Introduction

The Las Cañadas volcano on Tenerife, Canary Islands, generated at least seven major explosive eruptions during the Quaternary1,2,3. These events resulted in widespread deposition of pyroclastic material, with an estimated volume of >130 km3 (ref. 1,4). Despite considerable scientific interest in the volcano and associated hazards including major landslides5, little is known about the triggers for these large-scale events6. Early eruptions (~2 Ma) involved the explosive ejection of large volumes of phonolite magma, implosion of the volcanic edifice and development of the 16 km wide Las Cañadas caldera1. Products of this explosive phase are preserved within the caldera wall and the extensive pyroclastic apron of the Bandas del Sur (Fig. 1). More than seven ignimbrite units occur within the apron, each recording a separate Plinian eruption that culminated in a caldera collapse event1. Trachytes and phonolites of the Teide-Pico Viejo complex provide evidence that felsic magma remains beneath the centre of Tenerife3,5.

Figure 1
figure 1

Map of Tenerife showing locations of sampling sites.

(a) Regional map of Tenerife showing major geological features and the location of the Bandas del Sur pyroclastic apron. Inset shows the position of Tenerife within the Canary Islands. (b) Detailed map of the Bandas del Sur Region, with sampling locations shown (modified after Ref. 1).

Full size image

We have identified crystal cumulate nodules in ignimbrites from three major explosive eruptions: Fasnia, Poris and La Caleta1. Their distribution is laterally and stratigraphically variable within ignimbrite units, likely as a result of complex pyroclastic processes. These cumulates were not completely solidified before each eruption and retain layers containing abundant inter-cumulus melt <1 cm thick. Similar examples of crystal mush have been recorded from other effusive7 and explosive8 eruptions. However, here we use the ‘live’ cumulate nodules from Tenerife as a tool to examine the repeating magmatic processes that occurred prior to explosive volcanic eruptions. We present major and trace element chemical profiles across cumulate crystals to examine the magmatic processes that occurred during the development of the compositionally stratified magma chamber9, particularly during the final stages prior to eruption10,11.

Each of the mafic nodules contains medium-coarse grained cumulate crystals that are either in grain boundary contact as an adcumulate texture, or are partially separated by layers or domains rich in a microcrystalline, glassy groundmass. These melt-rich domains are bounded by crystalline layers and are quenched upon ejection from the magma chamber. As such, these nodules trap and have preserved the final liquid in contact with the cumulate which was actively forming at the margins of the magma chamber. It is the presence of discrete layers containing up to 80% interstitial melt that sets these apart from regular cumulates.

The most mafic nodules are wherlites, taken to represent material close to the chamber floor. Successively higher layers in the system are represented by gabbro, hornblende gabbro and foid gabbro to syenite. Within the gabbroic nodules, plagioclase (An50–88) is more primitive (mafic) than individual crystals in juvenile pumice from the same eruption6. Well-defined core regions occur within many cumulus plagioclases and are overgrown by oscillatory or simple zoned mantles, occasionally with well-developed sieve textures. Clinopyroxene compositions are comparatively limited, ranging En31–42Fs12–19Wo46–50, similar to, or slightly more Mg-rich, than those in juvenile pumice6. Typically, they have defined cores, with multiple and oscillatory zoned overgrowth mantles. Grain boundaries are generally well preserved, particularly in layers where the cumulus phases are separated by regions of interstitial melt. A key feature of both the plagioclase and clinopyroxenes adjacent to this melt is a thin, optically bright zone (<40 µm wide) at the crystal rim (Fig. 2).

Figure 2
figure 2

Three-component electron microprobe element maps of zoned cumulus phases.

(a) Variation in Na:Ca:K across a plagioclase cumulus crystal from the La Caleta Formation; note that lighter blue colours represent more albitic compositions. (b) Variation in Mg:Fe:Ca across a clinopyroxene cumulus crystal from the Fasnia Formation; note that darker orange colours represent more Fe-rich compositions. Colours are qualitative within each image and cannot be compared. Arrows highlight evolved rim zones at the crystal exteriors. Scale bar is 500 µm on both images.

