1 Introduction

Volcanic debris avalanches produced by large-scale (≥0.1 km3) volcano lateral collapse exist in all volcanic contexts around the world—from stratovolcanoes around the ring of fire, to submarine flanks of volcano islands, shield volcanoes, intracontinental calderas and other volcanic landforms with slopes that are susceptible to failure.

The 1980 eruption of Mount St. Helens, which was accompanied by a large lateral collapse and debris avalanche, prompted a re-evaluation of the frequency and characteristics of this phenomenon at volcanoes around the world. In the four decades since then, hundreds of papers have described individual debris avalanche deposits, with an average of about one regional or global inventory of collapse events per year (Table 1).

Table 1 List of inventories and datasets used to compile our new global database; arranged chronologically
Full size table

The first regional review was by Ui et al. (1986) for Japanese volcanoes, and several other inventories for Japan followed. Most regional surveys have focused on the volcanic arcs ringing the Pacific, including New Zealand, Papua New Guinea, Indonesia, Japan, Kamchatka, the Cascade Range, Mexico, Central America, and South America as study areas. Bathymetric studies demonstrate that collapse morphologies observed on land extend out to sea on a scale several orders of magnitude larger, and led to the recognition that a steep edifice slope angle was not a requirement for failure. The advent of bathymetric studies along the Hawaiian Islands (Moore et al. 1989) cued searches for similar events elsewhere (Holcomb and Searle 1991). Subsequent surveys of collapse events on island chains in the southeast Pacific, West Indies, the Aleutian arc and Atlantic Ocean island arcs such as the Azores, Madeira, Canary Islands, and Cape Verde Islands followed (see Table 1). Detailed bathymetric studies on Reunion Island in the Indian Ocean identified 27 submarine VDADs from Piton de la Fournaise and neighbouring Pleistocene Piton des Neiges and Alizes volcanoes (Oehler et al. 2008), the highest number of collapse events currently known at an individual volcanic complex.

Early global inventories (Ui 1983; Siebert 1984; Siebert et al. 1987) catalogued events similar to the one at Mount St. Helens. Bernard (2008) and Dufresne et al. (2008) expanded these inventories and generated global debris avalanche databases with detailed metrics of the characteristics of their deposits and source scar geometries. A dozen more datasets were generated since then, which prompted the compilation of an updated database by Siebert and Roverato (2020, this volume) and this study. We present this new database and analyse the global distribution of VDADs, perform statistical analyses of the geometric parameters that characterize their failure and runout, and compare them to the deposits of large, non-volcanic rock avalanches.

2 New Global Database for VDADs

Our new global database contains 1001 known and inferred VDADs (mainly larger than 0.1 km3) from 594 volcanoes in 52 countries. The database was created by compiling all previous inventories (Table 1) and adding hitherto unpublished data from our archive (Siebert, unpubl. data). Data acquisition of deposits and/or source areas in the individual inventories ranged from fieldwork to remote sensing on variable scales. The completeness and certainty of data entries in the individual inventories also cover a broad range and we discuss the completeness of data in the following.

Of all deposits in our database, 65% are identified by unique names. Ages are known or inferred for 66% (Table 2) and mostly range from Pleistocene to historical—dating methods cover the full spectrum from best guesses based on stratigraphic context to precise age dating methods to direct observations of historical events.

Table 2 Parameters and quality of database entries
Full size table

Our database is quantified by available morphometric parameters, such as the deposit volume and geometric dimensions, the volcano height, source scar metrics, sedimentological characteristics, etc. Here we present selected parameters relevant for runout and frequency analyses. Entries for deposit volume (V), drop height (H), runout distance (L), and deposit area (A) are available for 25–44%, whereas preliminary collapse trigger are known (or inferred) for only 16% (158 of 1001 events). Of these 158, 111 are associated with magmatic eruptions, 12 with phreatic eruptions, and 35 are stated to be without associated eruptions. However, these proportions are likely not representative due to the difficulty in determining the character or existence of eruptive activity for pre-historical events. For the parameter ‘volcano type’, we primarily utilize the list available in the Smithsonian’s Global Volcanism Program (GVP). For 79% of the volcanoes in the database, the volcano type is known. The majority of volcanoes producing VDADs are stratovolcanoes (74.7% of the entries with known volcano type), including all alternative terms used in the GVP for these types of volcanoes (i.e. composite, compound, and complex volcanoes). Shield volcanoes make up 10.5%, domes reach 4.1%, and 3.4% are listed for calderas as separate volcanic landforms. Cones, submarine volcanoes, volcanic field, fissure vents, volcano complexes, and large igneous provinces host the remaining 7.3%.

