Multi-stage volcanic island flank collapses with coeval explosive caldera-forming eruptions

Volcanic flank collapses and explosive eruptions are among the largest and most destructive processes on Earth. Events at Mount St. Helens in May 1980 demonstrated how a relatively small (<5 km3) flank collapse on a terrestrial volcano could immediately precede a devastating eruption. The lateral collapse of volcanic island flanks, such as in the Canary Islands, can be far larger (>300 km3), but can also occur in complex multiple stages. Here, we show that multistage retrogressive landslides on Tenerife triggered explosive caldera-forming eruptions, including the Diego Hernandez, Guajara and Ucanca caldera eruptions. Geochemical analyses were performed on volcanic glasses recovered from marine sedimentary deposits, called turbidites, associated with each individual stage of each multistage landslide. These analyses indicate only the lattermost stages of subaerial flank failure contain materials originating from respective coeval explosive eruption, suggesting that initial more voluminous submarine stages of multi-stage flank collapse induce these aforementioned explosive eruption. Furthermore, there are extended time lags identified between the individual stages of multi-stage collapse, and thus an extended time lag between the initial submarine stages of failure and the onset of subsequent explosive eruption. This time lag succeeding landslide-generated static decompression has implications for the response of magmatic systems to un-roofing and poses a significant implication for ocean island volcanism and civil emergency planning.

show evidence of coincidental lateral blasts or other types of explosive eruption 24,25 . Thus, terrestrial volcanoes do not show a consistent coincidence of flank collapse and eruptions, and so there are issues with invoking a cause-and-effect relationship. However, past events on Tenerife, in the Canary Islands, may elucidate the range of relationships between flank collapses and large explosive eruptions.
The Canary Islands is among the most active volcanic archipelagos. Indeed, Tenerife is one of the world's largest active volcanoes having experienced numerous Plinian eruptions and caldera-collapses, many of which have ejected volumes up to fifty times greater than the 1980 Mount St Helens eruption [26][27][28] . Furthermore, the Canary Islands, and in particular Tenerife, have incurred numerous large landslides over the last seven million years (roughly every 150 to 250 kyr) with volumes often over 300 km 3 (Fig. 1) [6][7][8] . There is growing evidence that suggests explosive eruptions and mass wasting processes have a cause-and-effect relationship in ocean island settings 7,12,[29][30][31][32][33][34] ; whereas some other terrestrial volcanoes show no such relationship 9,24,25 . During the development of the Canadas Upper Group (1.56 Ma to 0.17 Ma) on Tenerife, in the Western Canary Islands, the Canadas volcano underwent three distinctive phases of growth and destruction, each ending with both a caldera eruption and a coeval lateral flank collapse (Fig. 2) 7,8,[29][30][31][32][33][34][35] . Thus Tenerife presents an excellent opportunity to examine the relationships between these two destructive processes.
Previous studies postulate that caldera-forming and similar explosive eruptions may cause ground accelerations that trigger flank collapse 33,34 . For example, the 2002 explosive eruption of Stromboli Volcano caused a sequence of subsequent tsunamigenic landslides 36 . The involvement of caldera-forming eruptions on ocean islands influencing flank collapse is postulated with the specific reference to the formation of the Icod and Orotava Valleys on Tenerife [31][32][33][34] . However, preliminary evidence from distal marine deposits of the Icod landslide (165 ka) on Northern Tenerife and terrestrial deposits from the resulting landslide-tsunamis suggest that initial stages of multistage submarine flank failure were succeeded by the Abrigo caldera-forming eruption 12,37 . Here, we investigate this relationship between volcanic island landslides and large explosive, often caldera-forming, Plinian eruptions. We wish to understand whether landslides and eruptions are linked, and if so, which comes first. Past work has shown that large-volume volcanic island collapse can occur in multiple distinct stages or sub-units 12,38 . Therefore we also wish to understand whether particular stages of failure are associated with the eruptions and test previous hypotheses regarding the influence of ocean island flank collapses towards changing magma chamber pressure regimes at depth 39 .
Our study examines distal marine deposits, called turbidites, from the multistage Icod, Orotava and Roques de Garcia flank collapses on Northern Tenerife. The ages of these deposits coincide with the timing of Abrigo and Granadilla eruptions that generated the Diego Hernandez and Guajara calderas, respectively, and ignimbrite eruptions that created the Ucanca caldera ( Fig. 2; Supplementary 1) 7,12,27,[29][30][31][40][41][42][43][44] . We compare chemical compositions of volcanic glass from multiple turbidite layers (sub-units) of these individual slides deposited in the basins adjacent to the Western Canary Islands. These data represent newly analysed volcanic glasses and those collated from past studies 12,38 . We also analyse volcanic glasses from the coeval Abrigo and Granadilla eruptions recovered from terrestrial exposures near the town of Tajao, Southern Tenerife. We also collate previous analyses of volcanic glasses from a tephra recovered in ODP Core 953 dated to ~1.0 Ma synonymous with the caldera-eruptions at the end of the Ucanca Formation 45,46 . Comparison of compositions of volcanic glasses from the landslides and eruptions provides a uniquely detailed investigation of the landslide dynamics and their relationships with island volcanism (Supplementary 1, 2) 6,7,38 .
