Sie sind auf Seite 1von 22

Journal of Volcanology and Geothermal Research 140 (2005) 273 – 294

www.elsevier.com/locate/jvolgeores

A model for the numerical simulation of tephra fall deposits


T. Pfeiffer a,*, A. Costab,1, G. Macedonioc
a
Department of Earth Sciences - University of Aarhus, Denmark
b
Dip. di Scienze della Terra e Geologico-Ambientali, Univ. di Bologna, Italy
c
Osservatorio Vesuviano INGV, Italy
Received 1 September 2003; accepted 2 September 2004

Abstract

A simple semianalytical model to simulate ash dispersion and deposition produced by sustained Plinian and sub-Plinian
eruption columns based on the 2D advection–dispersion equation was applied. The eruption column acts as a vertical line
source with a given mass distribution and neglects the complex dynamics within the eruption column. Thus, the use of
the model is limited to areas far from the vent where the dynamics of the eruption column play a minor role. Vertical
wind and diffusion components are considered negligible with respect to the horizontal ones. The dispersion and
deposition of particles in the model is only governed by gravitational settling, horizontal eddy diffusion, and wind
advection. The model accounts for different types and size classes of a user-defined number of particle classes and
changing settling velocity with altitude. In as much as wind profiles are considered constant on the entire domain, the
model validity is limited to medium-range distances (about 30–200 km away from the source).
The model was used to reconstruct the tephra fall deposit from the documented Plinian eruption of Mt. Vesuvius, Italy,
in 79 A.D. In this case, the model was able to broadly reproduce the characteristic medium-range tephra deposit. The
results support the validity of the model, which has the advantage of being simple and fast to compute. It has the
potential to serve as a simple tool for predicting the distribution of ash fall of hypothetical or real eruptions of a given
magnitude and a given wind profile. Using a statistical set of frequent wind profiles, it also was used to construct air fall
hazard maps of the most likely affected areas around active volcanoes where a large eruption is expected to occur.
D 2004 Elsevier B.V. All rights reserved.

Keywords: ash fall; settling velocity; computer model; Vesuvius 79 A.D.

1. Introduction

Armienti et al. (1988) and Macedonio et al.


(1988) presented a numerical 3D model of volcanic
* Corresponding author.
E-mail addresses: tpfeiffer@decadevolcano.net (T. Pfeiffer)8 tephra fallout. The model assumed that sufficiently
costa@ov.ingv.it (A. Costa)8 macedon@ov.ingv.it (G. Macedonio). far from the vent, the eruption column can be
1
Now at Osservatorio Vesuviano INGV, Italy. regarded as a vertical line source and that the
0377-0273/$ - see front matter D 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jvolgeores.2004.09.001
274 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

motion of particles can be sufficiently described by ! 3D modelling is time intensive to compute,


atmospheric advection, diffusion, and settling by whereas this simplified semianalytical model can
gravity. The model numerically solves the 3D perform thousands of runs in few tens of
advection–diffusion–sedimentation equation, assum- minutes on a PC.
ing constant atmospheric eddy diffusion coeffi-
cients. It was applied to the 1980 Mt. St. Helens In addition, we tested the reliability of the model
and 79 A.D. Vesuvius eruptions successfully. and the dependence of the results against the quality/
Because of its 3D nature, the model requires quantity of the input data.
numerical simulation in three spatial and one
temporal dimension. This requires considerable
computing time and limits the user to a small 2. The physical model
number of possible runs with different parameters
each time. In this study, a modified version (called Beyond a certain distance from the dynamic
HAZMAP) is presented that further develops eruption column, the dispersion and sedimentation
Macedonio et al. (1988)’s model into a simplified of tephra is governed mainly by wind transport,
semianalytical 2D model useful for a first approach turbulent diffusion, and the settling of particles by
to reconstruct tephra fall deposit (for a more gravity (Armienti et al., 1988; Macedonio et al.,
detailed computational description of the HAZMAP 1988). The motion of particles can be described by the
model see Macedonio et al., (2004)). We have mass conservation equation as follows:
modelled the same Plinian fall deposits of the
Vesuvius 79 A.D. eruption, as Macedonio et al. BCj BCj BCj BUsj Cj
þ Wx þ Wy 
(1988). The main differences between the two Bt Bx By Bz
models are summarized as follows: B2 Cj B2 Cj B2 Cj
¼ Kx þ Ky þ Kz þ Sj ð1Þ
Bx 2 By 2 Bz2
! The 3D version of the diffusion–advection-settling
problem (Eq. (1)) takes account of the vertical where C j is concentration of particles; t the time; x,
component of atmospheric diffusion. In this study y, z the spatial coordinates; Wi the horizontal wind
vertical diffusion and the vertical wind component field (i=x,y); K x and K y are the horizontal
are neglected. atmospheric eddy diffusion coefficients assumed
! Macedonio et al. (1988) used a single parameter- equal so that K x =K y =K and U s the settling velo-
ization for all settling velocity classes to describe city; and S j is a source function. Eq. (1) is valid for
the settling velocity dependence with altitude. In each class j of particles having a given settling ve-
this study, we adopted a more general ash settling locity U sj.
velocity model based on the experimental data of In a first-order approach, we assume that the
Walker et al. (1971) and Wilson and Huang vertical diffusion component is negligible with
(1979). respect to the horizontal ones, and that the
! Macedonio et al. (1988) accounted for the tempo- horizontal wind components are constant in time
ral variation of the column height (by independent and within the horizontal domain. This assumption
models). We assumed the same eruption column holds for intermediate distances of the order of
height for each phase and an instantaneous mass 100 km or more but becomes less accurate over
release. large distances. Moreover, it is assumed that the
! Macedonio et al. (1988) used a scaled and eruption column is high enough to neglect terrain
rotated measured summer wind profile estab- effects.
lished by statistical observations in southern Under these assumptions, dividing the vertical
Italy. In this study, we use several wind computational domain into N layer layers on which
models and test a parameterized wind profile, settling and wind velocity is assumed constant, as
which is best fitted against the observed field in Macedonio et al. (2004), the total mass on the
data. ground is computed as the sum of the contributions
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 275