2.1 Global Distribution of VDADs

The global distribution of VDADs (Fig. 1) depends on the selection and definition of regional criteria (i.e. volcano regions versus political boundaries). When evaluating the data according to the volcanic ‘regions’ as used in the GVP database organization, we observe that just over half (51%) of all the VDADs in our database are situated in Japan/Taiwan, South America, Mexico/Central America, and Kamchatka/Kuril Islands (Figs. 1 and 2a). Considering strictly the national boundaries (including overseas territory), this distribution changes—Japan, the United States, Russia, Indonesia, France, and Mexico comprise 54% of all recognized VDADs (Fig. 2b). Of the 155 VDADs on United States territory, 62 are situated in Alaska and 21 on or surrounding the Hawaiian Islands. Both distributions, however, reflect the global distribution of the active volcano regions grouped around the circum-pacific ring of fire (Fig. 1).

Fig. 1
figure 1

Map of the global locations of volcanoes (red triangles; based on GVP) and the distribution of volcanic debris avalanche deposits currently in our database (white circles; based on the data in Fig. 2)

Full size image
Fig. 2
figure 2

Global distribution of VDADs in percent of total entries. a Distribution by region. b Distribution by country (including overseas territories) showing 88% of all entries. In both, a and b only those countries/regions are shown that each host at least 1% of the total number of global VDADs

Full size image

2.2 Recurrence Intervals of VDADs Since 1500 AD

Twenty-eight edifice collapse events producing VDADs ≥0.1 km3 have been reported since 1500 AD—an average of more than 5 per century (Table 3; cf. Siebert and Roverato 2020, this volume). Almost three quarters of these were associated with magmatic eruptions, mostly of VEI 4 or 5 magnitudes. Those at Mount St. Helens in 1980 and Bezymianny in 1956 had documented associated lateral blasts. Two earlier VDADs, e.g. Komagatake in Japan in 1640 (Yoshimoto and Ui 1998) and Augustine in Alaska ~1540 AD (Siebert et al. 1995) also had associated lateral-blast deposits. This specific variant of pyroclastic density currents requires precise coincidence of landslide onset and decompression-generated explosions of magma high in the edifice (Belousov et al. 1999, 2007; Siebert 2002).

Table 3 Debris avalanches ≥0.1 km3 since 1500 AD
Full size table

Extending observations to times pre-1500 AD, Tibaldi et al. (2005) observed that the recurrence intervals of sector collapse (involving the summit of the volcano) on geological timescales and in zones of high magma production rates can be on the order of 30 years to a few hundred years; with 2–14 collapse events on single volcanoes. In our database, the highest numbers of collapse events belong to Piton de Neiges (19; Indian Ocean), Taranaki (14; New Zealand), Augustine (12; Alaska), followed by Rainier (Washington) with 10, Iliamna (Alaska) with 9 (all consisting of much smaller avalanches), and Shiveluch (Kamchatka) with 8 collapses.

In historical times, a collapse of part of the caldera wall of Fernandina volcano at the time of a 1988 magmatic effusive eruption generated an 0.9 km3 VDAD that covered much of the caldera floor (Chadwick et al. 1991). The Bandai collapse in 1888 was associated with a strong phreatic eruption that produced 0.15 km3 of non-juvenile tephra (Nakamura 1978). Other collapse events, some of which were flank failures not involving the central core of the volcano, occurred in the absence of eruptive activity. The largest known collapse globally during the past 500 years, at Ritter Island in Papua New Guinea in 1888, was accompanied by explosive sounds, but only a “fine, almost imperceptible ash fall,” perhaps associated with minor phreatic eruptions or dust clouds from the landslide itself (Johnson 1987; Ward and Day 2003).

