Natural Hazards 20: 215–229, 1999.

215
© 2000 Kluwer Academic Publishers. Printed in the Netherlands.

Analysing Lava Risk for the Etnean Area:

Simulation by Cellular Automata Methods

G. M. CRISCI1 , S. DI GREGORIO2? , F. NICOLETTA3, R. RONGO2 and

W. SPATARO2
1 Department of Earth Sciences, 2 Department of Mathematics, 3 Department of Chemistry,
University of Calabria, I 87036 Arcavacata, Italy

(Received: 1 February 1998; in final form: 26 May 1999)

Abstract. The model SCIARA, based on the “Cellular Automata” paradigm, is a versatile instrument
whose scope is to analyse volcanic risk from lava flows.
The possible fields of intervention are:
(a) Long term forecasting of the flow direction at various eruption rates and points of emission by
locating potential risk areas and permitting the creation of detailed maps of risk;
(b) The possibility to follow the progress of an event and predict its evolution;
(c) The verification of the possible effects of human intervention on real or simulated flows in stream
deviation.
A risk scenario has been developed for the Etnean territories of the towns of Nicolosi, Pedara
and S. Alfio, simulating possible episodes with different vent locations along the fracture opened in
the 1989 eruption and successively activated in the 1991–1993 eruption.
The main characteristics of lava flows, that might be dangerous to the inhabited areas, have been
analysed on the basis of the carried out Cellular Automata.

Key words: lava, Cellular Automata, simulation, hazard, risk.

1. Introduction
Cellular Automata (CA), a paradigm of parallel computing, represent an alternat-
ive to differential equations for modelling and simulating some complex systems,
which can be described in terms of local interactions of their constituent parts.
Furthermore, CA are easily and naturally implementable on parallel computers
and exploit the power of parallelism without significant limits (Di Gregorio et al.,
1996).
The forecasting of lava flows was prevailingly limited to qualitative aspects,
before the introduction of methods connected to computer use; in fact, for real
events solving analytically lava flow differential equations is almost impossible
except in very simple cases.
? Address for correspondence: Prof. S. Di Gregorio, Dip. Di Matematica, Univ. della Calabria,
1-87036 Arcavacata (CS), Italy, e-mail: dig@unical.it
216 G. M. CRISCI ET AL.

The use of computational numerical methods correlated to the solution of the

differential equations governing lava flows is, however, very hard. They must be
particularly complex in order to consider, for example, that lava rheology can range
from approximately Newtonian liquids to brittle solids while cooling.
Few computational and numerical flow models have been developed in the past.
They are, however, becoming more significant as the available computer power is
increasing, especially with new parallel machines.
Crisci et al. (1982, 1986) and Barca et al. (1987) designed three-dimensional CA
models; the computation costs at that time did not permit the use of them for large
lava flows and so it was necessary to simplify them in the two dimensional version
SCIARA (Barca et al., 1988a, b, 1993, 1994). This CA approach, in which space
and time are considered discrete, overcomes many of the complexities associated
with differential equations and allows features, such as, multiple flow development
to be simulated.
Using an approach similar to Cellular Automata, Ishihara et al. (1988, 1989)
started from the Navier–Stokes equations and deduced numerical formulations for
discrete space and time intervals; their method, however, cannot be applied to
multiple flows or flows which are extruded intermittently.
Young and Wadge (1990) presented a clever simulation code, faster than those
cited above, but applicable only to simple flow fronts.
This paper presents the main outlines of the last version of the CA model
SCIARA, which is validated only on Etnean lava flows with satisfying results in
terms of spatial and temporal evolution of the lava flow path (the 1986 eruption
(Barca et al., 1994), the first and last phase of 1991–1992 Etnean eruption (Barca
et al., 1994; Crisci et al., 1997)).
Moreover, some significant simulations of possible lava flows are shown and
discussed. They were chosen among hundreds of possible eruption scenarios,
which have been simulated with different vent locations along the fracture opened
in the 1989 eruption and successively activated in 1991–1993. The eruption site is
in the area next to the towns of Nicolosi, Pedara, Tre Castagne and S. Alfio.
The main characteristics of lava flows dangerous to the inhabited environment
are discussed at the end of the paper and the conclusions are given.

