Insights into the recurrent energetic eruptions on Awu, one of the deadliest volcano on earth

The little know Awu volcano is among the deadliest with a cumulative dead toll of 11048. In less than 4 centuries, 18 eruptions were recorded, including two VEI-4 and three VEI-3 with worldwide impacts. The regional geodynamic is controlled by a divergent-double-subduction and an arc-arc collision. In that context, the slab stalls in the mantle, undergoes an increase of temperature and becomes prone to melting, a process that sustained the magmatic supply. Awu also has the particularity to host alternatively and simultaneously a lava dome and a crater lake throughout its activity. The lava dome occurred passively through the crater lake and induced strong water evaporation from the crater. A conduit plug associated with this dome emplacement subsequently channeled the gas emission to the crater wall. However, with the lava dome cooling, the high annual rainfall eventually reconstituted the crater lake and creating a hazardous situation on Awu. Indeed with a new magma injection, rapid pressure buildup may pulverize the conduit plug and the lava dome, allowing lake water injection and subsequent explosive water-magma interaction. The past vigorous eruptions are likely induced by these phenomena, a possible scenario for the future events.


Introduction
Awu is a little known active volcano located on Sangihe arc, northeast of Indonesia. It is the largest and the northernmost volcano of the arc with an aerial volume of ~27 km 3 that constitutes the northern portion of Sangihe Island (Fig.1). The edifice culminates at 1318 m above sea level and extends nearly 2000 m to the sea bed, on its western flank. The summit crater is 1500 m in diameter and 380 m depth from the rim. It is currently occupied by a cooled lava dome. Since early 1980s, numerous studies have pointed to Awu as the center of strong volcanic manifestations with global impacts (Robock, 1981;2000;Handler, 1984;Zielinski et al., 1994;Jones et al., 1995;Palmer et al., 2001;Donarummo et al., 2002;Guevra-Murua et al., 2015) although recent works have reviewed and declassified some of these events, including the 1641 event that was considered as responsible for 1642-1645 global cooling (e.g., Robock, 1981;Simkin et al., 1981;Jones et al., 1995) but later attributed to the Parker eruption of Jan. 4, 1641 (Delfin et al., 1997). Similarly, on the famous Edward Munch painting of 1893 -the "The Scream", the red sky was first considered as induced by the 1892 eruption of Awu (Robock, 2000) but was later attributed to Krakatau 1883 eruption (Olson et al., 2004) and then finally considered as inspired by the nacreous clouds (Fikke et al., 2017;Frata et al., 2018). But Awu 1812 eruption (VEI 4) has loaded a significant amount of ash and aerosols into the atmosphere leading to the global abnormal correlation between dust load in the atmosphere and solar activity (Donarummo et al., 2002). In 1856, another eruption on Awu (VEI 3) injected massive sulfate aerosols into the stratosphere, leading to an increase of stratospheric aerosol's optical depth to 0.06, sufficient to reduce the sea surface temperature and thus subsequently reducing the number of tropical cyclones following the eruptive event (Guevara-Murua et al., 2015). In contrast, Handler (1984) indicates that the 1966 eruption of Awu (VEI 4) loaded a notable amount of aerosols into the stratosphere resulting in a warmer eastern tropical Pacific Ocean over three consecutive seasons with subsequent influence on El Nino type events. On the regional and local scale, Awu eruptive activities have triggered at least two tsunamis, on Mar. 2, 1856 and on Jun. 7, 1892 Paris et al., 2014) with causalities. No less than 18 eruptions were reported on Awu volcano since 1640, thus about 1 eruption every ~20 years (Table 1). The latest eruption was a VEI 2 in 2004. In the database of volcanic eruption victims compiled by Tanguy et al. (1998), Awu eruptions claimed a total of 5301 victims, mainly following lahar events, including 963 casualties during 1812 eruption, 2806 during 1856 eruption and 1532 during 1892 eruption. This latter database did not take into account the 2508 victims of the 1711 eruption (Van Padan, 1983. Around 3200 inhabitants were also reported killed by lahar events following the 1822 eruption (Lagmay et al., 2007). The latest VEI 4 eruption on Awu in 1966 have killed 39 people, injured 2000 and forced the evacuation of 420,000 inhabitants (Withan, 2005). In total since 1711, Awu recurrent eruptive activities have caused a cumulative 11048 fatalities. Awu is thus one of the deadliest volcanoes worldwide but paradoxically very little is known about its activity. This works aims to highlight the intense eruptive character of Awu volcano and provide insights into the mechanism behind the recurrent energetic eruptions.

