The hazards of unconfined pyroclastic density currents : a new synthesis and 1 classification according to their deposits , dynamics , and thermal and impact 2 characteristics 3

26 Pyroclastic density currents (PDCs) that escape their confining channels are among the most 27 dangerous of volcanic hazards. These unconfined PDCs are capable of inundating inhabited areas that 28 may be unprepared for these hazards, resulting in significant loss of life and damage to infrastructure. 29 Despite their ability to cause serious impacts, unconfined PDCs have previously only been described 30 for a limited number of specific case studies. Here, we carry out a broader comparative study that 31 Non-peer reviewed preprint submitted to EarthArXiv 2 reviews the different types of unconfined PDCs, their deposits, dynamics and impacts, as well as the 32 relationships between each element. Unconfined PDCs exist within a range of concentration, velocity 33 and temperature: characteristics that are important in determining their impact. We define four end34 member unconfined PDCs: 1. fast overspill flows, 2. slow overspill flows, 3. high-energy surges, and 35 4. low-energy detached surges (LEDS), and review characteristics and incidents of each from 36 historical eruptions. These four end-members were all observed within the 2010 eruptive sequence 37 of Merapi, Indonesia. We use this well-studied eruption as a case study, in particular the villages of 38 Bakalan, 13 km south, and Bronggang 14 km south of the volcano, which were impacted by slow 39 overspill flows and LEDS, respectively. These two unconfined PDC types are the least described from 40 previous eruptions, but during the Merapi eruption the overspill flow resulted in building destruction 41 and the LEDS in significant loss of life. We discuss the dynamics and deposits of these unconfined 42 PDCs, and the resultant impacts. We then use the lessons learned from the 2010 Merapi eruption to 43 assess some of the impacts associated with the deadly 2018 Fuego, Guatemala eruption. Satellite 44 imagery and media images supplementing fieldwork were used to determine the presence of both 45 overspill flows and LEDS, which resulted in the loss of hundreds of lives and the destruction of 46 hundreds of buildings in inundated areas within 9 km of the summit. By cataloguing unconfined PDC 47 characteristics, dynamics and impacts, we aim to highlight the importance and value of accounting 48 for such phenomena in emergency management and planning at active volcanoes. 49


51
Pyroclastic density currents (PDCs) are the deadliest volcanic hazard, accounting for nearly a third 52 of all historical volcano-related fatalities (Brown et al. 2017). They are also some of the most complex 53 and unpredictable volcanic phenomena, which makes accurate forecasting of their occurrence, 54 characteristics and the area impacted difficult. In particular, the ability for PDCs to surmount 55 topography and travel outside of river valleys can place them in direct contact with communities on 56 the flanks of volcanoes. They can destroy whole towns and kill tens of thousands of people (e.g., St 57 Pierre, Martinique, 1902: Lacroix, 1904) but little is known about their internal dynamic processes, 58 and the ability to measure their dynamics in real time during eruptions does not yet exist. As a result, 59 they pose a significant challenge for emergency management planning at explosive volcanoes in 60 densely populated regions. and covering a widespread area (Rosi et al., 1993). Similar to blasts, the intensity of these PDCs wanes 277 on the margins, where they can display non-binary impacts, with survival of people and structures 278 indicating characteristics consistent with slow margins of dense PDCs as well as low-energy, dilute 279 surges (Rosi et al., 1993). For example, the 1902 column collapse from La Soufrière, St Vincent, 280 showed extensive damage in areas near source (Anderson andFlett 1903, Baxter 1990), but was not 281 capable of overturning trees or sturdy structures by the time it reached heavily inhabited areas, ~8 282 km from source (Baxter 1990). Despite this, there were over 1500 deaths as well as nearly 200 283 hospitalizations (with 80 subsequent deaths) largely from burns and the asphyxiating effects of the 284 ash (Will 1903, Baxter 1990. Similarly, in the 1902 Mount Pelée blast eruption (which had a death 285 toll of ca. 29,000), people in St Pierre, ~8 km from source, were severely burned by PDCs and fires 286 with many laying prone in the "pugilistic attitude" frequently associated with deaths due to 287 temperatures over 200 °C (Anderson and Flett 1903, Will 1903, Lacroix 1904, Baxter 1990. 