Modelling the rapid near-surface expansion of gas slugs

in low-viscosity magmas
M. R. JAMES, S. J. LANE & S. B. CORDER
Lancaster Environment Centre, Lancaster University, Lancaster, LA1 4YQ
(e-mail: [email protected])

Abstract: The ascent of large gas bubbles (slugs) in vertical cylindrical conduits and low-viscosity
magmas is simulated using 1D mathematical and 3D computational fluid dynamic (CFD) models.
Following laboratory evidence, the 1D model defines a constant rise velocity for the slug base and
allows gas expansion to accelerate the slug nose through the overlying fluid during ascent. The
evolution of rapidly expanding gas slugs observed in laboratory experiments is reproduced well
and, at volcano scales, predicts at-surface overpressures of several atmospheres without requiring
any initial overpressure at depth. The near-surface dynamics increase slug nose velocities through
the overlying magma by a factor of c. 2.5 and the gas expansion results in pre-burst magma surface
velocities of c. 35 m s21. To examine pressure distributions and the forces exerted on a conduit,
3D CFD simulations were carried out. At volcano scales, the vertical single forces during final slug
ascent to the surface are c. 106 N, two orders of magnitude smaller than those associated with very-
long-period seismic events at Stromboli. This supports a previous interpretation of these events in
which they are generated by gas slugs flowing through changes in conduit geometry, rather than
being the direct result of slug eruption processes.

At volcanoes characterized by low-viscosity complex conduit geometries (e.g. multiple inclined

magma, degassing can occur through the periodic dykes) will require large and detailed fluid dynamic
release of large gas slugs, resulting in Strombolian- models (which, in turn, will be non-trivial to vali-
style eruptions (Chouet et al. 1974; Blackburn et al. date experimentally), we start here by considering
1976). These types of magmatic system are often a straightforward vertical pipe geometry. This
open and persistently active, allowing long-term facilitates comparisons with laboratory data and
studies of degassing rates and conduit flow pro- allows models to draw on a significant body of
cesses. Continuous geophysical monitoring includ- engineering literature. We concentrate on the rela-
ing video and thermal video, gas, infrasonic and tively unexplored effects of the very rapid near-
seismic data collection (e.g. Aster et al. 2004; surface gas expansion and associated conduit-scale
Auger et al. 2006; Marchetti et al. 2006) is now liquid accelerations, leaving the incorporation of
carried out at several such volcanoes, providing a initial bubble coalescence and slug formation to
wealth of data from which significant advances in future work. Slugs are represented as large single
our understanding of degassing, conduit geometry bubbles, as would be anticipated if they formed
and near-surface magma rheology can be made. In by foam collapse or at a depth sufficient that gas
order to combine these data and to include other expansion would have caused coalescence of any
geophysical, geochemical (Oppenheimer et al. initial component smaller bubbles. At Stromboli,
2006; Burton et al. 2007) and petrological or tex- this is supported by gas chemistry data which
tural approaches (e.g. Polacci et al. 2006; Armienti suggest that slugs form at depths of up to 0.8 to
et al. 2007; Lautze & Houghton 2007), physical 2.7 km (Burton et al. 2007). Nevertheless, we
models of the formation, ascent and burst of the anticipate that, to first order, our results would
bubbles are required. Such models would thus rep- also be applicable to ascending foam rafts as long
resent a framework to enable an integrated under- as their density was significantly less than that of
standing of the dynamics of these persistent the surrounding liquid.
volcanic systems. Previous mathematical models of slug ascent for
The aim of the work here is to present two volcanic scenarios have been given (Vergniolle
models describing the final ascent of gas slugs (up 1998; Seyfried & Freundt 2000) but not widely
to, but not including, their burst) with emphasis investigated and these models could not simulate
on understanding the form and magnitude of any accurately the laboratory experiments given here.
geophysical effects produced, and to assess model More sophisticated 3D lattice-Boltzmann, diffuse
sensitivity to changes in system parameters. Given interface and Bhatnagar-Gross-Krook numerical
that a full understanding of flow processes in models are under development (D’Auria et al.

From: LANE , S. J. & GILBERT , J. S. (eds) Fluid Motions in Volcanic Conduits: A Source of Seismic and
Acoustic Signals. Geological Society, London, Special Publications, 307, 147 –167.
DOI: 10.1144/SP307.9 0305-8719/08/$15.00 # The Geological Society of London 2008.
148 M. R. JAMES ET AL.

2004; O’Brien & Bean 2004, 2006; D’Auria 2006) surface tension of 0.07 Pa, within a 3.4-cm diameter
but have yet to be quantitatively validated. Here we tube; see Wallis (1969) for the calculation of slug
use two approaches; developing a relatively velocities), giving a decompression rate of 2 kPa
straightforward 1D model, to provide a quick s21. For a timescale representative of the potential
solution to the first-order physics and using a 3D for gas expansion, one could consider the duration
CFD model (constructed with a commercial fluid required for a theoretical volume doubling during
dynamics package) for a more full description of decompression to surface pressure (i.e. from 200
the flow. The applicability, relative usefulness of to 100 kPa), in this case c. 50 s. In comparison,
the different models and, more importantly, the for a volcanological scenario, ascent rates may be
uncertainties involved with modelling these pro- c. 2 m s21, through pressure gradients of c. 20 kPa
cesses are demonstrated. m21 (e.g. steam slugs in magma with a viscosity
Modelling slug flow is important in industrial of 500 Pa s, density c. 2600 kg m23 and surface
settings where multiphase flows are used to trans- tension of 0.4 Pa, within a 4-m diameter conduit
port fluids and catalyse chemical reactions. Much (Seyfried & Freundt 2000) giving decompression
of the industrial research is oriented at horizontal rates of 50 kPa s21, and hence effective volume
or low-angle pipes, or flowing liquids, with these doubling timescales of c. 2 s.
scenarios being applicable to oil pipelines (e.g. In order to assess the accuracy of model results
Gregory & Scott 1969; Wallis 1969; Dukler & under relevant expansion conditions, laboratory
Hubbard 1975; Fabre & Liné 1992). Furthermore, experiments were designed to allow slug ascent
associated models often describe time-averaged par- under low ambient surface pressures to be observed.
ameters of continuous flow rather than the progress For the theoretical laboratory example presented
of a single gas slug. Nevertheless, considerable above (and neglecting any decompression-related
research exists on the ascent of individual gas slugs phase changes), a ‘volume doubling to surface
up liquid-filled vertical pipes (Davies & Taylor pressure’ timescale of 2 s could be achieved by
1950; White & Beardmore 1962; Brown 1965; using a surface pressure of c. 4 kPa. However,
Wallis 1969; Polonsky et al. 1999). Computational note that this analysis is presented as an illustrative
fluid dynamic (CFD) models of individual gas slugs example only, rather than a formal scaling relation.
have been developed (Mao & Dukler 1990, 1991; Actual fluid accelerations will be driven by the
Clarke & Issa 1997; Taha & Cui 2006a, b) but detailed magnitude of volumetric expansion (as opposed to
modelling of multiphase flows, including slug flow, is volume ratios) so, given a timescale similar to the
complex and an ongoing area of research. volcanological scenario, the smaller laboratory
When compared with engineering-oriented length scales will result in significantly smaller
simulations, volcanic scenarios have the added accelerations. Furthermore, the unsteady nature of
complexity of the large length scales applicable, the flow in these regions means that traditional
which result in highly non-steady conditions due dimensionless scaling arguments (which were
to extremely rapid near-surface accelerations. derived for steady-state conditions) cannot be guar-
Within the engineering literature, limited slug anteed to scale expansion-dominated laboratory
expansion due to gas decompression during ascent results accurately to volcanic dimensions.
(in standard laboratory-scale apparatus) has often
been noted. With one recent exception (Sousa
et al. 2006), the resulting effects have usually Methods and models
been corrected for (e.g. in order to calculate
steady-state slug nose velocities: Laird & Chisholm In this work, both laboratory- and volcano-scale
1956; Nicklin et al. 1962; White & Beardmore scenarios are modelled using the fluid and conduit
1962), rather than being the main focus of the parameters given in Table 1 (unless otherwise
work. The authors are unaware of any previous stated), with a Newtonian liquid phase being used
engineering research on the dynamics of rapidly throughout. Appropriate parameters for a generic
expanding slugs as studied here, and the only exper- basaltic magma are used for the volcano-scale scen-
imental data have been from our own preliminary ario, as used in previous experimental work
studies (Lane et al. 2005; James et al. 2006a) simu- (Seyfried & Freundt 2000).
lating the expansions applicable to near-surface In volcanological fluid flow problems, high
regions of volcanoes (e.g. in the last few tens of temperature basaltic systems can be regarded as
metres of ascent). As an example of the relevant some of the most tractable, due to the liquid phase
magnitudes, in a generic laboratory experiment being reasonably represented as a Newtonian fluid
under standard conditions, slugs may ascend at in many cases. Consequently, the use of Newtonian
c. 0.2 m s21 through a pressure gradient of liquids has been standard in order to facilitate
c. 104 Pa m21 (i.e. air slugs in water, with a related types of experiment and analysis
viscosity of 0.001 Pa s, density 1000 kg m23 and (Vergniolle & Jaupart 1990; Vergniolle 1998;
 MODELLING GAS SLUG EXPANSION 149

Table 1. Fluid and conduit parameter values

Laboratory experiments Volcanic scenario

Ultragrade19 vacuum oil (BOC Edwards) Basalt magma

Liquid phase
Density, r (kg m23) 862 2600
Viscosity, m (Pa s) 0.124 500
Surface tension, s (N m21) 0.032 0.4
Gas phase* Air H2O
Molecular mass (g mol21) 28.9 18.0
Specific heats ratio†, g 1.4 1.1‡
Temperature, T (K) 293 1370
Conduit radius, rc (m) 0.013 1.5
Initial slug depth, h0 (m) c. 1.78 100
Dimensionless parameters (for steady
slug ascent conditions)§
Morton number, Mo 0.08 109
Eötvös number, Eo 179 106
Froude number, Fr 0.321 0.315
Inverse viscosity, Nf 91 85
Slug ascent velocity, Us (m s21) 0.162 1.71
(calculated from equation 8)

*The models used assume negligible gas densities (with respect to those of the liquids) and consider only gas volume and pressure. The
values provided here are used only to calculate equivalent gas masses for comparative purposes.
†
The best model fits to experimental data were obtained assuming isothermal conditions. Consequently, unless specifically noted, g ¼ 1.0
was used rather than the values here, which would be used under adiabatic conditions.
‡
For hot gases (Lighthill 1978).
§
See equation 8 and adjacent text for definitions and descriptions of the dimensionless numbers.