Methodology
The available documents that refer to Awu volcano, as summarized in the introduction, are generally incomplete but most point to vigorous explosions and subsequent casualties. Thus to gain more insights into Awu's volcanic activity, a reconnaissance visit to the summit was carried out in July 2015 with thermal and gas measurements. Thermal imaging was performed using OPTRIS PI400, a miniature infrared camera that weighs 320g, including a lens of 62°x49° FOV, f=8 mm and a dynamic range equivalent to the radiant temperature of -20°C to 900 °C. The detector has 382 × 288 pixels and the operating waveband is 7.5-13 µm. The maximum frame rate is 80 Hz. The camera was first positioned on the crater rim ( Fig.2) observing the whole crater, then in the crater, looking at the two main hot surfaces (Fig.2). The radiant flux (Qrad) estimation is obtained using the following: Qrad = AƐσ(Ts 4 -Ta 4 ), where A is the hot surface, Ɛ is the emissivity (0.9 for andesite), σ is the Stefan-Boltzmann constant (5.67x10 -8 W m -2 K -4 ), Ts is the hot surface temperature and Ta is the ambient temperature. Thermal results are corrected for an oblique viewing angle of 30° following the approach detailed in Harris (2013)  atmospheric influence relying on ACPC (https://atmcorr.gsfc.nasa.gov/) and validated with closer thermal recording before integrated into the radiant flux calculation. Thanks to the high acquisition rate of OPTRIS, few series of continuous recordings were obtained on the most heated surfaces to retrieve the heat flow dynamic.
A portable Multi-GAS system from INGV (as used by Aiuppa et al. 2015;Bani et al., 2017; was deployed to measure the gas composition. The instrument was positioned at different locations in the crater (Fig.2) and simultaneously acquired concentrations of H2O, CO2, SO2, H2S, and H2 at 0.1 Hz. Data were processed using Ratiocalc (Tamburello 2015). The scanning DOAS was used for the gas emission budget. The instrument performed at fixed position in the crater (Fig.2). Further details on this Awu's gas measurement are provided in a separate paper and hereafter referred to as (Bani et al. submitted).
The lava dome was sampled during this fieldwork then analyzed for major and trace elements using ICP-AES.

Results
The whole-rock composition of the lava dome, obtained from ICP AES analysis, indicates a source with 52-56% SiO 2 and relatively low alkaline corresponding to basaltic-andesite melt source (Table 2 respectively with the corresponding gas composition equivalent to 82% of H2O, 15% of CO2, 2% of H2S, 0.05% of SO2 and 0.02% of H2. Assuming the above results are representative then H2O, CO2, H2S, and H2 emission rates would be 5800 t/d, 2600 t/d, 340 t/d and 0.1 t/d respectively. The gas equilibrium temperature obtained by resolving together the SO2/H2S vs. H2/H2O redox equilibria (see methodology in Aiuppa et al., 2011;Moussallam et al., 2017) is circa 380 °C (Bani et al., submitted) Thermal infrared recording from the rim highlights two main heated surfaces in Awu's crater, but both are located to the northern part of the lower crater wall, next to the lava dome (IR2, IR3, Fig.3). It is also evident from these thermal results that the lower crater wall around the dome is much hotter than the lava dome itself. The total radiant flux from the crater (IR1 , Table 4) is 27±12 MW, including 5.6±2.4 MW from the lava dome and 21±9 MW from the area surrounding the dome. The highest radiant flux per area (0.9 MW) is recorded in the IR3 zone where gas is released at a low frequency of 0.3 Hz with a thermal fluctuation amplitude of 0.1-0.3 MW.

Melt source
To sustain the recurrent strong eruptive activity on Awu, highlighted by one strong eruption every 20 years over the last 3.5 centuries (including two VEI 4 and three VEI 3), it requires sufficient magma supply rate. The alkaline vs. SiO2 diagram (Table 2. controlled by the divergent double subduction that the currently resulted in Sangihe forearc overriding the Halmahera forearc (Cardwell et al.,1980, Morrice et al., 1983, Hall and Wilson, 2000Jaffe et al., 2004;Zhang et al., 2017;Bani et al., 2018). The pattern obtained by normalizing the trace elements to N-MORB indicates high LILE (Cs, K, Rb, Ba and Sr) content and low abundance of HFSE, represented by Nb, typical of subduction melt source in which the mantle wedge has been contaminated by fluid released from the subduction slab (McCulloch and Gambke, 1991;Davidson 1996, Mcpherson et al., 2003. This result is coherent with Jaffe et al. (2004) who highlight low 3 He/ 4 He (5.4-6.4 RA) and high CO2/ 3 He ratios (64-180 x10 9 ) as well as high δ 13 C (≥-2‰) suggesting slab contribution into the magmatic fluids at Awu. Clor et al. (2005) further point out anomalous high N2/He (2852) coupled with low δ 15 N (3.3%) suggesting increased slab contribution, possibly by slab melting as collision stalls the progress of the subducting plate and allows it to become superheated (Peacock et al., 1994). This is supported by the slow-down of the subduction rate as evidenced by seismic studies (McCaffrey, 1983;Pubellier et al., 1991;Zhang et al., 2017). This particular double subduction and arcarc collusion have rendered the slab prone to melting that subsequently produces the magmatic source behind the recurrent strong eruptive activities on Awu. The mechanism also contributes to unusual slab carbon delivery into the mantle as highlighted by the extremely elevated CO2 (Bani et al. submitted).

