Volcano-Hydrologic Hazards from Volcanic Lakes

Abstract

Volcanic regions typically host multiple lakes developed in explosion craters, volcano-tectonic collapse structures, and valley systems blocked as a result of eruptive activity, their boundaries and dimensions shifting in response to renewed activity and modification by background processes of erosion, sedimentation and tectonism. Such water bodies are a potent source of a wide range of complex and inter-related hydrologic hazards owing to their proximity to active volcanic vents, the consequent potential for violent mixing of magma with water, and the frequent fragility of their impoundments. These hazards arise as a result of water displacements within or from the lake basin and can be broadly sub-divided into 3 main types: (I) phenomena sourced within the lake basin as a direct or indirect consequence of subaqueous or subaerial volcanic activity; (II) floods from volcanic lakes triggered by volcanic activity, including induced breaching; and (III) floods from volcanic lakes with a non-volcanic cause. Type I hazards include subaqueous explosive volcanism and associated Surtseyan jets, base surges and tsunamis, which can impact lake shorelines and displace water over basin rims and through outlets. This results in Type II lahar and flooding hazards. Both types have been historically responsible for significant losses of life at many volcanoes worldwide. Other rapid phenomena such as pyroclastic flows, debris avalanches, and large lahars from intra- or extra-lake volcanoes are potentially tsunamigenic (Type I), and/or displacing, and can hence also lead to secondary (Type II) hazards, as can seismicity-producing volcano-tectonic movements. Slower processes including volcano-tectonic movements, subaqueous lava dome extrusion, cryptodome intrusion, and magmatic inflation can potentially produce Type II flooding through volumetric water displacement over the outlet. Erosion of the outlet can be catastrophic, magnifying the size of flood events. Damming of the outlet itself can result in backflooding of the basin. Type III hazards, i.e. volcanic lake break-out floods; result from breaching of the barrier constraining a volcanic lake as a result of passive overtopping, piping, mechanical failure, or headward erosion of the natural dam. Such events range in scale from relatively minor outflows triggered by failure of crater walls or the breaching of riverine dams composed of pyroclastic, volcaniclastic, or lava flow material to catastrophic floods generated by the breaching of caldera rims. Palaeohydraulic reconstructions of some of the latter indicate that they are amongst the largest post-glacial floods on Earth, being exceeded only by late Pleistocene deluges associated with breaching of ice-dammed lakes and pluvial basins.

1 Introduction

Volcanic activity is a prolific producer of lakes due to the capacity of eruptions and volcano-tectonic activity to generate both positive and negative relief. In the strictest definition, a volcanic lake is a cap of meteoric water over the vent of an active volcano: according to this criterion 16 % of the 714 identified Holocene volcanoes world-wide host one or more, frequently ephemeral, lakes in explosion craters and subsidence calderas (Delmelle and Bernard 2000). Many typical crater lakes (Fig. 1a) contain 1–10 × 10⁶ m³ of water, often at elevations several km above the surrounding landscape (Casadevall et al. 1984; Rowe et al. 1992; Christenson and Wood 1993; Kempter and Rowe 2000). Hydrothermal and hydromagmatic (maar) eruption craters (Fig. 1b) are typically <2 km in diameter and comprise a central pit ringed by a raised ejecta rim (Lorenz 1973). The transition from purely magmatic explosion craters to volcano-tectonic depressions formed by a combination of explosive ejection of material and magma withdrawal occurs at c. 2.5 km diameter (Williams 1941). The largest volcanic impoundments comprise intracaldera lakes, either developed by collapse at the summit of shield or cone volcanoes such as Crater Lake in Oregon (Fig. 1c), which impounds 1.9 × 10¹⁰ m³ at an elevation of 1,882 m (Nelson et al. 1994), or superimposed on regional tectonic depressions, like Lake Taupo in New Zealand (Fig. 1d), which contains 6 × 10¹⁰ m³ (Lowe and Green 1992). Lake Toba in Indonesia is the world’s largest caldera lake and holds 2.4 × 10¹¹ m³ of water (Chesner and Rose 1991). A review of c. 200 Late Pleistocene or younger terrestrial calderas found that around half held one or more intracaldera lakes (Manville 2010), either with or without a surface outlet (Larson 1989).

Fig. 1

A selection of volcanic lakes from around the world. a Crater Lake, Mt. Ruapehu, New Zealand. This c. 400 m diameter lake lies at a surface elevation of c. 2,530 mASL. Explosive volcanic activity has generated frequent hazards, including base surges (deposits arrowed) and eruption-triggered lahars by displacement waves over the outlet (1) or Surtseyan jetting over the crater rim (2). This photo was taken the day after the 25 September 2007 eruption. (Image B. Christenson, GNS Science). b Pulvermaar in the volcanic Eifel region of Germany. Diameter is c. 700 m. c Crater Lake, Oregon. Formed by caldera collapse during the 7.7 ka Mazama eruption, this 10 km diameter lake lies 1882 mASL and has no surface outlet. (Wikipedia commons). d Lake Toba, 87 km long, lies in a volcano-tectonic collapse structure last modified during the 74 ka Toba super-eruption. (Landsat Image)

In addition, volcanism is the third most common dam-forming mechanism, accounting for 8 % of natural-dammed lakes globally (Costa and Schuster 1988). Volcanogenic dams include lava flows (Fenton et al. 2004), pyroclastic flows (Aramaki 1981; Macías et al. 2004), debris avalanches from the collapse of stratovolcanoes (Meyer et al. 1986; Capra and Macías 2002), and rapid aggradation by lahars (Umbal and Rodolfo 1996). These blockages may impound lakes in the source crater or caldera, adjacent ones, or local valley systems (Fig. 2). All such impoundments have the potential to generate catastrophic break-out floods through sudden failure of the volcanic barrier, in some cases triggered by the resumption volcanism. The resulting floods from large volcanic lakes rank amongst the largest post-glacial floods on Earth (O’Connor et al. 2013).

Fig. 2

Lakes impounded behind dams of volcanic material. a Lake Chungará, dammed at c. 6 ka by a debris avalanche from Parinacota stratovolcano, Chile. (Wikipedia commons). b Spaceborne view of the snow-capped cone of Parinacota: Lake Chunagará lies c. 8 km to the south. Numerous other lakes, including Cotacotani, are impounded by the irregular topography of the debris avalanche deposits to the west (NASA image ISS029-E-20003). c Lake Garibaldi (10 km², 1,500 m ASL, >250 m deep) dammed by lave flows from Mt. Price and Clinker peak, British Columbia, Canada. (Wikipedia commons). d Lake Mapanuepe, dammed by lahars in the Marella river, Pinatubo, Philippines 1991 (USGS image)

1.1 Classification of Hazards Associated with Volcanic Lakes

A wide range of hazards derive from volcanic lakes as a direct consequence of magma:water interactions at the vent, or through the interaction of volcanically generated mass-flows (pyroclastic flows, debris avalanches, lahars, and lava flows) with the water body (Fig. 3). Consequently, although only c. 3.5 % of recorded volcanic eruptions have occurred through lakes (Simkin and Siebert 1994), these have been responsible for approximately 15 % of recorded deaths (Table 1). In many cases, a primary hazardous phenomenon can trigger a series of other secondary effects in a process-chain (Fig. 3): for example, a subaqueous explosion that breaches the lake surface can generate tsunamis that overtop the lake outlet to form a lahar, which itself causes erosion of the spillway leading to further water release and downstream flooding (McGimsey et al. 1994; Waythomas et al. 1996). Under certain (poorly understood) conditions, magma:water interactions can increase the violence and explosivity of volcanic eruptions (Thorarinsson et al. 1964; Lorenz 1973; Sheridan and Wohletz 1983; Wohletz 1986). Factors that influence the explosivity of the interaction likely include magma type, mass eruption rate, nature of magma pre-fragmentation, and water depth (Wohletz 1986; Mastin 1995; Koyaguchi and Woods 1996). The net effect can be to increase the area of the primary hazard by up an order of magnitude through the generation of base surges (Moore 1967) and even more with tsunamis (Latter 1981). A 0.01 km³ monogenetic basaltic eruption in an arid environment would produce a localised scoria cone, the same magma volume erupted in wet environment can produce a 1 km diameter maar/tuff-ring with phreatomagmatic base surges extending to >3 km radius (Németh et al. 2012). The travel distance of tsunamis is effectively limited by the dimensions and bathymetry of the water body, in lakes this can be 10’s of km. Alternatively, magma and water can mix relatively non-violently (Batiza and White 2000). A review of historical eruptions through volcanic lakes shows that c. 2 % involved relatively passive growth of subaqueous to emergent lava domes (Table 1). Other hazards reported during eruptions through water include tsunamis and seiches generated by volcano-tectonic displacements of the lake floor, Surtseyan jets, tephra and ballistic fall, lahars, lightning (associated with wet ash clouds), and torrential rainstorms (Mastin and Witter 2000). Growth in population and infrastructure, particularly in the developing world, has increased exposure to hazards from volcanic lakes and a number of major cities are vulnerable to volcano-lake interactions including Managua, the capital of Nicaragua (population 2.2 million). Therefore, characterisation and understanding of the range of hazards is vital to their mitigation.