Table 1 sufficiently far from the vent. The results from


Definitions of the symbols used Macedonio et al. (1988) and Armienti et al. (1988)
Symbol Definition suggest that the critical distance is approximately
(W x ,Wy ) Horizontal wind field given by the height of the eruption column itself.
K=K x =K y Horizontal diffusion coefficient Variations of the eruption column with time are
U sj Settling Velocity of the j-th particle class
thus replaced by a time-averaged column described
Sj Source function for the j-th class
M G(x,y) Total mass on the ground at (x,y) by a vertical mass distribution function f h(z).
Mi Total mass emitted from the i-th point source Describing the vertical mass distribution in an
fj Mass fraction of the j-th settling velocity class eruption column has been a major goal of a recent
(x Gi,y Gi) Centers of the Gaussian mass distribution research, but in our model, a purely empirical
r Gi Variances of the Gaussian mass distribution
treatment is adopted. Modifying the formula sug-
N vs Total number of settling velocity classes
N sources Total number of point sources gested by Suzuki (1983) and used in Armienti et al.
Dz k Thickness of the k-th layer (1988) and Macedonio et al. (1988), the vertical
Dt k Residence time of particles in the k-th layer mass concentration is assumed the same for all
particle classes as given by:
from each of the point sources distributed above the n z ok
vent and from each settling velocity class: S ð x; y; z; t Þ ¼ S0 1  exp½ Að z=H  1Þ
H
 dðt  t0 Þdðx  xv Þdðy  yv Þ ð3Þ
XNvs NX
sources
Mi fj
MG ð x; yÞ ¼ exp where S(z)={1z/H exp [A(z/H1)]} k is the
j¼1 i¼1
2pr2Gi
" # vertical mass distribution function, z the altitude in
ðx  xGi Þ2 þ ðy  yGi Þ2 the eruption column, S 0 a normalization factor, H
  ð2Þ the maximum plume height, A and k are two
2r2Gi
dimensionless parameters, and d is the Dirac’s
P Y P distribution (filiform and instantaneous release
where xYGi ¼ xY0i þ k W k Dtk and r2Gi ¼ 2K k Dtk assumption). Eq. (3) is applied as an empirical
are the centre and the variance of the Gaussian, description of the vertical mass distribution within
respectively, calculated as in Macedonio et al. the eruption column with a purely geometric
(submitted for publication) (Dt k =(z k z k1)/v s,k ). meaning. The value of the parameter A (bSuzuki
Moreover, N sources indicates the source points, coefficientQ) describes the vertical position of the
N vs the total number of settling velocity classes, maximum concentration relative to the maximum
M i is the total mass
P emitted from the point source column height, located at (A1)/A of the max-
in the layer i ( i Mi ¼ Mtot , with M tot total mass imum plume height (Fig. 1a). The parameter k is
injected into the system) and f j is the fraction a measure of how closely the total mass is
of that
P mass belonging to the settling velocity class concentrated around the maximum at H(A1)/A
j ( j fj ¼ 1; for the computational details see (Fig. 1b).
Macedonio et al., (2003)). A summary of the Theoretical and empirical observations on buoy-
definitions of the symbols previously used is ant plumes (e.g., Morton et al., 1956; Sparks, 1986)
reported in Table 1. show that the ratio H B/H T between the height of
neutral buoyancy of the plume H B and the
maximum height H T is usually around 3/4. We
3. Parameterizations and input data try to account for this by considering A to be about
4 and varying k in the range 1–5.
3.1. Eruption column
3.1.1. Grain size and settling velocity
The simplification of the eruption column as a Rosin’s law is often used to explain the size
linear source limits the use of the model to areas distribution of erupted pyroclastic material (Suzuki,
276 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

Fig. 1. Column shape: mass distribution models inside column.

1983). This probability density function is given Consistent with Armienti et al. (1988), the
by: present model assumes that all particles are always
" # traveling at their terminal settling velocity. This is
1 ð U  lÞ 2 justified as long as terminal fall velocities are of the
f ðUÞ ¼ pffiffiffiffiffiffiffiffiffiffiffi exp  ð4Þ
2prU 2r2U order of a few tens of meters per second, or less.
In the case of typical volcanic particles in air,
where d=2U , mm is the particle diameter, l the buoyancy forces can be neglected. The terminal
median value, and r U the standard deviation of the size settling velocity of a particle is given by the basic
distribution. Using this parameterization for each type equation when the weight of the particle is balanced
of component, we are able to describe the settling by the aerodynamic drag force:
velocity spectrum with only two parameters for each
component (l and r U ).
1
The bulk of the erupted mass is split into a user- mg ¼ CD qa Acs Us2 ð5Þ
defined number of grain size and classes of 2
components. where m is the mass and A cs a typical cross-sectional
The settling velocity determines the deposition of area of the particle, C D the dimensionless drag
all tephra and has a first-order importance for tephra coefficient, q a the density of air, U s the settling
deposition. In any quantitative model of fallout, an velocity, and g the acceleration due to gravity. The
accurate description of the settling velocity of drag coefficient itself is a complex function of particle
particles is critical. shape and the Reynolds number Re=q adU s/g a, which
The settling velocity of volcanic particles is a is a measure of the relative importance of inertial
complex function of particle size, shape, and density. versus viscous forces governing turbulent and laminar
It also depends on the density and viscosity of the flow, respectively (d is a bcharacteristicQ particle
surrounding air. Its value can only be computed diameter and g a the viscosity of air).
approximately and relies heavily on experimental and At low Reynolds numbers (Reb0.1–1) viscous
empirical data. While for certain regular and industrial forces dominate. For spherical particles (Landau and
shapes a large amount of experimental aerodynamic Lifchitz, 1971):
data is available, only a few studies have directly
measured the fall velocities of volcanic particles, the
most significant ones being by Walker et al. (1971) B
CD ¼ ; B ¼ 24 ð6Þ
and Wilson and Huang (1979). Re
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 277