A conduit plug
Lava domes are formed when viscous lava extruded to the surface effusively then pile up around the vent. Such phenomena involve complex processes, including crystallization, bubble nucleation, growth, coalescence and outgassing, bulk magma deformation, crack propagation and healing (e.g., Ashell et al., 2015 and ref therein). It is the competition between these processes that either promote or prevent degassing, leading to explosions or stability of a lava dome (Klug and Cashmana, 1996;Takeuchi et al., 2005;Mueller et al., 2008). On Awu, the SiO2 content of the lava dome higher than 50 wt% as well as the perfect semi-spherical morphology of ~1.3x10 7 m 3 that extended from the middle of the crater suggests an endogenous growth that generally inflates the dome carapace through magma injection at depth. In such case, lava domes are known to induce variable porous and brecciated carapace surrounding a denser and coherent interior (Fink et al., 1992;Wadged et al., 2009, Ashell et al., 2015Newhall and Melson, 1983) suitable to form a plug in the upper conduit (Watts et al. 2002). The radiant thermal energy around the lava dome is much higher than the heat release from the dome representing 79% of the total 27 MW from the crater. Only ~6 MW is released through the lava dome itself. It also around the dome that much of the gas is released to the atmosphere ( Figure 5). The hottest surfaces also correspond to the main degassing points which suggest that heat is rather sustained by fluid circulations around the dome. With a conduit plug, the gas released at depth is thus forced to the periphery of the lava dome (Fig.5), similar to other dome-forming systems, including Rokatenda (Primulyana et al., 2018), Lascar (Matthews et al., 1997) or Soufriere Hills (Sparks, 2003).
It is thus obvious that the existence of a conduit plug may constitute a barrier to the gas flow, suitable for rapid pressure built up with new magma injection, a situation that can strongly contribute to the vigorous explosions on Awu.

Transition of heat to the surface controls the water accumulation
Out of the 17 recorded eruptive activities on Awu, 11 were tagged as phreatic and 6 other eruptions were phreatomagmatic (Table 1). It is thus unambiguous that water played a major role in Awu volcanic activity. Indeed, with an average annual rainfall of 3500 mm (Stone, 2010) and a crater area of 1.5 km 2 , the Awu summit is likely to accommodate 5.2x10 6 m 3 of water each year. Given that there is no visible water outlet from the crater, one can expect water accumulation and strong infiltration into the hydrothermal system which may then subsequently contribute to phreatic eruptions. But as highlighted in figure 6, surface water was not always present in Awu's crater. Crater lake existed in 1922, 1973 and 1995 whilst in 1931 and 1979 crater lake co-existed with a lava dome. In July 2015 (this fieldwork) there was no water in the crater and a lava dome occupied the central part of the crater. July is among the driest month of the year, however, the average monthly rainfall on Sangihe Island doesn't fall below 130 mm (Stone, 2010). Hence one can expect a cumulative water volume of at least 195x10 3 m 3 (equivalent to 7x10 6 moles or 1.3x10 8 g, using PV=nRT) into Awu's crater on that period of the year. But the absence of water as observed in July indicate that the water was efficiently infiltrated and evaporated away. In theory, if we assume that the infiltration is negligible, then it requires a heat energy of 8.0x10 11 joules (using mCpΔT; m is the water mass, Cp is the water's specific heat capacity, ΔT is the difference between boiling and ambient temperature) to bring the above volume to the evaporation temperature (100°C) and another 4.6x10 9 joules (using mL; L is the latent heat of vaporization) to convert it into water vapor. A total 8.1x10 11 joules is thus sufficient to dry out the July incoming water volume. With 27 MW of radiant flux from the crater, equivalent to 2.7x10 7 J s -1 , only 8 hours is necessary to heat the 7x10 6 moles of water from 16 °C to the evaporation temperature and transform it to water vapor. This duration should be considered maximum as the portion of water infiltration is ignored. Nevertheless, the above simple calculation suggests that the heat transfer to the surface from the magmatic source is largely sufficient to evaporate out the water and thus controlling the water accumulation in Awu's crater. This is coherent with the 95% water loss by evaporation from the 3.5x10 6 m 3 of the lake volume before to the 1992 eruption (Table 1). Similarly, in 2004 the lake water progressive dried out before the eruption. But if that the cooling trend in the crater continues, then ultimately the heat will no longer sufficient to dry out the incoming water from the rainfall. Water may then accumulate to form a crater lake as already observed in the past.

