Anda di halaman 1dari 11

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/318106134

Buildings vs. ballistics: Quantifying the


vulnerability of buildings to volcanic ballistic
impacts using field studies and p....

Article in Journal of Volcanology and Geothermal Research · July 2017


DOI: 10.1016/j.jvolgeores.2017.06.026

CITATION READS

1 94

6 authors, including:

George Williams Ben Kennedy


Nanyang Technological University University of Canterbury
2 PUBLICATIONS 2 CITATIONS 137 PUBLICATIONS 1,122 CITATIONS

SEE PROFILE SEE PROFILE

Tom Wilson Rebecca Fitzgerald


University of Canterbury University of Canterbury
133 PUBLICATIONS 1,012 CITATIONS 10 PUBLICATIONS 70 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Volcanic eruption clean-up View project

Using Games-Based Learning Strategies to Teach Field Observation Skills View project

All content following this page was uploaded by George Williams on 23 July 2017.

The user has requested enhancement of the downloaded file.


VOLGEO-06147; No of Pages 10
Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Journal of Volcanology and Geothermal Research

journal homepage: www.elsevier.com/locate/jvolgeores

Buildings vs. ballistics: Quantifying the vulnerability of buildings to


volcanic ballistic impacts using field studies and pneumatic
cannon experiments
G.T. Williams a,⁎, B.M. Kennedy a, T.M. Wilson a, R.H. Fitzgerald a, K. Tsunematsu b, A. Teissier a
a
Department of Geological Sciences, University of Canterbury, Private Bag 4800, Christchurch, New Zealand
b
Mount Fuji Research Institute, Yamanashi Prefectural Government, 5597-1 Kenmarubi Kamiyoshida, Fujiyoshida-shi, Yamanashi 403-0005, Japan

a r t i c l e i n f o a b s t r a c t

Article history: Recent casualties in volcanic eruptions due to trauma from blocks and bombs necessitate more rigorous, ballistic
Received 16 March 2017 specific risk assessment. Quantitative assessments are limited by a lack of experimental and field data on the vul-
Received in revised form 22 June 2017 nerability of buildings to ballistic hazards. An improved, quantitative understanding of building vulnerability to
Accepted 30 June 2017
ballistic impacts is required for informing appropriate life safety actions and other risk reduction strategies. We
Available online xxxx
assessed ballistic impacts to buildings from eruptions at Usu Volcano and Mt. Ontake in Japan and compiled
Keywords:
available impact data from eruptions elsewhere to identify common damage patterns from ballistic impacts to
Volcano buildings. We additionally completed a series of cannon experiments which simulate ballistic block impacts to
Volcanic hazards building claddings to investigate their performance over a range of ballistic projectile velocities, masses and
Risk energies. Our experiments provide new insights by quantifying (1) the hazard associated with post-impact
Structures shrapnel from building and rock fragments; (2) the effect of impact obliquity on damage; and (3) the additional
Fragility functions impact resistance buildings possess when claddings are struck in areas directly supported by framing compo-
Experimental volcanology nents. This was not well identified in previous work which may have underestimated building vulnerability to
ballistic hazards. To improve assessment of building vulnerability to ballistics, we use our experimental and
field data to develop quantitative vulnerability models known as fragility functions. Our fragility functions and
field studies show that although unreinforced buildings are highly vulnerable to large ballistics (N20 cm diame-
ter), they can still provide shelter, preventing death during eruptions.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction Ballistics are fragments of lava or rock which can be ejected during
any style of explosive eruption (Sparks et al., 1997). Fragments range
Active volcanoes are receiving increasing interest from tourism and from a few centimeters to tens of metres in diameter and are large
other societal activities which is leading to increasing numbers of peo- enough to become separated from the eruptive column and travel
ple and other assets exposed to near-vent hazards, such as volcanic bal- along near parabolic trajectories (Alatorre-Ibargüengoitia and
listics. This strong societal desire to access vent zones of active Delgado-Granados, 2006). Ballistics can be ejected at up to hundreds
volcanoes conflicts with the typical ballistic risk management approach of meters per second and can land over 10 km from the vent although
of using exclusion zones to restrict access. Thus, volcanic ballistic haz- typically they land within 5 km (Blong, 1984). The high kinetic and at
ards have emerged as a major contemporary risk management chal- times, thermal energies that ballistics have when they land makes
lenge for volcanology, particularly from small eruptions with little or them hazardous to people, buildings, infrastructure and other societal
no forecastable warning such as the 1993 Galeras eruption, Columbia assets (Booth, 1979; Wardman et al., 2012; Oikawa et al., 2016).
(Baxter and Gresham, 1997), the 2012 Upper Te Maari eruptions, New When they are ejected from the vent, fragments of lava (ballistic
Zealand (Fitzgerald et al., 2014) and the 2014 Ontake eruption, Japan bombs) can be over 1100 °C. Although they cool during flight, upon
(Yamaoka et al., 2016). These types of eruptions require revaluation of landing bombs will often retain sufficient thermal energy to burn cer-
how to assess and manage ballistic risk, in particular balancing the de- tain building materials, such as dry timber which has an ignition tem-
sire to retain access to volcanoes while ideally avoiding casualties and perature of only 250 °C (Vanderkluysen et al., 2012). Similarly for
other damage. kinetic energy, although ballistics may be ejected at over 300 m/s,
they slow during their flight path, with terminal velocities of ballistics
⁎ Corresponding author. typically b 150 m/s (Walker et al., 1971). In turn impact energy (kinetic
E-mail address: gtw24@uclive.ac.nz (G.T. Williams). energy at the moment of impact) is strongly controlled by the size of a

http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
0377-0273/© 2017 Elsevier B.V. All rights reserved.