Fatalities were recorded at about half of these events (Table 3), a percentage lower than would be expected because many were at volcanoes in sparsely populated or unpopulated regions. Most of the recorded fatalities resulted from associated avalanche-generated tsunamis rather than from the debris avalanches themselves. This proportion could be even higher if the unknown number of possible avalanche-generated tsunami fatalities from the 1883 Krakatau eruption was included. The climactic phase of the 1883 Krakatau eruption has been suggested to have begun with a partial subaerial lateral edifice collapse to the north producing a large submarine debris avalanche outside the 1883 caldera followed by collapse of Rakata volcano into the newly formed caldera near the end of the eruption (Camus et al. 1992; Deplus et al. 1995). Tsunamis of variable size occurred throughout the August 26–27 paroxysmal phase of the eruption, with possible origins from landslides, pyroclastic flows entering the sea, caldera subsidence, submarine explosions, or explosion-generated air-sea shock waves. The significance of hazards from avalanche-generated tsunamis has recently been underscored by the collapse and VDA of Anak Krakatau in December 2018, causing 431 fatalities (Gouhier and Paris 2019).

3 Deposit Morphometric Characteristics Scheidegger (1973)

Volcanic debris avalanches compare in deposit appearance and total runout distance (L) with large (>106 m3) rock slides and rock avalanches (cf. Dufresne et al. 2020, this volume). Herein, the latter two are simply referred to as ‘rock avalanches’ (RAs). In the following, we analyse and compare geometric parameters of VDA and RA deposits. The volume (V) encompasses the total size of the deposit, including bulking of the initial failure mass due to comminution and material added during runout by entrainment of valley-fill sediments. It is often calculated by multiplying the plan area (A) of the deposit with an approximated average thickness. The runout distance (L) and the drop height (H; total drop height as defined by Heim 1932) are measured in relation to the elevation at the back of the source scar and the elevation at the most distal, longitudinal deposit margin. For RAs, the pre-failure elevation often corresponds to the highest point of the source scar. For VDAs, this same parameter, also referred to as collapse height (Ui 1983), is not always easy to assign since many edifice collapses involve the pre-failure summit (Capra et al. 2002). For these cases, Ui et al. (1986) proposed to use the crater rim or current summit of the volcano. The numbers of entries for each of these geometric parameters vary strongly between databases and range between 29 and 65% completeness (Table 4). The inventory for rock avalanches is a combination of published case studies (compiled from the literature; Dufresne, unpubl. inventory) and a dataset from the Central Asian region collected by Strom and Abdrakhmatov (2018) and presented in Strom et al. (2019).

Table 4 Number and percent of entries per parameter and landslide type based on the new global database of VDADs (this study) and data of RA deposits compiled by Dufresne (unpubl. inventory) and Strom and Abdrakhmatov (2018); see text
Full size table

The data are plotted on log–log plots to cover the full, several orders of magnitude, range and data scatter. The latter is reflected in the coefficient of determination (R2) of regression analyses (Table 5). It ranges between 0.22 (for large scatter) and 0.86 (for good correlation of variables) and can be very different for RA and VDA deposits in the same bivariate space. For both landslide types, the deposit area (A) correlates best with volume (V) and with runout (L); Fig. 3a, b. Strom et al. (2019) observed the same good correlation between A and L for the Central Asian dataset and further identified differences in how parameters correlate best depending on runout path topography. In their study, L and A correlate best for unconfined RAs, whereas frontally confined RAs have best fits for V * H over L. Using A over L, and also A over V, as proxies for lateral debris spreading in our new database, both landslide types show parallel trends, with slightly higher A at comparable L or V for VDADs. This might be attributed to a general difference in runout path topography in the respective typical settings. However, only a more comprehensive analysis like the one conducted by Strom et al. (2019) can narrow down this assumption. The influence of runout path topography on debris spreading and total distance travelled is not investigated herein, nor are other factors, such as water content of the spreading mass, substrate conditions or syn-collapse eruptive activity.

Table 5 Coefficients of determination of the log–log regressions shown in Fig. 4; n is the number of deposits plotted
Full size table
Fig. 3
figure 3figure 3

Comparison between geometric parameters of volcanic debris avalanche deposits (from: Siebert and Roverato 2020; this study) and non-volcanic rock avalanche deposits (from: Dufresne, unpubl.; Strom and Abdrakhmatov 2018; Strom et al. 2019). Solid trend lines are for VDAs, dashed lines for non-volcanic events. a Deposit area as a function of volume is an indicator of debris spreading. b Area and runout distance show good correlations for both deposit types. c The trendlines suggest that VDAs attain longer runouts at similar volumes than RAs; however, note the several order and overlapping scatter in both datasets. d Large scatter and poor correlations in V versus H (see text). e The so-called Scheidegger plot of the apparent coefficient of friction (H/L) and deposit volume. f Drop height (H) and deposit volume (V) are dependent variables, and hence this plot of L over H shows similar correlations and trends as c