Fig. 3

Classification of the main hazards associated with volcanic lakes, including triggering mechanisms and process-chains. Examples of volcanoes where these hazards have been observed historically are shown in brackets; names are italicized where the hazard has been inferred

Table 1 Compilation of historical Type I and II hazardous events at volcanic crater and caldera lakes, with a focus on the best characterised or most deadly

2 Subaqueous Explosive Volcanism

The rapid conversion of thermal to kinetic energy during the violent mixing of magma and water in a subaqueous eruption fuel-coolant reactions (Wohletz 1986; Frost et al. 1994; White 1996) produces an expanding and buoyantly rising bubble of water vapour, magmatic gas, and magma fragments whose interactions with the ambient water column can cause ballistic ejection of most of the volume of a crater lake from its basin (Nairn et al. 1979), as well as generating base surges (Moore 1967; Waters and Fisher 1971) and impulsive tsunami in larger water bodies (Watts and Waythomas 2003). Volcanic tsunamis (Latter 1981), including explosion generated waves, are treated separately in this discussion due to their range of potential triggering mechanisms.

2.1 Surtseyan Jets

Surtseyan jets, also known as “cock’s (rooster) tail” or “cypressoid” jets (Fig. 4) are characteristic of subaqueous explosive volcanism (Thorarinsson et al. 1964), and have been observed in a number of lacustrine (Belousov et al. 2000; Belousov and Belousov 2001) and shallow marine settings (Morimoto 1948; Machado et al. 1962; Reynolds et al. 1980). They comprise black jets of tephra and water that are ejected on sub-ballistic trajectories to heights of up to 800 m at velocities exceeding c. 100 m/s, and are produced during discrete explosions in relatively shallow water (<200 m). As the jets cool, they become fringed with white clouds of condensing steam (Fig. 4).

Fig. 4

a Surtseyan jets generated during the eruption of a submarine volcano north of Tongatapu in the Tonga Islands rising to >800 m. Darker ‘cypressoid’ or ‘rooster-tail’ jets are composed of mixtures of water droplets, gas and pyroclastic fragments, lighter clouds are steam and condensate. A toroidal base surge is propagating out from the base of one of the eruption columns (Getty images). b Small Surtseyan jet and base surge generated during a minor phreatic eruption, typical of activity during the late 1970s and early 1980s at Ruapehu volcano, New Zealand (Image GNS archive)

By analogy with conventional (Cole 1948; Kedrinksii 2005) and nuclear (Glasstone and Dolan 1977) explosive tests, detailed observations and analysis of selected eruptions (Belousov et al. 2000; Kilgour et al. 2010), and numerical simulations (Morrissey et al. 2010) a subaqueous volcanic explosion forms a sub-spherical bubble filled with super-heated explosion products including water vapour, magmatic gases and pyroclastic material (Fig. 5). This rises buoyantly towards the surface, expanding as its internal pressure equilibrates with the confining hydrostatic pressure. Ahead of this, the explosion shock wave impinges on the water surface, typically forming a symmetrical spray dome by spallation and cavitation that may rise several hundred metres into the air but which contains a volumetrically insignificant volume of water. The ascending explosion bubble then deforms the water surface above it, forming a rapidly rising dome of water. When this breaches, the explosion bubble vents to the atmosphere. The sudden drop in internal pressure causes the water:gas interface at the bubble’s base to rebound, rapidly collapsing the transient cavity and generating vertically- and radially-directed jets as it ‘turns inside out’ (Fig. 5).

Fig. 5

Inferred sequence of events during a subaqueous explosive eruption. a Initial explosion during magma-water mixing: this fuel:coolant reaction generates a shock-wave whose energy is largely reflected off the air:water interface at the lake surface, but still manages to raise a dome of spray. b As the explosion bubble of eruption products and water vapour equilibrates with the ambient hydrostatic pressure it expands and rises buoyantly, lifting a semi-spherical dome of water above it. c The water dome breaches as it thins and stretches and as the explosion bubble breaches the surface, the sudden drop in internal pressure as the transient cavity vents to the atmosphere causes its base and sides to rebound, generating vertical and radial water jets that entrain the bubble contents and remnants of the water dome

2.2 Base Surges

Base surges are ring-shaped basal clouds (Figs. 4 and 5) that sweep outwards as a dilute density current from the base of a vertical explosion column (Moore 1967). They are produced during subaqueous explosive eruptions principally through disruption of the rim of the transient cavity (Fig. 5), and secondarily by collapse of Surtseyan jets, and move radially outwards as mixtures of pyroclasts, water droplets and steam at typical velocities of 20–65 m/s (Moore 1967; Belousov et al. 2000). Base surges generated at Taal volcano in the Philippines are likely to have been responsible for most of the deaths on Volcano Island during the 1911 eruption, while in 1965 they reached up to 8 km from the vent, depositing surge-bedded dunes proximally and wet-mud aggregates and coatings distally (Moore et al. 1966a, b; Moore 1967). Base surges generated by explosive eruptions through a postulated caldera lake have been suggested as playing a role in the 1790 AD Kilauea eruption that killed 80–300 people (Mastin 1997).

2.3 Seiches

Seiches are standing waves produced in semi- or fully enclosed bodies of water due to resonant amplification of waves initiated by another mechanism (i.e. fluctuations in atmospheric pressure, earthquakes, or tsunami). Under certain circumstances they may be higher and more persistent than the original tsunami: studies at intermontane Lake Tahoe indicate potential seiche run-ups of up to 10 m in response to seismogenic fault offsets of the lake bed (Ichinose et al. 2000). Seiches are also a demonstrated hazard at volcanic lakes: a tectonic earthquake at Taal caldera lake in 1749 destroyed villages and caused fatalities (Newhall and Dzurisin 1988), while seiches of up to 0.5 m were reported at Lake Taupo in response to seismic faulting during earthquake swarms in 1922 and 1983 (Ward 1922; Otway 1986; Webb et al. 1986). Seiches set up by volcanic explosions during the 1965 Taal eruption capsized a number of boats, resulting in a number of fatalities (Moore et al. 1966b).

2.4 Atmospheric Effects—Rain and Lightning

Phreatomagmatic eruptions through volcanic lakes can result in torrential local rain due to steam condensation, potentially aided by nucleation on ash particles, sometimes in association with intense lightning activity (Mastin and Witter 2000). This can contribute to local flood hazards and syn-eruptive reworking of pyroclastic deposits. Examples have been observed at Taal volcano (Moore et al. 1966a, b), and inferred at: (i) Lake Taupo during the Hatepe and Rotongaio phases of the 1800a Taupo eruption (Walker 1981; Smith and Houghton 1995); and (ii) deposition of the Rotomohana Ash during the 1886 eruption at Tarawera (Nairn 1979; Walker et al. 1984).

3 Volcanic Tsunamis

Tsunami triggered by volcanic processes account for 25 % of all volcano-related fatalities since 1783 (Witham 2005), mostly associated with the marine 1883 Krakatau eruption (Carey et al. 2001). A range of triggers have been implicated (Latter 1981), with wave generation typically accomplished by volumetric water displacement (Fig. 6), for example by: a gas-filled explosion bubble (Egorov 2007; Morrissey et al. 2010); pyroclastic flows (Waythomas and Neal 1998; McCoy and Heiken 2000; de Lange et al. 2001; Watts and Waythomas 2003); basin floor displacement; and subaerial and/or subaqueous mass flows including debris avalanches (Johnson 1987; Tinti et al. 2006a; Begét et al. 2008) and landslides (Ward 2001), and lahars (Chrétien and Brousse 1989; Walder et al. 2003). While the greater travel distance of such events is largely irrelevant in the context of confined lake basins, the generating mechanisms remain pertinent.

Fig. 6

Tsunami-generating mechanisms associated with volcanic lakes, including subaqueous explosive eruptions. Surface waves may be triggered by upward and lateral displacement of water due to: (i) an expanding explosion bubble; and (ii) entry, and continued subaqueous movement of an initially subaerial (or wholly subaqueous) gravitational mass flow (pyroclastic flow, landslide, debris avalanche, or major lahar). Other causes include downward and lateral displacement of water caused by: (iii) the sinking of material from a segregating pyroclastic flow travelling over the water surface; (iv) coupling with a base surge or atmospheric pressure wave; and (v) collapse of jetted material. Also, (vi) upward or downward displacements due to seismogenic (i.e. rapid) volcano-tectonic movements. Of these only (i), (ii) and (vi) are capable of generating large waves in the far field (Watts and Waythomas 2003)

3.1 Explosion-Generated Tsunamis

During subaqueous volcanic explosions, tsunamis are principally generated during the phase of up-doming of the water surface, including the period when the bubble breaches the surface and the uplifted rim of the transient cavity is pushed outwards by its expanding contents (Figs. 5 and 6). Smaller tsunamis are also generated by fallback of Surtseyan tephra jets and coupling of the water surface with atmospheric shock waves and base surges (Latter 1981; Mastin and Witter 2000; Watts and Waythomas 2003). Observations of tsunami waves at volcanic lakes are limited to the 1996 Karymskoye Lake eruption (Belousov et al. 2000), minor phreatic events at Ruapehu during the 1970s and 1980s (McClelland et al. 1989), Taal volcano in 1911 and 1965 (Moore et al. 1966a, b), and a number of other crater lakes (Table 1) mainly known for their eruption-triggered lahar record (Mastin and Witter 2000). During the 1996 eruption at Karymskoye Lake “a light gray ‘collar’ [which] appeared around the focus (epicentre) of the explosion, representing a nearly axially-symmetric elevation of the water surface 130 m high that propagated radially at about 40–20 m/s to form the tsunami” (Belousov et al. 2000). Run-up from these surface gravity waves reached 30 m on adjacent shorelines 1.3 km away from the eruption centre (Belousov et al. 2000; Torsvik et al. 2010). Tsunami run-up deposits have been identified c. 20 m above the contemporaneous lake level at Lake Managua in Nicaragua in association with the 6.3 ka Mateare Tephra eruption (Freundt et al. 2006, 2007), and at Fisher caldera in Alaska (Stelling et al. 2005).