Seyfried & Freundt 2000; James et al. 2004, 2006b) phase. With a vapour pressure of 1026 Pa at
to these presented here. At temperatures below 20 8C, the use of the vacuum oil allowed the appar-
c. 1130 8C, basalt rheology becomes increasingly atus to be run at low pressures without either boiling
non-Newtonian with further cooling (Shaw 1969). or vapour adding significant mass to injected air
However, at conduit temperatures, and for the bubbles as they expanded.
large flow motions investigated here, departures Liquid pressure at the base of the tube was
from a Newtonian model are likely to represent recorded by an ASG pressure sensor (BOC
only second order uncertainties with respect to Edwards). Flow-induced vertical motions of the
those from other poorly constrained parameters apparatus were measured by suspending the appar-
(for instance, the general magnitude of the apparent atus on a spring with the base resting on a force
viscosity). In the following section, the laboratory transducer (S-100, Strain Measurement Devices
experiments are described first, then details of the Ltd.), which was calibrated for displacement. All
1D and 3D CFD models are given. sensors were logged at 5 kHz by a 12-bit National
Instrument DAQ board in a PC. To carry out an
Laboratory experiments experiment, the surface pressure was set to the
required value and the pump was isolated from
The laboratory flow tube used was a c. 2-m-long the system prior to a pocket of gas (c. 6 ml, at
vertical borosilicate glass tube of internal diameter near hydrostatic pressure) being injected with the
c. 25 mm, sealed at the base (with the exception syringe into the base of the tube. This procedure
of a syringe for gas injection), and connected to a initiated a few transient oscillations in the gas and
vacuum pump at the top (Fig. 1a). A vacuum liquid column but these were rapidly damped out
chamber provided a buffer volume (c. 0.1 m3) to by the liquid viscosity as the slug ascended. The
maintain a near-constant pressure at the liquid ascent and expansion of the gas slugs was recorded
surface during slug expansion, and was linked to using two digital video cameras (Canon MV750i
the flow tube by a flexible connection including a and XL1 in Frame mode) to cover the tube region.
trap to prevent ingress of any liquid droplets into A Basler A602f Firewire camera (up to c. 400
the chamber. The tube contained a depth of frames per second) was also used to image the
c. 1.7 m of vacuum oil (Table 1) for the liquid rapid near-surface slug expansion and burst. The
150 M. R. JAMES ET AL.

(a) (b) (c)

Psurf
suspension
springs to vacuum
system
flow h
tube

liquid h0 P L
surface vacuum
trap chamber

s
L0 P0
slug
rs
nose
rc
~1.7 m

slug
base t=0 t>0

vector scale: 30 cm s–1

syringe air
ASG injection
Fz
z
x
Fig. 1. (a) Experimental apparatus, (b) canonical 1D model and (c) CFD model. During experiments (a), fluid
pressures were measured at the base of the apparatus by a pressure sensor (ASG) and apparatus vertical motion
was measured by a force sensor (Fz). In (b), the parameters used in the 1D model (equations 1 to 7) are shown at (t ¼ 0),
and after (t . 0), the model start. The FLOW-3Dw model is illustrated in (c), where a 13-cm-high y ¼ 0 slice through
the results of a quarter-tube simulation (mirrored in the x¼0 plane) are overlain by liquid velocity vectors on the
left-hand side and by a representation of the computational mesh on the right-hand side. The results depicted are 1.2 s
into the simulation of a laboratory experiment (with a surface pressure of 319 Pa), at which time the slug nose is c. 45
cm below the liquid surface (not shown).

camera images were synchronized to the sensor data 1 Pa) and a volcanologically relevant scenario (at
by including a binary counter linked to the DAQ 319 Pa). For the c. 1.7 m head of liquid, the static
board’s scan clock within the recorded field of pressure at the base of the liquid column was
view. Spatial scaling (to convert pixels to metres) c. 14.4 kPa greater than the pressure at its surface.
was achieved by using 10-cm-interval marks Thus, with a surface pressure of 319 Pa, the total
along the length of the tube. amount of gas expansion possible in the experiment
From an extensive suite of experiments carried (a factor of 46) is equivalent to that of gas rising
out, two (with surface pressures of 1 and 319 Pa) from a depth of c. 175 m under volcanological
are dominantly used here to assess the models; a conditions.
detailed discussion of the experimental data is left In Figure 2, example images are given from two
for other work. The specific experiments selected experiments, one carried out at atmospheric
provide an extreme example of expansion (at pressure (Fig. 2a), illustrating a slug which has
 MODELLING GAS SLUG EXPANSION 151

Fig. 2. Images of gas slugs approaching the liquid surface in laboratory experiments. In (a), a short slug (c. 14 mg of
gas) is expanding slowly in an experiment carried out at atmospheric pressure (c. 105 Pa). The image is backlit
through the flow tube and the slug outline is predominantly shown by refracted light from the gas–liquid interface. The
sketch to the right indicates the position of the liquid surface, slug outline and exterior of the tube (dashed lines).
Although the shape of the slug is horizontally distorted by refraction through the circular tube, the standard morphology
under mixed viscous and inertial control (a domed nose and base) can be easily observed. In (b), a similar image
shows the nose of a much longer slug (c. 1 mg of gas) expanding to a surface pressure of 319 Pa (the apparent
decrease in image resolution is caused by the wider angle camera lens used in order to capture more of the long slug
body). At this point in time, the slug nose was moving upward at c. 2.4 m s21. Note the longer curved length of the
nose and the upward doming of the liquid surface, testifying to the dynamic nature of the slug expansion.

undergone only a very small expansion and one results’ section). Additionally, although shape
carried out with a surface pressure of 1 Pa changes to the slug nose region were observed as
(Fig. 2b) demonstrating the effects of very high slugs approached the surface (Fig. 2), the thickness
rates of slug expansion. of the falling film was deemed to be effectively con-
stant (if anything, the shape changes at the slug nose
1D model could have been represented as a film thickening).
This observation implies that, as the slug expands,
Vergniolle (1998) described a dynamic model for the volume of liquid maintained within the falling
slug ascent within a vertical cylindrical conduit, in film must increase and, in order to do that without
which the slug nose rises at a constant velocity decreasing the slug base velocity, liquid flux past
with respect to the overlying liquid. The bubble the slug nose must also increase. Thus, a constant
length, and hence its volume and pressure, was cal- nose velocity with respect to the overlying liquid
culated by equating the rate of change of momen- should not be assumed for rapidly expanding
tum of the liquid above the slug to viscous, slugs and a more convenient reference frame to
gravitational and pressure forces. However, liquid use would be that of a constant-velocity slug base.
volume in the falling liquid film around the slug Furthermore, in previous work, the inertial force
was not considered and application of this model resulting from the acceleration of the liquid above
to the experimental data proved problematic for the slug has been calculated from the total rate of
several reasons. change of momentum (Vergniolle (1998) equation
In our low pressure experiments, despite the 10, and also implied by Seyfried & Freundt
rapid expansion of the slugs, the video data indi- (2000) equation 11). In this formulation, liquid
cated that the slug base velocity was constant (to mass initially above the slug, which then flows
within measurement error) throughout the ascent past the slug nose (i.e. loss of liquid mass from
(see the ‘Comparison of laboratory and model above the slug), is effectively considered to
152 M. R. JAMES ET AL.

‘leave’ its momentum above the slug nose. A realis- under Poiseuille flow (Batchelor 1967), to give
tic mechanism for this to occur cannot be envisaged
and one implication is that it permits regions of par- Fv ¼ 8pmlp Uf ð3Þ
ameter space in which slugs at pressures less than
the liquid surface pressure are still increasingly where lp is the pipe length and Uf is the mean flow
expanding; a clearly non-physical result. Although velocity along the pipe. An estimate of the viscous
any simplified model of the momentum flux drag experienced by the liquid above the slug can
around and above the slug nose is necessarily
thus be made by assuming that it flows at a
going to require significant assumptions, using an volume flux equal to that of the gas expansion,
alternative scheme in which the momentum flows giving
with the mass (and hence is lost from the liquid
above the slug) avoids this non-physical scenario.
Here, in line with these observations, we _ 0:
Fv  8pmhLA ð4Þ
describe a new 1D model, similar to that of Verg-
niolle (1998) in that it is based on the forces Equating the product of mass and acceleration
exerted on the liquid above the slug, but also (of the centre of mass of the liquid column directly
accounting for liquid volume surrounding the slug over the gas slug) to the sum of forces gives
and using a constant slug base, rather than nose, vel-
ocity. The slug is represented as a gas cylinder of d 2 ðUs t þ L þ 12hÞ
constant radius rs and initial length L0, within a ver- prs2 rh ¼ Fp þ Fg þ Fv : ð5Þ
tical cylindrical conduit of internal diameter rc dt2
(Fig. 1b). Maintaining this simplified geometrical
representation of slug morphology used by previous Substituting for the forces and expanding the
workers (Vergniolle 1998; Seyfried & Freundt differential (using equation 2 to define h in terms
2000) allows the first order dynamics associated of L) gives
with the depressurization expansion of long slugs
to be captured with relative ease because, for the 1 2
pr rhð1 þ A0 ÞL€ ¼ prs2 ðP  Psurf Þ
slug sizes of interest, variations from a cylindrical 2 s
shape (e.g. the curved or seiching base and domed _ 0
 prs2 rhg  8pmhLA ð6Þ
nose) are volumetrically relatively small.
With the liquid surface an initial distance h0
above the top of the slug, conservation of liquid which, if the slug is represented as a perfect gas
volume at any given time yields (PV g ¼ constant, where V is gas volume and g the
ratio of specific heats, in this case leading to
P ¼ P0(L0/L)g, for a constant radius cylinder and
h0 prc2 þ L0 pðrc2  rs2 Þ ¼ ðh þ sÞprc2 an initial slug pressure, P0), can be reduced to
þ Lpðrc2  rs2 Þ ð1Þ
1
rð1 þ A0 ÞL€ ¼ P0 Lg0 Lg h1  rg
2
where s is the distance between the slug base and its _ c2 :
initial position, and h is the height of fluid above the  Psurf h1  8mLr ð7Þ
top of the gas cylinder (Fig. 1b). Simplification of
equation 1 allows h to be expressed as Note that h is a function of L (equation 2) and of
the (constant) slug parameters Us and rs, the calcu-
lation of appropriate values for which is discussed
h ¼ h0  Us t  ðL  L0 Þð1  A0 Þ ð2Þ
below. For all the simulations carried out, the
initial bubble pressure was given by the static
where A0 ¼ (rs/rc)2 and, for a slug with a constant head only, i.e. P0 ¼ rgh0 þ Psurf. Equation 7 is
base velocity, Us, at time t, s ¼ Ust. solved numerically in Matlabw, using an explicit
The acceleration of the liquid above the slug can Runge-Kutta (4, 5) formula (the Dormand-Prince
then be defined in terms of the pressure, gravita- pair; Dormand & Prince 1980) and, from the sol-
tional and viscous forces acting on the liquid cylin- utions, the positions of the slug and liquid surfaces,
der, given by Fp ¼ pr2s (P – Psurf), Fg ¼ –pr2s rhg as well as the slug gas pressure, P, are obtained.
and Fv respectively, where P is the slug gas Starting conditions are defined by L0 and by an
pressure, Psurf is ambient surface pressure, r is the initial estimate of the rate of slug expansion, L̇0.
liquid density and g is the acceleration due to For this, the initial expansion rate is assumed to
gravity. For laminar flow in a pipe, the viscous be much less than the slug velocity (L̇0  Us) and
drag force can be derived from the pressure drop dominated by the ascent-related reduction in static
 MODELLING GAS SLUG EXPANSION 153