2. The CA Model SCIARA for Lava Flows

The CA model for lava flows, SCIARA (Smart Cellular Interactive Automata
Rheology of Aetnean lava flows, to be read as “shea ’rah”), is given by

ASCIARA = (R, L, X, S, σ, γ ),

where
– R = {(x, , y)|x, y ∈ N, 0 6 x 6 lx , 0 6 y 6 ly } is the set of square cells
identified by points with integer co-ordinates in the finite region, where the
phenomenon evolves. N is the set of natural numbers.
ANALYSING LAVA RISK BY CELLULAR AUTOMATA 217

– L ⊂ R specifies the lava source cells.

– The set X identifies the geometrical pattern of cells that influence the cell state
change. They are the cell itself and the “north”, “south”, “east” and “west”
neighbouring cells:
X = {(0, 0), (0, 1), (0, −1), (1, 0), (−1, 0)}.
– The finite set S of states of the ea:
S = Sa × (Sw × St × ST × Sf4 )1 ,
where 1 is the number of lava layers, of which the lava is not miscible because
of the temperature difference.

Sa represents the altitude of the cell;

St represents a parameter correlated to the lava thickness in the cell layer;
ST represents a parameter correlated to the temperature of the lava in the
cell layer;
Sf represents a parameter correlated to lava outflows from the layer toward
the four neighbourhood directions.

σ : S 5 → S is the deterministic state transition for the cells in R

γ : St × N → St specifies the emitted lava from the source cell at the time t.
In this case the set of natural numbers N represents the time intervals of the
CA.
At the beginning of time we specify the states of the cells in R, defining the
initial configuration of the CA. At each following step the function σ is applied to
all cells in R, at the same time the function γ corrects the substate Sh for cells in
L, so that the configuration is changed in time and the evolution of the ASCIARA is
obtained.

3. Main Characteristics of σ
The ASCIARA states are very numerous compared with the states usually involved
in CA simulations, which concern mainly microscopic phenomena. Eruptions are
macroscopic events, but they can be described with a finite number of states at any
approximation. In this case the number of possible altitudes, temperatures and lava
thicknesses is large, but finite.
σ could look as a finite-difference approximation method because of its com-
putational characters, but it has a diversity. Here we adopt principles of qualitative
physics.
The Navier–Stokes equation for the lava flow are not used here, but in the
model we introduce parameters, e.g., the adherence (it rules the viscosity), without
direct corresponding in physics, while other parameters, e.g., the temperature, are
managed in an orthodox way.
218 G. M. CRISCI ET AL.

The evolution of the system is due mainly to the “lava outflows” from the lava
layers of a single cell toward the neighbouring cells and the updating of the cell
“temperature”, which are computed by the transition function.
At each step of the CA, the transition function updates the values of all the
substates in each cell simultaneously.
The transition function for each substate is illustrated in the following, using
sentences, formulas and/or Pascal-like algorithms. Moreover, indexes 0, 1, 2, 3,
4 indicate for a specified neighbouring the central cell and neighbours “north”,
“east”, “west”, “south” respectively.

3.1. THE sa SUBSTATE

The cell altitude remains unchanged until lava begins to solidify. When the tem-
perature of the layer in contact with the terrain becomes equal or less than the
solidification temperature of the lava, the altitude is increased by the width of such
layer.

3.2. THE st SUBSTATE

Since the cell dimensions are constant, the amount of lava accumulated in the layers
can be described in terms of changes in lava thickness.
The new thickness of the lava layers in the cell at time t + 1 is given by the
thickness at time t, plus the contribution determined by lava inflows from the
neighbouring cells, minus the contribution determined by lava outflows from the
cell toward the neighbouring cells at time t.

3.3. THE sT SUBSTATE

Changes in the layer temperature are modelled as a two-step process to describe

changes due to lava motion through a cell and to thermal energy loss from the layer
surface.
The first step considers layer inflows and outflows and simplifies the real situ-
ation by averaging the temperature T [i] of the residual lava res_lava in the layer
and the incoming flows f i[i] of its neighbours.
P4
res_lava × T [0] + i=1 (f i[i] × T [i])
Tav = P4
res_lava + i=1 f i[i]

The second step estimates the temperature drop due to thermal energy losses at
the surface. Assuming that losses through the other sides of the cell are small in
comparison, the rate of energy loss is approximated by the following formula:
q q
T = Tav / 1 + (3T1 σ 1t)/ρcV = Tav / 1 + (T13 pA/V ),
3 3 3
ANALYSING LAVA RISK BY CELLULAR AUTOMATA 219

where ρ is lava density, c the specific heat, V the volume, σ the Stephan–
Boltzmann constant, T the absolute temperature of the surface, A the surface area
of the cell, and  is the surface emissivity. 1t (1t = t2 − t1 ), the time interval, is
the step of the CA, p = 3σ 1t/ρc is the “cooling parameter”.
3σ t/ρc describes the lava’s physical properties, 1t is dependent also on
the side cell dimension, p was determined empirically; furthermore, different
simplifications have been adopted in order to calculate A/V .