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
2 G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx

ballistic as this limits both its terminal velocity and mass. For example,
Alatorre-Ibargüengoitia et al. (2012) modelled impact energies for
relatively large ballistics, 0.2–0.6 m in diameter ejected during VEI
2–3 eruptions to be as high as 10 6 J, far exceeding the 4000–
12,000 J energy threshold required to penetrate reinforced concrete
slabs (Jenkins et al., 2014a). For smaller ballistics, impact energies can
be much lower with 0.1 m diameter blocks typically restricted to
b104 J (Walker et al., 1971).
Ballistic hazard assessment is developing rapidly and is a consider-
able focus of scientific inquiry (e.g. Fitzgerald et al., 2014; Tsunematsu
et al., 2016; Taddeucci et al., 2017). However, effective volcanic risk as-
sessment requires characterising: the hazard (e.g. volcanic ballistics),
what societal elements (e.g. people, buildings, etc.) are exposed to the
hazard, and how vulnerable the societal elements are to the hazard. Cur-
rently for ballistic risk comprehensive vulnerability assessment is lack-
ing (Biass et al., 2016; Tsunematsu et al., 2016), although some
promising early work has been completed by Blong (1981, 1984),
Pomonis et al. (1999), Jenkins et al. (2014a) and Biass et al. (2016). In
the case of ballistics impacting buildings, stronger buildings are clearly Fig. 1. Snapshot from reinforced concrete test with pixel change detector software
more resilient to ballistics hazards – both in terms of survivability for oc- highlighting block trajectory pre-impact and rock fragment trajectories post impact.
Velocity is calculated by counting the number of frames taken while an object travels a
cupants and cost of repair (Blong, 1981). Despite many instances of bal-
known distance. In this test the block took 12 frames to travel 1.6 m from its loaded
listic impacts to buildings being reported from historic eruptions (e.g. position to impact the concrete. The fragment trajectory highlighted by the arrow
Booth, 1979; Blong, 1984; Wardman et al., 2012), there is little empirical beneath it took 79 frames to travel 4.2 m.
evidence to inform quantitative risk assessment models to calculate life
safety estimates; recommend what building materials or designs are
most resilient to volcanic ballistics; or inform what the best advice is then added data from five timber-framed buildings with sheet metal
for people to survive when exposed to ballistic hazards (Jenkins et al., roofs that were impacted by ballistics during the 2000 eruption of Mt.
2014a; Fitzgerald et al., 2016). Yet this is critical information to inform Usu, and continued to populate the graphs with data from other erup-
risk management strategies. The first quantitative examination of im- tions documented in the literature (Fig. 2). We determined impact ener-
pact energy thresholds for ballistic blocks perforating building claddings gy by first calculating the mass of an impacting ballistic then calculating
was carried out by Blong (1981) although this study relied on data from its impact velocity using media footage of the eruption (Television
cyclone vulnerability studies. Further insights into ballistic impacts to Hokkaido Broadcasting Co. Ltd., 2000) coupled with trajectory model-
buildings have been made by Pomonis et al. (1999) and the first ballistic ling using Eject (Mastin, 2001).
fragility functions based on this data were presented in Biass et al. We conducted 112 experiments between September 2015 and Octo-
(2016). ber 2016 with a pneumatic cannon at the University of Canterbury,
Quantitative field measurements of ballistic impacts to buildings are using andesitic - basaltic blocks weighing 3–10 kg with densities from
rare because of the risk associated with close observation and because 1775 to 3030 kg/m3 as the projectiles, impacting at up to 79 m/s. The
taking measurements of ballistic hazard intensity such as ballistic im- use of these projectile weights, densities and impact velocities as a sim-
pact energy or number of impacts per unit area is highly challenging ulation of ballistic blocks is within the range of those of measured and
due to poor forecastability. To address these difficulties, we compiled modelled values for ballistics ejected during the 2014 Ontake eruption
available eruption data from literature, surveyed eight buildings im- (Tsunematsu et al., 2016) and the August 2012 Te Maari, Tongariro
pacted by ballistics during the 2000 eruption of Mt. Usu, Japan and com- eruption (Breard et al., 2014; Fitzgerald et al., 2014). Projectiles were
bined this data with full scale ballistic laboratory experiments in an shot at sections of roof and wall claddings fitted to timber framing ac-
approach similar to that adopted for other volcanic hazards (e.g., cording to New Zealand building code (NZS 3604: Methods and details
Wilson et al., 2012). for the design and construction of timber-framed buildings up to three
Experiments allow us to analyse how damage occurs and provide stories high). We used pixel change detecting software on 240–
additional data to verify our field measurements. We simulated ballistic 1000 frames per second video to analyse the velocity of projectiles
impacts to building materials using a pneumatic cannon capable of and shrapnel (Fig. 1). We conducted 14–40 experiments on four types
launching rocks at up to 79 m/s (Fig. 1). To our knowledge these are of building materials including sheet metal, 2 × 4 inch timber planks
the first published experiments designed to investigate impacts to (timber 2 × 4's), timber weatherboard and reinforced concrete (de-
buildings using realistic size and weight volcanic rocks as the projectiles. scriptions of the materials, including material thicknesses, can be
These data inform quantitative vulnerability models (known as fragility found in Table 1). To find thresholds for various damage severities we
functions) which can improve ballistic risk assessments by defining the used impact energies ranging from 90 to 7200 J. These impact energies
crucial relationship between hazard intensity and likely damage and the corresponding damage assessment provided data for the deri-
severity. vation of fragility functions using the methodology outlined in Wilson
et al. (in press). For each material tested, we derived three fragility func-
2. Method tions using a three tiered damage state scale. A damage state (DS) is a
description of a specific, measurable damage severity with higher dam-
Best practice development of fragility functions uses empirical data, age states corresponding to higher hazard intensities for a given materi-
such as from post-eruption impact surveys or laboratory experiments al. Damage states are numbered 1–3 (e.g. DS1–DS3) and their
(Rossetto et al., 2014). All the fragility functions we present here quan- descriptions are given in Table 2.
tify the relationship between ballistic impact energy (kinetic energy at Fragility functions take the shape of a cumulative density func-
the moment of impact) and the likelihood of a specific damage severity tion of a Normal distribution (Φ) and are expressed as a function of
being reached or exceeded. As a starting point, functions were produced the mean kinetic energy (Emean) and the coefficient of variation (σ).
by fitting curves to Blong's (1981) impact energy thresholds for the per- Following Jenkins et al. (2014a,b) and Biass et al. (2016), the probability
foration of different cladding types by perpendicular block impacts. We to equal or exceed a given damage state (Pexceedance) is expressed as