288 Detached pyroclastic surges that decouple from and/or outrun their parent flows can also maintain 289 high dynamic pressure despite their low concentration. These surges are more capable of 290 overcoming topographic barriers than their parent dense PDC, as was the case in the deadly June 291 1991 Unzen eruption, in which dilute surges detached from their parent flows and outran them by 292 0.8 km, unexpectedly reaching an inhabited area where 43 people were killed (Nakada and Fujii 293 1993). The dynamic pressures of these surges were high enough (up to 8 kPa in some parts of the 294 surges; Clarke and Voight 2000) to destroy 50 houses, flatten trees, and move cars tens of metres 295 Fujii 1993, Cooper 2018). Similarly, some high-energy detached surges in the 1994 296 Merapi eruption maintained dynamic pressures that remained high enough to topple masonry walls, down trees, strip roof tiles, and destroy bamboo huts 5 km from source (Abdurachman et al. 2000), 298 which we estimate requires dynamic pressures of at least 2 kPa. 299 Deposits from these dilute, but high energy, surges are generally quite thin, but can reach greater 300 thicknesses in depressions and valleys. Following the Unzen eruption, surge deposits were typically 301 no more than 20 cm thick (Nakada and Fujii 1993) and were sometimes only a few centimetres thick, 302 in contrast to the up to 10 m thick deposits from the parent flows (Miyahara et al. 1992 Hills ranged from a few cm to 3 m outside of channels, while deposits were up to several metres thick 306 in river valleys (Belousov et al. 2007). High energy blasts often leave a distinctive two-layer deposit 307 (e.g., Soufrière Hills 1997, Merapi 2010) consisting of a basal, poorly-sorted, coarse layer that 308 typically includes ripped up clasts of the underlying surface, overlain by a much finer-grained, better 309 sorted deposit with some internal bedding (Brown and Andrews 2015; Komorowski et al., 2013). 310

Low-energy detached surge (LEDS) 311
LEDS represent the low-velocity, low-concentration end of the unconfined PDC spectrum. These these surges allows them to easily overcome topographic barriers. In recorded events, these surges 315 are most commonly seen moving laterally from their confined parent flows and escaping channels, 316 leading to unexpected inundation of inhabited areas. How, why or where along the flow path a surge 317 detaches is typically related to a change in the underlying syn-eruptive topography and/or the 318 pulsative nature of the eruption, which can act to reduce channel capacity or redirect the channel 319 away from the straight-line flow inertia, as described in the Introduction. 320 Due to their low velocity and concentration, and thus low dynamic pressure (typically <2 kPa), LEDS 321 damage to buildings, infrastructure or vegetation is typically minor (with the exception of secondary 322 damage through fire). For example, in the June 1997 Soufrière Hills eruption, LEDS were not capable 323 of blowing down trees or poles at distances greater than 2 km from source (Cole et al. 2002) and 324 damage to buildings was caused almost exclusively by temperatures up to around 400 °C (Baxter et 325 al. 2005). The impact for humans can range from minor through to fatal burns injuries, with the 326 chances of survival influenced by the LEDS temperature and duration as well as how much skin is 327 and Nakada 1999). In some events a "sear zone" or "singe zone" of charred vegetation was seen to 336 extend up to 25 m beyond the distal margins of surge deposits (Loughlin et al. 2002a). 337 The deposits of LEDS are characterized by their relative thinness and poor preservation in the long-338 term geologic record. In most historical cases, these surges have been recorded as thin as a few 339 centimetres and no thicker than 20 cm, even when associated with metres-thick, channel-confined Single eruptive events may contain both high-and low-energy detached surges, as seen in the 1991 350 Unzen eruption. In the June event, deadly high-energy surges killed 43 people and had dynamic 351 pressures large enough to sweep away cars and trees in one area (Cooper 2018), while low-energy 352 surges were capable of burning, but not bending or breaking, trees in other areas (Nakada and Fujii 353 1993). Similarly, in the September Unzen event, high-energy surges in some locations were powerful 354 enough to sweep away cars and trees damaged in the earlier eruption, while in another location low-355 energy surges caused damage only though heat, melting vinyl and charring building windows on the 356 volcano-facing side of the buildings (Fujii and Nakada 1999). Dynamic pressures in these events may 357 be up to 8 kPa in the high-energy surges, and lower than 2 kPa in the low-energy surges (Clark and 358   Field visits