Full size image

Results

Major element concentrations of minerals vary in response to both melt composition and magma chamber conditions including pressure, temperature, volatile content and oxygen fugacity (fO2)12. However, trace element concentrations are almost entirely a function of melt composition and are largely independent of changes in intensive parameters11. With the exception of rare patchy crystals, concentric zonation is evident (Fig. 2), implying little alteration by post-crystallization diffusion12. Within crystal mantles, compositions oscillate (Fig. 3), but zonation generally shows increasing An# (Ca/(Ca+Na) × 100) in plagioclase and either flat or slightly increasing Mg# (Mg/(Mg+Fe) × 100) in clinopyroxene towards the outer rim (i.e. reverse zoning). This is also reflected in trace element transects by increasing FeO in plagioclase (Fig. 3). However, the most striking feature of the cumulus zoning is a sharp decrease in plagioclase An# and clinopyroxene Mg# at the rims of these crystals, with a corresponding drop in Fe content and Al/Ti over the same distance.

Figure 3
figure 3

Compositional profiles across representative plagioclase and clinopyroxene cumulus phases.

(a) plagioclase and (b) clinopyroxene crystals from the La Caleta Formation, (c) plagioclase crystal from the Poris Formation, (d) clinopyroxene crystal from the Fasnia Formation. Photomicrograph images above the compositional profiles show optical zoning and position of compositional transects, shown below, which were collected from core to rim. The boundary between core- and mantle-regions is shown with a dashed line, where applicable. The grey shading in compositional profiles highlights the evolved zone at the crystal rims.

Full size image

In plagioclase, An# correlates positively with melt temperature and H2O content, with changes in chamber pressure exerting only a minor control13. Additionally, fluctuations in clinopyroxene Mg# can occur in response to changes in melt fO2 (Ref. 14). Therefore, the oscillatory major element zoning may result from closed system processes, including crystal movement along thermal or compositional gradients, or open system processes, such as magmatic recharge11,13. However, the large magnitude changes in An# and Mg# approaching the rim of the cumulate minerals are unlikely to result from variations in intensive parameters alone12 and are more consistent with a sudden switch to a more evolved (felsic) melt composition.

As a trace element in plagioclase, Sr correlates negatively with An# through the crystal mantles (Fig. 3). Although bulk melt composition may influence plagioclase-melt Sr partitioning, particularly in more evolved systems10,13, the dominant control on Sr is its increasing compatibility in plagioclase with decreasing An#15. Thus, the observed Sr zonation pattern is predicted by its changing partition coefficient, in response to this crystal-chemical control. In contrast, melt composition has the greatest effect on plagioclase Fe content13. This is known to increase with melt fO2 (ref. 16) and correlates negatively with temperature and An#13,17. However, within the oscillatory zoned plagioclase mantles, An# correlates positively with FeO (Fig. 3). As such, An-content and temperature may not have had a large influence on the Fe content of plagioclase. fO2-induced variations in plagioclase-liquid Fe partitioning are also unlikely to have significantly influenced Fe zoning, as this can not simply result in the positive correlation between An# and FeO. Variations in FeO are more readily explained by changes in melt composition resulting from repeated recharge of the fractionating magma chamber.

Variation in Al/Ti is a useful indicator of melt evolution in Cr-deficient clinopyroxene crystals18. While Al and Ti concentrations may be affected by temperature, pressure and rate of crystal growth19, the Al/Ti ratio more strongly reflects changes in melt composition14. A minor increase in clinopyroxene Al/Ti with temperature may occur due to the stronger partitioning of both AlIV and AlVI19. Although temperature fluctuations could cause the positive correlation between Al/Ti and Mg# observed within oscillatory zoned clinopyroxene mantles (Fig. 3), Ti concentration also shows a well-defined anticorrelation with Mg#, which cannot result from variations in temperature alone19. Variable pressure is also an unlikely explanation for oscillatory Al/Ti zoning, as closed system convection would only cause small pressure changes (<1 kb)20. Increased crystal growth rates relate to the degree of undercooling and correlate positively with Al and Ti21. This could explain the anti-correlation between Ti concentration and Mg#, so increased growth rates cannot be fully discounted as the cause of chemical Al/Ti zonation. However, all the clinopyroxenes exhibit concentric, rather than hourglass zoning, which would be expected if growth rate strongly influenced chemical zonation21. As such, clinopyroxene trace element zoning more likely records changes in melt composition.

Concordance of the plagioclase and clinopyroxene zoning patterns through the mantles of the cumulate phases is found in each of the ignimbrite units. This is consistent with a fractionating magma chamber, periodically refilled by more primitive melt, rather than fluctuations in parameters such as pressure and temperature (e.g. refs. 22, 23). Petrological evidence, such as sieve textured plagioclase phenocrysts, compositionally distinct phenocryst cores and overgrowth mantles also suggest that open system mixing occurred during chamber development24. However, elemental oscillations cannot be correlated between crystals, indicating these events did not affect the whole chamber equally.