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx 3

Fig. 2. Fragility functions derived using data from Blong's (1981) study, previous eruptions and field work at Usu Volcano. The perforation thresholds from Blong's (1981) study are plotted
at the top of the graph. The data points from published literature on previous eruptions correspond to ballistic block impacts to metal roofs (e.g. La Soufriere (Blong, 1984) and Pacaya
(Wardman et al., 2012)) and timber roofs (e.g. Ontake (Tsunematsu et al., 2016)). RC = reinforced concrete.

a function of the ballistic impact energy I (J) with the following component of impact energy using Eq. (2) gave relatively accurate re-
relationship: sults when predicting DS2 damage from oblique impacts to reinforced
concrete slabs. However, we note that generally the normalization
P ðExceedanceǀI Þ ¼ ϕ ð ln ðIÞ; ln ðEmean Þ; σ Þ ð1Þ imparted by Eq. (2) tended to underestimate a cladding's impact resis-
tance. Additionally, oblique impacts increased the likelihood of blocks
To simulate the range of impact types observed in the field we sys- ricocheting off claddings, making them less susceptible to ballistic per-
tematically varied the impact obliquity (i.e. the angle between the tra- foration but increasing the distance that ballistics and building shrapnel
jectory of an impacting ballistic and a line perpendicular to the face of would travel post-impact (Fig. 3). When safety barriers were not used,
the cladding it impacts). Eq. (2) uses impact obliquity (Ө) to resolve blocks impacting oblique to concrete slabs ricocheted then bounced
the normal component of velocity (Vn), which was then used in place and rolled up to 15 m from the initial impact site. Similarly, oblique im-
of the true velocity (Vt), during impact energy calculations. This allowed pacts to weatherboard panels resulted in block ricochet followed by
us to compare damage from oblique and perpendicular impacts and also sharp fragments of timber, some over 40 cm in length, being shattered
to use our full suite of data to constrain fragility functions. For informa- and ejected laterally up to six metres during experiments.
tion on the obliquity range and number of oblique impact experiments In general, our experiments show that unreinforced building clad-
conducted refer to Table 1 in the accompanying electronic supplement. dings have a low impact resistance relative to ballistic impact energies
which can be orders of magnitude higher. However, even when weaker
V n ¼ V t  Cosθ ð2Þ cladding and framing materials are perforated relatively easily by ballis-
tics, they still have the ability to slow ballistics down and reduce their
ability to perforate successive layers of material. During both sheet
3. Results metal and timber 2 × 4 experiments blocks slowed down significantly
after perforation, causing kinetic energies to drop by 440–1000 J and
Damage caused by impacts in the cannon experiments shared many 1300–1500 J post perforation for sheet metal and timber 2 × 4's respec-
similarities with the damage observed during building surveys (Fig. 4). tively (Fig. 7).
Consistent with previous research by Blong (1981), higher impact ener- Also when impacts produced concrete or weatherboard shrapnel,
gies (x-axis of fragility functions) correlated with a higher probability of the velocity of the shrapnel was always lower than that of the projectile
perforation (y-axis of fragility functions) for all materials (Fig. 5). Rein- with the fastest concrete and weatherboard fragments moving at
forced concrete had the highest impact resistance of all the materials 23 m/s and 11 m/s respectively from ballistics impacting at 36 m/s. Frag-
tested with the minimum impact energy resulting in perforation being ments of reinforced concrete were ejected from the backface of impact-
2164 J compared to 288, 514 and 1452 J for timber weatherboard, ed panels in a pattern consistent with impacts from blunt nosed
sheet metal and timber 2 × 4's respectively. We established the mini- projectiles (Chen et al., 2004) and the velocity of concrete shrapnel
mum impact energy threshold for perforation using experiments was calculated to be 17–37% that of the ballistics (Fig. 8). The largest
where impacts were perpendicular to the cladding surface. Increasing concrete fragments from each test weighed 0.3–1.9 kg with the largest
the impact obliquity away from perpendicular decreased the probability of these having a 140 J impact energy, slightly over the 100 J threshold
of achieving higher damage states for a given energy and in the case of for lethal skull fracture suggested by Raymond et al. (2009). However,
reinforced concrete, oblique impacts were unable to reach DS3 even the vast majority of fragments were well below the lethal threshold. De-
when impact energies were double the threshold for perforation during spite impacts being able to create complete perforations through the RC
perpendicular tests (Fig. 6). Using a 3000 J energy impact to reinforced slabs, the steel reinforcing bars prevented the blocks themselves from
concrete as an example, a perpendicular strike is 64% more likely to re- passing through (Fig. 4D; Fig. 4F; Fig. 9); this also occurred at Usu
sult in DS2 than an impact with a 50° obliquity. Resolving the normal where a double layer of reinforcing bars in a 20 cm thick slab of concrete

Table 1
Descriptions of building materials used in experiments.