3.2 Pyroclastic Flow-Generated

Pyroclastic flows, comprising hot, gas-rich, debris-laden, ground-hugging free-surface gravity currents that travel laterally away from their sources (Sparks 1976; Druitt 1998), are a common phenomenon associated with explosive volcanic activity and are demonstrably tsunamigenic (Latter 1981; Carey et al. 1996, 2000; Hart et al. 2004).

Most studies of volcanic tsunamis are based on submarine and coastal volcanoes, including the 1883 AD Krakatau eruption in Indonesia (Self and Rampino 1981) which produced proximal tsunami waves up to 35 m high that caused c. 36,000 deaths in coastal areas of Java and Sumatra up to 100 km from the volcano, and were still lethal in Sri Lanka 2,500 km away (Latter 1981; Carey et al. 2000, 2001). Pre-historic tsunami generated by pyroclastic flows entering the sea are known from a number of other regions, including Alaska (Waythomas and Neal 1998) and the Mediterranean, where they are associated with the Minoan eruption of Santorini (McCoy and Heiken 2000).

Pyroclastic flow-triggered lacustrine tsunamis have not been observed in historical times. However, they have been inferred at Lake Managua in Nicaragua where the c. 24 ka Apoyo ignimbrite erupted from a caldera 20 km away is estimated to have entered the lake at volume fluxes of 3 × 10⁶ m³/s and velocities of 65 m/s (Freundt et al. 2006), and at Lake Tarawera in New Zealand during the c. 1315 AD Kaharoa eruption (Hodgson and Nairn 2005). They will certainly have occurred at many caldera lakes where voluminous explosive activity has produced pyroclastic flows from mid-lake vents, for example during several Holocene eruptions at Lake Taupo (Wilson 1993). Tsunamis with this source mechanism are also implicated in the triggering of break-out floods at a number of Alaskan calderas (Table 2). At Fisher, an eruption at c. 1.5 ka from an mid-lake vent apparently triggered breaching of the southern caldera wall by overtopping waves with run-ups exceeding 20 m (Stelling et al. 2005).

Table 2 Compilation of known break-out floods from volcanic lakes, including Type II and III events

3.3 Debris Avalanches/Landslide-Triggered

Debris avalanches generated by the gravitational collapse of a volcanic edifice are a common phenomenon at stratovolcanoes around the world, occurring on average 4 times per century (Siebert 1984, 1996) due to a range of causes (McGuire 1996). Such failures are highly tsunamigenic (Keating and McGuire 2000), as has been demonstrated at a number of coastal (Siebert et al. 1995; Begét et al. 2008) and island volcanoes (Johnson 1987; Tinti et al. 2006b; Chiocci et al. 2008). For example, the tsunami generated by the 1792 AD collapse of the Mayu-yama dome at Shimabara killed an estimated 15,000 people around the shores of the Sea of Japan (Neall 1996; Tanguy et al. 1998).

Lacustrine tsunamis generated by landslides, rockfalls and icefalls are well known, and are often implicated in the failure of natural landslide and moraine dams (Lliboutry et al. 1977; Hermanns et al. 2004; Kershaw et al. 2005). Volcanogenic examples are less frequent (Table 1), but the most notable historic case is the 18 May 1980 debris avalanche at Mount St. Helens, which entered Spirit Lake (itself the product of valley-damming by a prehistoric flank collapse) to generate a wave with a run-up of 260 m (Voight et al. 1981). In 1999, a relatively small (150,000 m³) landslide from the crater walls of the Quaternary Kasu tephra cone in the highlands of Papua New Guinea displaced 5–10 % of the lake’s volume, triggering a 15 m high wave that overtopped a low point on the crater rim killing one person (Wagner et al. 2003). At Fernandina volcano in the Galapagos, c. 1 km³ of the caldera rim collapsed forming a debris avalanche that displaced the existing 2 km wide shallow intracaldera lake to the N and NW as a tsunami in 1988 (SEAN 13:10). Collapse of a geothermally-weakened fault scarp at the southern end of Lake Taupo in 1846 AD generated a lethal debris avalanche that triggered a small tsunami when it entered the lake (Hegan et al. 2001).

Prehistoric tsunamis induced in volcanic lakes by mass movements are inferred at Lake Managua, Nicaragua, where two recent debris avalanche deposits from Volcán Mombacho penetrate up to 7 km into the lake to form small islands (Freundt et al. 2007).

3.4 Lahar-Triggered

Lahars entering volcanic lakes are also tsunamigenic if they are sufficiently rapid and voluminous (Walder et al. 2006): the 18 May 1980 Pine Creek lahar at Mount. St. Helens generated a 0.4 m high wave on Swift Reservoir 22.5 km away from its entry point (Pierson 1985). During the May 1902 eruption of Soufrière St. Vincent, a lahar triggered by collapse of the crater rim of L’Etang Sec generated a 3–4 m high tsunami when it reached the sea (Chrétien and Brousse 1989). Similarly, very large and fast jökulhlaups have produced waves in coastal waters around Iceland (Björnsson 2002).

3.5 Earthquake-Triggered

Tectonic movements on faults crossing volcanic lakes can cause displacements sufficient to generate tsunami or excite seiches, whether related to volcanic activity or not. Recorded examples include at Taal in 1749 AD, where villages were destroyed and lives lost (Newhall and Dzurisin 1988), and at Lake Taupo in association with historic earthquake swarms (Ward 1922; Otway 1986; Webb et al. 1986).

4 Limnic Gas Eruptions

A special hazard at volcanic lakes, and some other major volcanically-influenced water bodies such as Lake Kivu (Nayar 2009), is the explosive exsolution of dissolved magmatic gases, principally CO₂ and methane, which accumulate in the water column (Tazieff 1989). Eruptions can be driven by over-saturation, climatic conditions, landslides into the lake, or volcanic activity. In 1984 a gas burst from Lake Monoun killed 37 people (Sigurdsson et al. 1987), and was followed 2 years later by the Lake Nyos disaster in the same country (Kling et al. 1987; Barberi et al. 1989; Tazieff 1989). Lake Nyos is a small (1,925 × 1,180 m) water body hosted in a maar-diatreme crater. On 21 August 1986, a small landslide into the 208 m deep lake triggered overturn and the catastrophic release of 0.3–1 km³ of CO₂ which overflowed the crater outlet and travelled over 10 km downstream as a dense, suffocating flow. Water surges triggered by the gas release caused waves that washed up to 25 m above the lake shoreline and 6 m deep over the outlet (Kling et al. 1987).

5 Volcanogenic Floods from Volcanic Lakes

Volcanic lakes are prone to sudden and catastrophic releases of water, either directly as a result of explosive volcanic activity (Table 1), or due to failure of their impoundments following active (i.e. eruption-triggered) or passive (i.e. overfilling) overtopping and runaway erosion (Table 2). In either case, the combination of large volumes of water, significant topographic elevation, and abundant pyroclastic material available for remobilisation makes such events particularly hazardous as they tend to travel down existing river valleys, which are frequently the locus of extensive populations.

Crater lakes, developed in pits excavated by explosions above volcanic vents, are prone to explosive (ballistic) ejection or displacement (including by waves) during eruptive activity. Consequently, primary eruption-triggered lahars from such water bodies have been one of the most lethal historic volcanic hazards (Table 1).

5.1 Eruption-Triggered Lahars

The summit of the stratovolcano of Mt. Kelut (1,731 m) on the Indonesian Island of Java is occupied by a crater lake that has been the source of multiple eruption-triggered lahars that have claimed tens of thousands of lives over the past millennium. Initial explosive phases of VEI-4 eruptions, occurring at mean intervals of 13 years (range 8–18) have ejected up to 40 × 10⁶ m³ of water in a matter of seconds, before progressing to Vulcanian eruptions, sometimes accompanied by pyroclastic flows. The resulting lahars have devastated the densely populated Plains of Blitar, destroying villages and killing tens of thousands since records began in c. AD 1000 (van Padang 1951; Zen and Hadikusumo 1965). In May 1848 an eruption caused partial breaching of the crater rim, releasing 48.7 × 10⁶ m³ of water (Neall 1996). Similar events (Table 1) are reported at Galunggung, Kaba, Raung, and Ijen (Mastin and Witter 2000).