This relation is equivalent with Stokes’ law for the Stokes’ law implies that, as long as Reynolds
settling velocity: numbers are very low, terminal fall velocity does
not change with air density. This has a significant
Us ¼ qd 2 g=18ga ð7Þ effect. While larger particles increase their terminal
fall velocity rapidly with decreasing air density,
where q is particle density. i.e., increasing altitude, a small fraction of tephra
The typical cross-sectional area A cs in Eq. (5) remains in the regime of low Reynolds numbers
and the characteristic particle diameter d in the and falls much more slowly out of the plume.
Reynolds number definition are not easy to determine This effect is greater the higher the eruption
for irregular shaped bodies. They depend not only on column and could be a major reason for trans-
the shape of the particle but also on the direction of portation to very long distances of tephra in high
the air flow around it and its turning mode. An exact plumes.
treatment is impossible, and it is more convenient to For particles falling at high Reynolds numbers
define d and A cs as the diameter and cross-section (ReN1000), inertial forces are dominant, and the drag
of an equivalent sphere, which gives: A cs=Wd2/4. coefficient no longer depends on Reynolds number.
Generally, d is not the real mean diameter of the Typically, it assumes a constant value for Mach
particle defined by the average of the three main numbers below about 0.7 (Aschenbach, 1972). The
axes. For subspherical shapes, taking d as the value of C D depends on the shape and orientation of
diameter of an enclosing sphere or a sphere with the particle. From data in the literature (e.g, Mironer,
the same volume as the particle will be a good 1979), drag coefficients for some basic shapes at
measure (Wilson and Huang, 1979). For elongated ReN1000 are:
particles, d and A cs depend very much on the
orientation and rotation modes of the particles, and
their appropriate values might vary widely. ! 0.44 for spheres
Wilson and Huang (1979), who measured the ! 1.2 for cylinders
terminal fall velocities of hundreds of small volcanic ! 0.8–1.05 for cubes
particles in the range 0.1bReb100, showed that it is
sufficient to define the typical particle diameter as the Very few experimental data exist for irregular-
actual diameter if at least one additional parameter is shaped volcanic particles falling at high Reynolds
used to describe fall velocity. They introduced the numbers. The experimental studies of Walker et al.
shape factor F=(a+b)/2c, where abbbc are the three (1971), however, suggest that drag coefficients for
principal diameters of the particle. Then, they showed volcanic particles at ReN1000 are also more similar
that the drag coefficient C D of about 80% of hundreds to those of cylinders than spheres. This is confirmed
of investigated tephra particles could be expressed as a by an analysis of their experimental data (Fig. 2).
simple function of mean diameter d, shape function F, As a result, drag coefficients are expected to be
and Reynolds numbers: around C D=1. In the case of intermediate Reynolds
24 0:828 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi numbers, the description of the drag coefficient
CD ¼ F þ 2 1:07  F ð8Þ becomes very difficult. Most authors of similar
Re
studies have used intermediate analytical expres-
Suzuki (1983) proposed a modification to this sions between the settling laws for low and high
formula which appears to be a better fit to both Wilson Reynolds numbers regimes or approximations of
and Huang (1979)’s experimental data for submillim- experimental data; for instance, Bonadonna et al.
eter clasts and the (less comprehensive) data by (1998) used Kunii and Levenspiel (1969)’s analyt-
Walker et al. (1971) of centimeter-sized pumice and ical expressions and Macedonio et al. (submitted for
crystals from Fogo and Askja: publication) used the formula given by Arastoopour
et al. (1982).
24 0:32 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi In this study, we have adopted a model based on
CD ¼ F þ 2 1:07  F ð9Þ
Re Walker et al. (1971) and Wilson and Huang
278 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

Fig. 2. Experimental data of settling velocities from Walker et al. (1971).

(1979)’s data. Because it is based on the largest set the following fitting curves that were also adopted
of relevant experimental data, Wilson and Huang here:
(1979)’s equation (Eq. (8)) with a shape factor
F=0.43 (the mean value of their observed shape qa ¼ q0 ½1  0:0226zðkmÞ4:255 below 11 km ð10Þ
factors, calculated by Macedonio et al., 1988) was
applied for Reb100. This is the range of the qa ¼ q1 exp½  0:157zðkmÞ  10:8 above 10:8 km
observed Reynolds number in the experiments. For
high Reynolds numbers ReN1000, as suggested by
and
Walker et al. (1971)’s experimental results, C D 2 33=2
was taken as C D=1. This is significantly higher T0 þ 120
than in the other mentioned studies, but as ga ¼ g0 4 T0  6:5zðkmÞ 5
T0  6:5zðkmÞ þ 120 T0
discussed, it appears to be closer to the true
value. For intermediate Reynolds numbers, 100b ð11Þ
Reb1000, the drag coefficient was linearly inter-
polated (Fig. 2). where q 0=1.022 kg/m3 is the density of air at sea
level, q 1 is the air density at 10.8 km, g 0=1.79105
3.1.2. Variation of settling velocity with altitude. Pa is the viscosity at T 0=288 K, and z is the
Results altitude in km.
Air density and viscosity change significantly As recognized by Bonadonna et al. (1998), the
with altitude, and as a consequence, terminal- variation of settling velocity with altitude strongly
settling velocity generally increases (Fig. 3). On depends on Reynolds numbers and is therefore
the other hand, Reynolds numbers decrease with different for different particle size and density
increasing height, and thus, particles might be classes.
subject to several regimes during their fall (see Small particles falling at low Reynolds numbers
Fig. 4). show almost no increase in settling velocity with
In this model, settling velocity is calculated for altitude because they fall according to Stokes’ law
any altitude. Density and viscosity of air are (see Eq. (7)), where fall velocity varies only with
calculated assuming the known values of the viscosity (see Fig. 3). Above the tropopause, the
International Standard Atmosphere (Smithsonian temperature of the atmosphere remains fairly constant,
Institution, 1951). Armienti et al. (1988) presented and thus, so does the air viscosity.
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 279

Fig. 3. Comparison of different settling velocity models.

In contrast, particles at high Reynolds numbers A significant implication is that particles with
fall with a constant drag coefficient C D according the same settling velocity at ground level but with
to Eq. (5), where terminal fall velocity is proportional different densities and diameters could not be
to the inverse of the square root of air density. treated as one settling velocity class. Smaller
Therefore, an accurate computation of fall velocity particles with higher densities will pass through
has to take into account the different settling laws of the transition to lower Reynolds number regimes at
any granulometric class and also the changes of lower altitudes than larger, less dense ones and
settling velocity with altitude. therefore increase their terminal settling velocity
280 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

Fig. 4. Changing Reynolds number regimes and settling velocity of particles of different U sizes and densities during their fall. Note that, for
small particles, the settling velocity increases much less with altitude due to the low Reynolds numbers at which they fall from different heights.
Settling velocity has been calculated using the model of this study.