Building material Description Thickness (mm) Number of experiments

Sheet metal Grade G300, zinc coated, corrugated steel 0.5 36


Timber weatherboard Pre 1980's bevel-back, Rimu 22 24
Timber 2 × 4 plank H4 treated pine, 90 mm by 45 mm thick 0.9 m long plank 90 14
Reinforced concrete 1 m2 slabs with 35 mPa compressive strength, single layer of 10 mm diameter 75 40
reinforcing bars spaced 20 cm apart in two directions

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
4 G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx

Table 2 Table 4
Damage state descriptions for ballistic impacts to sheet metal, timber framing, timber Comparison of materials from Blong's (1981) study with those used in our experiments.
weatherboards and reinforced concrete slabs. DS0 is required to calculate the probability
of DS1 occurring for a given impact energy. Building materials Thickness (mm) Range of impact energies
required for perforation (J)
Damage state Damage description
Dp: timber board 4.5–12 60–500
Sheet metal Timber weatherboard 22 288–508
DS0 No damage – cosmetic scratches Ep: steel sheet 0.4–0.7 400–1000
DS1 Cosmetic denting Metal sheet 0.5 514–700
DS2 Tearing (at impact site or around nail holes) Fp: reinforced concrete slab 50–125 4000–12,000
DS3 Ballistic perforation Reinforced concrete slab 75 2164–3547

Timber 2 × 4 framing (clad in sheet metal)


DS0 No visible damage to framing
DS1 Framing dented building materials. Therefore, the fragility functions are best applied to
DS2 Partially – completely split or nail connections lost the specific type of cladding materials used in this testing. Despite
DS3 Ballistic perforation these limitations our fragility functions adequately predict the damage
Reinforced concrete slabs caused to sheet metal roofs impacted by ballistics from previous erup-
DS0 No damage - cosmetic scratches tions at Pacaya, La Soufriere and Usu (Fig. 2).
DS1 Front-face spalling and/or back-face cracking
DS2 Ejection of concrete fragments from the back-face (scabbed) 3.1. Results summary
DS3 Perforation of concrete

Timber weatherboards This study has quantified the vulnerability of four commonly used
DS0 No damage – denting building materials to ballistic block impacts through the development
DS1 Boards cracked but still in contact
of fragility functions (Fig. 5). The following points can be made to sum-
DS2 Boards cracked apart
DS3 Ballistic perforation marise the results of pneumatic cannon experiments.

• We conducted 112 experiments with volcanic blocks weighing 3–


was found to have halted a 60 by 40 cm block while all the surrounding 10 kg travelling at up to 79 m/s. These masses and velocities, although
concrete had been shattered away following a very high energy impact sufficient to determine the perforation thresholds for all the materials
(Fig. 4A). tested, represent the lower end of possible ballistic hazard intensities.
Our experimental data constrain our fragility functions well, but it is • Individual layers of unreinforced building materials are vulnerable to
important to acknowledge that due to a relatively low number of tests perforation by all but the lowest energy ballistic impacts (b2000 J) but
being performed the accuracy of functions could be improved with fur- these materials do have the capacity to substantially lower the veloc-
ther testing. All of the fragility functions from these experiments were ity and kinetic energy of blocks reducing their ability to perforate suc-
produced from 14 to 40 tests. Rossetto et al. (2014) suggest function de- cessive layers of material present in most buildings (Fig. 7).
velopment requires a minimum of 30 buildings and recommends over • Ballistics require higher impact energies to perforate claddings in
100. Two important sources of variability are the block's shape and me- areas directly supported by framing components. Ballistic impacts to
chanical strength. Shape controls the contact area between block and sheet metal supported by timber 2 × 4 framing required over twice
cladding upon impact. For example, the block used for all the sheet as much impact energy to cause perforation compared to impacts to
metal tests was a ~20 × 10 cm columnar joint and its four lowest impact unsupported sheet metal (Fig. 5).
energies resulting in perforation (500–700 J) all occurred when the • Impact obliquity (i.e. the angle between the trajectory of a ballistic
block impacted end on (Fig. 10). Side on impacts distributed stress and the face of a cladding) strongly influences whether an impact
over a larger surface area and were unable to cause perforations until will result in ricochet or perforation. Ricochet always occurred for im-
energies N 850 J were reached. A block's mechanical strength influenced pact obliquities over 65° and in the case of reinforced concrete, rico-
its damage causing ability especially during reinforced concrete experi- chet without perforation occurred for all impact obliquities over 45°
ments where the lowest density, highly vesiculated blocks (with densi- (Fig. 3).
ties of 1775–2213 kg/m3) were only able to cause DS2 damage even at • Block ricochet, rock shrapnel and building shrapnel increase the area
high impact energies (up to 3689 J). The field data, contains additional over which ballistic hazards occur. The velocity of concrete shrapnel
sources of uncertainty from impact energy estimations associated is broadly linked to the impact velocity of a ballistic (shrapnel velocity
with modelling (e.g. a range of possible ejection angles influencing the is on average, 25% of ballistic impact velocity) but most of the frag-
maximum height of a ballistic's trajectory) and from the variability of ments were calculated to be non-lethal at the speeds produced during
these experiments (Fig. 8).

Table 3 4. Discussion
Median and standard deviation values for fragility functions.
These cannon experiments and new field data advance our knowl-
Fragility function suite Damage Median Coefficient of variation/standard
state (Emean) deviation (σ) edge of ballistic impacts to buildings by validating and refining Blong's
(1981) study and by investigating damage from oblique impacts and
Timber weatherboard 3 360 0.35
2 100 0.2
the hazard posed by shrapnel. The new data forms the basis for basic
1 50 0.2 building design recommendations and advice on where to shelter in
Sheet metal 3 650 0.25 buildings during ballistic fallout (presented below). However, we note
2 300 0.25 that these recommendations are applicable mostly for relatively low
1 150 0.2
explosivity, unheralded eruptions, which have relatively modest ballis-
Timber 2 × 4 inch plank 3 1450 0.25
2 1000 0.2 tic hazard footprints and impact intensities. This comes with the inher-
1 650 0.2 ent assumption that for larger explosive eruptions there should be
Reinforced concrete 3 2750 0.3 sufficient warning and so evacuations or exclusions should be in place.
2 1750 0.3 We also note a major limitation of the fragility functions produced in
1 1300 0.3
this study is that they only provide analysis of damage to individual