pressure (i.e. Ṗ  rgḣ). Differentiating PV g ¼ 0.162 m s21. This is only slightly greater than a
constant then allows an initial rate of expansion to graphical estimate made from the Eo-Fr-Mo plot
be estimated as L̇0  rgUsL0/(gP0) and using this of White and Beardmore (1962) which gives
non-zero value minimizes initial oscillations of Fr ¼ 0.305, corresponding to Us ¼ 0.154 m s21
the slug at the start of simulations. (identical within measurement error to the value
measured from the video data). Hence, although
the measured values can be used for modelling the
Slug ascent velocity laboratory experiments, equation 8 is used for deter-
mining slug base velocity for volcanic scenarios.
For the flow regimes of interest, both inertial and
viscous forces are important and defining an appro-
priate value for Us is non-trivial. Existing work Slug radius and falling film thickness
(White & Beardmore 1962; Wallis 1969) provides
dimensionless relationships for slug nose velocity The other fluid-related parameter requiring a calcu-
which are valid for conditions of no gas expansion lated value for the 1D model is the slug radius, rs.
(and hence also represent slug base velocities For long slugs, the falling film surrounding the
under the same conditions). Given that in our slug approaches a constant thickness, d1, with
laboratory experiments, slug base velocities were increasing distance from the slug nose (Brown
not detected to change with expansion (and 1965). At this thickness, the liquid velocities
changed negligibly with surface pressure), we within the film are steady and the thin film is fully
can thus utilize these existing determinations of supported (i.e. is no longer accelerating) by the
nose velocity as a slug base velocity in our viscous shear stress on the tube wall. Batchelor
expanding-slugs model. (1967) showed that, in this case, and for films
A general approach is provided by Wallis (1969) sufficiently thin that their curvature round the slug
in which ascent velocity for a buoyant phase rising can be neglected (d1  rc), conservation of mass
through stagnant liquid in a vertical pipe is given in can be used to approximate the falling film thick-
terms of the dimensionless Froude number, Fr, a ness as
ratio of inertial to buoyancy forces. In the case of sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a gas slug with negligible density with respect to 3 3mrc Us
the surrounding liquid d1 ¼ : ð10Þ
2rg
pffiffiffiffiffiffi
Us ¼ Fr gD ð8Þ
This was used by Vergniolle (1998), along with
Ul ¼ 0.345(gD)1/2 for inertially controlled slugs.
where D is the internal pipe diameter. Other dimen- With the Us values calculated above, equation 10
sionless numbers are defined to characterize provides a range of values for the laboratory exper-
parameter space into regions of different control iments of 3.5 to 3.6 mm.
(Wallis 1969); the Eötvös number relating gravita- An alternative was used by Seyfried and Freundt
tional and surface forces, Eo ¼ rgD 2/s, and the (2000), who employed an expression for the equili-
Morton number, relating viscous and surface brium film thickness derived by Brown (1965) in
forces, Mo ¼ gm4/(rs 3), with the ratio (Eo 3/ which
Mo)1/4 also being used as a dimensionless inverse
viscosity, Nf (Fabre & Liné 1992). If surface pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tension effects are negligible (Eo . 100, applicable 1 þ 1 þ 2Nrc
d1 ¼ ð11Þ
for both the experiments and for volcanic scenarios) N
and Nf . 300, slugs are under inertial control and
Fr ¼ 0.345. In contrast, for Eo . 100 and Nf , 2, where
slugs are under viscous control, and Fr ¼ 0.01 Nf sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(Wallis 1969). Over the intervening region of 3 r2 g
mixed viscous and inertial effects N ¼ 14:5 2 ð12Þ
m
 
Fr ¼ 0:345 1  eNf =34:5 : ð9Þ and typographical errors in Seyfried and Freundt
(2000, their equations 5 and 6) are corrected here.
For the experiments described here (Table 1), Brown (1965) gives a surface-tension-derived
Nf ¼ 91, Eo ¼ 179, putting slugs in this region of limit for the application of this of
mixed control with negligible surface tension Eo . 5(1 2 d1/rc)22 and, from comparison with
effects. Applying equations 8 and 9 to the exper- experimental data, a viscous limit of N . 30/rc,
imental conditions gives a slug ascent velocity of below which it was assumed that the potential
154 M. R. JAMES ET AL.

flow was being significantly altered by viscous –10

effects. For the experiments, this approach gives Mo = 10
0.3
d1 ¼ 3.2 mm. This is valid under the surface 10
–4

Froude number, Fr
tension criterion (the right-hand side of equation –2
10 evaluates to 8.8 and Eo ¼ 179), but N is calcu- 0.2
10
0
lated to be 1890 and the viscous limit of 30/rc 10 1
10
evaluates to 2300, so strictly this exceeds, although 2
10
is close to, the limit of applicability. 0.1
For the experiments, the effective thin film
thickness can be calculated by applying constant
liquid volume criteria (i.e. equation 1) to measure- 0.0
ments of the slug length and the liquid surface pos- 10 100 1000
Eötvös number, Eo
ition during the ascent of a slug. This approach
provided a best fit value of rs ¼ 9.5 mm, i.e. Fig. 3. Dimensionless slug ascent velocities. The grey
d1 ¼ 3.5 mm, for experiments carried out at lines represent curves of constant Morton number
atmospheric pressure in which slug nose shape (labelled with the appropriate Mo values) in Eo– Fr
changes are minimized. The models thus provide space, redrawn from White & Beardmore (1962). The
a good estimate of this, with equation 10, evaluated black symbols joined by dashed curves (purely to guide
using the measured slug ascent velocity, being the reader), represent the results of 3D CFD simulations
the closest. Consequently, we retain equation 10 carried out with a liquid of density 1000 kg m23, surface
tension of 0.1 N m21, with combinations of viscosity
for calculation of slug radii in volcanological
values of 1.79  1023, 56.5  1023, 0.179, 0.565, 1.79
scenarios. and 5.65 Pa s and tube diameters of 0.0101, 0.0143,
0.0226, 0.0319, 0.0639 and 0.101 m. For each
3D CFD model simulation, Fr was calculated from the resulting base
ascent velocity of the slug. The grey-shaded areas are
3D computational fluid dynamics simulations were regions of parameter space for which the slug velocities
carried out using FLOW-3Dw (release 9.2, http:// are either consistently under- or overestimated, with the
www.flow3d.com), a general purpose finite differ- upper, high-Fr area representing the region of strong
ence (finite volume) CFD program. FLOW-3Dw is inertial control.
based on solving the Navier-Stokes equations over
a specified Cartesian grid and specializes in
free-surface flows. parameter space for model validation (Fig. 3, and
Given the strong contrast in density and vis- see next section).
cosity between the liquid and the gas phase, the The liquid was represented as an incompressible
gas phase can be modelled as a ‘void’ region, gov- Newtonian fluid and turbulence was accounted
erned by PV g ¼ constant. For such void regions, for with either a Prandtl mixing length model (a
internal fluid flow, and consequently also shear standard simplified parameterization in which
stresses at boundaries, is not simulated. Thus, in turbulence is estimated from the local shear rate
contrast to analytical approaches, there is no con- and a mixing length describing the distance over
stant stress condition across the interface between which velocity changes are equal to the mean
the gas (void) and the liquid regions. Such free turbulent fluctuation) or a renormalized group
surface interfaces are tracked using the (RNG) model. Gas slugs have been successfully
volume-of-fluid approach (Hirt & Nichols 1981) modelled previously with an RNG turbulence
in which the fluid volume fraction for each cell is model (Taha & Cui 2006a) but in FLOW-3Dw,
stored. Free surfaces are then indicated by cells the RNG option is only usable with explicit
partially filled with fluid (or between full and calculation of viscous stresses, which can con-
empty cells) and their slopes and curvatures can siderably slow calculations. Using the Prandtl
be computed from the fluid volume fractions in model generally decreased run times by allowing
neighbouring cells. The interface positions are implicit (successive under-relaxation) viscous
controlled by the gas and liquid static pressures, solution. A no-slip condition was applied at the
liquid dynamics and surface tension. Although liquid –solid boundary and heat exchange between
surface tension plays little role in slug ascent the phases was neglected. The pressure –velocity
within the experiments (for important surface equations are iterated by an implicit successive
tension effects, Eo , 100 (Wallis 1969) but, for over-relaxation solver, which includes a self-
the experiments, Eo ¼ 179) and could be neglected corrective procedure (Hirt & Harlow 1967) to
entirely at the volcano scale (Eo ¼ 106, Table 1), it minimize accumulation of iteration errors. The
was included within the model to allow a compari- time step, dt, convergence and stability criteria
son of slug ascent velocities over a wide region of (which include maintaining dt smaller than
 MODELLING GAS SLUG EXPANSION 155