The summit crater of Mt. Ruapehu, New Zealand’s largest and most active onshore andesitic stratovolcano, hosts a hot acidic lake at an elevation of c. 2,530 m. Primary lahars have accompanied all large historic eruptions through the explosive ejection of Crater Lake water over the rim (Fig. 1a), and volumetric displacement (Healy et al. 1978; Nairn et al. 1979; Cronin et al. 1997), sometime forming unusual snow slurry lahars (Kilgour et al. 2010). Secondary lahars have been triggered by heavy rain on ash deposits (Cronin et al. 1997; Hodgson and Manville 1999), while break-out lahars have followed magmatic eruptions by a decade (Manville et al. 2007) (see below).

5.2 Floods Triggered by Tsunami

During the 1996 eruption of Karymskoye in Kamchatka, tsunamis triggered by subaqueous explosive eruptions caused water to surge over the outlet from the lake as a series of waves, triggering flooding along the Karymskaya River with a maximum discharge of 500 m³/s (Belousov and Belousov 2001). Similar lahars have been observed at Ruapehu where explosions have also displaced water over the outlet area when the lake has been full during eruptions (Cronin et al. 1997; Kilgour et al. 2010).

5.3 Floods Triggered by Lake Floor Displacement

Volcano-tectonic uplift of basin floors has been implicated as a factor in the generation of a number of volcanic floods. This may occur through emplacement of a subaqueous lava dome or shallow intrusion, or through inflation of a shallow magma chamber, although it has been argued that even a significant magma intrusion event would produce negligible surface deformation at a caldera lake like Taupo (Ellis et al. 2007). Rates of water displacement by subaqueous intrusion of lava domes are limited by the effusion rate, which is typically 1–10 m³/s. Higher rates may be associated with inflation of a shallow magma chamber, either due to injection of a hot slug of basaltic magma or runaway vesiculation at its top, or thermal expansion of groundwater.

A historical flood of this type from an intracaldera lake occurred at Ilopango Caldera, El Salvador (Goodyear 1880; Newhall and Dzurisin 1988; Richer et al. 2004). The present lake occupies a nested rectangular caldera most recently modified by eruption of the Tierra Blanca Joven ignimbrite in AD 479 (Rose et al. 1999). In January 1880, subaqueous eruption of a dacitic lava dome raised the level of the 108 km² lake by 1.2 m (Goodyear 1880) over a period of 5 days. Increased overflow through the narrow outlet channel of the Río Desagüe caused rapid downcutting, with the lake level falling by 9.2 m between 12 and 20 January. Peak discharges are estimated at c. 3 × 10³ m³/s based on drawdown rates of c. 0.1 m/h, resulting in destruction of the downstream town of Atusulca (Newhall and Dzurisin 1988; Richer et al. 2004). By 8 February the lake had fallen by an additional 2.6 m, corresponding to a total water loss of 1.2 × 10⁹ m³.

Displacement of the floor of Crater Lake, Ruapehu, during the 1995 eruption sequence is recorded by a stage hydrograph on the single outlet channel 58 km downstream at Karioi. The data shows a complex series of transient, spiky, primary explosion-triggered lahars superimposed on a broad trend lasting c. 14 days with a peak discharge of c. 30 m³/s that is inferred to represent steady uplift of the lake floor by rising magma (Cronin et al. 1997). Total displacement by this uplift is estimated at c. 6 × 10⁶ m³, representing half the volume of the pre-eruption lake, and a volume equivalent to that expelled by explosive ejection over the rim and outlet. An alternative explanation is that this represents superposition of the exponential waning limbs of individual eruption-triggered flows. Water displacements are also associated with the October 2006 (Kilgour et al. 2007) and September 2007 (Kilgour et al. 2010) phreatic events at Ruapehu, where increased lake level has also been associated with injection of water from the sub-lake hydrothermal system (Christenson and Wood 1993; Christenson et al. 2010).

5.4 Jökulhlaups

A special class of potentially hazardous volcanic lake comprises sub-glacial water bodies, break-outs from which, termed jökulhlaups, have been identified as the most frequently occurring volcanic hazard in Iceland where a number of large basaltic calderas filled with subglacial lakes lie beneath the Vatnajökull and Mýrdalsjökull icecaps (Björnsson 1975, 1992; Tómasson 1996; Björnsson 2002). Most such floods are relatively small, and result from semi-periodic accumulation and drainage of geothermally sustained subglacial lakes. However, larger, less frequent events with peak discharges in the range 10³–10⁵ m³/s are associated with major subglacial volcanic eruptions that create transient lakes or overfill existing ones (Tómasson 1996; Gudmundsson et al. 1997; Tómasson 2002; Carrivick et al. 2004). Despite being functionally a class of glacier-lake breakout because the outflow hydrograph is controlled by the glacier not the volcano (Walder and Costa 1996; Clarke 2003), they are included here since the lake is the result of volcanic activity. Volcanogenic floods generated by the melting of snow and ice (Pierson et al. 1990) are however excluded unless a transient lake is developed (Waitt et al. 1983; Pierson 1997). In the latter case, a directed blast from the Mount St. Helens lava dome on 19 March 1982 dynamically mixed hot pyroclastic debris with snow and ice in the crater basin. A meltwater lake developed within tens of minutes to cover an area of 0.3 km² to a depth of 8–15 m, as indicated by rafted pumice blocks and mud-lines, before simultaneously breaching through outlet channels on either side of the lava dome. The lake held c. 4 × 10⁶ m³ of water at its highstand, but the total outflow was considerably more as contributions from snowmelt continued during lake drainage. Combined peak discharge for the two spillways is estimated at 2.6–4.4 × 10³ m³/s: bulking and transformation to debris flow increased this to 9 × 10³ m³/s by 6 km downstream.

6 Non-volcanogenic Floods from Volcanic Lakes

In addition to the volcanogenic triggering of break-out floods from volcanic lakes, these water bodies are also vulnerable to non-volcanic breaching of their impoundments, resulting in partial or total drainage of the whole basin. Newly created volcanic lakes are particularly vulnerable to breaching as they fill with meteoric water due to the inherent instability of their dams, which are typically constructed of unconsolidated, often low density material, and without the benefit of engineered spillways. Failure of the volcanic dam or basin rim can occur due to either overtopping or piping from within (Waythomas et al. 1996; Massey et al. 2010) or headward erosion of an external drainage (Karátson et al. 1999). A review of the geological literature reveals that such floods are a relatively common phenomena worldwide (Table 2).

6.1 Crater Lakes

The best characterised example of a passive break-out from a Crater Lake is at Mt. Ruapehu, New Zealand. Following the 1995–1996 sequence that emptied the Crater Lake during a series of initially phreatic, phreatomagmatic and then purely magmatic eruptions (Nakagawa et al. 1999), the lake basin refilled with precipitation run-off and fumarolic condensate over the following 11 years. On 18 March 2007, the rising Crater Lake breached the tephra dam following destabilisation of its downstream face by piping flow (Massey et al. 2010); releasing c. 1.3 million m³ of water in less than 2 h (Manville and Cronin 2007). The flood bulked up through entrainment of snow, ice, colluvium, and older lahar deposits in the Whangaehu Gorge to form one of the largest lahars of the historical sequence (Carrivick et al. 2009; Graettinger et al. 2010; Procter et al. 2010; Lube et al. 2012). No lives were lost during this event owing to improvements in knowledge and monitoring, in contrast with the 1953 Tangiwai lahar that claimed 151 lives (Healy 1954; O’Shea 1954). This break-out was triggered by collapse of an unstable barrier composed of volcanic material deposited during the 1945 eruption and buttressed by the Crater Basin Glacier (O’Shea 1954; Manville 2004). Approximately 1.8 × 10⁶ m³ of 26 °C water was released into the headwaters of the Whangaehu River over a period of 1–2 h following piping failure of the tephra dam (Fig. 7a). The flow rapidly entrained snow, ice, and volcanic debris and alluvium to transform into a debris- to hyperconcentrated flow that critically damaged a railway bridge 39 km downstream at Tangiwai (Fig. 7b), resulting in the loss of 151 lives (O’Shea 1954; Stilwell et al. 1954). Palaeohydraulic analysis indicates that the peak discharge from Crater Lake was c. 350 m³/s, bulking to 2,000 m³/s by 10 km downstream (Manville 2004) and attenuating to c. 590–650 m³/s at Tangiwai.

Fig. 7

a Breach in tephra dam/glacier ice resulting from the 24 December 1953 break-out of Ruapehu’s Crater Lake. Mouth of ice tunnel is c. 50 m wide and 30 m high. Remnants of tephra dam visible to left of breach. (GNS archive). b Oblique aerial view of Tangiwai, 38 km downstream from Crater Lake. The lahar came from the top of the picture and partially destroyed the railway bridge just before the Wellington-Auckland express train arrived [Archives New Zealand, R5, W2279/13, 300/1607]

Other historic non-volcanic floods from crater lakes include the 1875 non-volcanic failure of 60 m of the western rim of Mt. Kelut (Table 2), which collapsed during heavy rain releasing a lethal cold lahar and changing the principle lahar path to the southwest (Suryo and Clarke 1985). Elsewhere in Indonesia, break-out of a crater lake has been correlated with a devastating flood in AD 1638 at Raung (van Padang 1951).