Changes in trace element concentrations accompany major element variations observed at crystal rims. In one plagioclase (La Caleta), the drop in An# at the rim correlates with a substantial (≤ 11%) decrease in FeO concentration. This suggests that the rim zone reflects a significant change in melt composition, with lower FeO and An# indicative of crystallisation from a more evolved liquid. In contrast, rim zones in other plagioclase crystals from the three ignimbrites show an increase in FeO concentration, mirroring the drop in An#. This can be explained by cooling and rapid crystal growth, potentially accompanied by an increase in melt fO2, associated with a change to more evolved melt compositions. During rapid crystal growth, a chemical boundary layer, enriched in plagioclase incompatible elements such as Fe, may form at the crystal-melt interface13. Although such kinetic effects could contribute to the high Fe concentrations at crystal rims, they cannot account for the concurrent drop in An#. Contrasting Fe enrichment and depletion trends observed within rims of different plagioclase crystals are likely to result from varying degrees of undercooling. Both require a significant change in melt composition, regardless of the concentration shift direction. A large drop in Al/Ti accompanies the decrease in Mg# at the rim of most clinopyroxene phenocrysts analysed in this study and is taken as a further indication of a large-scale change in melt chemistry.

To test if the interstitial melt is in equilibrium with Tenerifian basaltic liquids or more evolved phonolitic compositions, we recovered and analysed interstitial material from nodules in each unit. Figure 4 shows that these interstices are displaced towards more phonolitic compositions relative to the liquids in equilibrium with the cumulus crystal-forming melts (excluding rims). As such, the “frozen” final liquids within these cumulates confirm that mixing occurred between phonolitic and basaltic magmas before each eruption.

Figure 4
figure 4

Chondrite normalised rare earth element systematics of the Las Cañadas volcanics.

Rare earth data are shown for: 1. Nodule forming liquids (calculated to be in equilibrium with the bulk cumulus phases, excluding rim), 2. Groundmass material, representing the final interstitial liquid and 3. Juvenile pumice. Each of these categories was measured for the La Caleta, Poris and Fasnia Formations. For comparison, Tenerifian basanites-basalts and phonotephrites-phonolites are plotted as the pink and blue data fields respectively. Evolution of Tenerifian liquids is shown as the black line. All data were normalised using the C1 chondrite in ref. 38. Details of the data sources and modelling are presented in the methods section.

Full size image

Discussion

Rim zones are thin (<40 µm), thus final growth is likely to represent only a brief period prior to eruption, probably on the order of 7—132 days25. Low An# and Mg# in these rims suggests final growth in the presence of a significantly more evolved liquid and the sharpness of this change precludes normal fractional crystallization. This is supported by the presence of an intercumulus liquid within the final “frozen” nodule, which has a more evolved composition than the liquid from which the cumulus phases crystallised. Two scenarios could explain the observed zonation: (1) The magma chamber was recharged with felsic melt, sourced from a separate, more evolved chamber. This has been recorded elsewhere26 but on Tenerife would require a third magma reservoir, separate to the Las Cañadas chamber and the source of mafic recharge magma; or (2) The stratified magma chamber became destabilised, causing mixing between evolved material close to the chamber roof with primitive material at the base (Fig. 5)27. Such an overturn may be driven by heating at the base of the magma chamber28, sinking of cold, dense plumes from an upper cupola layer10 or cooling and de-volatilisation of mafic magma close to the boundary with the overlying felsic material causing a density decrease29. Pre-eruptive mixing scenarios are supported by the presence of banded pumice within each formation1.

Figure 5
figure 5

Schematic diagram showing the repeating development of the Las Cañadas magma chamber.

(a) Explosive eruption of the Las Cañadas volcano is triggered by destabilisation of the stratified magma chamber. (b) Incorporation of nodules from all magma chamber units into major pyroclastic density currents1 and resulting caldera collapse.

Full size image

Similar evolved rims have been identified within phenocrysts from Tenerifian lavas and these have been related to overturn of a stratified magma chamber10. The lack of any well-defined resorption surface between the rim and mantle suggests rim growth did not involve a significant change in magmatic temperature, so the eruption cannot easily be explained by felsic recharge10. Cumulate crystals analysed in this study also lack a clear resorption surface between the oscillatory mantle and rim zone and testify to a similar overturn-mixing scenario. Furthermore, there is a lack of evidence for two substantial, high-level, evolved magma chambers coexisting between Plinian eruptions.