3.4. THE sf SUBSTATE

The lava outflow into an adjacent cell from a layer of the cell can be directed only
to a correspondent layer with a temperature difference not exceeding a threshold
value, depending of the lava type. This value corresponds to the temperature dif-
ference that permits the mixing of different temperature lava. If the correspondent
layer does not exist, a new layer in the neighbour will be created on the top.
The outflow depends on the hydrostatic pressure gradients across the cell, due
to differences in altitudes and lava thickness, compared with the correspondent
layers, if any, of the neighbouring cells. The effects of variation in lava thickness
are accounted for by minimising the total difference of height (cell altitude plus
lava thickness of the layer and below layers) between a cell and its four neighbours
after each time interval.
During cooling, a lava rheological resistance is strongly dependent on temper-
ature. The resistance increasing as temperature decreases. Because of complexities
inherent in specifying lava rheology and its variation with temperature, we have
chosen to model rheological resistance in terms of an “adherence parameter” v,
which represents the amount of lava (expressed as a thickness, as discussed above)
that cannot flow out of a cell because of rheological resistance. v is assumed to vary
with temperature T according to a simple inverse exponential relation v = ae−bT ,
where a and b are constants describing lava rheology. Results from earlier simu-
lations of non-Etnean flows (Barca et al., 1987, 1988a, b) and preliminary studies
of Etnean flows, other than the main 1986–1987 flow field, suggest that the Etnean
examples are reasonably represented by v = 0.7 m for T = 1373 ◦ K and v = 7 m
for T = 1123 ◦ K (the nominal solidus). For these limiting conditions, the constants
a and b become 109 m and 0,0154 ◦ K−1 , respectively. A further approximation
(linear function) was effectively utilised in the simulations.
The lava distribution algorithm is based on the following data: la the central cell
lava amount, which may be distributed, z[0] the height of the central cell minus la,
z[i], the heights of the neighbours.
The cells, where the lava cannot flow in, are individuated initially considering
the average height given by
!
X
4
av_height = la + z[i] /5,
i=0
220 G. M. CRISCI ET AL.

Figure 1. Example of the lava distribution algorithm.

where: z[i] greater than av_height indicates that lava cannot flow towards i; so i
must be eliminated from the distribution and from av_height computations.
This computation is iterated with the remaining cells, calculating the new
av_height and eliminating cells with z[i] greater than av_height until there are
no more cells to be eliminated. Then the quantity av_height-z[i] for the remaining
neighbour cells represents the possible lava inflow to the neighbour cell i.
The following example (Figure 1) is considered: la = 15, z[0] = 2, z[1] = 30,
z[2] = 7, z[3] = 13, z[4] = 3.
The Pascal-like procedure in Figure 2 better illustrates the algorithm. eliminat[i]
is a logical variable true when cell i is eliminated, z_sum specifies the sum of
the z plus the quantity of lava to be distributed, new_control is a logical variable
true, when a new av_height must be calculated, poss_fe[i] is the possible outflow
towards the cell i, count counts the remaining cells.
Note that the length of the flow path depends also on the slope, while in the
two-dimensional model the “height” is not given explicitly; in this case, a trivial
correction is introduced in the previous algorithm in order to account for that
problem.

4. Global Parameters
In the previous specification of the transition function appear parameters that char-
acterise globally the CA and, therefore, lava flow in its physical properties. So the
model is endowed with a great flexibility which allows the modelling of many
different types of lava flows.
ANALYSING LAVA RISK BY CELLULAR AUTOMATA 221

Figure 2. Pascal-like procedure up-dating the substate Sf .