0.45  the Courant limit) are automatically and Fig. 1c) that were entirely consistent with those
dynamically adjusted throughout each simulation observed in the laboratory (Fig. 2). In particular,
to optimize solution. the topography of the slug base provides a sensitive
Although FLOW-3Dw can work within a cylind- indicator of Fr, and this was well simulated by the
rical coordinate system (which could have been uti- 3D CFD model. In order to more quantitatively
lized with the straightforward geometries used investigate the accuracy of the simulations, slug
here), a Cartesian mesh was selected to enable ascent velocities over different regions of parameter
more complex conduit geometries to be investi- space were compared with previously published
gated comparably in the future. In order to reduce values. To reduce simulation time, only relatively
the total mesh size, and hence minimize compu- small lengths of tube were simulated (sufficient
tational time, axial symmetry was utilized, with only to establish a stable slug velocity) with most
quarter-tube simulations being defined with the Z quarter-tube simulations being between 173 and
axis along the conduit axis, and symmetry boundary 400 cells in Z, depending on the anticipated
conditions prescribed at the fluid or void-filled mesh ascent velocity. For Mo , 1022 and low values of
boundaries (X¼0 and Y ¼ 0). All force results pre- Eo (indicating an increasing importance of surface
sented have been subsequently rescaled back to full- forces with respect to viscous forces), surface
tube conditions. For most of the scenarios of inter- tension was resolved better, and hence velocity
est, a cubic mesh with 88 cells in X–Y produced accuracies were increased, by using 14  14 cells
results very similar to those of more densely in X–Y, with appropriate decreases in cell size in
meshed simulations, so this relatively coarse Z and corresponding increases in run time. Slug
meshing was used to minimize run times. expansion was unwanted, so full atmospheric
However, for some simulations in which surface pressure was prescribed (105 Pa) at the
tension was important or conduit shear forces liquid surface. Nevertheless, slug velocities
needed to be accurately resolved, a denser mesh were ascertained from the slug base to avoid any
of 14  14 cells in X– Y (and a corresponding remaining expansion-related velocity increases
increase in Z ) was used. that the slug nose velocity would have been
In the simulations carried out, the starting con- susceptible to.
ditions consisted of a vertical, cylindrical tube, In Figure 3, the results are compared with the
closed at the base and open at the top, containing benchmark Fr-Eo-Mo dimensionless graphical cor-
liquid to a specified height and with an initial relations determined by White & Beardmore (1962)
bubble represented as a smaller, concentric cylind- from multiple laboratory datasets (grey curves). It
rical void region at the base of the tube, as indicated can be seen that for most Froude numbers ,0.25,
in Figure 1b for the 1D model. The closed base the ascent velocities calculated reproduce the pre-
faithfully models the laboratory experiments but is viously published laboratory data well. However,
clearly less directly relevant within a volcanic scen- velocities are underestimated for larger
ario. However, its presence does not alter the flow Froude numbers, an issue that was confirmed by
after stable slug ascent has been established and it consultation with the manufacturer of FLOW-3Dw
provides a convenient mechanism for calculating (but note that constant slug base velocities
vertical pressure forces. In a volcanic scenario, emerged from the CFD model, in line with the
these forces would be coupled to the surrounding experimental observation). Consequently, slug
rock at any non-vertical region of the conduit ascent under conditions of strong inertial control
(James et al. 2006b), so the model can be viewed is not reproduced as accurately. At the lowest
as providing the magnitude, but not the location, Eötvös and Froude numbers (strong viscous
of the vertical component of pressure forces control), velocities are somewhat overestimated,
exerted on the conduit. but this region of parameter space is far from the
Simulations were run on a desktop PC (a laboratory or volcanological scenarios considered
300 MHz Pentium 4, with 2 Gb of ram) and gener- in this work.
ally took a few hours to complete, with larger, It is worth noting that Taha & Cui (2006a)
denser simulations taking up to several days. demonstrated a CFD model capable of fully repro-
Forces and pressures at multiple points within the ducing the White & Beardmore (1962) results.
system were output at defined time intervals (nor- However, such simulations (carried out with
mally every 0.1 s) throughout the simulations. FLUENTw) used a relatively small translating
computational domain, moving so that it was
stationary with respect to the slug. This efficient
3D CFD slug velocities approach is not applicable here where the interest
in slug expansion and the associated liquid
The 3D CFD simulations produced slug shapes motions requires the entire fluid-filled region to be
(e.g. domed nose and falling film thicknesses, modelled simultaneously.
156 M. R. JAMES ET AL.

Comparison of laboratory and model ascent, the model can reproduce the slug nose and
results liquid surface positions well. The model was fitted
(by eye) by varying g (as a proxy for variable
In Figure 4, the ascent of a gas slug is illustrated by degrees of heat transfer) and the mass of gas
laboratory data (symbols) and the two different injected (i.e. changing L0). In Figure 4, the results
models (lines) for a gas slug expanding into are shown using gas masses which best fit the data
surface pressures of 1 Pa (a, b) and 319 Pa (c, d). when g ¼ 1.0, representative of isothermal gas
The laboratory data were obtained from the digital expansion. Although the masses of gas involved
video and illustrate the position of the liquid are very small (c. 1 mg) and the surface to volume
surface, the slug nose, and the slug base (the ratios of the slugs are very high (potentially allow-
lowest point on the slug), as indicated in ing heat to be absorbed rapidly), expansion is unli-
Figure 1a. During slug ascent, the data illustrate a kely to be completely without temperature change.
constant base velocity and the rapid slug expansion However, there is a trade-off between the values of
as it approaches the surface. g and L0 such that the result of increasing g can be
For the 1D model, the base velocity of the gas somewhat offset by also increasing L0. The choice
slug was defined by the value obtained from the lab- of g ¼ 1.0 was made in this case because of its
oratory video data and, for the majority of the real physical meaning and the fact that the gas

(a) 1 Pa (c) 319 Pa

2.0
surface surface
1.8
Height (m)

1.6

1.4

e ose
1.2 nos gn
slu
g
bas
e slu ba se
slug slug
1.0
0 1 2 3 4 0 1 2 3 4
Time (s) Time (s)
(b) (d)

2.0
surface surface

1.8
Height (m)

1.6

e
nos
e nos
1.4 slug slug
ba se
ba se slug
slug
1.2
1.2 1.3 1.4 1.5 1.6 1.2 1.3 1.4 1.5 1.6
Slug base height (m) Slug base height (m)

γ =1.0 γ =1.0
Video data 1D model { γ =1.1
3D CFD model { γ =1.1
Fig. 4. Plots of gas slug ascent video data (symbols) with 3D CFD (black curves) and 1D (grey curves) model
results overlain, for experiments carried out with surface pressures of 1 Pa (a, b) and 319 Pa (c, d). The solid curves
give the best fit results for isothermal conditions (i.e. g ¼ 1.0), obtained by varying the slug gas mass. For the
parameters appropriate to the laboratory model, the 3D model underestimates the ascent velocity by c. 14% (a, c) but
plotting the same data against slug base height rather than time (b, d), demonstrates that the slug expansions are
well reproduced. The dashed curves show results for simulations carried out using g ¼ 1.1 in order to demonstrate the
sensitivity of the results to variations in the degree of heat exchange.
 MODELLING GAS SLUG EXPANSION 157

Table 2. Parameter values for the experiments and models in Figure 4

Laboratory experiment 1D model 3D CFD model

Slug radius, rs (cm) 0.95 0.95 0.987

Slug base velocity, Us (m s21) 0.151 0.151 0.131
0.154 0.154 0.131
Effective gas mass* (mg) 1.1† 1.33 1.33
1.1† 1.41 1.45

Figures in upright type denote predetermined constant values, determined either from the experimental apparatus or as inputs to the
models. Figures in italics represent measurements or model outputs. Where two figures are given the upper one is for a surface pressure of
1 Pa and the lower one for a surface pressure of 319 Pa.
*Both models assume negligible gas densities and consider only gas volume and pressure. For comparative purposes, volumes have been
converted to masses using the parameters given in Table 1.
†
Minimum values calculated from injection volumes and assuming hydrostatic pressure only.

masses implied by the model were larger than those varying L0 as previously, and can reproduce the
directly estimated for the experiments, from the observed expansions well. In Figure 5c, similar
volumes injected with the syringe (Table 2). Note ascent profiles determined for volcanic-scale slugs
that these estimated injected masses will be of different gas mass are illustrated.
minimum values (a one-way valve on the injection A further advantage of the 1D model (over the
mechanism means that the injection pressures, and CFD approach) is that the solver allows the defi-
hence masses, will have been slightly larger than nition of event criteria, at which a solution is
calculated, assuming static pressure alone), so specifically calculated, allowing run results to be
some cooling is indeed likely but attempts to fit accurately compared at identical points of the mod-
the data assuming adiabatic expansion (i.e. using elled process. Output from the 3D CFD model
g ¼ 1.4, Table 1) produced significantly poorer occurs at a defined time interval, so comparisons
results. To illustrate the sensitivity of the model to between model runs are dependent on the relative
the degree of heat transfer, the same scenario was timing of the process with respect to the output
also run with g ¼ 1.1 to represent a nearly isother- intervals and can thus appear artificially noisy if
mal system, with the results shown by the dashed single points (e.g. maxima or minima) of rapidly
grey curves in Figure 4. changing parameters are of interest.
The 3D CFD model was fitted to the data (again, Here, for a comparative point in the slug ascent
by eye) by assuming isothermal conditions (for process, we define a ‘slug burst’ criterion based on
direct comparison with the 1D results) and by the thickness of the remaining liquid over the slug.
varying the initial gas volume (i.e. L0). The most For the volcanological scenario, a thickness of
obvious result is that the slug ascent velocity (as 0.1 m of magma is used. This is a reasonable
measured at the slug base and which was indepen- value based on clast sizes ejected in Strombolian
dent to the relatively small changes in L0) is under- eruptions (e.g. Chouet et al. 1974) and of the
estimated by c. 14% (Fig. 4a, c) consistent with the order of magnitude (i.e. .0.01 m and ,1 m)
results shown in Figure 3. Although this produces a appropriate for draining basalt films. However, it
very poor fit to the data when plotted against time, is well within the region of interaction between
plotting the data against slug base position the velocity field around the slug nose and the
(Fig. 4b, d) shows that the expansion of the slug liquid surface (h  2rc), where the 1D model
is reproduced with some accuracy. The similarities assumptions are likely to be poor representations
between the 3D CFD results and those from the 1D of the strongly 2D flow. Nevertheless, the use of a
model, including the estimates of gas mass involved consistent point of evaluation will allow relative
(Table 2), provide mutual support between these comparison of runs and absolute accuracies can be
different approaches. assessed by observation of the sensitivity of the
results to parameters.
Variations with gas mass In Figure 6, slug overpressures, slug length,
surface height changes and slug nose velocities
The 1D model has a simplicity and speed that evaluated for this ‘burst’ point are given for slugs
makes it suitable for easy investigation of parameter of different gas masses, ascending in different
space and, in Figure 5, ascent profiles for a range of radii conduits. Although at depth, overpressure is
gas slug sizes (Fig. 5a) and surface pressures defined as P-rgh-Psurf, at the burst point, in order
(Fig. 5b) are compared with laboratory exper- to reflect surface measurable effects, overpressure
iments. For each run, the model was fitted by is given by P-Psurf. All grey shaded regions
158 M. R. JAMES ET AL.