to the Merapi area three weeks after the 5 November 2010 event, and several times over 387 the years that followed, allowed some of the authors [SFJ, SJC, JCK, and PJB] to collect detailed field 388 data on confined and unconfined PDCs produced during the eruption. The geology, dynamics and 389 impacts associated with the directed blast that affected a large swathe of the upper flanks to ~8 km on the generation mechanisms, deposits, impacts and inferred dynamics associated with i) slow 392 overspill flows, and ii) low energy detached surges in villages along the Gendol river channel more 393 than 10 km to the south of the summit. Fast overspill flows were also observed along the Gendol, but 394 impacts were total, with all buildings, vegetations and victims buried with no observable remains. 395 Uniquely, our field studies combined geological and engineering expertise in collecting and 396 interpreting data on the deposits and physical impacts of the unconfined PDCs, which could be cross-397 referenced with medical data on the nature of burns injuries to victims. Some of the geology 398 previously; here we present data not included in these studies. We hope that these data and 401 interpretations are valuable for emergency management planning in providing the first multi-402 disciplinary case study of unconfined PDCs and their impacts. 403

Case study sites 404
We focus on two distinct type of unconfined PDC, for which we discuss the associated generation 405 mechanisms, dynamics, impacts and deposits at two villages ( Figure 2

Dynamics 478
The height of the unconfined PDC as it was emplaced remained above 10 m throughout Bakalan, as 479 evidenced by palms and trees that were singed to their full height (e.g., Figure 3a). Just to the west of  Jenkins 25 days after impact on 30 November 2010. 528

Impacts 529
Direct damage from the LEDS was minimal, with buildings remaining largely intact but interior and 530 exterior plastic melted, furniture charred and paper singed. Although the LEDS were not hot enough 531 to directly ignite these flammable objects, fire was the cause of total and partial destruction of 532 buildings in the village. Of the 48 buildings impacted by LEDS in Bronggang, seven timber buildings 533 were completely destroyed and a further five masonry buildings, with timber frame and tiled roofs, 534 partially destroyed, all by fire rather than direct damage from the LEDS (Figure 5a). Firebrands 535 (embers from burning logs within the parent PDC) carried within the surge ignited flammable 536 materials such as hay in animal sheds and sticks and coconut husks in outside lean-to wooden 537 kitchens (Figure 5b), with fires beginning in these flimsy wooden structures and then rapidly 538 spreading into the adjacent houses. The large ventilation gaps also allowed firebrands to travel with 539 the ash inside several houses, as evidenced by ignited mattresses or sofas, which smouldered without 540 causing the houses to catch fire ( Figure 5c). In one building, a small rupture in the gas tank of a 541 motorbike stored inside greatly increased the availability of fuel leading to a fire that completely 542 destroyed the building. 543 Prior to the eruption, ~400,000 people were rapidly evacuated to emergency shelters (Surono et al. (c) are marked by the corresponding letters in (a). 568

Deposits 569
We identified a complex stratigraphy at the base of the sabo wall, just inside the village of Bronggang, 570 consisting of four different main depositional units ( Figure 6): 571 • At the base of the sequence, there were patches of dry very fine grained, very well-sorted, loose 572 grey ashfall deposit 1 cm thick, which we interpreted as pre-5 November tephra erupted between 573 • The third unit was a dark grey, fine ash, massive, well-sorted and normally graded 4 cm thick unit 580 with chunks of charcoal and a locally erosive lower contact. We interpreted this unit as a second 581 LEDS deposit contemporaneous or correlated to one of the surge units seen in the main Gendol 582 channel. It was again overlain by a 1 cm thick, very fine grained, pinkish-tan, ashfall layer. 583 • The uppermost unit was a very poorly sorted, massive, compact, pinkish brown, normally to 584 symmetrically and even reversely graded, 35-45 cm thick fines-rich unit. This unit contained 585 large dense clasts up to 23 cm in diameter scattered on the top surface and also formed a central 586 coarser clast-rich zone with a more pinkish matrix. We interpreted this unit as resulting from a 587 minor overspill lobe of a valley-confined, block-rich PDC. Field evidence suggests that this PDC 588 was not very mobile and was stopped by the ~30 cm tall stone-wall curb of the main village road 589 on the Gendol side ( Figure 6). This unit was correlative and thickened to a 93 cm thick sequence 590 directly on top of the sabo wall. It was overlain by a 5-6 cm thick, very well-sorted, massive, fine 591 pinkish tan ashfall layer with a vesicular texture and perhaps some poorly preserved 592 accretionary lapilli. 