The global parameters are:

(a) Lava flow temperature at the vent Tv = 1373 ◦ K, at solidification Tsol = 1123
◦
K.
(b) Lava adherence at the vent adv = 3 m and at solidification adsol = 12 m.
(c) Cooling parameter, p = 5.15e−14 .
(d) Cell edge length, L = 10 m.

The initial and boundary constraints are as follows:

(e) Lava discharge rate.

(f) Ground slope, topography and location of vents.

5. Application Examples Using the SCIARA Program

During the eruptive crisis of 1989 (CNR–GNV, 1990) a fracture was opened be-
ginning from the central vent going towards SSE until height 1500 m near the
222 G. M. CRISCI ET AL.

Figure 3. Topographic map of the study area. Boxes in the lower part show the inhabited
areas. The full line represents the terminal part of the fracture of 1989. The dotted line shows
the morphological barrier (named Tre Monti) which protects the areas from lava flows emitted
in the north of the volcano.

northern side of Mt. Salto del Cane. How it appears from Figure 3, the area that is
interested by the terminal part of the fracture of 1989 is situated slightly northward
of an inhabited area, namely Nicolosi, Pedara, S. Alfio and Tre Castagni. During
the eruptive crisis of 1991/93 there was a reactivating of the superior part of the
1989 fracture. Therefore, it appears important to estimate if a possible activation of
the terminal part of the fracture might, in some way, generate lava flows that could
be dangerous to the inhabited areas of Nicolosi, Pedara, S. Alfio and Tre Castagni.
However, in each case, we try to individualise the eruptive scenario that may be
more dangerous for the specified inhabited areas.
ANALYSING LAVA RISK BY CELLULAR AUTOMATA 223

Figure 4. Simulation of the event that has its origin on the north east side of Mt. Salto del
Cane. The parameters used are shown in the box. The gray shades indicate the different periods
of the lava flow.

To verify our assumption we simulated more than one hundred different epis-
odes initially located in the area between Mt. Salto del Cane, Mt. Serra Pizzuta e
Mt. Arcimis, the terminal zone of the fracture of 1989. Moreover, a certain number
of episodes were simulated south of Mt. Salto del Cane, in the hypothesis of an
extension of the fracture of 1989.
In Figures 4–6 we refer to some results of carried out simulations. Figures 4 and
5 show two examples of simulations realised with the vents situated north of Mt.
Salto del Cane, while Figure 6 shows a simulation with the eruptive centre situated
south of Mt. Salto del Cane.
It appears evident from the examples referred to above that all the eruptive
events simulated have assumed a flux direction towards east and south-east. In no
instance does it involve the inhabited area between Nicolosi and Tre Castagne.
Analysing the areas involved in the simulation, we clearly observe the existence
of a morphological barrier along the direction Tre Monti, Mt. Serra di Falco, east of
Mt. Difesa, west of Mt. Salto del Cane (Figure 3), that prevents lava flows north of
the volcano from intercepting the inhabited areas considered in the present work.
To verify the effects of the lava flow, on the areas considered, coming from
south of the morphological barrier, another series of simulations were done.
Figure 7 shows results from analysis of lava flows originating very near the
morphological barrier, north of Mt Difeso. The parameters used in this simulation
were an average flux of 3 m3 /s for 70 days and for a total volume emitted of about
20 million m3 of lava.
The flow divided into several branches, overtook Mt. Difeso, and took a south-
ward direction towards Pedara (Figure. 7). From the 10th day the various flows
got together and the lava invaded the area beneath Mt. Serra di Falco. The analysis
has revealed a valid basin for the accumulation of lava coming from Mt. Difeso. In
224 G. M. CRISCI ET AL.

Figure 5. Simulation of the lava flow originated north of Mt. Salto del Cane, along the 1989
fracture. The parameters used are shown in the box. The gray shades indicate the different
periods of the lava flow.