with height less sharply. For instance, particles of while particles of 750 kg/m3 with the same settling
2500 kg/m3 density with fall velocity of 1.5 m/s velocity at sea level only cross this boundary at
pass through the Re=1 boundary at around 30 km, around 40 km (see Fig. 3). The effect is very small
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 281

in as much as the transitions typically occur very model of Rose et al. (1983). Their estimates for the
gradually. This should allow a parameterization of White and the Gray layers, respectively, are 2.51012
settling velocity altitude dependence for each and 6.51012 kg of tephra, or 1 and 2.6 km3 DRE,
velocity class at sea level. assuming a magma density of 2500 kg/m3. Based on
field data with at least 10 data points for each level,
partly published in Sigurdsson et al. (1985), Carey
4. Applications of the model to documented and Sigurdsson (1987) applied Carey and Sparks
eruptions: Mt. Vesuvius, Italy, 79 A.D. eruption (1986)’s quantitative model of the relationship
between column height, wind speed, tephra distribu-
4.1. Introduction tion, and grain size characteristics of fall deposits to
determine the column height and its variation with
The 79 A.D. Plinian eruption of Mt. Vesuvius is time for both the White and the Gray Pumice phase.
certainly the most well-known ancient eruption. It is The maximum downwind and crosswind ranges of
famous for the destruction of the Roman towns of lithics were determined from the maximum lithic
Pompeii and Herculaneum and for the detailed isopleths based on all available field sites (in total 45).
account by the Roman lawyer Pliny the Younger These isopleths should correspond to the maximum
who described the events where his famous uncle heights of the eruption column reached in each phase.
Pliny the Elder died. The volcanological aspects of the There were sufficient data to establish the critical
79 A.D. eruption have been described by numerous dimensions of the 6.4, 3.2, and 1.6 cm lithic isopleths,
authors, most notably Sigurdsson et al. (1985) and which could be compared with the Carey and Sparks
Carey and Sigurdsson (1987). On the basis of the (1986) model. Carey and Sigurdsson (1987) found
historic account by Pliny and volcanological studies that the data were in good agreement with the model
of the well-preserved tephra deposits, the mentioned predictions for the 10-year average summer wind
authors presented a qualitative and quantitative profile for Brindisi, Italy, reported in Cornell et al.
reconstruction of the eruption. The eruption com- (1983). To establish the temporal evolution of the
prised several phases: column, they subdivided each layer into several
normalized chronostratigraphic levels and applied
! A short initial vent-opening phase with the the same technique.
deposition of breccia and phreatomagmatic According to their results, the column increased
products. from about 15 to 26 km during the White Pumice
! A purely magmatic phase with a high sustained phase. During the Gray phase, it increased to a
eruption column. During this phase, a graded maximum of 32 km and then decreased to about 27
Plinian pumice fallout layer was deposited, which km. Using Sparks (1986)’s model, Carey and Sparks
is divided into a lower layer of white phonolitic (1986) estimated magma discharge rate from the
pumice (called bWhite pumiceQ) and an upper height of the eruption column. The duration and the
tephritic–phonolitic pumice fall (bGray pumiceQ) temporal evolution of the fallout phases (ca. 7 h for
deposit. This phase ended column collapse and the the White Pumice, 11 h for the Gray Pumice) could
emplacement of pyroclastic flows. then be evaluated from the independently obtained
! Phreatomagmatic activity and the emplacement of values of the total mass and discharge rate. This result
lithic-rich breccia and pyroclastic surges, inter- was then compared and refined with the time scale
preted as the consequence of decreasing pressure deduced from the accounts of Pliny made by
in the vent and magma–water interaction. Sigurdsson et al. (1982, 1985).
! Increasingly bwetQ deposits as wet surges and mud
flows were formed as a consequence of decreasing 4.2. Input parameters
intensity and magma/water ratio.
To compare the model response to various input
Sigurdsson et al. (1985) estimated the volumes of parameters, a series of runs with different wind and
the White and the Gray Pumice fallout layers from the original grain size distribution input parameters were
282 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

performed. The ranges used in the best fitting sampling point is representative of the eruption
procedure for the different input parameters are column that generated the flow. In this model, both
discussed in the following. The final used values are this and a best fitted settling velocity distribution Eq.
summarized in Table 2. (4) according to Rosins law were applied.
(1) Eruption column height. In as much as the (5) Settling velocity of particles. As in Macedonio
height of the eruption column varied during the et al. (1988), the model described both of the two
eruption, the model was given the following ranges principal components present in the deposits, i.e.,
that represent the lower and higher ends of the juvenile pumice and lithic fragments. Particles from
estimates by Carey and Sigurdsson (1987): 10–27 each component were divided into 10 (if the pumice
km for the White pumice phase and 20–33 km for the flow composition was used) or 20 grain size classes
Gray phase. covering the interval from U=5 (32 mm) to U=5
(2) Total erupted mass. Sigurdsson et al. (1985) (0.032 mm) in the case of the Gaussian model.
estimated the erupted volumes of the White and Gray The density of each component grain size class was
pumice phases as 1 and 2.6 km3 dense rock obtained by interpolation of the measured values
equivalent, assuming a density of 2500 kg/m3 for given in Macedonio et al. (1988). For each class, the
both pumice and lithic components. This corresponds settling velocity was calculated using the model
to a mass of 2.51012 and 6.51012 kg, respectively. previously described (see Section 3.1.1).
These values were also used by Macedonio et al. (6) Diffusion coefficients. Macedonio et al. (1988)
(1988) in their modelling. In this model, the total observed the best fitting for horizontal diffusion
erupted mass is automatically obtained by performing coefficients of K=3000 m2/s. In this study, the best
the best fitting of the calculated deposit with the fitting diffusion coefficients greatly exceeded the
observed one. highest value permitted to the model simulations in
(3) Mass distribution along the column. Macedo- the range 1000–10,000 m2/s. Concerning the horizon-
nio et al. (1988) used the modified formula of Suzuki tal diffusion coefficient K in HAZMAP, it is not
(1983) and obtained the best fitting with the factor properly the turbulent atmospheric diffusion coeffi-
A = 4.2 for both the White and Gray phases. In this cient, but when it is fitted by the observed thicknesses,
model, the parameter A was fixed at A = 4. Thus, the it represents an empirical parameter of the model that
position of maximum mass concentration was defined accounts for all the physical effects that dispersed the
at 3/4 of the maximum column height. To improve the volcanic particles. When the amount of the field input
model, the modified version Eq. (3) was used, where data (thicknesses, granulometries, etc.) is not suffi-
the additional factor k was introduced. k was varied in cient to well constrain the model parameters, the
the range 1VkV5. model could require K values belonging not just to
(4) Grain size population in the eruption column. physical ranges, affecting also the reliability of the
Macedonio et al. (1988) determined the grain size other parameters. Therefore, in our case, its value was
distribution of the eruption column on the basis of the fixed at 5000 m2/s as typical values for the involved
(approximately Gaussian-shaped) grain size distribu- scales.
tion observed in a pumice flow emplaced immediately (7) Wind field. The wind field at the time of the
after the end of the Gray pumice phase. In as much as eruption is unknown and can only be vaguely
this flow has been interpreted as related to the total reconstructed from field data. The observed fallout
column collapse, it is reasonable to consider this axis of the deposit indicates winds from the NW in the
composition of the flow deposit at the sampling point troposphere during the eruption. The best studied and
as a representative of the original column that described part of the deposit is proximal (within 10
discharged the Gray pumice. From the overall km), too close to yield reliable information about
characteristics of the deposits, they inferred that the high-level winds, and the direction of the dispersal
White pumice grain size distribution was similar to axis beyond the proximal area is only defined by very
that of the Gray pumice with a slightly higher pumice/ few data points all within about 50 km distance. Thus,
lithics ratio. It is, however, difficult to assess whether there is no immediate information about the direction
the component grain size composition of an individual of the upper level winds that transported most of the
Table 2
Summary of input and output parameters
Run: V1 V2 V3 V4 V5 V6
White Gray White Gray White Gray White Gray White Gray White Gray