A mudflow and flood in 1541 AD that destroyed part of the former capital of Guatemala, Ciudad Vieja, claiming >1,300 lives has been attributed to collapse of the crater rim of Agua following a period of torrential rain (Mooser et al. 1958; Neall 1996). No lake was reported in the crater prior to this event, although it is possible that an ephemeral water body formed and failed due to the intense precipitation.

An unusual mechanism for generating a crater lake break-out has been inferred for the c. 2.5 km diameter Albano composite maar in Italy (Funiciello et al. 2003; De Benedetti et al. 2008). In Roman times, the lake stood some 70 m higher than its current surface elevation of 293 m, coincident with the low point of the topographic rim, which also forms the apex of a 60 km² outwash fan of laharic deposits. This fan indicates at least two catastrophic water releases from the lake during the Holocene (Funiciello et al. 2003), while Roman archives record an overspill event in 398 BC (De Benedetti et al. 2008). Both authors speculate that during climatic highstand periods, seismically-triggered injections of CO₂-rich hydrothermal fluid into the groundwater and lake bed triggered dramatic rises in lake level accompanied by overspill and partial failure of the crater rim.

6.2 Caldera Lakes

Historical examples of break-out floods from caldera lakes include Pinatubo, in the Philippines, in 2002. The 5–7 km³ DRE eruption of Pinatubo in June 1991 formed a 2.5 km wide summit caldera that accumulated a lake over the following decade (Stimac et al. 2004). As the lake approached the low point on the rim (960 mASL) a crisis developed due to recognition of the potential for catastrophic breaching and the release of a huge lahar into the heavily populated Balin-Baquero catchment. During August-September 2001 a trench was dug by hand through the unconsolidated 1991 pyroclastic deposits forming the upper part of the rim in an attempt to trigger a controlled break-out, or at least prevent a further rise in lake level (Lagmay et al. 2007). However, outflow through the trench was too slow to produce a controlled breach. Failure of the caldera rim was delayed until 10 July 2002 when Typhoon Gloria delivered 740 mm of rain to the Pinatubo area (Bornas et al. 2003). Approximately 65 × 10⁶ m³ of water was released at up to 3,000 m³/s, lowering the lake level by 23 m and eroding a V-shaped notch in the caldera rim. Downstream bulking of the flow with abundant pyroclastic debris from the 1991 eruption generated the largest lahar of the entire post-Pinatubo sequence, aggrading the Bucao River valley by 3–5 m and threatening the town of Botolan (population 40,000).

Subaqueous eruption of a tuff ring in Karymskoye lake, Kamchatka, on 3 January 1996 generated not only primary lahars through the explosive displacement of lake water, but also temporarily blocked the outlet to the 4 km diameter Akademiya Nauk caldera (Belousov and Belousov 2001). After rising by 2.6 m, the 12.5 km² lake breached the pyroclastic barrier on 15 May 1996, triggering a flood that deposited a small fan downstream of the outlet gorge.

The c. 10 km diameter Aniakchak caldera was formed by a major ignimbrite emplacing eruption at c. 3.4 ka on the Alaskan peninsula (Miller and Smith 1977). Post-eruption, the closed topographic depression (Fig. 8) was partially filled by an intracaldera lake that rapidly drained following failure of the caldera rim, releasing c. 3.7 × 10⁹ m³ of water at a peak rate of 7.7 × 10⁴ to 1.1 × 10⁶ m³/s and cutting a 183 m deep notch (McGimsey et al. 1994; Waythomas et al. 1996). Highstand terraces indicate that the lake was close to its maximum capacity before the breakout, suggesting failure was initiated by overtopping, possibly triggered by tsunamis or water displacement during an intracaldera eruption. Similar scenarios have been reconstructed at the Fisher (Stelling et al. 2005) and Okmok (Almberg et al. 2003) calderas in Alaska, and at Towada volcano in Japan (Kataoka 2011) on the basis of geological and geomorphic evidence (Table 2).

Fig. 8

Aniakchak caldera in Alaska formed at c. 3.4 ka: the intracaldera lake later breached to produce the largest known Holocene terrestrial flood, releasing 3.7 × 10⁹ m³of water at a peak discharge of up to 1 × 10⁶ m³/s (Waythomas et al. 1996). a Aniakchak caldera looking east, diameter c. 10 km (Image M. Williams, National Park Service 1977). b ‘The Gates’ the c. 200 m deep breach cut in the rim of the Aniakchak caldera by the break-out flood (Image C.A. Neal, USGS)

Lake Taupo, the largest lake in the central North Island of New Zealand partially occupies a volcano-tectonic collapse structure whose present configuration largely reflects caldera collapse during the 26.5 ka Oruanui eruption and faulting and downwarping on regional structures (Davy and Caldwell 1998; Wilson 2001). The 530 km³ DRE Oruanui eruption destroyed a long-lived Pleistocene lake system and deposited hundreds of metres of unwelded pyroclastic deposits around the collapse caldera to produce a closed topographic depression (Wilson 2001; Manville and Wilson 2004). Following the eruption, water in Lake Taupo accumulated to reach a highstand elevation of c. 500 m, as marked by an irregularly preserved shoreline terrace (Grange 1937; Manville and Wilson 2003, 2004). Overtopping caused an initial overspill and 10–20 m drop to a level controlled by a resistant sill of welded ignimbrite, allowing development of another palaeoshorelines terrace. Some time before 22 ka, headward erosion through non-welded pyroclastic flow deposits 20 km to the east breached the caldera rim, releasing an estimated 60 km³ of water in a break-out flood peaking at 3.5 × 10⁵ m³/s (Manville and Wilson 2004). During the 1800 a Taupo eruption, the outlet gorge from the lake was blocked by voluminous pyroclastic flow deposits (Wilson and Walker 1985) while much of the pre-eruption lake was expelled, evaporated, or drained into a sub-rectangular caldera collapse structure (Davy and Caldwell 1998). Over a period of several decades (Wilson and Walker 1985; Smith 1991a), the lake refilled to a mean highstand level of +34 m above the modern elevation of 357 m, as marked by a semi-continuous, tectonically warped and offset wave-cut bench and highstand shoreline deposit (Wilson et al. 1997; Riggs et al. 2001; Manville and Wilson 2003). Failure of the 12 km long unconsolidated pumiceous pyroclastic dam was initiated by overtopping, and is inferred to have been catastrophic based on the absence of intermediate shoreline terraces between the highstand and a wave-cut bench at +2–5 m (Manville et al. 1999), releasing c. 20 km³ of water in a single phase. Flood deposits can be traced for over 220 km downstream from Lake Taupo (Manville et al. 1999; Manville 2002), and include: a 12-km-long vertical-walled spillway immediately downstream of the outlet floored with lithic gravel and boulder lags; streamlined landforms sculpted from older deposits and exhumed river terraces; bouldery fan deposits and expansion bars downstream of valley constrictions; fine-grained slackwater deposits in off-channel embayments and hydraulically ponded depressions; valley-wide erosional unconformities; and buried forests in distal areas.

6.3 Pyroclastic Dams

As well as impounding water bodies within the depressions of erupting volcanic centres, pyroclastic material emplaced laterally by pyroclastic flows and surges, or by vertical fallout from Plinian or ballistic eruptions can dam existing valley systems at varying distances from the active volcano.

The 1991 Pinatubo eruption in the Philippines generated proximal pyroclastic flow deposits up to 200 m thick that locally dammed numerous catchments (Scott et al. 1996a). Most lakes were shallow and short-lived, failing by overtopping during the 1991 wet season, but others filled and failed repeatedly as secondary pyroclastic flows and/or lahars rebuilt barriers at late as 1994 (Table 2).

Pyroclastic flows emplaced by the March-April 1982 eruption of El Chichón in southern Mexico partially filled the drainage network on and around the volcano, blocking the adjacent Río Magdalena with deposits up to 30 m thick and 300–400 m wide that extended for c. 1 km along the valley (Silva et al. 1982; Macías et al. 2004). By the time the dam failed by overtopping on 26 May it impounded a 5 km-long lake with a volume of 48 × 10⁶ m³ that reached boiling temperatures due to its contact with 650 °C pyroclastic material. Drainage is estimated to have taken c. 2 h, generating a scalding flood that cooled from 82 °C at 10 km downstream to 52 °C at a hydroelectric plant 35 km downstream where a worker was killed. Peak discharge was estimated at 1.1 × 10⁴ m³/s (Macías et al. 2004), while the break-out deposits covered 5 km² and thinned from 8.5 to 2 m distally. Deposit sedimentology indicates that the flow evolved from clear-water, through hyperconcentrated flow to debris flow (Costa 1988). This event resembles an oral legend of a hot flood in the same area centuries before (Duffield 2001).