593 We interpreted these deposits as representing three different minor PDC overspills, and associated 594 ash cloud fallout, from the Gendol main channel: the lower two represent deposition from the dilute 595  Earth imagery, acquired ~5 months after the eruption, showed that most roofing material (metal 722 sheets) in San Miguel los Lotes had been scavenged from PD and NVSD buildings in the intervening 723 months (Fig 8a). From satellite and media imagery of the two overspill locations, we were able to 724 identify at least two types of unconfined PDC, and their impact:   increased debris (e.g., branches, bricks) that they carried, ii) evidence of deposit surface 750 remobilisation, iii) that they were clearly wet in comparison with photos of pristine PDC deposits, and iv) the correlation with mechanical impact above the flow (lahars not imparting any mechanical 752 impact above their flow surface: Figure 9a). 753 Approximately two-thirds of the northwestern area of the village of San Miguel Los Lotes was almost 754 totally destroyed by overbank PDCs on 3 June (Figure 8a), which contained large boulders many 755 metres in diameter inside an ash matrix and left massive, poorly sorted deposits up to 2 or more 756 metres deep (Figure 9d and e). A small number of buildings at the northern end of the town escaped 757 total destruction, and buildings at the edge of the zone of total destruction towards the south and 758 east of the town mostly suffered partial damage, but their structures were still identifiable on satellite 759 imagery (Figure 8a

Inferred dynamics 784
The maximum height of the PDCs was not easy to determine from media images. There were a 785 number of tall (>10 m) trees in both affected areas that appear singed by the PDC, implying a current 786 height of more than 10 m as the PDC entered the town and golf resort. Media videos of the PDC 787 flowing past a bridge in the adjacent channel just to the south of San Miguel los Lotes suggest that 788 current heights maintained similar heights in the channel. However, trees in the southern half of San 789 Miguel Los Lotes and towards the peripheries of impacted areas at La Reunión golf resort were not 790 singed to their full height, with canopies still showing as green and unaffected in satellite imagery 791 while the full height of buildings was affected. Thus, the total PDC height decreased from >10 m to 792 between 2 to 10 m as the flow slowed down and came to a stop. 793 At the golf resort, the energy of the overspill flows was low enough that they could be largely blocked 794 by buildings, and where the flows entered buildings, they only moved objects such as chairs a few 795 tens of centimetres away with the associated surges leaving countertop items such as bottles coated 796 in ash but upright (Figure 9a). This suggests that both the dense and dilute PDC components were 797 traveling at ~1 m/s (certainly less then 5 m/s based on the categories outlined in Figure 1 Lotes, leading to partial or total roof collapse in places. It is not clear from remotely derived imagery 817 if isolated fires were the result of embers carried within the PDC, as at Merapi, or related to the heat 818 of the deposits. Flow deposits at Fuego contained large boulders, but field studies showed that most 819 of these boulders were not juvenile, and therefore unlikely to have provided a concentrated heat 820 source that may also have triggered fire. burns that have penetrated below the skin layer to involve the limb muscles (at least 200 °C) (Baxter, 836 1990). Not all casualties displayed this attitude, despite wearing similar clothing (t-shirt and 837 trousers) and being affected by LEDS in the same town. Since there was no observed evidence of fires 838 near these casualties, the temperatures necessary to cause these injuries can be attributed to the 839 LEDS. The evidence from thermal effects therefore suggests that there was variability in temperature 840 and/or duration of the impact across the impacted area, reflecting the uneven inundation of PDCs 841 across the village area. 842 While remotely assessing impacts visible in satellite, aerial and media images in this case study was 843 valuable, access to photos and on-the-ground experience in the aftermath of the Fuego eruption 844 provided vital detail that allowed us to confirm or refute inferences made from media images and 845 added information not visible in remote imagery. Ideally, remote