The Hazardous situation
The contact of liquid water with a hot surface is widely accepted as a process that can trigger explosive watermagma interactions (Wohletz, 1986;2002;Zimanowski et al., 1995;Thiéry and Mercury, 2009). However, according to a review of historical eruptions through volcanic lakes, ~2% involved relatively passive growth of subaqueous to emergent lava domes (Manville, 2015). This was the case on Kelud volcano (Java) in 2007 where a lava dome emerged in the middle of a huge crater lake without any vigorous explosion as suspected (Hidiyati et al., 2009). But only 7 years later, in 2014 that a VEI 4 eruption was witnessed on the volcano (Kristiansen et al., 2014). On Awu given the limited information on its past activities, it is difficult to provide a rigorous description of the successive lava dome emplacement. However, the coexistence of lava dome and crater lake in 1931 and 1979 were preceded by crater lake solely and no eruption in between (Fig.6). Such sequence suggests that Awu lava domes possibly occurred through the crater lake without explosive water-magma interactions. Further, the cooling and the relatively small and steady volume were the average rate of dome growth is approximately 10 4 m 3 day -1 with a mean volume of 5x10 7 m 3 and some with unstable slopes (Newhall and Melson, 1983). The hazardous situation on Awu is thus more related to the presence of the conduit plug and a crater lake since this latter may further increase the potential of violent eruptions (Sheridan and Wohletz 1983;Wohletz 1986). Similar to Kelud 2014 eruption, the vigorous VEI 4 eruption witnessed on Awu in 1966 occurred 35 years following the lava dome emplacement. Also, the 1992 eruption has pulverized a lava dome formed 13 years earlier. The current lava dome on Awu has developed 16 years ago, just after the 2004 eruption. T he prevalence of H2S over SO2, the low SO2 emission budget of 13 t/d and the low equilibrium temperature of circa 380 °C obtained by resolving together the SO2/H2S vs. H2/H2O redox equilibria (see methodology in Aiuppa et al., 2011;Moussallam et al., 2017) indicate that the current activity on Awu is sustained by a degassed magma. Hence, assuming that the processes that lead to vigorous eruption will repeat themselves then the subsurface volcanic system has to develop more than 3.1 MPa of pressure (Pressure (Pa) = F/Area (m 2 ); F (N) = masse (kg) x 9.8 (m/s 2 ); density of 2700 kg/m 3 ) to destabilized the 3.5x10 10 kg of lava dome. The total pressure than can be developed in the conduit or reservoir generally reaches tens of magapascals (e.g., Gudmundsson, 2012), largely sufficient to clear-up the conduit and pulverize the lava dome. Thus a new magma injection into the reservoir may be necessary to induce a rapid pressure buildup, capable to pulverize the lava dome and the conduit plug. This latter scenario may constitute an opening phase of volcanic eruption, creating a favorable condition for a rapid external lake water injection that could lead to magma-water interaction. Events of particularly high intensity described as phreato-Plinian eruptions can only be explained with the involvement of surface water in the eruption dynamics (Areva et al., 2018). A presence of water in Awu's crater may thus increase the potential of intense eruptive manifestations, a possible scenario behind the past vigorous eruptions. The most hazardous situation on the Awu volcano would be the new magma injection into the reservoir beneath a conduit plug and a sufficient water volume in the crater lake (Fig.7).

Conclusion
Awu is the northernmost active volcano of the Sangihe arc with 18 eruptions over the last 3.5 centuries, including 2 with VEI 4 and 3 with VEI 3 with local, regional and even worldwide impacts. Awu is also one of the deadliest volcano on earth with a cumulative dead toll of 11048 following its recurrent eruptive activity. Paradoxically, very little is known about this volcano. As emphasized in this work, the regular magma supply that sustained the activity of Awu is possibly linked to the peculiar geodynamic context of the region, controlled by the divergent double subduction and the subsequent arc-arc collision. Awu also has the particularity to host alternatively or simultaneously a lava dome and a crater lake. Lava domes at Awu seem to occur passively through crater lake and the heat that accompanied this emplacement has shown to be sufficient to dry out the water. The emplacement of the lava domes also appears to be associated with a conduit plug development that forces the degassing to the crater wall. With time the lava dome cools down, allowing progressively the crater lake formation with the high annual rainfall in the region. This scenario may ultimately constitute the most hazardous situation if a new magma injection occurs at deep. Indeed this latter may induce a rapid pressure buildup that can pulverize the conduit plug and the lava dome, creating a favorable condition for large water injection and a subsequent explosive water-magma interaction. Such a scenario likely resulted