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx 5

Fig. 3. Experimental data illustrating the relationship between impact obliquity, impact energy and whether or not an impact resulted in ricochet. No perforations were observed at
obliquities over 65° regardless of the material used. Reinforced concrete slabs could not be perforated by oblique impacts at the energies measured during experiments.

components of a building caused by individual ballistic impacts, rather 4.1. Comparison of laboratory data with previous work
than considering how multiple ballistic impacts may impact a building
system. The specific damage descriptions presented here are useful A direct comparison of the fragility functions produced using Blong's
when considering the safety of building occupants, but this information (1981) penetration thresholds and our fragility functions produced
is difficult to compare with the broad damage descriptions to entire using similar building materials has been made in Fig. 11. This shows
buildings which are required to estimate building repair or replacement that both functions for metal sheets match closely, whereas timber
costs. weatherboards in our testing have a ballistic impact resistance nearly

Fig. 4. Ballistic impacts observed during building surveys at Usu Volcano (A–C) compared with impacts from the pneumatic cannon experiments (D–F). A and D illustrate damage state
three, reinforced concrete impacts where the slab has been perforated, shrapnel was ejected from the backface but blocks have not fully penetrated the cladding. B and E show damage
state three perforation of fibre-reinforced boards and timber weatherboards respectively. C illustrates a ballistic impact to sheet metal where the damage state could be either damage state
two or damage state three based on the presence of a block suspended within the damaged section of the roof. F illustrates damage state three to sheet metal.

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
6 G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx

Fig. 7. Experimental data on kinetic energy reduction of blocks after perforation during
sheet metal and timber 2 × 4 tests.

studies measuring the energy required for bullets and metal pipes to
fully perforate and pass through slabs, whereas our experiments used
rocks as the projectiles and measured the energy required for a perfora-
tion to form in the slab, regardless of whether or not the rock was able to
pass between the spaces of reinforcement bars. This comparison high-
lights the need to conduct ballistic-building vulnerability assessments
Fig. 5. Ballistic fragility functions for different cladding types. A: timber weatherboard, B: using realistic volcanic rock projectiles rather than bullets or pipes and
sheet metal, C: timber 2 × 4 inch planks, D: reinforced concrete. Horizontal bars span
also that the impact energy thresholds for perforation can vary for dif-
the range of impact energy values for a specific damage state within each impact energy
bin following Wilson et al. (in press). The points indicate median bin values and points ferent thicknesses and arrangements of the same basic building materi-
with exceedance probabilities of 0 or 1 have no horizontal bars. A lone tick mark with al. However, we gratefully acknowledge the usefulness of Blong's
no horizontal bar indicates there was only one impact energy where that particular (1981) study which provides essential foundational knowledge and a
damage state was reached within that impact energy bin range. The medians and point of comparison in the absence of other knowledge.
standard deviations which can be used to reproduce these functions are presented in
Table 3.
Our findings suggest previous vulnerability studies, which only con-
sider perpendicular impacts to areas unsupported by framing, overesti-
mate cladding vulnerability while underestimating overall ballistic risk
double that of the timber boards used in Blong's, 1981 study and con- by not accounting for the secondary hazard associated with shrapnel
versely the reinforced concrete slabs used in our testing had perforation and ballistic ricochet. With the exception of three reinforced concrete
thresholds under half that of the same materials from Blong's (1981) tests, the impact energy of the shrapnel produced in these experiments
study. We expected our timber weatherboards to have a higher impact was below the threshold for lethality. However, considering shrapnel
resistance due to their effective thickness being doubled in areas where velocity and ballistic impact velocity were linked and ballistics can at-
each weatherboard is fixed with a partial overlap to the weatherboard tain impact velocities over twice as high as those reached in these
below it and because the individual weatherboards tested in our exper- tests, building shrapnel must be accounted for when modelling the po-
iments were almost twice as thick as those tested in Blong's (1981) tentially lethal area associated with ballistic impacts (e.g. Fitzgerald et
study (Table 4). Comparing the reinforced concrete between the two al., 2014).
studies shows our slabs are more prone to perforation than those
from Blong's (1981) study. This is, at least in part, due to the RC slabs 4.2. Implications for sheltering in buildings
in Blong's (1981) study being considerably thicker than those used in
our experiments (125 mm compared to 75 mm), highlighting this as Although claddings such as timber and sheet metal have relatively
an important factor for impact assessment. Also different projectiles low energy perforation thresholds, our experiments support the use of
and different definitions of perforation were used compared to our ex- any cladding for shelter in a volcanic eruption as they will afford at
periments. The thresholds given in Blong (1981) are derived from least some degree of protection. While this is a relatively obvious

Fig. 6. Comparison of impact resistance between perpendicular and oblique impacts to reinforced concrete slabs. Functions derived using the normal component of velocity from oblique
impacts are also shown. Note that damage state three is not plotted as it was never observed during oblique shots.