Tenerife's flank collapses and their relationship to caldera eruptions. The Las Cañadas volcanic edifice on Tenerife has a history punctuated by both effusive lava flows and explosive eruptions of phonolitic ignimbrites 31,47 . Indeed, recent volcanism has occurred within three cycles, each culminating in a caldera-forming eruption, these include the most recent Abrigo (160-185 ka) at the end of the Diego Hernandez Cycle, Granadilla (560-600 ka) ignimbrites at the end of the Guajara Cycle, and a series of ignimbrites (1.0-1.  multistage Icod (165 ± 5 ka), Orotava (535 ± 5 ka) and Roques de Garcia (1.15 ± 0.01 Ma) landslides, which broadly coincide with the age of the Abrigo, Granadilla and terminal Ucanca caldera-forming eruptions ( Fig. 2; Supplementary 1) 6,7,12,39 , suggesting that the collapses and eruptions are likely linked 7,12,29,30,37 ; although the nature of that link remained unresolved with certainty.
Turbidites were deposited beneath dilute turbulent sediment flows formed as the slide mass moved downslope under gravity and disaggregated 7,8,12,38 . Canary Island landslide deposits in adjacent deep sea basins, such as those of the Icod, Oratava and Roques de Garcia landslides, are unusual as they comprise a series of stacked graded turbidite sand and mud intervals, called sub-units ( Fig. 3) 12,38,48,49 . These sub-units are compositionally distinct and record landslide emplacement into the ocean in a series of retrogressive stages 12,38,49 . In these distal deep-sea environments over 500 km from the islands, each turbidite sub-unit is often separated by a 5-75 cm layer of clay and fine silt. These fine-grained intervening sediments settled relatively slowly from the waning turbidity current, potentially reflecting substantial time lags (representing at least tens to hundreds of hours) between individual successive sub-unit failures (Fig. 3) 12 . The initial sub-unit(s) within the distal landslide deposits are thicker and contain marine bioclastic sediment and altered volcaniclastics in contrast to later sub-units 12,38,49 . This suggests that the initial stage(s) of each landslide comprised most of the failure volume, and were sourced from submerged parts of the volcanic island 12,38 . Later sub-units are smaller-volume and are composed almost entirely of unaltered volcanic glass from the flanks of the active subaerial edifice 12,38,49 , implicating progressive sub-unit failures migrated retrogressively from the submerged to the entirely subaerial flanks of the volcano 12,38,49 .
We compared the major element composition of volcanic glasses in the sub-units of the Icod and Oratava landslide-turbidites with the volcanic glass compositions of terrestrial ignimbrite deposits of the Abrigo and Granadilla caldera-forming eruptions. These two caldera-related ignimbrites have distinct glass chemical compositions (Fig. 4). Our work confirmed that only the uppermost sub-unit of the Icod deposit (sub-unit 7), the last stage of multi-stage flank collapse, contains volcanic glasses of the same composition as the Abrigo ignimbrite (Fig. 4). A similar relationship is seen for the Orotava deposit, where only the uppermost sub-units (sub-units 4 and 5) contain volcanic glasses that are consistent with the Granadilla ignimbrite (Fig. 4). Importantly, glasses of compositions representing these ignimbrites are not found in initial sub-units of either Icod or Orotava failure (Fig. 4). This implies that the lattermost stages of landsliding during each multi-stage landslide took place following the deposition of the respective ignimbrites; so the caldera-related eruptions preceded these last subaerial landslide stages in a temporal sequence, but succeeded an initial series of submarine flank landslides.
The Abrigo and Granadilla eruptions had volumes of 10 and 20 km 3 , respectively 28 , and likely rapidly covered the subaerial part of the island with erupted material within a matter of hours. Tephra fallout from the eruption columns and co-genetic pyroclastic density currents would emplace erupted material across the island's submerged slopes most likely within hours to days from the start of the eruption 50,51 . Thus should the eruptions have occurred prior to commencement of flank collapse, there should have been both the volume and spatial distribution for that material to be present in the initial stages of the Icod and Orotava flank collapses. Our results imply that the initial failure(s) occurred prior to and thus potentially triggered the final most explosive eruptions in each respective eruption cycle, which in these cases were caldera-forming eruptions. Whilst the landslides do not specifically cause caldera-collapse eruptions, as the magmatic system would have independently developed conditions towards such an eruption, the landslide occurrence may instead escalade the final eruption or induce its occurrence earlier than expected.