In 1912, the largest pyroclastic eruption of the 20th Century occurred in the Valley of Ten Thousand Smokes in Katmai National Park in Alaska (Hildreth 1983). A chain of secondary phreatic explosion craters impounded a 1.5 km long supra-ignimbrite lake that failed in the summer of 1912 or 1913. The resulting flood scoured the surface of the pyroclastic flow deposit to a depth of 1–2 m, transported 0.5 m diameter blocks of welded tuff over 20 km, and aggraded the lower parts of the valley by 1–8 m (Hildreth 1983). The eruption at Katmai also created a small caldera, now occupied by a lake. As the rate of filling has slowed markedly since the 1950s a break-out is considered unlikely (Motyka 1977).

The eruption of Asama, Japan, in 1783 AD dammed the Agatuma-gawa River with a block-and-ash flow. Breaching of the dam resulted in a catastrophic flood that destroyed over 1,200 houses and killed an estimated 1,300 people (Aramaki 1981; Yasui and Koyaguchi 2004).

Emplacement of the 30 km³ Taupo ignimbrite over an area of 20,000 km² radially disposed around the vent also blocked numerous other catchments, leading to the development of multiple supra-ignimbrite lakes (Smith 1991b; Manville et al. 2005), the largest of which impounded c. 1.5 × 10⁹ m³ of water over an area of 190 km² in the Reporoa Basin (Manville 2001). Further lakes formed in areas where river valleys were dammed by distal pyroclastic flow lobes (Segschneider et al. 2002). Overtopping failure of these lakes likely generated break-out floods and is believed to have contributed to the re-establishment of drainage systems (Meyer and Martinson 1989; Manville 2002). In more distal areas, aggradation by lahars also dammed tributaries to form numerous shallow, ephemeral lakes.

The eruption of the Laacher See volcano at 12.9 ka, the largest late Quaternary eruption in central and western Europe, blocked the Rhine River to a depth of at least 15 m by a combination of primary pyroclastic flows that travelled down tributary valleys and reworked pyroclastic fall material from the lower Neuwied Basin (Park and Schmincke 1997; Baales et al. 2002). Six km upstream of the gorge blockage site an ephemeral lake developed in the Neuwied Basin proper, covering 80–140 km² at its maximum as marked by highstand rafts of floated pumice c. 15 m above the pre-eruption land surface. The lake is inferred to have failed catastrophically by earthquake shaking during late stage explosions from Laacher See (Park and Schmincke 1997), releasing an estimated 9 × 10⁸ m³ of impounded water in a flood that scoured channels floored by dense lag deposits in the Neuwied Basin as it drained and deposited fluvially-reworked pyroclasts for over 50 km downstream as far as Bonn.

6.4 Lava Flow Dams

Lava flows are eminently capable of blocking river valleys owing to their downslope flow and physical properties and dimensions. Lava flows emplaced during the 1846–1847 eruptions of Volcán Quizapu in Chile blocked a number of tributaries of the Rio Blanquillo (Hildreth and Drake 1992). In 1932 a pumiceous lahar accompanying an eruption triggered break-out and permanent drainage of the Lagunas del Blanquillo, causing much damage downstream on the Río Maule: concerns had been expressed about the dam-break hazard from these lakes as early as 1916.

Eruption of a rhyolitic lava flow at 23 ka dammed the 25 × 15 km Laguna del Maule intracaldera lake basin in Chile, raising the lake level by 160 m. A single sharp strandline suggests rapid break-out of the lake, while its catastrophic nature is indicated by a narrow gorge cut through the lava and boulder bars and scablands below the gorge outlet (Hildreth, pers. comm. 2004).

Basaltic lava flows have dammed a number of rivers in the western United States (Ely et al. 2012). As well as rivers in Arizona, Idaho, Oregon, and Washington, the Colorado River in the western Grand Canyon in Arizona has been dammed at least 13 times in the last 1.2 Ma (Hamblin 1994; Fenton et al. 2002). Evidence for catastrophic failure of some of these dams has been presented by Fenton et al. (2002, 2003), who argued that emplacement of basaltic lava erupted from the Uinkaret Volcanic Field adjacent to the canyon on unstable talus slopes and alluvium deposits, in addition to the hyaloclastic foundations of the flows formed through interaction of the lava with river water, made the blockages inherently unstable. At least 5 lava dams failed catastrophically between 100 and 525 ka, depositing flood gravels 20–110 m thick between 53 and 200 m above current river level and up to 53 km downstream from the inferred dam site (Fenton et al. 2003). Sedimentary evidence of large scale floods from these break-outs includes downward fining boulder deposits with maximum clast dimensions of up to 35 m, giant cross-beds (up to 45 m high foresets), slackwater deposits, and high-water markers that decline exponentially in elevation downstream of the dam site. The Qfd4 unit flood at 165 ± 18 ka resulted from failure of a 280 m high dam impounding an estimated 9 km³ of water: peak outlet discharge is estimated as 2.8–4.8 × 10⁵ m³, declining to 1.6 × 10⁵ m³/s by 59 km downstream (Fenton et al. 2003). The youngest flood event, Qfd5 at 104 ± 12 ka, resulted from failure of a 180 m high dam with a reservoir volume of 2.7 km³ to give an estimate peak discharge of 1.4–2 x 10⁵ m³/s. Non-catastrophic breaching of many valley lava flow dams occurred over thousands of years following filling of the impoundment with sediment (Crow et al. 2008; Ely et al. 2012).

Late Pleistocene basaltic lava flows erupted from the Fort Selkirk area in Canada dammed the Yukon River to a depth of c. 30 m, representing the most recent volcanic blockage of this drainage: distinctive pillow lavas are interpreted as representing lava emplacement into standing water (Huscroft et al. 2004). A coarse unit composed of basalt boulders >1 m in diameter outcropping 6 km downstream of the inferred dam site are interpreted as the product of a catastrophic breaching of the barrier.

6.5 Debris Avalanche Dams

The growth of a volcanic edifice is typically limited by the gravitational stability of its cone, with periods of instability experienced as a consequence of both endogenous and exogenous factors (Siebert 1984; McGuire 1996). Structural failure can then be triggered by other factors such as tectonic or volcanic earthquakes (Vidal and Merle 2000) or pore fluid pressure changes induced by magma intrusion (Iverson 1995). Such failures typically generate massive landslides, or debris avalanches (Glicken 1998; Siebert 1996), with the largest terrestrial examples having volumes of c. 30 km³ (Stoopes and Sheridan 1992).

The May 1980 eruption of Mount St. Helens in the western USA was accompanied by emplacement of a 2.3 km³ debris avalanche deposit following gravitational collapse of the volcanic edifice (Lipman and Mullineaux 1981; Glicken 1998). The debris avalanche deposits impounded or modified three large lakes (Coldwater Creek, South Fork Castle Creek, and Spirit Lake) and several smaller ones (Jennings et al. 1981; Meier et al. 1981; Youd et al. 1981; Meyer et al. 1986; Glicken et al. 1989). Failure of the blockage to Maratta Creek on 19 August 1980 transferred water to the Elk Rock impoundment, which itself overtopped and breached on 27 August, releasing 0.3 × 10⁶ m³ of water into the North Fork Toutle River with a peak discharge of 450 m³/s (Jennings et al. 1981; Costa 1985). Modelling of the potential peak discharges from failure of the three largest lakes (Jennings et al. 1981; Dunne and Fairchild 1983; Swift and Kresch 1983) suggested clear-water peaks > 7.5 × 10⁴ m³/s were possible, prompting intervention by the U.S. Army Corps of Engineers to artificially stabilise their levels (Costa 1985). A pre-historic break-out from an ancestral Spirit Lake similarly dammed by a debris avalanche occurred at c. 2.5 ka, generating a peak discharge estimated at 2.6 × 10⁵ m³/s (Scott 1989).

A number of prehistoric break-out floods from debris avalanche dammed lakes have been identified globally. In Alaska the upper Chakachatna valley was dammed to a depth of 150 m by a Holocene debris avalanche from the Spurr volcanic complex, impounding a lake estimated to have held 1.2 × 10⁹ m³ of water (Waythomas 2001). Other examples include a major debris avalanche from the Antuco volcano at c. 9.7 ka that dammed the outlet to the Río Laja in Chile, impounding a large lake that subsequently failed to generate an outwash fan covering an area of 50 × 60 km (Thiele et al. 1998), and collapse of the eastern flank of the Nevado de Colima Volcano in Mexico at 18.5 ka which produced a voluminous debris avalanche that dammed the Naranjo River 20 km away to a depth of 150 m. Obstruction of the drainage produced a temporary impoundment of c. 1 km³ of water, which eventually breached the dam through overtopping, generating a peak discharge estimated at 5.7 × 10⁵ m³/s (Capra and Macías 2002). Bulking of the flow through entrainment of material from the dam and along the flood path increased the peak discharge by a factor of 6 and deposited c. 10 km³ of debris.