T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294
Wind modela One direction, wind speeds from autumn wind profile from Rotated and scaled summer (V3) wind profile (W3) Parameterized wind profile (W2)
Cornell et al. (1983) (W1)
Wind direction or 3228 NW 3118 NW 3228 NW 3178 NW 418 cw 378 cw 608 cw 608 cw 2408 NW 3558 NW 3398 NW 3558 NW
rotation of profile (b11 km) (b11 km) (b11 km) (b11 km)
2718 W 2708 W 2748 W 2738 W
(N15 km) (N15 km) (N15 km) (N15 km)
Wind scaling factor 1.0 1.1 0.8 0.9 1.1 1.3 0.4 0.4 – – – –
Wind scaling at 11 21/16 m/s 23/17 m/s 17/12 m/s 19/14 m/s 31/10 m/s 37/12 m/s 11/4 m/s 11/4 m/s 27/25 m/s 24/25 m/s 24/10 m/s 27/25 m/s
km/N15 km
Initial grain-size Pumice flow Gaussian U-distribution, Pumice flow (Macedonio et al., 1988) Gaussian U-distribution,
populationb: (Macedonio et al., 1988) l(U) / r(U) / wt.% l(U) / r(U) / wt.%
pumice (P) P: 1/2/85% P: 1/1.5/64% P: 5/1.5/88% P: 1/2/79%
and lithics (L) L: 3/3/15% L: 1.5/2/36% L: 1.5/1/12% L: 1/1/21%
Setting velocity This modelc Macedonio et al. (1988) This model
Total mass [kg] Obtained by best fittingd Fixed Obtained by best fitting
9.61011 6.61011 6.81011 7.51011 1.01012 1.11012 2.61012 6.51012 8.31011 1.21012 5.01011 7.51011
Column height 27 km 33 km 27 km 33 km 20 km 33 km 23 km 30 km 21 km 33 km 27 km 33 km
Suzuki factors A/k 4/1 4/1.5 4/2 4/3 4/1 4/1 4/1 4/1 4/2.5 4/1 4/4 4/1
Diffusion coefficient 5000 m2/s 5000 m2/s 5000 m2/s 5000 m2/s 5000 m2/s 5000 m2/s 3000 m2/s 3000 m2/s 5000 m2/s 5000 m2/s 5000 m2/s 5000 m2/s
Number of 10 20 10 20
U-classes
r. source pts. 50 300 50
Number of data 16+2*5=26 17+2*5=27 26 27 26 27 26 27 26 27 26 27
Degree of freedom 20 21 15 16 20 21 26 27 18 19 13 14
v 2 total deposit 38.01 71.61 14.45 49.03 18.73 64.10 628.7 698.9 17.01 33.59 15.44 26.88
2
v grain-size dist 10.52 12.19 3.86 14.52 11.09 10.87 15.98 133.1 6.62 11.71 3.87 9.84
v 2 weighted 43.03 71.92 16.19 56.65 28.30 64.32 522.74 718.8 21.56 40.92 17.00 33.31
v 2 reduced 2.15 3.42 1.08 3.54 1.42 3.06 20.1 26.6 1.20 2.15 1.31 2.38
a
For a detailed description of the different wind models, see text.
b
For runs V1, V3, V4, V5, the grain size distribution of the pumice flow deposit at the end of the Gray pumice phase reported in Macedonio et al. (1988) was taken as input grain size distribution for the
model eruption column.
c
For a detailed description of the settling velocity model in all runs except V4, see previous chapter. For run V4, the same simplified formula given in Macedonio et al. (1988), which uses an approximate
exponential altitude-dependent settling velocity: V set(z)=(4dq pg/3Dq a)1/2 exp(0.024 z), where z is altitude in km, d particle diameter, q p particle density, q a air density, g the gravitation constant, and D the
drag coefficient taken as 3.6.
d
For detailed description of the weighting and error procedure used for the best fitting, see text.

283
284 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

fine material. The date of the eruption, 24 August 79 models have been applied (see Fig. 5) based on both
A.D., places it in the last month of summer, the average summer and autumn wind profiles.
suggesting a wind profile similar to either a typical
summer or autumn profile for the region. A compar- ! Wind model 1 (W1): In W1, only one wind
ison of seasonal summer and winter profiles recorded direction is considered. The absolute values of
at Brindisi in southern Italy (reported in Cornell et al., wind speed of the autumn wind profiles are taken,
1983) shows that both profiles are generally similar rotated, and scaled uniformly (the corresponding
but differ significantly in the direction of the upper runs were called V1 and V2). In the summer
wind levels above 15 km (see Fig. 5). In the lowest profile, wind speed increases rapidly to about 30
levels of the atmosphere, the most common directions m/s near the tropopause, about 50% greater than
are west–southwest during winter and spring, and the autumn winds, and then decreases rapidly to
northwest–west during summer and autumn. There- values between 5 and 10 m/s, whereas the upper
fore, both the summer and the autumn wind profiles wind speeds in the autumn profile drop only to
are in qualitatively good agreement with the deposit. about 10 m/s and then rise gradually again to about
For the numerical reconstruction of the deposit and 20 m/s. Not considering the effects of changing
the evaluation of the model, three different wind direction with altitude, both profiles produce