(a) 2.4 (a) 7

3.10

1.82
gas mass 6.38

1.09
0.73
Distance from tube base (m)

0.37
(mg)

Burst overpressure (×105 Pa)

6 1m
2.0 surface rc =
5

1.6 4

ose 3 1.5 m
gn
slu
1.2
2
ba se surface pressure = 319 Pa 2m
slug 1
0.8
6 7 8 9 10 11
Time (s) 0
(b) (b) 120
2 kPa
1 kPa

7.4 kPa
2.0
3 kPa
surface

101 kPa
1 Pa
Distance from tube base (m)

pressure
100
1m
1.8 rc =

Slug length (m)

80
1.5 m
1.6 60

40
1.4 2m
20
1.2
8 9 10 11 12 0
Time (s)
(c)
(c) 40
20 gas mass 70 60 50
Surface height change (m)

m
(kg) 40 30 rc = 1
10 20
10
Distance from initial

5
surface height (m)

0
30 1.5 m

–10
20 2m
–20

–30 10
–40 surface pressure = 105 Pa
–50 0
0 5 10 15 20 25
Time (s) (d)
70
Fig. 5. Comparisons of laboratory gas slug ascent and m
Slug nose velocity (m s–1)

60 rc = 1
results of the 1D model. For (a) and (b), video data
(symbols) are shown with the model results for the slug 50
nose (black curves) and liquid surface (grey curves) 1.5 m
positions. For clarity, time offsets have been applied to 40
align the positions of the slug base, shown only by the 2m
30
video data and indicated with dashed lines. In (a), the
results of experiments carried out with a surface pressure 20
of 319 Pa and different-sized slugs are shown. The
curves are labelled by their best fit model gas masses, 10
with corresponding laboratory experimental masses of 0
6.38, 3.10, 1.82, 1.09, 0.729 and 0.365 mg, as calculated 0 50 100 150 200
from nominal injection volumes of 35, 17, 10, 6, 4 and 2 Gas mass (kg)
ml respectively. Note that video data are not available
for heights .2.18 m above the tube base. In (b), similar Fig. 6. Changes in slug characteristics at the time of
results are given for a suite of experiments carried out burst with varying gas mass (as given by the 1D model,
with different surface pressures using an injected volume for volcanological conditions). The grey shaded areas
of 6 ml (c. 1 mg) for all pressures except 101 kPa, in show the results calculated for the labelled conduit
which 10 ml (c. 14 mg) was used. In (c), slug ascent radius and degrees of heat transfer between full
curves are given for different equivalent gas masses (isothermal conditions, g ¼ 1.0, upper black bounding
under volcanological conditions (Table 1). As an curves) and none (adiabatic conditions, g ¼ 1.1, lower
illustration, the position and length of a 70 kg slug is black bounding curves). For these simulations, rs and Us
given schematically by the grey shaded region at 15 s. were calculated from equations 10 and 8 respectively.
 MODELLING GAS SLUG EXPANSION 159

represent simulations carried out with the viscosity surface height changes (Fig. 7b) demonstrate
and magma density of the standard volcano-scale much more linear increases.
parameters (Table 1, 500 Pa s and 2600 kg m23 In Figure 8, the relative independence of burst
respectively), with the upper bound being isother- overpressures to the initial bubble conditions
mal conditions (i.e. g ¼ 1.0) and the lower bound, (depth and any starting overpressure) are shown.
adiabatic conditions. In Figure 6a and 6c, the For the volcanological conditions given in
effects of reducing the viscosity to 300 Pa s (a Table 1, an initial depth ,30 m is required in
value suggested to be appropriate for Stromboli order to significantly reduce burst pressures, indi-
(Vergniolle et al. 1996)) are given by the hatched cating that it is the final c. 30 m of ascent in
regions. Note that this decrease in viscosity has neg- which most of the dynamic pressurization must
ligible effect on slug length (so has not been plotted occur. In Figure 8b, this is further demonstrated
in Fig. 6b) and, counter-intuitively, decreases the by the fact that for sufficient start depths (here,
calculated overpressure at burst. modelled at 100 m, and for isothermal systems,
Given that the burst criterion of 0.1 m is poorly i.e. g ¼ 1.0) burst pressures do not reflect the start-
constrained and previous work (Vergniolle & ing pressure. Initial overpressures will cause oscil-
Brandeis 1996) has suggested that even smaller lations of the slug (Vergniolle 1998), but will be
values could be appropriate, based on the smaller effectively dissipated by the time the slug reaches
(centimetric) clast sizes imaged by Chouet et al. the surface. Figure 8 also shows the adiabatic case
(1974) during Strombolian eruptions, simulations (g ¼ 1.1, Table 1) to illustrate the sensitivity of
were also carried out using a ‘burst criterion’ of the results to variations in heat transfer.
h ¼ 0.01 m. The results only showed significant
differences from the previous runs for the calculated
slug nose velocities (given by the hatched regions
in Fig. 6d and reflecting the large accelerations
(a) 140
occurring just prior to burst), but not for the other
Burst overpressure (×105 Pa)

parameters plotted. Thus, although the poor con- 120

straints on an appropriate h value to define slug
50 00

15 100
1

Ascent time (s)

burst are recognized, most characteristics appear
80
15

relatively independent of this variable.

0

10 15 60
10 0
0
50 40
Variations with conduit radius and starting
5
conditions
The different curves in Figure 6 indicate that 0
conduit radius has a strong influence on burst con- 40
(b)
ditions. In Figure 7, variations of ascent time (for
Surface height change (m)
the slug nose ascending through 50 m of magma),
Slug nose velocity (m s–1)

overpressure at burst, liquid surface height change 30

and slug nose velocity are shown for slugs of gas 150
masses 50, 100 and 150 kg in conduits of radius 150 100 20
0.5 to 2 m. Both ascent time and overpressure 60 100
50
(Fig. 7a) are shown to increase significantly for 50 10
conduit radii ,1 m, whereas nose velocity and 40

20
Fig. 6. (Continued) Slugs were deemed to burst when 0.5 1.0 1.5 2.0
the fluid depth above the slug, h, was 0.1 m (with the Conduit radius (m)
slug nose velocities given being the average velocity Fig. 7. Changes in slug characteristics at the time of
between h ¼ 0.2 m and 0.1 m). Further simulations were burst with varying conduit radius (as given by the 1D
carried out by individually changing (i) magma viscosity model for isothermal volcanological conditions).
to 300 Pa s and (ii) by using a burst condition of Characteristic parameters have been evaluated for slugs
h ¼ 0.01 m. Where the results showed significant of three different gas masses, which label the curves (in
deviation from the original solutions, they are shown by kg). In (a), the overpressure at burst (solid curves) and
hatched regions. Decreasing viscosity produced ascent time (dashed curves) are shown to be strong
significant changes only in overpressure (a), and surface functions of conduit radius at small radius values. In (b),
height (c), and changing the burst criterion produced slug nose velocity (solid curves) and the surface height
significant change only to the final slug nose velocity change (dashed curves) are shown to vary more linearly
(d), now calculated between h ¼ 0.1 and 0.01 m. with conduit radius.
160 M. R. JAMES ET AL.

(a) 2.8 due to pressure-related gas volume changes would

significantly outweigh those associated with
Burst overpressure (×105 Pa)

2.6 conduit volume changes.

isothermal In Figure 9a, the pressure and force transients
2.4 generated by the ascent and burst of a slug within
the laboratory experiments are illustrated. The
2.2 pressure trace (recorded at the base of the tube)
shows the effect of increasing slug length (as
2.0 adiabatic increasing pressure drop), slug burst (as high-
frequency transient) and drainage of the liquid
1.8 film surrounding the burst slug (as a slow pressure
0 50 100 150 200 recovery). The net force on the apparatus, as illus-
Initial depth, h0 (m) trated by apparatus displacement, shows no discern-
(b) 2.8 ible effect until a sudden upward transient at slug
burst, followed by a damped oscillation of the
Burst overpressure (×105 Pa)

2.6 apparatus. The details of these data are expanded

isothermal in Figure 9b, with model results added. To enable
2.4 direct graphical comparison, the model-time has
been scaled to account for the underestimation of
2.2 the ascent velocity by the 3D CFD model (i.e.
model times have been multiplied by the ratio of
2.0 adiabatic modelled to observed slug velocities, 0.131/
0.154). Furthermore, only the last c. 0.6 m of slug
1.8 ascent was modelled (rather than the full c. 1.7 m
0 0.2 0.4 0.6 0.8 1 of the experiment) so the model time was offset to
Initial overpressure (×106 Pa) coincide the slug burst points.
The plot shows that the pressure drop below the
Fig. 8. Burst overpressures as functions of the initial slug is reproduced accurately in both form and mag-
depth of the slug (a) and the initial overpressure (b), nitude by the model. After this rapid pressure
calculated by the 1D model for volcanological
conditions. Curves are given for both isothermal
decrease, oscillations are shown in the model-
(g ¼ 1.0) and adiabatic (g ¼ 1.1) conditions to pressure that appear to be similar to those in the
demonstrate the relative effect of heat transfer; all experimental data. In the experiments, the pressure
other parameters as given in Table 1. oscillations are induced by the vertical resonant
oscillation of the entire apparatus on the spring
and force transducer but, as conduit motion is not
Conduit force components included in the 3D CFD model, this cannot be the
case for the simulation. No physical process has
Although the 1D model reproduces the observed been attributable to the model oscillations and it is
slug expansions well (as indicated by the compari- thus assumed that they are numerical artefacts trig-
sons with experimental data, Figs 4–5) and can be gered by the instantaneous nature of the bubble
used for relatively straightforward investigations depressurization on burst (when the overlying
of parameter space, it cannot be used to characterize liquid membrane first ruptures), produced because
pressures and forces on the conduit because the internal flow of the gas phase was not simulated.
dynamics of the liquid flow around the slug are Although the model does not include apparatus
not calculated. In contrast, the 3D models allow displacement, the net vertical force calculated
calculation of full moment and single force over the domain (Fig. 9b, bottom panel) is similar
components at any position. Note that here, the in form to the observed displacement, with a
conduit is assumed to be effectively rigid, so solid single main upward transient. The rapid nature of
deformations are not calculated. This is in contrast the apparatus transient (which appears to be faster
to analytical approaches which have been employed than the resonant frequency of the apparatus) pre-
to look at deeper seismic events such as tremor, in vents accurate determination of the force involved,
which, with only a single incompressible fluid but the data suggest that the magnitude is c. 1 N,
phase, oscillations are stimulated through defor- which is not deemed to be significantly greater
mation of the conduit walls (Julian 1994; Balmforth than the modelled 0.3 to 0.4 N.
et al. 2005). The use of a rigid conduit is thought to In Figure 10, the vertical forces acting on a
be a reasonable assumption for near-surface two- conduit are shown for the standard volcanological
phase scenarios, given that the volume and relative scenario (Table 1). Figure 10a gives a time– height
compressibility of the gas determine that liquid flow representation of the development of the vertical
 MODELLING GAS SLUG EXPANSION 161

(a) slug injection slug burst Shear force magnitude (×105 Nm–1)
0 0.1 0.4 0.9 1.6 2.5
slug ascent
16
Pressure (kPa)

15 (a) 20

Distance from initial surface height (m)

14 10
magma surface downward shear
13 0

Displacement (µm)
4 –10 upward shear
2
–20
0 se
–30 no
g
–2 slu e
–40 as
gb
5 10 15 20 sl u
–50
Time (s) (b) 2

Force (×106 N)
(b) pressure 1
force
Pressure (kPa)

15 0
shear
14 –1
force
(c)

Force (×106 N)
13 1
Displacement (µm)

lab. experiment 4 net force

CFD simulation
2 0

0
0 10 20 30 40
–2 Time (s)