A question remains what ultimately triggered the initial submarine stages of each multi-stage landslide leading to the subsequent caldera-collapse eruption and further subsequent subaerial mass wasting? Indeed, could earlier pre-eruptive seismic activity or smaller escalating eruptions trigger the multi-stage flank collapse and commence the aforementioned scenario? Each of the Abrigo and Granadilla terrestrial volcanic successions show a series of pumaceous lapilli fall deposits before the final ignimbrites and tephras representing the caldera collapse eruptions 28,42,52 . Thus eruptive activity prior to and escalating towards the final caldera-forming eruption may be responsible for triggering the initial submarine stages of flank collapse and inducing the aforementioned scenario.
The Roques de Garcia landslide from Northern Tenerife occurred at 1.15 ± 0.01 Ma, dated from an event bed in the offshore Madeira Abyssal Plain 7,8 , supports evidence from the Icod and Orotava landslides. This slide is also represented in the deep sea by a series of stacked turbidites representing multistage collapse, similar to the Icod and Orotava deposits (Fig. 3) 38 . This flank collapse occurs coincidentally with a tephra recovered in ODP Cores 953, 954 and 956 (~1.00-1.15 Ma) that represents an ignimbrite eruption at the end of the Ucanca Formation (1.05 Ma) 45,46 . Volcanic glasses of terminal Ucanca ignimbrite eruptions have compositions only present in the uppermost subunit (sub-unit 4) (Fig. 4). This supports evidence from the Icod and Orotava landslides, whereby the initial submarine stage(s) of flank collapse likely triggered by volcanic activity or related seismicity are followed by a major explosive eruption, in this case caldera-collapse, and finally the last stage(s) of subaerial flank collapse that contain material from the preceding eruption.

Discussion of landsliding influence on explosive volcanism. Our study importantly demonstrates
that whilst early eruptive activity may trigger initial large-volume submarine stages of flank collapse on Tenerife, these initial submarine stages of flank collapse occurred before caldera collapse and potentially induce the eventual explosive eruption. This is demonstrated by only the last stage(s) of the Icod, Orotava and Roques de Garcia failures containing glasses of the respective coeval eruption, thus occurring post-eruption, whilst initial stages are devoid of volcanic glasses of such compositions and are pre-eruption (Figs 3-4). This data supports models that suggest the significant pressure changes involved with large-scale removal of surface material during flank collapse may be responsible for inducing caldera-collapse eruptions 39,[43][44][45] . This is through means of forced decompression on the magma chamber by removal of overburden by landsliding.
One suggestion is the volcanic island flank collapse leads to decompression of relatively shallow magma chambers likely present during the Ucanca, Granadilla and Abrigo caldera-forming eruptions. Although whilst the landslides induced the final eruptions, the style and explosivity of those eruptions were likely constrained by the properties of the magma chamber that had developed. However, the scenario may be more complicated, at least for the 165 ka phonolitic Abrigo eruption. The Abrigo magma was generated at a pressure of 130 ± 50 MPa at a relatively shallow depth of 4-5 km, but was originally water and volatile undersaturated during magma storage 53 . It is  suggested that an input of volatile-rich mafic magma from depth was likely needed to trigger the caldera-collapse eruption. Numerical modelling demonstrates that landslide-induced unloading of about 3% of an island mass can cause a pressure reduction of several MPa at shallow depths below ocean islands 39 . This pressure reduction deteriorates exponentially with depth but is still significant at 18-26 km depths beneath Tenerife where primitive basaltic magmas likely accumulate and differentiate 39,[53][54][55][56] . The landslide decompression may have had the effect of instigating the basaltic magma to rise. Despite landslide-induced decompression the shallow Abrigo magma was water and volatile undersaturated (~3 wt% H 2 O), and may not have erupted explosively 53,55 . Mixing with the deeper ascending volatile-rich basaltic magma may have instigated such an explosive eruption represented by the Abrigo caldera-collapse [53][54][55][56] . Effusion of less evolved lavas with high volatile contents follow large flank collapses and ignimbrite eruptions on Tenerife, supporting the notion of deep magma ascent following induced decompression 29,39,53,54,57 .
For the Icod, Orotava and Roques de Garcia landslides on Tenerife we suggest the following scenario. Firstly, early volcanic activity may initially trigger lateral volcanic island collapse of the submarine flank. Static depressurisation effect caused resulting from flank collapse decompress a shallow magma chamber and induces eruption, and in these cases caldera-forming eruptions. However, where magma compositions were too volatile-undersaturated for such eruptions to commence, static decompression may have caused liberation of mafic, volatile-rich magmas at great depths that ascended and mixed with the shallower volatile under-saturated magma, and finally triggered an explosive eruption 29,39,54 . The static decompression affects of landslides on ocean island magmatic systems is not restricted to the Canary Islands, with similar effects evidenced from multistage collapses on Montserrat, in the Lesser Antilles 57,58 .