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx 7

Fig. 10. Experimental data from sheet metal experiments highlighting the influence of
Fig. 8. Shrapnel velocity compared to ballistic impact velocity in reinforced concrete block orientation on damage state in the ~500–700 J impact energy range. Higher long
experiments. For a given impact velocity, damage state three impacts generally produce axis inclinations are associated with force being exerted onto the metal over a smaller
higher shrapnel velocities than damage state two impacts. Shrapnel velocity increases surface area (higher stress).
with increased ballistic impact velocity but low R2 values for trend lines indicate that
this is not the only factor controlling shrapnel velocity.
energies N 38,000 J were found lodged in hut roofs without fully pene-
suggestion, we note the protection timber roofed mountain huts pro- trating the building envelope (Tsunematsu et al., 2016). These impact
vided to exposed people during the 2014 eruption of Mt. Ontake in energies are 1–2 orders of magnitude higher than the critical perfora-
Japan (Oikawa et al., 2016). The impacts of this eruption and the associ- tion threshold for timber cladding suggested by Blong (1981) and our
ated reports of hikers and journalists provide a crucial case-study to test data. We suggest that the performance of the mountain hut roofs at
against our functions (Oikawa et al., 2016). Impacts of blocks on a Ontake, benefited from non-perpendicular impacts, a large proportion
mountain hut's timber roof can clearly be heard in videos taken by of blocks being too small to cause severe damage, the presence of mul-
hikers during the eruption. Spectrum analysis of the audio identified tiple layers of timber boards in roofs and a large number of impacts to
77 distinct impacts over an 11 second time period. Assuming all the areas of the roof supported by framing components. Additional factors
impacts identified from spectrum analysis were caused by ballistics that require further investigation may include, the armouring effect of
(and not lapilli or surge material) then if this rate of impact was wind protection boulders on the roof, and the potential for a layer of
maintained for the initial 6-minute duration of ballistic fallout (from ash that accumulated during the eruption to act as an energy absorption
~11:53:30 pm to 12:00 pm), this hut and its immediate surroundings barrier (Fig. 12). Corroborating this observation, on-going ballistic can-
would have sustained over 2500 impacts. Our experiments show tim- non experiments have found a 20 cm layer of scoria substantially
ber weatherboard material can be perforated by impacts b1000 J and increases the impact resistance of reinforced concrete slabs while
would almost certainly be perforated by impacts of N 5000 J but despite adding a load of b3 KPa.
this there are b20 visible perforations in each hut's timber roof, imply- Our data illustrate a reduced velocity of ballistics following building
ing very few ballistics were above the 5000 J threshold for this site perforation and a relatively low impact energy of associated shrapnel.
during this eruption. Despite this, ballistic blocks with modelled impact This offers an explanation for why everyone who took shelter inside

Fig. 9. A series of frames from 1000 frames per second video footage taken during a reinforced concrete test. Photos show shrapnel being ejected from the underside of the slab despite the
block shattering on impact and not passing through the slab. The elapsed time between the first frame to the last frame is 43 ms. The cartoon on the right shows the typical dispersal
patterns of rock and concrete shrapnel formed during experiments and illustrates the concepts of block inclination and impact obliquity.

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
8 G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx

Fig. 11. Fragility functions derived from the perforation thresholds in Blong (1981) compared with fragility functions derived from laboratory experiments.

buildings at Ontake survived despite several ballistics perforating roofs. following design recommendations for buildings based on our field
Additionally, online at ONTAKE: Eyewitness of Eruption, testimonies and laboratory studies:
from survivors indicate they sheltered at the lowest level within the
huts, put backpacks on their heads or wore helmets and even cooking • Use high strength cladding where possible, such as reinforced con-
pots which were stored in the hut (NHK, 2015; http://www.nhk.or.jp/ crete. At Usu Volcano, reinforced concrete slabs performed well with
d-navi/link/ontake2014/index.html). This created multiple layers of large 60 cm diameter blocks being caught by a 20 cm thick slab
cladding and personal protection, increasing the number of barriers which had 2 layers of 1 cm diameter reinforcement bars. Add a high
which could dissipate the energy of ballistics and shrapnel, as shown tensile strength layer on the interior side of reinforced concrete
in our experiments. Similarly at Usu, both multi-story reinforced con- slabs to catch fragments which scab off during impacts.
crete buildings we surveyed had no visible perforations in the floor de- • Increase the thickness of and number of framing members used in
spite having several in their roofs. buildings, particularly in the roof. This will increase the chance of bal-
listics being halted when they hit framing supported areas of cladding
4.3. Design recommendations for buildings within ballistic hazard zones as well as increasing the structure's load bearing capacity, allowing it
to withstand heavier tephra loads without collapsing.
Results indicate the ideal building to shelter from ballistic impacts • For buildings which have roofs with relatively high load bearing ca-
would have a reinforced concrete roof, with multiple stories or suffi- pacities (N4 KPa), a layer of rocks, scoria, or a material with similar en-
ciently thick with an additional layer on the interior side of the building ergy absorption properties could be placed onto roofs to armour them
to protect from any shrapnel. To further protect themselves, we advise against ballistic impacts.
people take shelter in a position that puts as many layers of material • In cases when buildings cannot be retrofitted, we suggest building an
as possible (including back packs, robust furniture, mattresses etc.) be- easily accessible concrete bunker into the ground either within the
tween themselves and incoming ballistics (Fig. 13). building or immediately outside as a life safety mitigation measure.
We suggest the most effective risk reduction strategy for volcanic
ballistics is to minimize exposure of people within potential ballistic 4.4. Linking damage to repair costs
hazard zones. We note that although buildings can offer some protec-
tion against ballistic impacts, their presence within ballistic hazard The fragility functions presented in Fig. 5 predict the severity of
zones may increase exposure by encouraging people to stay for longer damage caused by an individual ballistic to an individual component
periods of time. Where exposure is unavoidable, we provide the of a building. Although broad repair or replacement cost estimates can
be linked to damage from these fragility models, it is unlikely to give
an accurate representation of the repair costs to the property as a
whole. A large number of factors will control the costs associated with
repairing or replacing buildings damaged by ballistic impacts. For in-
stance building repair costs will vary in different parts of the world
where there are different standards of construction and habitability
(Yepes-estrada et al., 2016). Even in a single eruption case, the costs
will also be dependent on contextual factors such as how long a building
remains exposed post-impact (for instance, this may depend on dura-
tion of evacuation), the total number of ballistic impacts to the building
(e.g. Hill, 1970), any mitigation measures applied pre-, syn- or post-im-
pact, and the damage caused by other volcanic hazards (e.g. at Usu Vol-
cano, rooms of a building where ash and rain had entered through
ballistic perforations had deteriorated much more severely than adja-
cent rooms with no ballistic perforations). Laboratory studies such as
this one are useful for closely examining specific aspects of ballistic im-
pacts but using them to estimate building repair costs will require
modelling which then relies on quantitative field studies for calibration
(Jenkins et al., 2014b). Therefore the best source of information for
linking ballistic damage to building repair costs is likely to be empirical
Fig. 12. The roof of a mountain hut hit by ballistics b500 m from the craters formed in the
2014 eruption of Mt. Ontake. A row of wind protection boulders which cover the entire
data gathered during field-studies. Such studies should seek to develop
roof can be seen near the eves beneath a layer of ash which blanketed the roof. a second set of fragility functions which build on the ones presented
Photo taken by Kae Tsunematsu in 2015. here by considering the damage caused to a building by multiple