0.2 Fig. 10. Conduit forces modelled for the volcanic

Force (N)

scenario. A gas slug equivalent to 40 kg gaseous H2O is

0 initiated with the slug base at 50 m from the magma
–0.2 surface. In (a), the magnitude of the vertical shear (force
13 14 15 16 per vertical metre of the conduit) acting on the conduit
Time (s) wall is plotted over the time period of the gas slug ascent
(from 0 to c. 22 s), the slug burst, and subsequent onset
Fig. 9. Pressure changes and vertical displacements of of film drainage. The solid black curves show the
the apparatus during laboratory experiments. (a) Fluid vertical position of the liquid surface, the slug nose and
pressure at the base of the apparatus (upper curve) and the slug base, with the nose and surface data close to the
vertical apparatus displacement (lower curve) recorded time of burst omitted for clarity. Above the slug, the
during the ascent and burst of a gas slug (mass c. 1 mg) liquid motion exerts an upward shear on the conduit
into a 319 Pa atmosphere. The data show the initial wall, and surrounding and just below the slug, the
pressure transients associated with injection of the gas conduit experiences a downward shear. In (b), the sum
slug and the large displacement offset (at 5 to 6 s) of these shear forces on the conduit are plotted with time
produced when the injection syringe was replaced on the and the change in the pressure force exerted at the
apparatus. After these initial transients cease, the fluid (notionally closed) base of the conduit. In (c), the sum of
pressure below the slug decreases as the slug expands, these components, and thus the net vertical force over
reflecting the increasing volume of liquid within the the entire conduit is plotted, showing the same upward
falling film surrounding the slug, effectively being transient as observed in the laboratory models. In this
supported by viscous shear forces on the tube walls. No example, the transient magnitude is c. 106 N.
net displacement of the apparatus is detected until a
rapid upward transient, starting within 0.1 s of slug
burst, excites a damped oscillation of the apparatus. In
(b), data from around the time of the slug burst are shear forces acting on the conduit during the slug
enlarged to show further detail and for comparison with ascent. Above the slug nose, liquid being accelerated
3D CFD model results (grey curves). To facilitate direct upward by the slug expansion induces upward shear
comparison, model-time has been rescaled and offset to on the conduit, which significantly increases in
coincide the times of slug burst (see text for more magnitude within the region of rapid slug expansion
details). The 3D CFD model does not calculate
apparatus displacements so, in the lowest panel, the
near the surface (c. 22 s). In the region surrounding
calculated net vertical force on the computational the slug, the falling liquid film produces downward
domain is plotted as a proxy. The single upward force shear on the conduit. The implication of this is
transient agrees well with the vertical apparatus that, in the vertical direction, the conduit experiences
displacement recorded. the passage of the slug as an ascending region of
162 M. R. JAMES ET AL.

dilation (at the slug nose), followed by a region of During the bulk of the ascent, this is almost fully
compression at the slug base. compensated for by the shear, as illustrated by the
In Figure 10b, the sum of the vertical shear force zero net force in Figure 10c. However, just prior to
components is provided at each time step (grey burst, an upward force transient is observed, just as
curve), along with changes in the pressure force in the simulations of the laboratory experiment
experienced at the base of the conduit (black (Fig. 9b), but, at the volcano-scale, the force magni-
curve). Initial transients excited as the slug starts to tude is c. 106 N. Note that this is strongly related to
ascend can be seen in both curves and numerical the intense upward shear on the conduit imparted
noise is notably higher amplitude in the pressure by the liquid at and ahead of the slug, rather than a
force trace (calculated from significantly fewer straightforward pressure-loss effect occurring at
cells than the net shear force). As the slug expands burst. An implication of this is that reduction or
and more liquid is viscously supported on the damping of these forces due to magma compressibil-
conduit wall, the pressure below the slug decreases ity and any deformation of the lower conduit is likely
(inducing the upward pressure force illustrated). to be negligible, or at least a second order effect.
Carrying out simulations for varying gas masses
allows assessment of the force magnitude change.
(a) In Figure 11, ascent curves produced from the 3D
CFD model for slugs of different gas mass are illus-
Distance from initial surface height (m)

20 60 50
70 40 30 trated, and the magnitude of the force transients
20 associated with the slug burst are shown
10 magma 10
surface (Fig. 11b). Due to the rapid nature of the transient,
0
the precise values (but not the general magnitude or
–10 the trend) of the forces determined vary with the
output frequency of the model. With forces output
–20 ose
gn at 0.1 s intervals, forces are smaller but less noisy
slu
–30 than if obtained at 0.01 s. In Figure 11b, a compro-
e
–40 bas mise of 0.01-s-data, subsequently averaged by a
slug
seven-point sliding window, was used.
–50
5 10 15 20 25 30
Time (s)
(b) 2.0 Discussion
The 1D and 3D CFD models presented have proved
1.5 complementary and capable of reproducing the fun-
Force (×106 N)

damental aspects of slug ascent and expansion

1.0 under laboratory conditions. Given that the 1D
model contains no end-effects relating to the inter-
action of the liquid velocity field with the liquid
0.5
surface, near-surface expansions are simulated sur-
prisingly well (Fig. 4). Extending the model to
0.0 include end effects would involve significant com-
0 20 40 60 80
Mass of H2O gas (kg) plications (and there are no appropriate parameteri-
zations to describe changes in the slug nose shape),
Fig. 11. (a) Slug ascent profiles calculated by 3D CFD, so are deemed unnecessary at this stage. The 3D
labelled by equivalent gas mass, for the volcanic CFD model has proved invaluable in allowing accu-
scenario (Table 1). The position of the liquid surface rate evaluation of the conduit force and pressure
(grey curves), the slug nose (black curves) and the slug changes involved, although issues remain concern-
base (dashed lines) are shown. Simulations were
initiated with slugs at the base of a 50-m depth of magma
ing the velocity underestimation at high Froude
(as shown in Fig. 10), and results are shown for times numbers. Simulations involving a non-Newtonian
after initial transients have been damped out. The fluid are a future extension of work in this area
individual simulations have been shifted (in time) to and would build on similar slug research within
align the ascent of the slug base. Slug base velocities the engineering field (Carew et al. 1995; Welsh
during the different simulations are sufficiently similar et al. 1999; Xie et al. 2003; Sousa et al. 2004, 2005).
that the paths overlay on the plot but there is a small but
consistent increase with gas mass, from 1.49 to 1.53 m
s21 for slugs of 10 and 70 kg respectively. In (b), the Ascent and expansion
magnitude of the upward force transient (experienced by
the conduit) calculated at the burst of these slugs is One of the potential observables at volcanoes is the
shown against equivalent gas mass. position and motion of the fluid surface during slug
 MODELLING GAS SLUG EXPANSION 163

ascent, which can be calculated from both models such near-surface geophysical data (Harris &
(Figs 4, 5 and 11). As a slug ascends and expands, Ripepe 2007).
the liquid surface rises at a rate which reflects the At the time just prior to burst, magma in the
volumetric expansion of the gas. Due to the liquid film above the slug nose will be also be travel-
dynamic pressurization of the slug only becoming ling upward, but at a lower velocity than the slug
significant in the upper few tens of metres, at nose due to the increased downward liquid flux
depths greater than these expansion is dominated into the falling liquid film around the slug. For
by the reducing static load. Consequently, initial nose velocities much greater than slug base velocity,
liquid surface motions reflect the slug position, the liquid surface velocity can be approximated by
ascent velocity (a function of conduit geometry the product of the nose velocity and the ratio of
and liquid parameters) and the gas mass. If slugs the slug and conduit cross-sectional areas (c. 0.5
start to rise at depths sufficient to establish relatively for the volcano-scale scenario used). This suggests
steady ascent, then the final burst conditions are also that magma in the liquid film can have velocities
strong functions of gas mass, as shown by Figure 6. of several tens of metres per second, a considerable
Thus, detailed records of magma surface motion fraction of measured ejecta velocities which are
could be used to ascertain gas masses, ascent vel- typically up to c. 100 m s21, but can be significantly
ocities and rheological parameters by constraining less (Harris & Ripepe 2007). Thus, final (but pre-
ascent models. Note that the greatest surface vel- burst) slug expansion velocities need to be
ocities only occur in the last few seconds, so accounted for when calculating slug parameters
measurements would have to be relatively frequent. from measurements of ejecta velocity (Chouet
Such liquid surface velocity data have been et al. 1974; Blackburn et al. 1976; Ripepe et al.
obtained by Doppler radar at Erebus (Gerst et al. 1993; Hort et al. 2003; Dubosclard et al. 2004).
2007, 2008) for slugs busting at the surface of In the work described here, the emphasis has
a lava lake. Although the flared nature of the been on rapid slug expansion and the associated
conduit (i.e. rapid near-surface opening into dynamics, but similar volumetric analyses could
the lake) permits unobstructed observation of the be equally applied to the much more gradual
burst, the straight cylindrical conduit geometry changes during gas pistoning events (e.g. such as
modelled here means that our results cannot be those documented at Hawaii; Swanson et al. 1979;
appropriately applied directly to the Erebus data. Tilling 1987; Johnson et al. 2005). In descriptions
In cases where the magma surface cannot be of gas pistoning, it is often unclear as to whether
observed, model estimates of the magnitude of the observed magma levels reflect the ascent of
surface movement (Figs 6c and 7b) can provide many small bubbles (possibly to be trapped by a
constraints for its position, with applications for geometric or rheological boundary, in a ‘gas
the interpretation of seismic and infrasonic data. accumulation’ phase) or the relatively slow ascent
As slugs reach the surface, the expansion-related of a large slug-bubble. Further insight could be
accelerations result in slug nose velocities that (in gained by comparing surface position data with
an external reference frame) are approximately an appropriate slug ascent models.
order of magnitude greater than the slug base vel- For accurate time-based comparisons between
ocities (Figs 6d and 7b). In the reference frame of simulations and field data, reliable slug rise vel-
the liquid above the slug, nose velocity also ocities are critical. In the 1D model, velocities are
increases and, depending on the overall duration determined using tried-and-tested engineering para-
of slug ascent, this could significantly reduce esti- meterizations (Wallis 1969), but this restricts accu-
mated ascent times (which are likely to naturally rate comparisons to idealized scenarios of vertical
reflect nose rather than base velocities). For cylindrical conduits. In contrast, although the 3D
example, ascending through the last 10 m of CFD model provides the flexibility to vary
magma in our volcano-scale scenario, the 1D conduit geometry, its performance in reproducing
model indicates an average slug nose velocity of slug velocities for high Froude numbers
4.4 m s21 (compared to the slug base velocity of (Fr . 0.25) was disappointing. However, for the
1.7 m s21). A similarly elevated apparent slug vel- generic volcanic scenario used (which includes
ocity of 3.7 m s21 was interpreted by Ripepe m ¼ 500 Pa s, giving Fr ¼ 0.32), Froude numbers
et al.(2001) from analysis of image, seismic and are less than 0.25 if liquid viscosity is .950 Pa
infrasonic data from Stromboli. Such velocities s. This is well within the range of viscosities
were difficult to reconcile with calculated ‘standard’ which may be relevant at many appropriate volca-
slug velocities of 1.8 m s21 (Ripepe et al. 2001), but noes, for example, viscosity estimates at Erebus
are in line with the nose velocity increases are between 1 and 10 kPa s (Dibble et al. 1984)
calculated here. This supports the importance of and, at Stromboli, geochemical investigations indi-
this aspect of our 1D model and indicates that cate 14 kPa s (Metrich et al. 2001) and 100 kPa has
expansion should be accounted for when combining been suggested if the presence of small bubbles in
164 M. R. JAMES ET AL.