Figure 6. Simulation of a lava event that originated south of Mt. Salto del Cane. The para-
meters used are shown in the box. The gray shades indicate the different periods of the lava
flow.
ANALYSING LAVA RISK BY CELLULAR AUTOMATA 225

Figure 7. Simulation of a lava event that originated north of Mt. Difeso, just beneath the
morphological barrier of Tre Monti. The parameters used are shown in the box. The gray
shades indicate the different periods of the lava flow. The lava flow invests the northern area
of the town S. Alfio.

fact, for about 50 days, the lava accumulated in the basin. Near the 60th day (Figure
7) the lava exceeded the morphological threshold south of Tre Monti, proceeding
toward the east. After 68 days from the beginning of the eruption, lava proceeded
to the northern part of the town of S. Alfio (Figure 7).
On the basis of the eruptive analysis we can state that:
(a) events that develop south of the Tre Monti morphological barrier have a
preferential direction toward the east;
(b) the existence of a basin capable of containing large quantities of lava between
Mt. Serra di Falco and Tre Monti (at least 14 million cubic meters of lava);
(c) the northern zone of the area of S. Alfio is a lava flow risk area;
(d) the most suitable site for the construction of a barrier is next to the saddle south
of Tremonti with the twofold aim to postpone the advance towards S. Alfio and
to increase the volume of the “reservoir” existing between Mt. Serra di Falco
and Tremonti.
In Figure 8 we show three moments of a second simulation with an origin more
westward. The vent is positioned barely west of Mt. Arso. This simulation lasted
for 31 days, and the lava flux increased with time. In the first 8 days the flux was
226 G. M. CRISCI ET AL.

Figure 8. Simulation of the lava event that originated east of Mt. Arso. The parameters are
indicated in the box. The gray shades indicate the different periods of the lava flow. The main
branch reaches the north part of S. Alfio, while the secondary branches point towards the town
of Pedara.

averaged at 3.3 m3 /s. The flux was increased slightly (3.4 m3 /s) in the following 23
days. The final flux was 4.14 m3 /s.
From the beginning, the flow has a direction eastward, pointing towards Tre
Monti (Figure 8) passing east of Montepeloso. It is interesting to observe that the
flow does not undertake the direction west of Montepeloso due to a morphological
barrier blocking its advancement. After 20 days (Figure 8) the principal flux goes
towards east to the basin situated between Mt. Serra di Falco and Tre Monti. A
second branch, less fed, develops and follows the preceding one, slightly more
southward. At the 31st day the major branch overtakes the threshold south of Tre
Monti, and flows to the zone north of the town of S. Alfio. The southern branch,
barely alimentated, created a number of minor fluxes that move towards the south,
pointing towards the western part of the town of Pedara (Figure 8). From the results
of this simulation a vent, positioned more westward in areas north of Nicolosi,
does not set the area at risk since the flows are not capable of passing west of
Montepeloso.
In Figure 9 we simulate an eruptive event with a vent west of Mt. Gervasi. In
this case we have assumed initially (Figure 9) that the eruption starts with a high
ANALYSING LAVA RISK BY CELLULAR AUTOMATA 227

Figure 9. Simulation of the lava event that originated east of Mt. Gervasi. The parameters are
indicated in the box. The gray shades indicate the different periods of the lava flow.

flux (12 m3 /s), the flux later decreases rapidly to a very low value and eventually
increases again to 1 and 2 m3 /s.
If the fluxes remain high, they tend to move only towards the east. Quick vari-
ations of the flow imply successions of high and low flows. In conditions of very
low feeding, the cooling produces solidification or, at least, the impossibility of
mixing. This occurs when the flow starts again at high rate, since the old lava
flow represents in these conditions a morphological barrier for the new high rate
lava flow. In our simulation such a barrier prevented lava flows going around Mt.
Peloso towards the west, and thus fluxes went south towards Nicolosi. The town of
Nicolosi was only reached marginally since it was protected on the north-eastern
flank by a big incision. The valley works as a natural defence, towards the east of
the town of Nicolosi. In fact, even if the simulated event reaches a long duration
(127 days), the town of Nicolosi is not invaded by the lava fluxes.

6. Conclusions
Simulations with the SCIARA code, based on the CA paradigm, prove its po-
tentiality of applications in the field of lava hazard mapping. The possibility to
228 G. M. CRISCI ET AL.