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx 9

Fig. 13. Cartoon illustrating likely building damage for ballistic impacts to timber and reinforced concrete buildings. Damage severity is indicated by line colour with red, yellow and blue
corresponding to DS3 DS2 and DS1 respectively. Pairs of figures illustrate appropriate actions (blue) compared to actions which do not increase life safety (red). (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)

ballistic impacts, perhaps using the number of ballistic impacts per m2, Alatorre-Ibargüengoitia, M.A., Delgado-Granados, H., Dingwell, D.B., 2012. Hazard map for
volcanic ballistic impacts at Popocatépetl volcano (Mexico). Bull. Volcanol. 74:
instead of impact energy as their measure of hazard intensity. 2155–2169. http://dx.doi.org/10.1007/s00445-012-0657-2.
Baxter, P.J., Gresham, A., 1997. Deaths and injuries in the eruption of Galeras Volcano, Co-
5. Conclusions lombia, 14 January 1993. J. Volcanol. Geotherm. Res. 77 (96):325–338. http://dx.doi.
org/10.1016/S0377-0273(96)00103-5.
Biass, S., Falcone, J., Bonadonna, C., Di, F., Pistolesi, M., 2016. Great balls of fire: a probabi-
The volcanic ballistic fragility functions presented herein are the first listic approach to quantify the hazard related to ballistics — a case study at La Fossa
to be derived using experimental data. The experiments refine previ- volcano, Vulcano Island, Italy. J. Volcanol. Geotherm. Res. 325:1–14. http://dx.doi.
org/10.1016/j.jvolgeores.2016.06.006.
ously unvalidated estimates of ballistic-cladding perforation thresholds Blong, R., 1981. In: Self, S., Sparks, R.S.J. (Eds.), Some effects of tephra falls on buildings.
which are based largely on cyclone vulnerability studies (Blong, 1984). Tephra Studies, Reidel, Dordrecht, pp. 405–420.
Experiments highlight the hazard associated with ballistic shrapnel and Blong, R., 1984. Volcanic Hazards: A Sourcebook on the Effects of Eruptions. Elsevier.
Booth, B., 1979. Assessing volcanic risk. J. Geol. Soc. 136:331–340. http://dx.doi.org/10.
the need to model variable impact obliquities during ballistic risk as- 1144/gsjgs.136.3.0331.
sessments. The most severe damage occurs when ballistics strike per- Breard, E.C.P., Lube, G., Cronin, S.J., Fitzgerald, R.H., Kennedy, B., Scheu, B., Montanaro, C.,
pendicular to cladding and the minimum impact energies required to White, J.D.L., Tost, M., Procter, J.N., Moebis, A., 2014. Using the spatial distribution
and lithology of ballistic blocks to interpret eruption sequence and dynamics: August
cause perforation of the claddings used in these experiments were as 6 2012 Upper Te Maari eruption, New Zealand. J. Volcanol. Geotherm. Res. http://dx.
low as 290 J for timber weatherboard and as high as 2160 J for rein- doi.org/10.1016/j.jvolgeores.2014.03.006.
forced concrete roof slabs. Although ballistics can land with impact en- Chen, X.W., Fan, S.C., Li, Q.M., 2004. Oblique and normal perforation of concrete targets by
rigid projectiles.pdf. Int. J. Impact Eng. 30:617–637. http://dx.doi.org/10.1016/j.
ergies well over the perforation threshold for typical building materials,
ijimpeng.2003.08.003.
multiple layers of building materials significantly reduce the velocity of Fitzgerald, R.H., Tsunematsu, K., Kennedy, B.M., Breard, E.C.P., Lube, G., Wilson, T.M., Jolly,
ballistics. This suggests the safest place to shelter from ballistics in unre- A.D., Pawson, J., Rosenberg, M.D., Cronin, S.J., 2014. The application of a calibrated 3D
inforced buildings is on the far side of the building relative to the vent ballistic trajectory model to ballistic hazard assessments at Upper Te Maari, Tongari-
ro. J. Volcanol. Geotherm. Res. 286:248–262. http://dx.doi.org/10.1016/j.jvolgeores.
with as many layers of material as possible placed between the person 2014.04.006.
and incoming ballistics. Fitzgerald, R.H., Kennedy, B.M., Wilson, T.M., Leonard, G.S., Tsunematsu, K., Keys, H., 2016.
The communication and risk management of volcanic ballistic hazards. In: Fearnley,
C., Bird, D., Jolly, G., Haynes, K., McGuire, B. (Eds.), Observing the Volcano World: Vol-
Acknowledgments cano Crisis Communication. Advances in Volcanology. Springer International Publish-
ing, p. 350.
Funding for this study was provided by DeVoRA (Determining Vol- Hill, M.R., 1970. Mount Lassen is in Eruption and There is No Mistake about That: State of
California. Division of Mines and Geology.
canic Risk in Auckland) and Natural Hazards Research Platform contract Jenkins, S.F., Spence, R.J.S., Fonseca, J.F.B.D., Solidum, R.U., Wilson, T.M., 2014a. Volcanic
C05X0804 (TW, BK). We thank the New Zealand Earthquake Commis- risk assessment: quantifying physical vulnerability in the built environment.
sion (EQC) for on-going support and acknowledge Biennial Grant 16/ J. Volcanol. Geotherm. Res. 276:105–120. http://dx.doi.org/10.1016/j.jvolgeores.
2014.03.002.
727 (BK, TW). RHF is supported by a doctoral scholarship from the Ngāi Jenkins, S.F., Wilson, T.M., Magill, C.R., Miller, V., Stewart, C., 2014b. GAR Technical Back-
Tahu Research Centre. We acknowledge the University of Canterbury, ground. pp. 1–39.
Geological Sciences Department technicians and Callum Boot and Mastin, L.G., 2001. A simple calculator of ballistic trajectories for blocks ejected during
volcanic eruptions. U.S. GEOLOGICAL SURVEY Open-File Report 01-45, Version 1.2.
George Daly for assisting with experiments and Peter Jones for design- NHK, 2015. NHK Ontakeyama “Testimony of the Eruption”. http://www.nhk.or.jp/d-navi/
ing and building the cannon. Many thanks to Hiromu Okada for guiding link/ontake2014/index.html (June 21, 2017).
us through fieldwork around Mt. Usu and for sharing many of his spec- Oikawa, T., Yoshimoto, M., Nakada, S., Maeno, F., Komori, J., Shimano, T., Takeshita, Y.,
Ishizuka, Y., Ishimine, Y., 2016. Reconstruction of the 2014 eruption sequence of
tacular photos of buildings damaged during the eruption. Ontake Volcano from recorded images and interviews. Earth Planets Space 68 (1):
79. http://dx.doi.org/10.1186/s40623-016-0458-5.
References Pomonis, A., Spence, R., Baxter, P., 1999. Risk assessment of residential buildings for an
eruption of Furnas Volcano, Sao Miguel, the Azores. J. Volcanol. Geotherm. Res. 92:
Alatorre-Ibargüengoitia, M.A., Delgado-Granados, H., 2006. Experimental determination 107–131. http://dx.doi.org/10.1016/S0377-0273(99)00071-2.
of drag coefficient for volcanic materials: calibration and application of a model to Po- Raymond, D., Van Ee, C., Crawford, G., Bir, C., 2009. Tolerance of the skull to blunt ballistic
pocatépetl volcano (Mexico) ballistic projectiles. Geophys. Res. Lett. 33:1–5. http:// temporo-parietal impact. J. Biomech. 42 (15):2479–2485. http://dx.doi.org/10.1016/j.
dx.doi.org/10.1029/2006GL026195. jbiomech.2009.07.018.