the magma is accounted for (Ripepe & Gordeev (which is not usually the case), then these
1999). These greater viscosities would imply infrasonic-derived overpressures will be minimum
ascent velocities one or two orders of magnitude values, and post-burst gas expansion within the
less than those that are commonly accepted and conduit would also need to be accounted for. The
consistent with seismics, infrasonic data and ability to determine gas mass directly from infra-
general observations of the dynamic nature of the sonic data would allow degassing rates from these
system (c. 2 m s21). Consequently, this viscosity events to be monitored for comparison with more
range and the resulting inconsistencies underscore traditional gas flux measurements such as FTIR
the requirement for an integrated approach to spectroscopy (e.g. Burton et al. 2003, 2007;
improve our current understanding. Oppenheimer et al. 2006) and to allow continuous
measurement through periods when other tech-
Burst overpressures niques cannot be used, for example in cloud or
at night.
Slug burst overpressures have been previously cor- It is intuitive to anticipate that a dynamic, low
related with initial slug depth (Ripepe et al. 2001), viscosity system would produce greater overpres-
the duration of slug ascent (Lautze & Houghton sures than a more viscously controlled one (e.g.
2007) and gas flux (Marchetti et al. 2006), with Lautze & Houghton 2007). However, this is not
initial slug overpressure (Vergniolle 1998) often the case for the 1D model (Fig. 7a), which reflects
also regarded as having a role. In our models, all the importance of viscous forces in retaining
slugs are initiated at depths sufficient to allow pressure within ascending slugs. Note that changing
stable slug flow to establish itself and consequently, viscosity from 500 to 300 Pa s also had no discern-
initial depth has no significant effect on burst over- ible effect on final slug length (Fig. 7b) so, from a
pressure (Fig. 8a). Furthermore, initial overpres- volumetric point of view, the variation in calculated
sures were explicitly excluded by initializing slugs burst overpressure with viscosity is not reflecting
at the local static pressure (the sum of the surface slug length changes, but variations in the thickness
pressure and the liquid static head). Nevertheless, of the falling film around the slug.
the volcano-scale simulations indicate that
at-surface excess slug pressures of c. 105 Pa are
readily achievable (e.g. with slugs of c. 50 kg in a Conduit forces
1.5-m-radius conduit, Fig. 6a) purely as a result of Although much slower to run than the 1D model,
dynamic pressurization during ascent. This is in the 3D CFD model was required in order to quantify
good agreement with order of magnitude estimates the forces and pressures exerted on the apparatus in
made at Stromboli of between 104 Pa for the rela- the laboratory experiments. For volcano-scale
tively passive burst of small (c. 0.5-m-sized) simulations, the forces involved (c. 106 N) are
bubbles (Ripepe et al. 2001; Marchetti et al. 2006) several orders of magnitude lower than those inter-
to 105 Pa for Strombolian eruptions (Vergniolle & preted from VLP seismic data from Stromboli
Brandeis 1996; Ripepe et al. 2001; Ripepe & (Chouet et al. 2003). Interestingly, this is in agree-
Marchetti 2002; Vergniolle & Caplan-Auerbach ment with previous estimates based on first order
2006), mainly from infrasonic data. scaling arguments (James et al. 2004; Lane et al.
The variation of overpressure with conduit 2005) and supports the hypothesis that VLP
radius shown in Figure 7 could also help explain signals are produced as a result of flow regime
infrasonic measurements made by Ripepe & instability as slugs pass though a change in the
Marchetti (2002) at Stromboli, in which smaller conduit geometry at some depth (James et al.
slugs appeared to be more overpressured than 2006b) rather than reflecting near-surface slug
larger ones. Their data indicated that explosions expansion or the burst process itself. Nevertheless,
occurring at the NE crater were eight times more as instrumentation at volcanoes increases and
overpressured, but half the volume of, those from improves, and seismic data analysis allows for
the SW crater. Thus, converting the overpressure moving and distributed source functions, seismics
values given (4  105 and 0.5  105 Pa) into absol- generated by slug expansion and burst may also
ute pressures (5  105 and 1.5  105 Pa) suggests play a part in estimating the gas masses involved.
that the ratio of gas masses involved was c. 1.7.
Figure 7a indicates that for a gas mass ratio of
two (50 and 100 kg), an overpressure ratio of Conclusions
nearly five can be achieved by the smaller slug
ascending a 0.5-m radius conduit and the larger The 1D model reproduced bulk expansions well and
one a 2-m radius conduit. We consider only ratios can be used to investigate relationships between
here, rather than absolute values because, unless ascent parameters and surface geophysical data
slugs are bursting at the very top of the conduit acquired during slug bursts. This has highlighted
 MODELLING GAS SLUG EXPANSION 165

the relationships between overpressure, gas mass, Stromboli Volcano, Italy. Geophysical Research
conduit radius and liquid surface motions. Allowing Letters, 33, L04301.
the velocity of the slug nose through the overlying B ALMFORTH , N. J., C RASTER , R. V. & R UST , A. C. 2005.
liquid to vary during the ascent is a key difference Instability in flow through elastic conduits and volca-
nic tremor. Journal of Fluid Mechanics, 527,
to previous models, and is supported by laboratory 353– 377.
and field measurements. B ATCHELOR , G. K. 1967. An Introduction to Fluid
For the volcanic scenario used, the models have Dynamics. Cambridge University Press, Cambridge.
shown that a gas slug of c. 100 kg will be overpres- B LACKBURN , E. A., W ILSON , L. & S PARKS , R. S. J.
sured at the surface by c. 2  105 Pa. This is the 1976. Mechanics and dynamics of Strombolian
result of dynamic expansion-related compression activity. Journal of the Geological Society London,
which dominantly occurs in the last c. 30 m of 132, 429–440.
ascent and is effectively decoupled from the B ROWN , R. A. S. 1965. The mechanics of large gas
bubble pressure at greater depths. Close to the bubbles in tubes I. Bubble velocities in stagnant
liquids. The Canadian Journal of Chemical Engineer-
surface and just prior to burst, gas expansion will ing, 43, 217–230.
have accelerated the liquid above the slug nose to a B URTON , M., A LLARD , P., M URÈ , F. & O PPENHEIMER ,
velocity of c. 30 m s21, with the slug nose rising at C. 2003. FTIR remote sensing of fractional magma
an additional c. 4.4 m s21 within it (greater than degassing at Mount Etna, Sicily. In: O PPENHEIMER ,
two-and-a-half times the stable slug base velocity). C., P YLE , D. M. & B ARCLAY , J. (eds) Volcanic
Ultimately, computational fluid dynamic models Degassing. Geological Society, Special Publications,
will provide powerful tools, capable of incorporat- 213, 281–293.
ing most of the parameters measurable at volcanoes. B URTON , M., A LLARD , P., M URE , F. & L A S PINA , A.
In particular, they are the route to interpreting LP 2007. Magmatic gas composition reveals the source
depth of slug-driven Strombolian explosive activity.
and VLP seismic source functions determined in Science, 317, 227– 230.
terms of fluid flow processes. Although the scen- C AREW , P. S., T HOMAS , N. H. & J OHNSON , A. B. 1995.
arios modelled here were specifically chosen to rep- A physically based correlation for the effects of power
resent the most straightforward cases, FLOW-3Dw law rheology and inclination on slug bubble rise vel-
has the capabilities to simulate complex conduit ocity. International Journal of Multiphase Flow, 21,
geometries and liquid rheologies, to include 1091.
thermal effects (including convection, radiative C LARKE , A. & I SSA , R. I. 1997. A numerical model of
cooling and rheological variation) and could also slug flow in vertical tubes. Computers & Fluids, 26,
be used to track geochemical mixing. Conse- 395– 415.
C HOUET , B., H AMISEVICZ , N. & M C G ETCHIN , T. R.
quently, such CFD tools are well placed to form 1974. Photoballistics of volcanic jet activity at Strom-
the backbone of integrated approaches but, as illus- boli, Italy. Journal of Geophysical Research, 79,
trated by the low slug velocities determined 4961– 4976.
here, care needs to be taken to validate C HOUET , B., D AWSON , P., O HMINATO , T. ET AL . 2003.
models appropriately. Source mechanisms of explosions at Stromboli
Volcano, Italy, determined from moment-tensor inver-
MRJ was funded by the Royal Society and SBC by Lan- sions of very-long-period data. Journal of Geophysical
caster University. J. Phillips and L. D’Auria are thanked Research-Solid Earth, 108.
for thorough reviews which significantly improved the D’A URIA , L. 2006. Numerical modelling of gas slugs
manuscript. B. Chouet is also thanked for comments on rising in basaltic volcanic conduits: inferences on
an early version of the text. The CFD models were Very-Long-Period event generation. International
improved by various valuable contributions from Workshop: The Physics of Fluid Oscillations in Volca-
A. Chandorkar, D. Souders (Flow Science Inc.) and nic Systems, Lancaster, U.K. http:// www.es.lancs.ac.
K. Doyle (CFD Solutions Ltd.). uk/seismicflow/.
D’A URIA , L., C HOUET , B. & M ARTINI , M. 2004. Lattice
Boltzmann modeling of a gas slug rise in a volcanic
References conduit: inferences on the Stromboli dynamics.
IAVCEI 2004 General Assembly, Pucon, Chile.
A RMIENTI , P., F RANCALANCI , L. & L ANDI , P. 2007. Abstr. # s08b_pf_119.
Textural effects of steady state behaviour of the Strom- D AVIES , R. M. & T AYLOR , G. 1950. The mechanics of
boli feeding system. Journal of Volcanology and large bubbles rising through extended liquids and
Geothermal Research, 160, 86–98. through liquids in tubes. Proceedings of The Royal
A STER , R., M C I NTOSH , W., K YLE , P. ET AL . 2004. Real- Society of London Series A-Mathematical and Phys-
time data received from Mount Erebus volcano, Ant- ical Sciences, 200, 375– 390.
arctica. EOS Transactions of AGU, 85, 97–101. D IBBLE , R. R., K IENLE , J., K YLE , P. R. & S HIBUYA , K.
A UGER , E., D’A URIA , L., M ARTINI , M., C HOUET , B. & 1984. Geophysical studies of Erebus Volcano, Antarc-
D AWSON , P. 2006. Real-time monitoring and tica, from 1974 December to 1982 January. New
massive inversion of source parameters of very Zealand Journal of Geology and Geophysics, 27,
long period seismic signals: An application to 425– 455.
166 M. R. JAMES ET AL.

