apatite. (b) Ternary plot of volatile compositions (in mol. %) of 2014 apatite point analyses. Below this plot is ternary plot showing cores and rims compositions of apatite profiles. X-ray elemental maps show Cl and F distributions. (c) Cpx-hosted inclusions (2014); (d) Amph-hosted apatite (2014), and (e) groundmass apatite (1990)

Full size image

Apatite from the 1990 and 2014 deposits mostly overlap in composition (Fig. 3), with some exceptions. The 1990 main-stage apatite volatile contents largely overlap irrespective of their textural positions (groundmass or inclusions), with 0.9–1.6 wt. % F, and 0.6–0.9 wt. % Cl. Apatite in the late-stage pumice is, however, enriched in F (2.0–2.4 wt. %), and in Cl (1.1–1.3 wt. %) compared to the main-stage apatite (Fig. 3a). This is consistent with previous evidence of the late-stage interstitial melt being more evolved than in the main-stage pumice (Utami et al. 2021). The 2014 apatite volatile compositions overlap with those of the 1990 main-stage apatite in F (1.0–1.7 wt. %) and Cl (0.6–1.7 wt. %). However, there are some chemically distinct populations in the 2014 eruption (Fig. 3b). For example, the amphibole-hosted apatite crystals have significantly higher Cl concentrations (1.7 wt. %) compared to the rest (0.6–1.0 wt. %).

None of the apatite crystals from the 1990 nor the 2014 pumices are visibly zoned in volatile elements with the range of analytical techniques we have used (Fig. 3b-e). The 2007 dome apatite crystals are different from the 1990 and 2014 eruption in that they are reversely zoned in Cl, F, and OHc over lengths of about 9–25 µm. Rim compositions of zoned apatite are 0.9–3.2 wt. % F and 0.5–1.6 wt. % Cl, whereas core compositions have 1.4–1.7 wt. % F and 0.6–0.9 wt. % Cl. We identified two main types of zoning in the 2007 apatite crystals (Fig. 4a-d vs Fig. 4e-f). One common feature in the groundmass apatite is reverse zoning in Cl, which increases towards the rims (reverse zoning) (Fig. 2). F and OHc are mirror images in concentration of those of Cl from core to rim (Fig. 4a-d). Other apatite crystals show a reverse zoning with an increase in F towards the rims (Fig. 4e-h), alongside a slight increase in Cl, whereas OHc decreases. This zoning is found in crystals that are not exposed to the melt or only partly exposed (e.g. wedged between phenocrysts, melt tubes/re-entrants, or unsealed melt pockets in glomerocrysts). Only the apatite crystals from the 1990 late-stage pumice, and the amphibole-hosted apatite from 2014 display significantly different compositions. In summary, we found that apatite from the 1990 and 2014 eruption are unzoned in volatile elements, whereas apatite from the 2007 dome displayed two types of zoning patterns. The differences in the presence or absence of zoning is likely related to the different modes of magma degassing between the two eruption styles, such that explosive eruptions involve magma ascent rates that are considerably faster than that of domes. In the following sections we interpret these zoning patterns in the context of volcanic processes and historical unrest at Kelud.

Fig. 4
figure 4

Two representative examples of compositional profiles of apatite crystals for the 2007 dome, with diffusion timescales calculated using ApTimer (Matlab version) from Li et al. (2020). Blue dashed line of each profile is the initial profile before diffusion. Red solid line shows the diffused model profile. In type 1 zoning, XCl increases from core to rim: (a) SEI image of groundmass apatite, with red arrow showing the profile. Inset shows a pole figure with the c-axis and a-axes. Orientation of the start of the profile is given as ‘trv’. (b) XCl profile from rim to core. (c) XF profile from rim to core. (d) XOH profile from rim to core, which is a mirror image of XF. In type 2 zoning, XF increases from core to rim: (e) BSE image of apatite embedded in phenocrysts, with red arrow showing the profile. Inset shows pole figure with similar information as in (a). (f) XCl profile from rim to core. (g) XF profile from rim to core. (h) XOH profile from rim to core, which increases as in XF