Fig. 5. Different wind models used in the different runs. Wind intensity is shown on the left and wind direction on the right.
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 285

almost identical results in the model. Output for each class (including particle density and shape
values differ typically by less than 5% if the same errors).
set of input parameters is used. For W1, the values To account for the small number of available
of the autumn profile with its less pronounced grain size data with respect to the total deposit
peaking behavior at the tropopause have been data, the cumulative total deposit error was
applied. weighted 0.8 against 1.2 for the cumulative error
! Wind model 2 (W2): W2 uses a rotated and scaled of grain size data. It was estimated that the model
user-defined base profile. The summer wind completely fails to be justified for distances less
profile has been adopted as the base profile. Runs than 10 km from the vent. One data point (location
which use this profile were called V3 and V4. 1) was therefore deliberately excluded. Points
However, in as much as the autumn wind profile between 10 and 30 km away from Vesuvius were
has a near constant wind direction, its use is almost progressively weighed, and full weight was applied
equivalent to wind model W1. to all points beyond 30 km. The individual weights
! Wind model 3 (W3): W3 (run V5, V6) is a were normalized.
parameterized wind profile that was developed in In as much as the exact evaluation of the relative
order to test the possible influence of different error which affects the input data is not possible, it is
wind directions at high altitude. A constant wind possible that our error considerations overestimate the
direction h 1 for all levels up to 11 km and a btrueQ one, but in any case, this is not relevant for our
different wind direction h 2 for levels above 15 km aim. The v 2 method was chosen to be a reasonable
are free parameters. Wind speed increases linearly minimization procedure for the fitting. The final
to a maximum value W s1 at the 11 km level. function that we have chosen to minimize is:
Above 15 km, it assumes a constant value W s2.
Wind speeds and directions at the intermediate v2 ¼ 0:8v2deposit þ 1:2v2granulometric ð12Þ
levels are interpolated.
4.3. Results
4.2.1. Error considerations for the fitting procedure
For the purpose of fitting, suitable error intervals A general summary of inputs used and results is
for all variables used must be established. For both given in Table 2. Figs. 7–9 show the results of the
units, the deposit thickness values at 18 and 19 sites computed isopachs, and in Fig. 10, the comparison
for the White and Gray deposits and grain size between simulated and observed granulometric dis-
analysis data at three individual sampling points tributions in three points where they are available is
were used, as reported in Macedonio et al. (1988). shown.
To calculate the cumulative error between the model Using a single wind direction (wind model W1,
versus the observed deposit, the following procedure runs V1 and V2), the model produces a deposit that
was performed: at first, calculated mass accumula- fits better the White Pumice deposit (see Fig. 7). The
tion values were transformed into thickness values best fitting axis for the Gray Pumice is displaced to
by summing up the volume of all components in the the east and is a product of the fact that only a few
considered site with respect to their individual data points were available in the more heavily
densities. This volume would be the completely weighted range beyond 30 km to constrain the eastern
compacted volume and, for the sake of simplicity, no side of the deposit. The modeling improves slightly
consideration of the unknown void volume between when W3, which accounts for a different wind
individual clasts has been taken. The relative error of direction at higher altitudes, is used. The best fitting
the measured values reported in Macedonio et al. directions of lower and upper wind levels are very
(1988) was assumed to be 30%, with a minimum similar in every simulation for both the White and
error of 1 cm. The grain size analyses for three Gray phases. Larger differences appear as to the best
sampling points, also reported in Macedonio et al. fitting wind speeds at higher levels, whereas the best
(1988), were assumed to be affected by a total fitting upper wind directions appear to be consistently
relative error of 50% and a minimum error of 3 wt.% from the west. Wind model W2 results in another
286 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

slightly different wind profile for the model. The congruence between calculated and observed grain
results are comparable to the results of simulations size distributions in three reference points, however, is
running with W3. Compared to Macedonio et al. comparably good in all results, for both eruption
(1988), who achieved the best fitting with a rotation of distributions (pumice flow from Macedonio et al.,
the summer wind profile by 608 clockwise, the best 1988 or Gaussian distribution from Eq. (4)). Obvi-
fitting rotation angle here was 378 and 418 for the ously, more data points are needed to allow a more
White and Gray phases, respectively (runs V3, Fig. 8). detailed judgement of which distribution is more
It is apparent that the limited amount of field data, likely to represent the true eruption column. A general
especially in distal parts, impacts heavily on the agreement exists only in the fact that the best fitting
conclusions that can be drawn from the model results. weight fraction of lithics during the Gray phase is in
The best model results given by V5, Fig. 9, were all runs about 10–20 wt.% larger than for the White
obtained using the parameterized wind profile W2. phase, which is in agreement with the overall field
But also in the other cases (using W1 and W3), the evidence.
fitting goodness was comparable (see also the A discrepancy exists in the best fitting masses,
summarizing Table 2). which are between about 5 and 10 times smaller
The best fitting Gaussian distributions of U-grain than the estimates by Carey and Sigurdsson (1987)
size for pumice and lithics (run V6, Fig. 10) result in derived from applying the estimates of the eruption
quite different distributions from the distribution of column height together with the model of magma
the Gray Pumice flow used in the other runs and by discharge rate as a function of column height and
Macedonio et al. (1988) (see Fig. 6). The overall estimates of the duration of the eruption. The

Fig. 6. Comparison between the fitted Gaussian grain size distribution in the eruption column (run V2) and the Pumice flow and distribution
used in Macedonio et al. (1988).
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 287