D ORMAND , J. R. & P RINCE , P. J. 1980. A family of Volcano. Journal of Geophysical Research-Solid

embedded Runge-Kutta formulae. Journal of Compu- Earth, 110, B11201.
tational and Applied Mathematics, 6, 19–26. J ULIAN , B. R. 1994. Volcanic tremor – nonlinear exci-
D UBOSCLARD , G., D ONNADIEU , F., A LLARD , P. ET AL . tation by fluid-flow. Journal of Geophysical Research-
2004. Doppler radar sounding of volcanic eruption Solid Earth, 99, 11859–11877.
dynamics at Mount Etna. Bulletin of Volcanology, L AIRD , A. D. K. & C HISHOLM , D. 1956. Pressure and
66, 443 –456. forces along cylindrical bubbles in a vertical tube.
D UKLER , A. E. & H UBBARD , M. G. 1975. Model for Industrial and Engineering Chemistry, 48,
gas-liquid slug flow in horizontal and near horizontal 1361– 1364.
tubes. Industrial & Engineering Chemistry Fundamen- L ANE , S. J., J AMES , M. R. & C ORDER , S. B. 2005.
tals, 14, 337–347. Experimental investigation of volcano-seismic forces
F ABRE , J. & L INÉ , A. 1992. Modelling of two-phase generated during gas-slug expansion in Strombolian
slug flow. Annual Review of Fluid Mechanics, 24, eruptions. EOS Transactions of AGU, 85, Fall Meet.
21–46. Suppl., Abstr. V31D– 0652.
G ERST , A., H ORT , M., J OHNSON , J. B. & K YLE , P. R. L AUTZE , N. C. & H OUGHTON , B. F. 2007. Linking vari-
2007. The first second of a Strombolian eruption: able explosion style and magma textures during 2002
Doppler radar and infrasound observations at Erebus at Stromboli volcano, Italy. Bulletin of Volcanology,
volcano, Antarctica. Geophysical Research Abstracts, 69, 445–460.
9, 07280. L IGHTHILL , M. J. 1978. Waves in Fluids. Cambridge
G ERST , A., H ORT , M., K YLE , P. R. & V OGE , M. 2008. University Press, Cambridge.
4D velocity of Strombolian eruptions and man-made M AO , Z. S. & D UKLER , A. E. 1990. The motion of Taylor
explosions derived from multiple Doppler radar instru- bubbles in vertical tubes—I. A numerical simulation
ments. Journal of Volcanology and Geothermal for the shape and the rise velocity of Taylor bubbles
Research, doi: 10.1016/j.jvolgeores.2008.05.022. in stagnant and flowing liquids. Journal of Compu-
G REGORY , G. A. & S COTT , D. S. 1969. Correlation of tational Physics, 91, 132– 160.
liquid slug velocity and frequency in horizontal cocur- M AO , Z. S. & D UCKLER , A. E. 1991. The motion of
rent gas-liquid slug flow. AIChE Journal, 15, Taylor bubbles in vertical tubes: II. Experimental
933– 935. data and simulations for laminar and turbulent flow.
H ARRIS , A. & R IPEPE , M. 2007. Synergy of multiple Chemical Engineering Science, 46, 2055–2064.
geophysical approaches to unravel explosive M ARCHETTI , E., R IPEPE , M., U LIVIERI , G. & D ELLE
eruption conduit and source dynamics – a case study D ONNE , D. 2006. Degassing dynamics at Stromboli
from Stromboli. Chemie Der Erde-Geochemistry, volcano: Insights from infrasonic activity.
67, 1 –35. EOS Transactions of AGU, 87, Fall Meet. Suppl.
H IRT , C. W. & H ARLOW , F. H. 1967. A general corrective Abstract. V43B–1796.
procedure for the numerical solution of initial-value M ETRICH , N., B ERTAGNINI , A., L ANDI , P. & R OSI , M.
problems. Journal of Computational Physics, 2, 2001. Crystallization driven by decompression and
114– 119. water loss at Stromboli volcano (Aeolian Islands,
H IRT , C. W. & N ICHOLS , B. D. 1981. Volume of fluid Italy). Journal of Petrology, 42, 1471– 1490.
(VOF) method for the dynamics of free boundaries. N ICKLIN , D. J., W ILKES , J. O. & D AVIDSON , J. F. 1962.
Journal of Computational Physics, 39, 201– 225. Two phase flow in vertical tubes. Transactions of the
H ORT , M., S EYFRIED , R. & V OGE , M. 2003. Radar Institute of Chemical Engineers, 40, 61– 68.
Doppler velocimetry of volcanic eruptions: theoretical O’B RIEN , G. S. & B EAN , C. J. 2004. A discrete numerical
considerations and quantitative documentation of method for modeling volcanic earthquake source
changes in eruptive behaviour at Stromboli volcano, mechanisms. Journal of Geophysical Research, 109,
Italy. Geophysical Journal International, 154, 515–532. B09301.
J AMES , M. R., L ANE , S. J., C HOUET , B. & G ILBERT , J. S. O’B RIEN , G. & B EAN , C. 2006. Seismicity generated by
2004. Pressure changes associated with the ascent and gas slug ascent: a numerical investigation. Inter-
bursting of gas slugs in liquid-filled vertical and national Workshop: The Physics of Fluid Oscillations
inclined conduits. Journal of Volcanology and in Volcanic Systems, Lancaster, U.K. http://www.es.
Geothermal Research, 129, 61–82. lancs.ac.uk/seismicflow/.
J AMES , M. R., C ORDER , S. B. & L ANE , S. J. 2006a. Lab- O PPENHEIMER , C., B ANI , P., C ALKINS , J. A., B URTON ,
oratory investigations of possible gas-slug related M. R. & S AWYER , G. M. 2006. Rapid FTIR sensing
seismic source processes in low viscosity magmas. of volcanic gases released by Strombolian explosions
International Workshop: The Physics of Fluid Oscil- at Yasur volcano, Vanuatu. Applied Physics B-Lasers
lations in Volcanic Systems, Lancaster, U.K. http:// and Optics, 85, 453–460.
www.es.lancs.ac.uk/seismicflow/. P OLACCI , M., C ORSARO , R. A. & A NDRONICO , D. 2006.
J AMES , M. R., L ANE , S. J. & C HOUET , B. A. 2006b. Coupled textural and compositional characterization
Gas slug ascent through changes in conduit diameter: of basaltic scoria: insights into the transition from
Laboratory insights into a volcano-seismic source Strombolian to fire fountain activity at Mount Etna,
process in low-viscosity magmas. Journal of Geo- Italy. Geology, 34, 201– 204.
physical Research-Solid Earth, 111, B05201, P OLONSKY , S., S HEMER , L. & B ARNEA , D. 1999. The
doi:05210.01029/02005JB003718. relation between the Taylor bubble motion and the vel-
J OHNSON , J. B., H ARRIS , A. J. L. & H OBLITT , R. P. 2005. ocity field ahead of it. International Journal of Multi-
Thermal observations of gas pistoning at Kilauea phase Flow, 25, 957.
 MODELLING GAS SLUG EXPANSION 167

R IPEPE , M. & G ORDEEV , E. 1999. Gas bubble dynamics T AHA , T. & C UI , Z. F. 2006b. CFD modelling of slug flow
model for shallow volcanic tremor at Stromboli. inside square capillaries. Chemical Engineering
Journal of Geophysical Research-Solid Earth, 104, Science, 61, 665– 675.
10639– 10654. T ILLING , R. I. 1987. Fluctuations in surface height of
R IPEPE , M. & M ARCHETTI , E. 2002. Array tracking of active lava lakes during 1972– 1974 Mauna Ulu erup-
infrasonic sources at Stromboli volcano. Geophysical tion, Kilauea Volcano, Hawaii. Journal of Geophysi-
Research Letters, 29. cal Research, 92, 13721– 13730.
R IPEPE , M., R OSSI , M. & S ACCOROTTI , G. 1993. Image- V ERGNIOLLE , S. 1998. Modelling two-phase flow in a
processing of explosive activity at Stromboli. Journal volcano. 13th Australasian Fluid Mechanics Confer-
of Volcanology and Geothermal Research, 54, ence, Aristoc. Offset, Monash University,
335–351. Melbourne, Australia.
R IPEPE , M., C ILIBERTO , S. & D ELLA S CHIAVA , M. V ERGNIOLLE , S. & J AUPART , C. 1990. Dynamics of
2001. Time constraints for modeling source degassing at Kilauea Volcano, Hawaii. Journal of
dynamics of volcanic explosions at Stromboli. Geophysical Research-Solid Earth and Planets, 95,
Journal of Geophysical Research-Solid Earth, 106, 2793– 2809.
8713–8727. V ERGNIOLLE , S. & B RANDEIS , G. 1996. Strombolian
S EYFRIED , R. & F REUNDT , A. 2000. Experiments on explosions. 1. A large bubble breaking at the
conduit flow and eruption behavior of basaltic volcanic surface of a lava column as a source of sound.
eruptions. Journal of Geophysical Research-Solid Journal of Geophysical Research-Solid Earth, 101,
Earth, 105, 23727– 23740. 20433–20447.
S HAW , H. R. 1969. Rheology of basalt in the melting V ERGNIOLLE , S. & C APLAN -A UERBACH , J. 2006. Basal-
range. Journal of Petrology, 10, 510–535. tic thermals and subplinian plumes: constraints from
S OUSA , R. G., N OGUEIRA , S., P INTO , A., R IETHMUL- acoustic measurements at Shishaldin volcano,
LER , M. L. & C AMPOS , J. 2004. Flow in the negative Alaska. Bulletin of Volcanology, 68, 611–630.
wake of a Taylor bubble rising in viscoelastic V ERGNIOLLE , S., B RANDEIS , G. & M ARESCHAL , J. C.
carboxymethylcellulose solutions: particle image 1996. Strombolian explosions. 2. Eruption dynamics
velocimetry measurements. Journal of Fluid Mech- determined from acoustic measurements. Journal
anics, 511, 217– 236. of Geophysical Research-Solid Earth, 101,
S OUSA , R. G., R IETHMULLER , M. L., P INTO , A. & 20449–20466.
C AMPOS , J. 2005. Flow around individual Taylor W ALLIS , G. B. 1969. One-Dimensional Two-Phase Flow.
bubbles rising in stagnant CMC solutions: PIV measure- McGraw-Hill, New York.
ments. Chemical Engineering Science, 60, 1859–1873. W ELSH , S. A., G HIAAISAAN , S. M. & A BDEL -K HALIK ,
S OUSA , R. G., P INTO , A. & C AMPOS , J. 2006. Effect of S. I. 1999. Countercurrent gas-pseudoplastic liquid
gas expansion on the velocity of a Taylor bubble: two-phase flow. Industrial & Engineering Chemistry
PIV measurements. International Journal of Multi- Research, 38, 1083–1093.
phase Flow, 32, 1182– 1190. W HITE , E. R. & B EARDMORE , R. H. 1962. The velocity
S WANSON , D. A., D UFFIELD , W. A., J ACKSON , D. B. & of rise of single cylindrical air bubbles through
P ETERSON , D. W. 1979. Chronological narrative of liquids contained in vertical tubes. Chemical Engin-
the 1969–1971 Mauna Ulu eruption of Kilauea eering Science, 17, 351–361.
volcano, Hawaii. US Geological Survey Professional X IE , T., G HIAASIAAN , S. M., K ARRILA , S. &
Papers, 1056. M C D ONOUGH , T. 2003. Flow regimes and gas
T AHA , T. & C UI , Z. F. 2006a. CFD modelling of slug flow holdup in paper pulp-water-gas three-phase
in vertical tubes. Chemical Engineering Science, slurry flow. Chemical Engineering Science, 58,
61, 676. 1417– 1430.