Full size image

Discussion

Interpretation of the volatile zoning patterns in apatite

We propose that the type of zoning of Cl, F, and OH in apatite during magma ascent can be related to the combined effect of magma degassing (e.g. water loss from the melt) during ascent and the associated microlite crystallization (e.g. Blundy and Cashman 2000; Couch et al. 2003; Martel 2012). In the previous sections, we showed that volatile zoning profiles from the 2007 dome apatite population display two types of zoning: one where Cl increases from core to rim, with variable OH, and F (type 1) (Fig. 4a-d); and another where F increases from core to rim, with variable Cl and OH (type 2) (Fig. 4e-f). However, due to the coupled partitioning and uphill diffusion of OH, Cl, and F into the same crystallographic site of apatite (Li et al. 2020), their concentration gradients are not straightforward to interpret. Water loss from the volatile exsolution typically leads to crystallization of microlites of minerals including plagioclase, pyroxenes, and Fe-Ti oxides (e.g. Blundy and Cashman 2000; Toramaru et al. 2008; Bernard and Bouvet de Maisonneuve 2020). Given that these minerals do not incorporate significant amounts of Cl and F, we propose that the concentrations of these volatiles would increase in the melt, which would also increase in apatite such as in type 1 zoning of zoned apatite crystals. The amount of increase may depend on several interrelated factors, including the value of the partitioning coefficient of volatile elements between apatite and the surrounding melt, and their dependence on intensive melt parameters such as temperature (Li and Costa 2019). The relationship could also change if there is pressure- and temperature-dependent partitioning of these elements between the melt and apatite in the presence of an exsolved fluid phase or brine (Li et al. 2021). Despite this range of possible scenarios, we hypothesize that the volatile zoning in Kelud apatite is likely driven by coupled magmatic-volatile processes, including degassing, microlite crystallization, and in the case of the 2007 zoned dome apatite (which records slower timescales than 1990/2014), diffusion from slow ascent coupled with degassing (Utami et al. 2021). This lack of zoning likely reflects fast magma ascent, and thus limited transport time available to develop measurable chemical zoning.

Apatite crystals with F reverse zoning (type 2) are found as inclusion in other minerals, or only partly open to the melt may behave similarly to volatile trends in melt inclusion embayments (e.g. Myers et al. 2019, 2021; Saalfeld et al. 2022). In this case, the apatite volatile zoning may not be fully controlled by degassing-induced crystallization. Rather, the zoning could reflect a localized temperature decrease, as F partitioning between apatite and melt is more sensitive towards temperature dependence than both Cl and OH (Li and Costa 2019).

Other studies of zoned apatite have shed light on the role of volatile on pre-eruptive processes. For example, Li et al. (2021) compared the volatile budget of zoned apatite from the 2006 versus unzoned apatite from 2010 rocks of Merapi volcano, and proposed that input of deep volatile into shallow magma column before the 2010 eruption led to fast magma ascent rates, and thus higher degree of explosivity. At other magmatic systems including Campi Flegrei and Laacher See, volatile zoning in apatite are interpreted to reflect the progressive volatile saturation in the system as a result of fractional crystallization, and re-organization of the magmatic system (Stock et al. 2018; Humphreys et al. 2021). Even so, few studies use apatite as a tool to explore magmatic timescales and ascent rates. For example, differences in magma ascent dynamics have been observed in stratovolcanoes such as Merapi (Li et al. 2021), and caldera eruptions, such as Rabaul caldera (Bernard et al. 2022). Our case study thus complements existing literature on apatite as an important tool in volcano petrology.

Apatite timescales and their relation to volcanic unrest

The 2007 Dome eruption.

A total of 17 diffusion timescales (with typical diffusion lengths of 10–30 µm) were obtained from 12 different crystals of the 2007 dome rocks. We found timescales of up to 175 days were found, although 65% of them are ≤ 60 days (See Table 1 and Fig. 5). Most of the zoned apatite crystals from the 2007 eruption were in contact with the melt. Moreover, one apatite crystal was included in an amphibole pseudomorph (timescale of around 1 day), which may record time since amphibole breakdown, and the apatite’s exposure to the melt. We believe that such a short timescale reflects the apatite crystal’s limited exposure to the melt with different volatile contents before they were partially to fully enclosed as inclusions in other minerals, compared to those fully exposed in the melt These observations supports our assumption of an initial homogeneous composition, and the hypothesis that the zoning we observe in apatite is related to ascent and associated changes in the melt.

Fig. 5
figure 5

Histogram of timescales of zoned crystals from the 2007 dome of Kelud. The timescales are divided into two groups: timescales where the acute angles of the profile is closer to the c-axis (light green), and timescales where the acute angles of the profile is closer to the a-axis (dark green)

Full size image

We also found that apatite compositional profiles closer to the crystallographic a-axis systematically record timescales that are at least 2.5 times longer (63–175 days) than those closer to the c-axis (1–58 days). This difference is most noticeable where multiple profiles were measured on the same crystal. Such discrepancy could stem from a paucity in the knowledge and limited data on diffusion coefficients of volatile elements along the a-axis. We further observe no systematic difference between the times obtained from crystals that are mainly reversely zoned in Cl vs. those that are mainly zoned in F. Due to the potential kinetic delays between magma degassing, microlite crystallization and reaction of the apatite’s F-Cl-OH concentrations to these processes, the calculated times are likely to be the minima especially for crystals that are not directly in contact with the external melt.