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
discrepancy can partly be explained by taking into 4dq=ð3CD qa Þ , with a very high drag coefficient
account that the model best fit is based on C D=3.6 as opposed to C D=1 in all other runs. It has
granulometric data that only describe the relatively been pointed out that such a high value is
coarse grain size fraction and does not need to take unrealistic in the range of high Reynolds numbers
the mass contained in the fine fraction into account. (ReN1000).
pffiffiffiffiffiffiffi Mathematically, it results in a fall time
In fact, the overall fitting of the computed deposit 3:6 times larger, and in fact, the required wind
thickness (or mass accumulation) is in good agree- speed to carry particles to the same distance, and
ment with field data, but there are not many data hence produce a similarly good fit, is only half the
points in the distal zone, and this could also help to value of the model runs where C D=1 is applied.
explain the mass discrepancy (see Section 5). The relatively low wind speeds used by Macedonio
Comparison between different runs (see Table 2) et al. (1988) are considered suspicious and seem
shows that the best fitting masses vary greatly in unlikely to truly represent the actual wind at the
the case of the White Pumice, from 5.01011 kg in time of the eruption. Within the limitations of both
V6 against 1.01012 kg in V3, and from 6.61011 models, this observation is regarded as a confirma-
kg in V1 against 1.21012 kg in V5 for the Grey tion of the proper use of drag coefficients C D near
Pumice. 1 rather than 3.6.
Run V4 has been made for the sake of To summarize, we stress:
comparison between the present 2D model and
the reconstruction presented by Macedonio et al. ! Using suitable input parameters, both the 3D model
(1988) who used a similar 3D version. All the by Macedonio et al. (1988) and the present one
input parameters were chosen the same or very successfully reconstruct the deposit and achieve a
similar to those used in Macedonio et al. (1988), generally good agreement between total thickness
including the calculation of settling velocity and its and grain size population in three field locations of
altitude dependence, which differ considerably from both fall deposits (see Figs. 7–10).
the models used in the other runs (V1, V2, V3, V5, ! The best fitting parameters required by the present
V6). Surprisingly, there is no match between the model generally agree between different assump-
results obtained by the two models. In both models, the tions on wind and original grain size distribution.
same mass is used. The 2D model completely fails to The resulting differences are too small, compared
produce a deposit that resembles the actual ground with the possible error and the small number of
deposit, unlike the 3D model. Using the same input available field data, to allow a refinement of
parameters, the 2D model works better with 10% of the estimates made by other models.
mass used in Macedonio et al. (1988). Obviously, the ! The results show that a smaller mass is required
3D model has much more ability to distribute mass. to explain the observed deposit than that
The only significant difference between the models is estimated, e.g., in Sigurdsson et al. (1985).
considered to be the neglecting of vertical diffusion. It The discrepancy with other mass estimates
does seem unlikely that this could explain such a big (Sigurdsson et al., 1985) could suggest that the
divergence. Another possible cause of the difference mass contained in the 79 A.D. eruption fall
between the models is related to the numerical deposit might be smaller than assumed. How-
diffusion of the numerical scheme used to solve the ever, no information is given by the model to
advection–diffusion equation in Macedonio et al. how much of the total mass is contained in the
(1988). fine particle fraction.
It is interesting to compare the effect of the ! The computed deposit of the White phase is
settling velocity calculations. Nearly all particles significantly better reproduced than the Gray
that fall in the range of study settle at high phase. This reflects in parts the more precise
Reynolds numbers. As mentioned earlier, for the data on total deposit thickness, in as much as
settling velocity, V4 uses the formula given by this phase of the fall deposit is completely
Macedonio et al. (1988), which is: U s(z)=U(0) exp preserved in almost all locations, whereas in
(0.024z ), where z is in km, and Us ð0Þ ¼ some sites, parts of the Gray deposit were
288 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

Fig. 7. Isopachs reproduced by RUN V1 (on the left) and RUN V2 (on the right).
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 289

Fig. 8. Isopachs reproduced by RUN V3 (on the left) and RUN V4 (on the right).
290 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

Fig. 9. Isopachs reproduced by RUN V5 (on the left) and RUN V6 (on the right).
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 291

Fig. 10. Settling velocity spectra obtained from RUNS V1, V2, V3, V4, V5, and V6.

missing, and the original thickness had to be When only the proximal area is used as the btargetQ
estimated. for a fitting procedure, a suitable set of input
parameters can usually be found, but in this case, they
may depart significantly from the parameters found by
5. Discussion considering also the distal area. For example, in
another study (Pfeiffer, 2003) where the model was
Our study confirms that the simple advection– applied to reconstruct the deposit of the Mt. Spurr
diffusion-settling model is valid as a first approach to 1992 eruption, it was observed that the reconstruction
reconstruct tephra fall deposits. of the proximal area deposit requires only about 10%
In the case of the Vesuvius 79 A.D. eruption as of the total mass calculated by the reconstruction of the
well as in the study by Bonadonna et al. (2002), the total deposit. This implies that the model cannot be
model did reproduce quite well the fallout in medium- used to calculate total tephra volumes if a significant
proximal areas, but in all of these cases, the modeling part of the deposit is not exposed.
was optimized for this area, and there was no concern This probably reconciles the differences in the
for the more distal deposit. total mass estimates between the values predicted
292 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

Table 3 code can be used in the probability mode to


Comparison between proximal and distal total mass fitting for the produce a hazard map (see Macedonio et al.,
18 August 1992 eruption of Mt. Spurr, Alaska
submitted for publication). The previous HAZMAP
Distance Distal fitting Proximal fitting
version considered only the case of constant settling
0–10 km 3.2108 5.5108 velocity that is realistic only for a low eruption
0–20 km 1.5109 1.7109
column. In fact, when a realistic altitude settling
0–30 km 3.1109 2.5109
0–50 km 6.4109 3.2109 velocity dependence is taken into account, the
0–100 km 1.11010 3.4109 probability values that the deposited mass on the
0–400 km 4.31010 3.4109 ground exceeds a given threshold can be different
Total mass 6.41010 3.4109 from the case with constant settling velocity. As an
example, we carried out two different simulations
by this model in the Vesuvius 79 A.D. case as using HAZMAP in the probability mode: the first
compared to other estimates. For example, for the considering a constant settling velocity model as in
18 August 1992 eruption of Mt. Spurr, Alaska, a the previous HAZMAP version and the second
brief comparison between the masses obtained by accounting for the altitude settling velocity depend-
fitting the observed thicknesses in the distal area ence. For both, the same input parameters used to
and in the proximal one is shown in Table 3, which reproduce the Gray Pumice deposit in the RUN V5
illustrates the different mass dispersal, i.e., deposited were considered. Moreover, a set of 3125 wind
mass contained within a given distance from the profiles recorded at Brindisi (Italy) during about 10
vent (Pfeiffer, 2003). From this comparison, it years was used. For a considered settling velocity
becomes evident that, in terms of tephra volume, class, the altitude settling velocity dependence of
the differences in the mass contained in the very pumices and lithics does not change dramatically
proximal areas are insignificant compared to the with the kind of particle (see Fig. 4). Hence,
total mass of the deposit better estimated by the run considering that this model is intrinsically a first
performed fitting the distal data. order approach model, in order to reduce the
Finally, another important question to discuss computational time, we used a parameterization
concerns the implications of the settling velocity for each settling velocity class at the ground
model for the hazard assessment. The HAZMAP irrespective of the kind of particle. In summary,