Multistage collapse, explosive eruptions and tsunamis.
Here, we present evidence that triggering eruptions by multistage flank failure is not always instantaneous as observed at the lateral blast of the 1980 Mount St. Helens eruption (Fig. 5). Analysis of stacked turbidites of these landslides recovered in sediment cores from the Madeira Abyssal Plain identified between 5 and 75 cm of 2-4 μm clay is deposited between individual sub-units. It has been suggested that to first deposit and then consolidate these fine-grained intervening sediments to resist erosion by successive sub-unit flows would require time lags on an order of hundreds of hours 12 . These time lags have a geodynamic and associated geohazard implication, whereby several days to weeks may pass between successive stages of flank collapse and that similar prolonged time may pass before a triggered explosive caldera eruption may occur (Fig. 5).
Volcanic island flank collapses can also produce destructive tsunamis 7,13,37,56,57,59 . This and other Canary Island flank collapses were multistage, dividing the landslide mass among numerous staggered failures, resulting in reduced tsunami wave amplitudes compared to equivalent single-block subaerial failures, because the volume is the primary control on tsunami amplitude 12,13,38,58 . Modelling of landslide-tsunamis suggest that time lags of at least greater than 20-50 seconds between successive stages of collapse removes effects of wave inference that would otherwise cause increased tsunami wave amplitudes 12,13,[60][61][62][63][64] . Our study suggests that time lags on the order of days likely exist between the individual stages of last three significant failures from the northern flank of Tenerife, and as a result the tsunami waves were likely discrete without interference and thus had lower comparative amplitudes [60][61][62][63][64] .

Methods and Data
The sediment piston cores of this study form a component of previous published research and an established basin stratigraphy 7,8,65 . Examples of the biostratigraphy and hemipelagite lithostratigraphy of these are included in supplement 1. Coccolith biostratigraphy provided a series of datum horizons 7 . Species were counted using a Hitachi TM1000 SEM at the British Ocean Sediment Core Research Facility (BOSCORF) and the methodology of Hunt et al. 7 . Counts were completed at five centimetre intervals and above and below turbidites or changes in hemipelagite lithology.
Volcanic glasses from the individual sub-units of the event beds was completed using the SwiftEd EDS as part of the Hitachi TM1000 SEM at BOSCORF and the methodology of Hunt et al. 38 (Supplementary 2). Samples from each sub-unit were sieved at 63 μm and subjected to acetic acid (0.1 M) leaching to remove carbonate. For the Icod slide sub-units were sampled from core D11813 and included 283 new analyses across all sub-units. The Orotava slide sub-units were sampled from core D18813 with 150 new analyses, the Roques de Garcia slide sub-units were sampled from core D11814 with 42 new analyses, and the Cumbre Nueva slide sub-units were sampled from core D11818 with 28 new analyses. Grains were mounted on semi-conductor pads and 30-50 volcanic glasses 90-125 μm and >45 μm thick were analysed for 120 s at 15 kV. Samples were recovered from the El Abrigo and Granadilla ignimbrites on Tenerife and analysed using the SwiftEd EDS as part of the Hitachi TM1000 SEM at BOSCORF and the methodology of Hunt et al. 38 (Supplementary 3). These data were validated for accuracy and precision using a series of international standard reference materials (SRMs) as analytical glasses, produced using methodology of Hunt et al. 38 (Supplementary 4). Concentrations between 1 and 2 wt% had precisions of 5.0-10.2% of the value, concentrations between 2 and 10 wt% had precisions of 1.0-7.5%, while those values >10 wt% had precisions within <1.5% of the value. Accuracies of SEM EDS results compared to certificated data were generally within 0.3-10.0% of the SRM certified value for values >1 wt%. Specifically, SiO 2 that ranged from 45.35 to 53.41 wt% among SRMs had accuracies to within 0.3-3.7% of the certified value. For Na 2 O that ranged from 1.13 to 3.64 wt% among SRMs had accuracies to within 5.8% of the certified value. Values of K 2 O among the SRMs only ranged from 0.25 to 2.60 wt%, including three SRMs <0.35 wt%, and had precisions on average within 7.2% of the certified value. Generally, values of Na 2 O and K 2 O within volcanic glasses analysed were >2.5 wt% and >4.0 wt%, respectively and thus yielded both improved precision and accuracy. Data availability statement. All geochemical data pertaining to the compositions of volcanic glasses from each of the landslide sub-units and their associated caldera-forming eruptions is supplied within the supplementary data. Any addition data can be made available upon request of the lead author.