and field approaches are combined, 846 with remote approaches providing information on the immediate post-impact situation and for areas that cannot be easily accessed as well as providing the larger scale overview, which can then be 848 refined and ground-truthed with informed field visits that provide more detailed information, 849 background and context. Studies relating impacts and PDC dynamics with the deposits provide an 850 evidence base from which likely PDC dynamics and impacts can be forecast. This is particularly  The range of observed characteristics across different eruptions but within unconfined PDCs of the 902 same type (Table 1) can be related to a few potential factors. The size of the magma batch and the 903 volume of erupted material may affect PDC temperatures at their generation, leading ultimately to 904 differences in source temperatures between eruptions, before PDCs become unconfined. The PDC 905 generation mechanism appears to affect the temperature of ensuing PDCs: collapsing lava domes 906 (e.g., Soufrière Hills 1997, Merapi 1994Merapi , 2006Merapi , 2010 are correlated with higher initial PDC 907 temperatures than a sector collapse (e.g., Fuego 2018), and this appears to play a stronger role than 908 distance from the volcano. For example, PDCs at the affected sites near Fuego appear to have been 909 either similar or lower temperature despite being closer to the volcano (~8 km) than the sites at 910 Merapi (~13 km). 911 The velocity (and therefore the dynamic pressure) of PDCs seems to be strongly influenced by the itself. In all recorded LEDS, localised fires were ignited and at Merapi, this could be attributed to 951 embers (firebrands) carried within the LEDS (Jenkins et al., 2013). It is reasonable to infer that in 952 situations with fewer firebrands present (e.g., fewer trees consumed in the surge path), the likelihood 953 of building damage from LEDS may be lower. Building typology is also a factor affecting the level of 954 damage sustained during LEDS, with timber buildings much more likely to be damaged in a LEDS-955 caused fire than masonry buildings. Considering these factors, fire damage resulting from LEDS is By their nature, unconfined PDCs are difficult to forecast because they inundate areas beyond the 963 topographic lows that are typically given priority in volcanic hazard planning. As numerical models they may be better able to recreate, and therefore forecast, the path and dynamics of unconfined 966 PDCs. In the meantime, one approach in mitigation planning has been to apply a 'buffer' (e.g., Neri et 967 al., 2015) around a PDC-prone channel to highlight threatened populations and infrastructure, with 968 the aim of implementing long-term land-use or short-term proactive evacuation measures for 969 communities close to topographic lows. The extent of this buffer is difficult to define, and is a function 970 of the channel topography, PDC volume and local PDC mass flux/velocity as well as preceding events 971 in the eruption sequence (e.g., the infilling by previous PDC deposits). For directed blasts, a buffer is 972 clearly not appropriate because of their wide-reaching and topography-mantling nature, in these 973 cases an energy cone model that defines distance from the summit may be useful. For those high-974 energy surges that are not unconfined from origin, e.g., Unzen 1991, this type of model is less useful 975 as it is unable to identify locations of surge detachment. For overspill PDCs, we found they reach a 976 maximum lateral distance of 800 m (Table 1) from the flowpath channel. However, we recognise that 977 buffer extents are likely to be unique to the specific eruptions and require consideration of the 978 topography, channel path and likely eruptive style. Reliance on geological deposits for defining 979 buffers and potentially hazardous areas must be cognisant of the thinner deposits that reflect 980 unconfined PDCs that cannot be preserved but are still deadly. 981 Volcanic hazard and risk assessment relies upon empirical data from past eruptions and their 982 impacts. However, we are often limited in the amount of data that can be collected shortly after an 983 event, while deposits and impacts are preserved, because of safety and access limitations. In this 984 study, we have used lessons learned from remote and ground surveys of PDC dynamics following the 985 Merapi 2010 eruption to provide a similar assessment for Fuego 2018. Remote assessment at Fuego 986 using satellite imagery and media images to supplement a field study allowed for many similar 987 determinations of PDC dynamics and resultant impacts as at Merapi. In both cases, through imagery Bull Volcanol 33, 600-620. https://doi.org/10.1007/BF02596528