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
10 G.T. Williams et al. / Journal of Volcanology and Geothermal Research xxx (2017) xxx–xxx

Rossetto, T., Ioannou, I., Grant, D.N., Maqsood, T., 2014. Guidelines for Empirical Vulnera- Wardman, J., Sword-Daniels, V., Stewart, C., Wilson, T.M., 2012. Impact assessment of the
bility Assessment Report Produced in the Context of the Vulnerability Global Compo- May 2010 eruption of Pacaya volcano, Guatemala. GNS Science Open-File Report
nent Project. 2012/09.
Sparks, R.S.J., Bursik, M.I., Carey, S.N., Gilbert, J.S., Glaze, L.S., Sigurdsson, H., Woods, A.W., Wilson, T.M., Stewart, C., Sword-Daniels, V., Leonard, G.S., Johnston, D.M., Cole, J.W.,
1997. Volcanic Plumes. John Wiley and Sons, Chichester, United Kingdom (GBR). Wardman, J., Wilson, G., Barnard, S.T., 2012. Volcanic ash impacts on critical infra-
Taddeucci, J., Alatorre-Ibargüengoitia, M.A., Cruz-Vázquez, O., Bello, E. Del, Scarlato, P., structure. Phys. Chem. Earth 45–46:5–23. http://dx.doi.org/10.1016/j.pce.2011.06.
Ricci, T., 2017. In-flight dynamics of volcanic ballistic projectiles. Rev. Geophys. 55. 006.
http://dx.doi.org/10.1002/2017RG000564. Wilson, G., Wilson, T.M., Deligne, N.I., Blake, D.M., Cole, J.W., Framework for developing
Television Hokkaido Broadcasting Co. Ltd., 2000. HTV News 31/03/2000. volcanic fragility and vulnerability functions for critical infrastructure. J. Appl.
Tsunematsu, K., Ishimine, Y., Kaneko, T., Yoshimoto, M., Fujii, T., Yamaoka, K., 2016. Esti- Volcanol. (in press).
mation of ballistic block landing energy during 2014 Mount Ontake eruption. Earth Yamaoka, K., Geshi, N., Hashimoto, T., Ingebritsen, S.E., Oikawa, T., 2016. The phreatic
Planets Space 68 (1):88. http://dx.doi.org/10.1186/s40623-016-0463-8. eruption of Mt. Ontake volcano in 2014. Earth Planets Space 68 (1), 175.
Vanderkluysen, L., Harris, A.J.L., Kelfoun, K., Bonadonna, C., Ripepe, M., 2012. Bombs be- Yepes-estrada, C., Silva, V., Eeri, M., Rossetto, T., 2016. The Global Earthquake Model Phys-
having badly: Unexpected trajectories and cooling of volcanic projectiles. Bull. ical Vulnerability Database. 32, 4:pp. 2567–2585. http://dx.doi.org/10.1193/
Volcanol. 74:1849–1858. http://dx.doi.org/10.1007/s00445-012-0635-8. 011816EQS015DP.
Walker, G.P.L., Wilson, L., Bowell, E.L.G., 1971. Explosive volcanic eruptions-I the rate of
fall of pyroclasts. Geophys. J. Int. 22 (4):377–383. http://dx.doi.org/10.1111/j.1365-
246X.1971.tb03607.x.

Please cite this article as: Williams, G.T., et al., Buildings vs. ballistics: Quantifying the vulnerability of buildings to volcanic ballistic impacts using
field studies and pneumat..., J. Volcanol. Geotherm. Res. (2017), http://dx.doi.org/10.1016/j.jvolgeores.2017.06.026
View publication stats