6.6 Lahar Dams

Rapid sedimentation by lahars is capable of damming rivers at confluences where one stream aggrades its bed more rapidly than the other (Rodolfo et al. 1996). The dominant stream builds a steep-fronted delta, fed by sediment-laden underflows, which progrades into the subordinate channel as the stagnation zone at the flow confluence migrates. Damming occurs when the rate of aggradation is faster than the rate of water level rise in the impounded channel.

Remobilisation of voluminous pyroclastic flow deposits at Pinatubo in the Philippines following the June 1991 eruption led to multiple instances of lahar-dammed lakes and subsequent breakouts in a number of catchments (Pierson et al. 1992; Newhall and Punongbayan 1996). The largest such impoundment, Lake Mapanuepe (Fig. 2d), covered an area of 6.7 km² and held 75 × 10⁶ m³ of water at its maximum extent in 1991 (Umbal and Rodolfo 1996). Eighteen break-outs, peaking at 400–650 m³/s, were recorded in 1991 alone as the lake repeatedly overtopped and partially breached the impounding lahar deposit, which was rapidly rebuilt by aggradation of 7–20 m/year (Rodolfo et al. 1996). In total, the dam reached a height of c. 25 m before engineering intervention stabilised the lake elevation at 111 m. Numerous other lakes filled and spilled between 1991 and 1994, generating large lahars and entrenching drainage routes that became conduits for subsequent damaging flows (Newhall and Punongbayan 1996; Rodolfo 2000). Similarly, the 1902 eruption of Santa Maria in Guatemala resulted in the creation of at least 16 lahar-dammed lakes large enough to be mapped in the decades after the eruption (Kuenzi et al. 1979). However, it is not recorded whether any of these failed catastrophically.

A number of small, short-lived lakes were impounded by lahar dams in the upper Chakachatna River valley, following historic eruptions in 1953 and 1992 of the Spurr volcanic complex in Alaska (Waythomas 2001). The 1953 lahar dam was c. 20 m high, 600 m long and 200 m wide; the 1992 lahar dam was approximately 10 m high, 200 m long and 50 m wide: lake volumes are estimated at 10⁷–10⁸ m³ from modern topography. Lahar dams up to 60 m high have also formed at least twice in the late Holocene, impounding up to 4.5 × 10⁸ m³ of water.

In the aftermath of 1886 AD basaltic plinian eruption of Tarawera, New Zealand (Walker et al. 1984), the level of Lake Tarawera rose by 12.8 m, before a rain-triggered break-out in November 1904 reduced it by 3.3 m, producing a flood estimated to have peaked at c. 780 m³/s 24 km downstream (White et al. 1997). The post-eruptive rise has been attributed to construction of a small alluvial fan across the outlet channel by flash-flood induced remobilisation of 1886 pyroclastic and older debris in the Tapahoro gully (Hodgson and Nairn 2000).

As well as multiple pyroclastic-flow dammed lakes, numerous temporary impoundments were created in the aftermath of the 1.8 ka Taupo eruption through volcaniclastic resedimentation and laharic aggradation of up to 10s of metres in distal areas (Kear and Schofield 1978; Manville 2002; Segschneider et al. 2002).

6.7 Composite Dams

Composite dams are those formed by a combination of mechanisms, most commonly primary and secondary pyroclastic flows and lahars (Table 2). One such example occurred at Lake Tarawera following the c. 1315 AD Kaharoa eruption (Nairn et al. 2001). Lake Tarawera lies within the 64 ka Haroharo caldera in the Okataina Volcanic Centre in New Zealand (Nairn 2002), and is bounded by the western rim of the caldera and the resurgent lava dome complexes of Haroharo and Tarawera to the east. The 5-7 km³ DRE Kaharoa eruption formed much of the Tarawera dome complex and deposited widespread plinian falls (Nairn et al. 2001). Primary block-and-ash flows and pyroclastic debris remobilised by flash floods off the Tapahoro dome blocked the narrow lake outlet, infilling a narrow channel cut through a c. 5 ka pyroclastic/volcaniclastic fan implicated in a previous highstand. Lake Tarawera rose by c. 30 m above its pre-eruption level before tsunami generated by late-stage pyroclastic flows overtopped the dam and triggered its catastrophic failure (Hodgson and Nairn 2005). Approximately 1.7 × 10⁹ m³ of water was released into the head of the Tarawera valley as the lake level fell by >40 m, excavating a 300 m wide and 3 km long spillway before overtopping the 70 m high Tarawera Falls. Flood deposits including boulders up to 13 m in diameter and giant bars extend c. 40 km from the lake; approximately 700 km² of the Rangitaiki Plains was resurfaced and the shoreline advanced by c. 2 km (Pullar and Selby 1971). Peak discharge at the outlet was estimated at 1.5 × 10⁵ m³/s assuming instantaneous breach formation, while boulder flow-competence relations further downstream indicate flows in the 10⁴–10⁵ m³/s range (Hodgson and Nairn 2005).

7 Discussion

A review of hazards from volcanic lakes due to eruptive activity or other causes reveals a great deal of complexity in triggering mechanisms and hazardous phenomena. Well-studied historic examples such as the 1996 eruptions at Karymskoye (Belousov and Belousov 2001) and Taal (Moore et al. 1966a), the 1991 Pinatubo eruption and its aftermath (Newhall and Punongbayan 1996), and lahar events at Ruapehu (Cronin et al. 1997; Kilgour et al. 2010) and Kelut (Suryo and Clarke 1985) demonstrate this variety and complexity: in many cases individual hazards contribute to subsequent effects in a process-chain. Examination of the geological record reveals a fuller range of both hazards and event magnitudes.

Moderate-sized (10⁷–10⁸ m³) crater lakes developed at the summits of andesitic-dacitic stratovolcanoes in relatively humid arc environments are potentially the most dangerous volcanic lakes due to their capacity for almost instantaneous expulsion of large water volumes during subaqueous explosive eruptions, which can occur with little or no effective warning (e.g. Mt. Kelut, Ruapehu). Smaller water bodies are either evaporated or the expelled water is too widely dispersed during eruptions to generate significant lahars. In larger diameter lakes the Surtseyan jets that are the main displacers of water cannot reach their rims, and water depths >100 m suppress the ability of subaqueous explosions to breach the surface (Mastin and Witter 2000; Morrissey et al. 2010).

At crater and caldera lakes more than 2 km in diameter, base surges and tsunamis generated by volcanic activity, or collapse of the caldera walls (Ramos 1986), are the most significant hazard. Examples with populated shorelines that fall into this category include Taal in the Philippines, where historic eruptions have resulted in major loss of life, Taupo and Rotorua in New Zealand, Ilopango in El Salvador, and Toya and Towada in Japan. Volcano-tectonic lakes facing similar hazards include Lakes Managua and Nicaragua in Nicaragua. Tianchi (Baitoushan) at the summit of Mount Paektu/Changnai on the North Korea/China border is vulnerable to both renewal of volcanic activity and collapse of the steep walls of this very young (c. 1 ka) 5 km diameter lake (Wei et al. 2003, 2004). Lakes Quilotoa and Cuicocha in Ecuador are also potentially hazardous (Gunkel et al. 2008).

Volcanic lakes have lifespans measured in minutes to thousands of years. Intervals between creation and failure are a function of the geometry and stability of the lake-forming blockage, and the rate of lake level rise, which is itself a function of the catchment area, the proportion occupied by the lake, climatic effects, and seepage losses (Capra 2007). Examination of Tables 1 and 2 shows that a number of hazardous volcanic lakes have been destroyed in the past few hundred years while new ones have been created. Natural dams composed of unconsolidated pyroclastic deposits are extremely vulnerable to break-outs triggered by non-volcanic processes, such as erosion following overtopping (El Chichón), internal piping (Ruapehu), or headward erosion. Where barriers are composed of indurated material such as lava flows or welded ignimbrites, lakes can persist for many thousands of years: some, such as Crater Lake, Oregon, never reach overtopping level because seepage losses match water inflows as the lake level rises and hydrostatic pressures increase. Others, such as Aniakchak (McGimsey et al. 1994; Waythomas et al. 1996) and Fisher volcano (Stelling et al. 2005) in Alaska appear to have reached a stable highstand level controlled by a hard rock sill. Failure required active breaching of the rim by tsunamis generated during intra-lake eruptions, in the case of Fisher volcano this was 7.9 ka after caldera formation (Stelling et al. 2005).

The largest volcanic flood hazards derive from break-outs from intracaldera lakes (Taupo, Okmok, Aniakchak, Towada), and from valleys dammed by debris avalanches (Colima, Mount St. Helens) because of the potential volumes of water involved. Avalanche-dammed lakes are particularly hazardous not only because they can be voluminous but also because flood discharge can be amplified by the erosion and entrainment of unconsolidated material from the blockage (Capra and Macías 2002).

Other hazards can arise at crater lakes due to acid brine seepage, which can weaken the edifice (Reid et al. 2001), potentially leading to sector collapse debris avalanches and break-out lahars, e.g. at Rincon de la Vieja, Costa Rica (Kempter and Rowe 2000), and Copahue, Argentina (Varekamp et al. 2001). Acid brines can also contaminate drinking and irrigation water, while acid gas aerosols released from hyper-acidic crater lake waters are an unusual potential hazard along flow paths (Schaefer et al. 2008).