In relating the apatite timescales with the pre-eruptive unrest signals observed before the 2007 eruption, we interpret the cluster of timescales < 30 days as the final stage of magma ascent that began in early October and continued until the eruption date (see Supplementary Material Figs. S1, S2, S3). These dates overlap with the onset of different types of unrest in October 2007, such as seismic swarms, increases in seismic energy released, and inflation/deflation (Hidayati et al. 2019) (Fig. S1, S2). The relationships between the timing from the apatite zoning and the monitoring data thus indicate magma movement and ascent towards the surface prior to the dome emplacement (Fig. S3). Moreover, timescales > 60 days belong to profiles that were measured along the a-axis. Given the limited experimental constraints on the diffusion coefficients along this crystallographic axis, we are cautious about directly correlating these timescales with any unrest signal and ascribing them to magma ascent.

For the 2007 dome, we considered the possibility that volatile diffusion in apatite could continue after/during the lava dome emplacement, as it cooled much more slowly that the explosive eruptions. For this we calculated closure temperature of Cl in apatite, as described in Costa (2021). For a 50 µm crystal and a range of cooling rates (0.001–1 °C/year), we found the closure temperature varies from 554 to 704 °C. This suggests that in principle, Cl re-equilibration could have continued during dome cooling. However, these temperatures are lower than that of a nearly dry solidus of the system, meaning at that point there would be no melt left that the apatite could re-equilibrate and exchange with. Thus we believe that there should be limited or no effect of dome cooling on the retrieved time scales.

The 1990 and 2014 eruptions.

The apatite crystals from the 1990 and 2014 deposits (we observed six in total) are unzoned in Cl, F, and OH at the scale of the EPMA maps and profiles. This lack of zoning likely reflects fast magma ascent, and thus limited transport time to be able to develop measurable chemical zoning. Notwithstanding, we can only calculate the time that would be required to generate an observable zoning length with the analytical techniques we have used. We found that using a length of 4 μm (i.e. the spatial resolution of our measurements) implies a maximum time of diffusion of 7–8 h, which means that the magmas would have been transported from their reservoir to the surface in less than this time. These maximum timescales are shorter than the 1990 and 2014 total unrest (1–3 months—Lesage and Surono 1995; Hidayati et al. 2019; Utami et al. 2021) but are of the same order of magnitude as the increase in seismic unrest directly before the eruptions (12–48 h; Lesage and Surono 1995; Hidayati et al. 2019). The relationship between the maximum timescales and the duration of unrest suggests that for the 1990 and 2014 eruptions, degassing and magma ascent occurred within 12–48 h before eruption. Such short timescales of ascent likely account for the fast escalation of seismicity before eruption.

Contrasting magma ascent rates for explosive and dome eruptions

We can estimate the magma ascent rates of Kelud by combining the timescales from apatite and the magma storage pressure estimated from phase equilibria experiments, mineral geobarometers, and volatile saturation pressures (Utami et al. 2021, 2022). The pressure was converted into depths below the edifice using an averaged density of the crust of 2640 kg m−3, which was previously proposed for the upper crust under Kelud (Jeffery et al. 2013). Utami et al. (2021; 2022) reported a range of storage pressures from 160 to > 400 MPa depending on the minerals, which suggests a melt and mineral distribution across a range of depths underneath Kelud, with most crystals around 100–200 MPa. Here we chose an intermediate pressure of 175 MPa (average of 150–200 MPa) according to phase equilibria experiments (Utami et al. 2022), that best matches the phase assemblage and phenocryst content of the three magmas and corresponds to a depth of ~ 6–7 km below the crater.