realise hundreds of simulations of different vents and with different eruption rate
permits us to verify potential risk areas. For example, the simulations show that the
towns of Nicolosi, Pedara, Trecastagni and S. Alfio cannot be reached from flows
coming from the northern part of the volcano. Flows that are originated south of
the morphological barrier Tre Monti, set the zones north of S. Alfio at high risk.
Nicolosi is a low risk area since a valley situated from north-east to east protects
it naturally. A detailed analysis implies that only a high lava flow rate eruption,
generated in a small area towards the north close to the town, could generate serious
problems to the inhabited area.
The Pedara case seems intermediate. Such locality may collide with secondary
branches of lavas that point in other directions. The simulation of Figure 6 has
indicated the zones where these secondary branches pass, permitting, in this case,
to order an adequate defence in the area.
This offers the possibility to plan interventions in the territory, as the building
of earthen barriers, channels, etc. The flows that invest S. Alfio pass a thin mor-
phological threshold situated south of Tre Monti. The creation in this area of an
earthen dike should delay the advancement of the lava flow for many days.
It is clearly evident that if the flow rate were approximately constant throughout
the event, it would have had an evolution similar to the Figure 8 case. The south-
ward direction is mainly due to the rapid changes in the alimentation of the flow.
The simulated case shows manifestly the extreme importance that the history event
has on the evolution of the lava flow.

References
Barca, D., Crisci, G. M., Di Gregorio, S., and Nicoletta, F. P.: 1987, Lava flow simulation by cellular
automata: Pantelleria’s examples, Proc. AMS Int. Conf. Modeling and Simulation, Cairo, Vol.
4A, pp. 9–15.
Barca, D., Crisci, G. M., Di Gregorio, S., Marabini, S., and Nicoletta, F. P.: 1988a, Nuovo modello
cellulare per flussi lavici:colate di Pantelleria, Bollettino GNV, 4, pp. 41–51.
Barca, D., Crisci, G. M., Di Gregorio, S., Marabini, S., and Nicoletta, F. P.: 1988b, Lava flow simula-
tion by Cellular Automata and Pantelleria’s example, Proc. Kagoshima Int. Conf. on Volcanoes,
pp. 475–478.
Barca, D., Crisci, G. M., Di Gregorio, S., and Nicoletta, F. P.: 1993, Cellular Automata methods for
modeling lava flow: simulation of the 1986–1987 Etnean eruption, In: C. Kilburn and G. Luongo
(eds), Active Lavas (Chapt. 12), UCL Press London (1993), pp. 283–301.
Barca, D., Crisci, G. M., Di Gregorio, S., and Nicoletta, F. P.: 1996, Etnean lava flow simulation
by a cellular automata model, Proc. of 1st European Congress on regional cartography and
information systems, Bologna (Italy), June 13–16 1994 , Vol. 1, pp. 47–49.
Barca, D., Crisci, G. M., Di Gregorio, S., and Nicoletta F. P.: 1994, Cellular Automata for simulating
lava flows: a method and examples of the Etnean eruptions, Transport Theory and Statistical
Physics 23(1–3), 195–232.
CNR-GNV: 1990, In: F. Barberi, A. Bertignani and P. Landi (eds), Mt. Etna: The 1989 Eruption.
Crisci, G. M., Di Francia, A., Di Gregorio, S., Nicoletta, F. P., Rongo, R., and Spataro, W.:
1997, Sciara.2: A Cellular Automata Model for Lava Flow Simulation, Proc. IAMG97, Vera
Pawlowsky Glahn (ed.), Addendum, pp. 11–16.
ANALYSING LAVA RISK BY CELLULAR AUTOMATA 229

Crisci, G. M., Di Gregorio, S., Ranieri, G.: 1982, A cellular space model of basaltic lava flow, Proc.
Kit. Conf. Applied Modelling and Simulation 82, Paris, France, Vol. 11, pp. 65–67.
Crisci, G. M., Di Gregorio, S., Pindaro, O., Ranieri, G.: 1986, Lava flow simulation by a discrete
cellular model: first implementation, Int. J. Modelling and Simulation, 6, 137–140.
Di Gregorio, S., Rongo, R., Spataro, W., Spezzano, G., and Talia, D.: 1996, A parallel cellular tool for
interactive modeling and simulation, IEEE Computational Science & Engineering 3(3), 33–43.
Ishihara, K., Iguchi, M., and Kamo, K.: 1988, Numerical simulation of lava flows at Sakurajima,
Proc. Kagoshima Int. Conf. on Volcanoes, pp. 479–482.
Ishihara, K., Iguchi, M., and Kamo, K.: 1989, Numerical simulation of lava flows on some volcanoes
in Japan, In: J. Fink (ed.), IAVCEI Proc. in Volcanology, Vol. 2, Springer-Verlag.
Young, P. and Wadge, G.: 1990, Flowfront: simulation of a lava flow, Computer & Geosciences 16(8),
1171–1191.