7.1 Mitigation

A variety of engineering interventions have been applied to prevent break-outs of unstable lakes and to mitigate lake break-out floods (Schuster 2000). Successfully applied techniques include: (i) reinforced spillways and check dams on weak blockages; (ii) drainage and diversion channels or tunnels across or through stable (bedrock) abutments; and (iii) pumps or siphons to lower lake level. These methods vary in expense and difficulty, and some are reliant on suitable topography or an adequate timeframe and success is not guaranteed. Diversion tunnels are probably the most long-term (and expensive solution), but despite their success at some volcanoes (e.g. Kelut and Mount St. Helens), they require maintenance and are vulnerable to destruction by renewed volcanic activity. There are also considerable eruption and other risks to workers during construction phases. Downstream mitigation measures against lahars and floods include construction of check dams, bunds and dykes, and raising and strengthening of bridges.

7.1.1 Engineering Interventions

Engineering interventions to reduce the hazard from volcanic lakes pre-date the Romans: at Albano maar, a tunnel dug by the Etruscans to drain the lake below its overspill level was rebuilt by the Romans in c. 396 BCE (Funiciello et al. 2003; De Benedetti et al. 2008). This 1.5 km long tunnel lowered the lake by 70 m and still functions today.

Mt. Kelut in Indonesia is probably the best example of a successful, although protracted, expensive, and technologically challenging intervention (Zen and Hadikusumo 1965; Suryo and Clarke 1985). The first attempt to control eruption-triggered lahars from the summit crater lake was construction of a diversion dam on the Badak ravine in 1905 following the 1901 eruption. This was overwhelmed by the 1919 lahars when all water was explosively expelled from the lake, triggering flows up to 58 m deep that travelled up to 38 km from source, inundating 131 km², destroying more than 100 villages and killing over 5,000 people. Following this disaster, an earlier plan to drain the lake was implemented in September 1919. This involved excavating a 955 m long tunnel with a 3 % gradient designed to intercept the bottom of the lake. Tunnelling was begun at both ends as the lake was dry following the eruption earlier in the year, but was stopped at the upstream end by lava ascent a year later. After tunnelling upwards for 735 m, volcanic debris in the inner crater wall was encountered and the tunnel needed to be lined with concrete. Meanwhile, the crater lake had been refilling until by early April 1923 it held 22 × 10⁶ m³ of water: at this point leakage into the tunnel killed 5 workers. This led to modification of the plan to include a series of tunnels through which water would be siphoned at progressively lower levels (Fig. 9). The first tunnel was driven just above the then lake level of 1,185 m, preventing any further rise, and the lake was progressively lowered to 1,129 m. The volcano erupted again on 31 August 1951, but the small (<2 × 10⁶ m³) volume of lake water was largely evaporated rather than explosively expelled, leading to minimal lahars. Unfortunately the tunnels were blocked by the eruption and the crater deepened by c. 70 m (Zen and Hadikusumo 1965). In 1954, work was begun to reactivate the tunnels, with 3 being cleared by 1955 when the lake volume was limited to 23.5 × 10⁶ m³. A new tunnel was then driven 20 m below the lowest old tunnel, and two galleries and further adits were then driven from it towards the lake in the hope that seepage would drain the water. Unfortunately the desired drainage rate was never achieved and the project was abandoned in March 1963. On 26 April 1966 an eruption expelled the c. 20 × 10⁶ m³ of water in the lake, generating lahars that destroyed the rebuilt Badak diversion dam, killing 208, and again damaging the tunnel system. In June 1966 a new tunnel was dug linking the then empty crater to an older unfinished drive. Completion of this by the end of December 1967 limited the lake to a maximum volume of 4.3 × 10⁶ m³ with the result that the 1990 eruptions produced only minor primary lahars that killed only 32 (Thouret et al. 1998).

Fig. 9

Cross-section of the crater lake of Mt. Kelut in Indonesia showing the tunnel system used to lower the lake level and reduce the hazard from eruption-triggered lahars (after Zen and Hadikusumo 1965)

At Mount St. Helens, the levels of the three largest lakes impounded by the 1980 debris avalanche were stabilised by the U.S. Army Corps of Engineers by the construction of permanent spillways and bedrock drainage tunnels (Costa 1985). The greatest assessed risk was from Spirit Lake: numerical studies suggested that failure of the debris avalanche dam could generate a flood with a peak discharge of 7.5 × 10⁴ m³/s (Swift and Kresch 1983). To gain time, a 20-pump facility was installed to remove up to 5 m³/s of water through a 1.5 m diameter pipe buried in the dam crest while a 2.59 km long, 3.4 m diameter tunnel was bored through the ridge to the west of the lake, ultimately stabilising the lake at 1,048 m (Schuster and Evans 2011). Meanwhile, Coldwater Lake was stabilised by construction of a permanent spillway across a resistant bedrock abutment, and Castle Lake was stabilised by construction of an armoured spillway.

Repeated break-out floods from the lahar-dammed Mapanuepe Lake were a feature of the early years at Pinatubo in the Philippines following the 1991 eruption (Rodolfo et al. 1996; Umbal and Rodolfo 1996). Its level was stabilised in November 1992 by excavation of a bedrock spillway through an adjacent rock spur. Geological and anthropological data indicate that a similar lake occupied this site on two previous occasions (Rodolfo and Umbal 2008).

Emptying of the summit Crater lake of Mt. Ruapehu during the 1995–1996 eruptions (Johnston et al. 2000), coupled with deposition of a c. 8 m thick blanket of tephra over the former outlet area, raised the possibility of a future break-out lahar in a repeat of the 1953 Tangiwai lahar scenario, which resulted in New Zealand’s worst volcanic disaster (Healy 1954; O’Shea 1954). Given the decade-long delay between formation of the tephra barrier and the subsequent breakout in March 2007 (Manville and Cronin 2007; Massey et al. 2010; Lube et al. 2012) there was adequate time to quantify the relative risks of intervention at the crater rim versus downstream mitigation efforts. The situation was complicated by the status of the summit of Ruapehu as a World Heritage Site, its location in a National Park, and not least Crater Lake’s special spiritual significance to the local Maōri population. Numerical modelling of the potential magnitude of any lahar event (Hancox et al. 2001) and a residual analysis of the risks of intervention versus non-intervention (Taig 2002) resulted in the decision to install a real-time telemetered warning system on the tephra dam and downstream channel, a bund at the apex of the Whangaehu Fan to prevent overspill into vulnerable northern drainages, raising and strengthening of the State Highway 49 road bridge at Tangiwai, a system of automatic lights and gates to close roads close to or crossing the predicted lahar path, and a public education campaign (Keys 2007; Becker et al. 2008; Keys and Green 2008). In the event, the lahar occurred as predicted with no loss of life or injuries and minimal infrastructural damage.

7.1.2 Limnic Gas Eruptions

Following the gas eruptions from Lakes Monoun and Nyos, warning systems comprising CO₂ monitors, sirens and public education were implemented, as well as a pipe system designed to safely degas the lake water (Kusakabe et al. 2008). Reinforcement of the outlet to Lake Nyos to stabilise the natural dam is also currently underway (Aka and Yokoyama 2013), in response to the potential for breaching to catastrophically release the upper 40 m of lake water. This would not only cause flooding into Nigeria over 200 km downstream, but also release another burst of CO₂ due to depressurisation of the gas-saturated deep water (Lockwood et al. 1988).

8 Conclusions

Although hydrologic hazards from volcanic lakes are over-represented in terms of fatalities per eruption, improvements in volcano monitoring and increased awareness are mitigating the risk at many volcanoes around the world. Timely and appropriate engineering interventions have been effective in minimising deaths at a number of volcanoes (Kelut) and preventing disasters at others (Mount St. Helens, Lake Nyos). Although currently very few volcanic lakes are monitored on a regular basis, changes in a lake’s appearance, thermal, and/or chemical characteristics could be diagnostic of impending volcanic activity (Oppenheimer 1993; Hurst and Vandemeulebrouck 1996; Christenson 2000; Christenson et al. 2010). Break-outs associated with collapse of blockages are less predictable, unless the lake level is approaching the overflow point, or seepage on the downstream face of the dam is obvious and sediment-laden (Turner et al. 2007; Massey et al. 2010).

Most fatalities due to volcanic activity occur between 10–30 km away from the vent (Auker et al. 2013). This reflects the combination of (i): this distance being within the range of the heaviest tephra falls; (ii) the typical travel distance of pyroclastic flows, lahars and intra-lake tsunamis; and (iii) the tendency for populations to be concentrated on the gentler slopes and flat alluvial (often lahar) terraces and plains flanking and downstream of volcanoes, (volcanic) lakeshores, and along coastlines vulnerable to volcanogenic tsunamis. However, this is also far enough away for a combination of effective monitoring and warning systems, public education, land-use planning, and engineering interventions to mitigate many of the hazards presented by volcanic lakes. In the future, a combination of clearer understanding and better practice are likely to be most effective in alleviating the associated dangers.