Full size image

From Fig. 3a, c one might conclude that VDAs are generally larger than non-volcanic rock avalanches. However, the data used in these analyses is biased since the focus of data assemblage for VDAs has been deposits larger than ~0.1 km3 in volume, whereas RA inventories typically include deposits starting from 106 m3. Yet the impression that VDADs reach larger volumes than RAs holds true in so far as only 18% of the catalogued RAs exceed volumes of 0.1 km3, and only 5% are 1 km3 or larger. Hence, non-volcanic slopes reach, with few exceptions, an upper collapse volume which volcano slopes can easily exceed; see also Scott et al. (2001) and Devoli et al. (2009). This can be further explained by the parameter combination of deposit volumes (V) and drop height (H) (Fig. 3d). Even though the statistical correlation is poorer than for other parameter combinations (compare R2 values in Table 5), the inference that volcano slopes produce larger-volume landslides at the same drop height than rock slopes is reasonable from a geometric point of view. VDAs generally fail along deeper rupture surfaces, have more deeply-seated and bowl-shaped scars, and longer flanks (Fig. 4a) compared to RA source scars (Fig. 4b). Hence, they have an overall greater potential to produce larger failure volumes at comparable slope heights.

Fig. 4
figure 4

source geometry between a volcano edifice and b rock slope failures explains higher possible landslide volumes for VDAs. Profiles after Siebert (1984), Prager et al. (2009; Köfels), Qi et al. (2011; Donghekou), and Dykes and Bromhead (2018; Vaiont). Note difference in scale between the two slope failure types. Bandai and Iriga are non-magmatic collapse events, the others volcanic failures are associated with magmatic activity

Differences in

Full size image

These two different correlations between V and H for the different landslide types are then, of course, also noticeable in the classic ‘Scheidegger’ plot of H/L versus V (Fig. 3e). The latter is known as the travel angle, apparent coefficient of friction, angel of reach or, originally, Fahrböschung (Heim 1932), and thought to describe the mobility of a landslide. However, the above findings on the strong dependence of V on source scar geometry joins the previously published statements that H/L (Fig. 3f) is mechanically meaningless for mobility considerations (Hsü 1975; Davies 1982; Legros 2002; Dufresne and Geertsema 2019). A more useful approach might be to utilize the centre of mass for determining H and L (as originally proposed by Heim 1932); however, these are difficult to impossible to determine for deposits for which both, pre- and post-failure source and runout path topographies are not known. Furthermore, V and H are geometrically dependent, with different failure geometries controlling the two different landslide types (as discussed above and illustrated in Fig. 4).

4 Summary and Conclusion

The new database compiled herein contains the basic spatial and geometric data of VDADs, their source scar geometry, and volcano characteristics. Large-scale volcano lateral collapses are primarily grouped around the circum-pacific ring of fire and reflect the distribution of the most active volcano regions. Multiple-collapse events are common, with 2–19 collapses from single volcanic edifices. The ‘historical’ recurrence of ~five events per century continues the prehistoric trend previously published in the literature. Volcanic debris avalanche deposits compare in their appearance with large (>106 m3) deposits from non-volcanic rock slope failures. Our data suggest that they also have similar spreading behaviour with an only slightly larger area covered by volcanic material at comparable runout distances of rock avalanches. This is likely due to the prevalence of unconfined runout conditions in volcanic settings as opposed to narrow valleys typical for rock avalanche environments. The most significant empirical difference between the two landslide types is the volume excavated at the source: at the same (slope) height, volcanic lateral collapses can produce greater volumes since their scars are deeper-seated and more bowl-shaped than common for rock slopes. This is demonstrated very well in drop height versus volume plots and in source profiles from case studies. This relationship shows that, due to the typical collapse geometries, rock slopes reach an upper limit of potential excavated volumes that volcanic slopes can easily exceed. To investigate more empirical relationships and study the differences between these two landslide types, increased efforts in consistent data collection are essential—which is the focus of our database research network aimed at closing this gap.