We find that the calculated ascent rates for the 2007 dome in this manner vary between 0.4–78 × 10–3 m s−1 but most are between 2.6–39 × 10–3 m s−1 for the 2007 dome (Table 1). These ascent rates from the apatite crystals are comparable or faster than the extrusive rates from other dome building eruptions such as Mt St Helens and Soufriere Hills (1–8 × 10–3 m s−1) (Devine et al. 1998). Our calculated ascent rates are only estimates of their mean values, as magma ascent rate changes depending on processes occurring as the magma moves towards the surface, and the magma storage depth is for the 2007 is less well defined than for the 1990 and 2014 eruptions (Utami et al. 2022). Moreover, Utami et al. (2022) has shown that the 2007 magma re-equilibrated as it moved towards the surface to lower melt water contents of around 1 wt. %. In such a scenario, we calculated a saturation pressure of about 77 MPa with MagmaSat (Ghiorso and Gualda 2015) using the calculated matrix composition given in Utami et al. (2022), along with 336 ppm CO2 from Cassidy et al. (2019). This corresponds to about 77 MPa and 3 km depth, which would imply even slower ascent rates. For an apatite crystal that recorded a timescale of 25 days, the reduction in depth to 77 MPa would result in a threefold reduction in the ascent rate from 3 × 10–3 m s−1 to 1 × 10–3 m s−1. With our longest diffusion timescale of 175 days, we would even reduce the ascent rate to 2 × 10–4 m s−1.

For the 1990 and 2014 magmas we only obtained minimum magma ascent rates of about 3.0 × 10–1 m s−1 from our estimated maximum transfer times, as discussed in the previous section. These are more than an order of magnitude higher than those of the 2007 dome magma. Such a difference in ascent rates would account for the contrasting eruption styles, where the magmas feeding explosive eruptions would be able to retain their dissolved volatile content, but the slow ascent would allow the dome to release the volatiles through permeable degassing (Gonnermann and Manga 2013). These ascent rates are significantly lower than the rates of 4–5 m s−1 previously proposed for the 2014 eruption of Kelud (Cassidy et al. 2019), but we note that the difference in ascent rates also depends on the chosen pre-eruptive magma storage depths (~ 7 km used in our study from Utami et al. (2021, 2022) vs 2–3 km in Cassidy et al. (2019)). Nonetheless, our estimates are still comparable with the range of ascent rates for explosive eruptions from volcanoes including Mt St Helens and Soufrière Hills (Rutherford 2008). We find that ascent rates for explosive eruptions of Kelud overlaps with rates from other historical eruptions from basaltic andesite volcanoes, such as 1974 eruption of Volcan de Fuego, and the 1986 eruption of Izu-Oshima (Cassidy et al. 2018).

Relating ascent rates to eruptive styles

Many volcanoes show a range of eruptive styles, and these are likely controlled by magmatic processes resulting in rapid changes in pre-eruptive and syn-eruptive parameters (e.g. Cassidy et al. 2018; Annen and Burgisser 2021; Annen et al. 2023). Ultimately, magma fragmentation is involved in the explosivity of a given eruption (Dingwell et al. 1993), a process that is highly dependent on the physical properties of the magma and fluids, including the effective viscosity. We hereby attempt to relate the ascent rates to the effective viscosities to investigate the dominant processes controlling explosive and effusive eruptions at Kelud volcano, as proposed by Gonnermann and Manga (2013).

The magma viscosities of 1990 and 2014 explosive events are quite different from the 2007 dome, mainly because of the much higher phenocryst content of the latter (Table 3; Fig. 6). The magma viscosity models of Faroughi and Huber (2015), Costa et al. (2009), and Caricchi et al. (2007) yield comparable effective viscosity ranges for the 1990 (103.8 -107.7 Pa s) and 2014 (103.6–107.2 Pa s) eruptions for a minimum magma ascent rate of > 3 × 10–1 m s−1, respectively. These magma viscosities are also somewhat comparable to estimates from the viscosity model of Moitra and Gonnermann (2015, Moitra et al. 2018) for the 1990 (105.8–109.3 Pa s) and 2014 (104.5–108.7 Pa s), which give slightly higher estimates compared to the previous models, but still overlap with those calculated using the Einstein-Roscoe equation for the same eruptions (1990 =  > 104.2–107.8 Pa s; 2014 =  > 104.0–107.5 Pa s). In contrast, the 2007 dome span a wide range of effective viscosities for the given magma ascent rate (10–4 to 10–2 m s−1). The magma viscosity models of Faroughi and Huber (2015), Costa et al. (2009), and Caricchi et al. (2007) yield effective viscosities for the 2007 dome of 108.0–1013.3 Pa s. The viscosity for the 2007 dome (106.4 -1011.0 Pa s) using the model of Moitra and Gonnermann (2015, Moitra et al. 2018) give slightly lower effective viscosities relative to other models. Despite these range of viscosities between the different models, most of the models either give consistent ranges (Faroughi and Huber 2015; Costa et al. 2009; and Caricchi et al. 2007), or overlap with the other models as in the case of Moitra and Gonnermann (2015, Moitra et al. 2018). Hence we shall consider all models in our interpretation (See Table 3 and Fig. 6).