Fig. 11. Comparison of the probability maps that the deposited mass on the ground exceeds a threshold of 300 kg/m3 obtained using a constant
settling velocity model (on the left) and parameterising the results showed in Fig. 4 (on the right).
T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294 293

for each settling velocity class at the ground obtained by different settling velocity models shows
V s0=V s(0), we introduced a different parameter- that, in the case of high Plinian eruption columns,
ization function f = f (V s0,z) that basically reproduce the role of the settling velocity models is not
the behavior described by the model discussed in negligible.
Section 3.1.1: The satisfactory results obtained in the reconstruc-
tion of the 79 A.D. Vesuvius eruption and in other test
Vs ð zÞ ¼ Vs0 f ðVs0 ; zÞ ð13Þ cases support the validity of the model as a fast first
order approach to predicting the distribution of
with hypothetical or real ash fall.
f ðVs0 ; zÞ ¼ exp½bðVs0 Þz if Vs0 5 m=s
and Acknowledgments
2
f ðVs0 ; zÞ ¼ 1 þ a1 ðVs0 Þz þ a2 ðVs0 Þz
This work was partially supported by the bEU
þ a3 ðVs0 Þz3 if Vs0 b 5 m=s Research Project EXPLORISQ N. EVR1-2001-00047
and by the Gruppo Nazionale per la Vulcanologia
where the parameters b(V s0) and a i (V s0) were chosen
(INGV). Tom Pfeiffer was supported by the bEU
by best fitting the results shown in Fig. 4.
Research Training Network in VolcanologyQ N.
The results clearly show that the probability that
HPRN-CT-2000-00060.
the deposited mass on the ground exceed a given
threshold reported in Fig. 11, obtained using a
constant settling velocity model or a more realistic
settling velocity model, are quite different, specially in References
the case of the highest probability curves.
Arastoopour, H., Wang, C.H., Weil, S.A., 1982. Particle–particle
interaction in a dilute gas–solid system. Chem. Eng. Sci. 37 (9),
1379 – 1386.
6. Conclusion Armienti, P., Macedonio, G., Pareschi, M.T., 1988. A numerical
model for simulation of tephra transport and deposition:
The simple semianalytical model for ash disper- application to May 18, 1980, Mount St. Helens eruption. J.
sion HAZMAP (Macedonio et al. submitted for Geophys. Res. 93, 6463 – 6476.
Aschenbach, E., 1972. Experiments on the flow past spheres at very
publication) was improved introducing a more high Reynolds numbers. J. Fluid Mech. 54, 565 – 575.
realistic settling velocity model and more general Bonadonna, C., Ernst, G.G.J., Sparks, R.S.J., 1998. Thickness
parameterizations. variations and volume estimates of tephra fall deposits: the
Strong and weak points of this approach were importance of particle Reynolds number. J. Volcanol. Geotherm.
stressed, and the model was tested by reconstructing Res. 81 (3–4), 173 – 187.
Bonadonna, C., Macedonio, G., Sparks, R.S.J., 2002. Numerical
the isopachs of the well-known 79 A.D. Vesuvius model of tephra fallout with dome collapses and Vulcanian
eruption, considering a set of different input param- explosions: application to hazard assessment on Montserrat.
eters (e.g., wind model, total mass). Mem. Geol. Soc. Lond. 21, 517 – 537.
More general parameterizations for eruption col- Carey, S., Sparks, R.S.J., 1986. Quantitative models of the fallout
umn shape and particle distribution in the eruption and dispersal of tephra from volcanic eruption columns. Bull.
Volcanol. 48, 109 – 125.
column were introduced. Carey, S., Sigurdsson, H., 1987. The eruption of Vesuvius in A.D.
The dependence of volcanic ash settling velocity 79: II. Variation in column height and discharge rate. Geol. Soc.
on altitude was extensively investigated using a Am. Bull. 99 (2), 303 – 314.
settling velocity model based on the available experi- Cornell, W., Carey, S., Sigurdsson, H., 1983. Computer simulation
ment data, and the results of different settling velocity and transport of the Campanian Y-5 ash. J. Volcanol. Geotherm.
Res. 17, 89 – 109.
models were compared. Kunii, D., Levenspiel, O., 1969. Fluidization Engineering. J. Wiley
Finally, implications for hazard forecasting were and Sons.
considered. Comparison of the probability maps Landau, L., Lifchitz, E., 1971. Mècanique des Fluides. Mir.
294 T. Pfeiffer et al. / Journal of Volcanology and Geothermal Research 140 (2005) 273–294

Macedonio, G., Pareschi, M.T., Santacroce, R., 1988. A numerical Sigurdsson, H., Cashdollar, S., Sparks, R.S.J., 1982. The eruption of
simulation of the Plinian fall phase of 79 A.D. eruption of Vesuvius in A.D. 79: reconstruction from historical and
Vesuvius. J. Geophys. Res. 93, 14817 – 14827. volcanological evidence. Am. J. Archaeol. 86, 39 – 51.
Macedonio, G., Costa, A., Longo, A., 2004. Hazmap—computer mo- Sigurdsson, H., Carey, S., Cornell, W., Pescatore, T., 1985.
del of volcanic ash fall-out from erupting sustained columns and The eruption of Vesuvius in A.D. 79. Natl. Geogr. Res. 3,
hazard assessment. Comput. Geosci. submitted for publication. 332 – 397.
Mironer, A., 1979. Engineering Fluid Mechanics. McGraw-Hill, Smithsonian Institution, 1951. Smithsonian Meteorological Tables,
New York, 592 pp. 6th ed. Washington, DC.
Morton, B.R., Taylor, G., Turner, S., 1956. Turbulent gravitational Sparks, R.S.J., 1986. The dimensions and dynamics of volcanic
convection from maintained and instantaneous sources. Proc. R. eruption columns. Bull. Volcanol. 48 (1), 3 – 15.
Soc. Lond. 234, 1 – 23. Suzuki, T., 1983. A theoretical model for dispersion of tephra. In:
Pfeiffer, T., 2003. Two catastrophic volcanic eruptions in the Shimozuru, D., Yokoyama, I. (Eds.), Volcanism: Physics and
Mediterranean: Santorini 1645 BC and Vesuvius 79 AD. PhD Tectonics. Arc, Tokyo, pp. 95 – 113.
Thesis, Department of Earth Sciences, University of Aarhus, Walker, G.P.L., Wilson, L., Bowell, E.L.G., 1971. Explosive
Denmark. volcanic eruptions: I. Rate of fall of pyroclasts. Geophys. J.
Rose, W.I., Wundermann, R.L., Hoffmann, M.F., Gale, L., 1983. R. Astron. Soc. 22, 377 – 383.
Atmosperic hazards of volcanic activity from a volcanologist’s Wilson, L., Huang, T.C., 1979. The influence of shape on the
point of view: Fuego and Mount St. Helens. J. Volcanol. atmospheric settling velocity of volcanic ash particles. Earth
Geotherm. Res. 17, 133 – 157. Planet. Sci. Lett. 44, 311 – 324.

Das könnte Ihnen auch gefallen