Multiple Plinian eruptions punctuated the volcanic development of Tenerife through the Quaternary, separated by quiescent periods ranging between 10 ka and 300 ka. Approximately 170 ka has elapsed since the last Plinian eruption, but this may be irrelevant if magmatic systems can reach critical eruptive states within decades30. Regardless of the timescales, the pre-climactic phase of magma chamber development appears to be consistent and systematic. Large-scale felsic-mafic magma interaction, preserved in partially developed crystal cumulates, appears to be the repeating trigger for destruction of the Las Cañadas magma chamber.

Methods

Electron microprobe analysis

Mineral major element compositions, plagioclase FeO and clinopyroxene TiO2 and Al2O3 concentrations were determined using a Cameca SX-100 five spectrometer electron microprobe in Earth and Environmental Science at the Open University, operating in wavelength-dispersion mode. All analyses were collected using a 20 kV accelerating voltage and 20 nA beam current. Measurements were made along a linear transect outwards from the crystal cores, with a 10 µm beam diameter. Count times range between 20 and 80s per element. Calibration standards include: jadeite (Na), bustamite (Ca), hematite (Fe), forsterite (Mg), K-feldspar (Al) and rutile (Ti). Relative reproducibility estimates (2 sd) obtained from repeat analysis of a kaersutite reference material are ≤ ±1% (CaO), ≤ ±2% (MgO, FeO, Al2O3), ≤ 3% (TiO2) and ≤ 5% (Na2O3).

Solution ICP-MS analysis

Juvenile pumice and intercumulus material were analysed for trace elements by inductively coupled plasma-mass spectrometry (ICP-MS), using a Thermo X-series instrument in Ocean and Earth Science at the University of Southampton. Following HF and HNO3 digestion, samples were diluted by 2000-4000 and introduced via a microconcentric nebuliser. REE ratio precision is estimated to be better than 2% relative (2 sd) based on repeat analyses of rock standard JB-2.

Laser ablation ICP-MS analysis

Plagioclase Sr and clinopyroxene REE concentrations were determined using a Thermo X-Series II ICP-MS interfaced with a New Wave 193 excimer laser ablation system in Ocean and Earth Science at the University of Southampton. Typically measurements were conducted using a 5 Hz laser repetition rate with an 85% output. Count times were 20s and an Ar carrier gas was used. Data were collected along linear transects, equivalent to the previous microprobe measurements, using a 20 µm laser spot size, with a 3 µm bridge between analyses. ICP-MS results were calibrated using NIST 610 and NIST 614 reference glasses. Relative reproducibility estimates for Sr, La, Sm, Dy and Yb are ±3–4% (2 sd).

Data were excluded from both microprobe and LA-ICP-MS transects where analyses sampled inclusions or cracks within crystals. Additionally, points were discarded at the crystal rims where results showed evidence for any incorporation of intercumulus material.

Crystal fractionation modeling

A liquid evolution curve for the Tenerife alkaline magmas was calculated using a starting composition of basaltic lava DH97-28A31. Liquids were calculated using the modal assemblages of basaltic, phonotephrite and phonolite lavas of Ablay et al.32 and the partition coefficients of Fujimaki et al.33, Fujimaki34, Neuman35 and Nielsen et al.36. An initial 60% crystallisation used an assemblage with modal fractions olivine: clinopyroxene: amphibole: magnetite: apatite = 0.296: 0.561: 0.006: 0.115: 0.012. The final liquid of this stage was then crystallised by a further 40% using olivine: clinopyroxene: plagioclase: magnetite: apatite = 0.050: 0.589: 0.198: 0.146: 0.016. Subsequently, this liquid was crystallised by 50% using olivine: clinopyroxene: plagioclase: amphibole: magnetite: apatite = 0.016: 0.190: 0.334: 0.295: 0.138: 0.047.

Data sources for Las Cañadas volcanics

For comparison with our measured data we have compiled rare earth element data for the Las Cañadas volcanics (Fig. 4), throughout the explosive phase of Tenerifian development. Samples include lavas, dykes and explosive ejecta which were sourced from the Bandas del Sur region and the Las Cañadas caldera wall. They are grouped according to their major element composition (i.e. primitive and evolved). Data sources are as follows:

Basalts and basanites:

data sourced from ref. 6,23,31

Phonotephrites to phonolites:

data sourced from ref. 23,31,37