Fig. 6
figure 6

Viscosity vs. ascent rate and mass discharge rate, modified after Fig. 4.10 of Gonnermann and Manga (2013). The magma ascent rates were derived from apatite diffusion modelling and viscosity calculated for Kelud magmas in Table 3. Coloured boxes represent the region outlined by magma viscosity vs. ascent velocity observed for the 1990 (grey), 2007 (red), and 2014 (blue) eruptions for a range of dissolved water content in the melt. Boxes with dashed lines are the average to maximum mass discharge rates determined by Maeno et al. (2019)

Full size image

We used the relationship proposed in Gonnermann and Manga (2013) to connect the magma viscosity and magma ascent rates to eruptive styles (Fig. 6). We find that the initial processes dominating the 1990 and 2014 explosive eruptions are inertial fragmentation and bubble nucleation, straddling the Hawaiian and sub-Plinian fields. In the 1990 and 2014 sub-Plinian eruptions (e.g. Maeno et al. 2019), we propose that syn-eruptive processes such as degassing and microlite crystallization would promote an increasingly viscous magma that rapidly rose to the surface with buoyancy driven ascent due to fluid accumulation. The magma would then eventually evolve towards sub-Plinian field (assuming to proceed from left to right on Fig. 6), where shear localization, viscous heating, and brittle fragmentation dominate (Gonnermann and Manga 2013). These processes contrast with the dominant process for the 2007 dome, which is primarily permeable outgassing. The data shows some overlap with the sub-Plinian field, but the combination of slow ascent and outgassing would make the 2007 magma so degassed that processes such as brittle fragmentation were unlikely to happen.

The ascent rates we calculated for the 1990 and 2014 events imply mass discharge rates > 105.5 kg s−1, which overlaps with discharge rates (107–108 kg s−1) independently determined for the same eruptions by Maeno et al. (2019). Combining the magma viscosity with the discharge rate for the 2007 dome (5.8 × 103—1.9 × 104 kg s−1) (Maeno et al. 2019), we find that the corresponding velocity becomes slightly higher at around 10–1.5 m s−1. This velocity overlaps with the upper range of the ascent rates determined from apatite, but as the MDR is estimated at the vent exit conditions, the corresponding exit velocities of the 1990 and 2014 eruptions are expected to be significantly higher than the one we estimated with apatite. Moreover, these relationships do not typically consider the presence of an exsolved volatile phase at pre-eruptive conditions as we have proposed for the 1990 and 2014 eruptions (Utami et al. 2022), and which should have an influence on the viscosity, magma ascent rates, and eruptive style relations (Cassidy et al. 2018). Notwithstanding these limitations, we find it remarkable that magma ascent rates derived from apatite are consistent with other parameters such as discharge rates, which potentially enables us to characterize different eruption styles.

Conclusions

The duration and rates of magma ascent are considered to be key parameters that control the eruption style. Here we show that using the volatile zoning (F-Cl-OH) in apatite from the last three eruptions of Kelud volcano (1990, 2007, 2014), it is possible to gain insights into explosive and effusive eruptions at Kelud volcano, Indonesia using apatite. Unzoned apatite from the explosive eruptions of 1990 and 2014 record evidence for minimum ascent rates of > 3.0 × 10–1 m s−1. In contrast, reversely zoned apatite from the 2007 dome rocks record evidence for considerably slower ascent rates (> 0.4 × 10–3 m s−1). The estimated ascent rates for the 1990 explosive eruptions overlap with magma ascent rates constrained through the onset of exponential increase in seismicity directly prior to the eruption as previously observed in Lesage and Surono (1995). Our ascent rates are consistent with ascent rates from other volcanoes with effusive and explosive eruptions (e.g. Mount St Helens, Soufriere Hills, Unzen) (Rutherford 2008), and thus we can establish a lower bound on the ascent rates of explosive eruptions for Kelud and other analogous volcanoes. The reversely zoned apatite from the 2007 dome allows us to estimate timescales of buoyancy driven magma ascent towards dome emplacement, which overlaps with the seismic and geodetic unrest (Lesage and Surono, 1995; Lubis 2014). Magma viscosity and ascent relationships show that whereas the 1990 and 2014 eruptions are typical sub-Plinian eruptions with dominating processes including brittle fragmentation, bubble nucleation, and shear localization, dome emplacement is likely dominated by permeable outgassing. Our study demonstrates the utility of apatite to shed light on variations in eruption style for volatile dominant processes at active volcanoes such as Kelud to record timescales of ascent and rates that are consistent with eruptive style at active volcanoes.