ARTICLE IN PRESS

VOLGEO-04446; No of Pages 9
Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx

Contents lists available at ScienceDirect

Journal of Volcanology and Geothermal Research

j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / j v o l g e o r e s

Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian
eruption at Vesuvius (Italy) through a new best-fit procedure
R. Bonasia a,⁎, G. Macedonio a, A. Costa a,1, D. Mele b, R. Sulpizio b
a
Istituto Nazionale di Geofisica e Vulcanologia, Sezione “Osservatorio Vesuviano”, Napoli, Italy
b
CIRISIVU, Dipartimento Geomineralogico, Università degli Studi di Bari, Bari, Italy

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

Article history: A simple semi-analytical model for ash-fall deposit was applied to reconstruct the tephra deposits of the sub-
Received 1 July 2009 Plinian 472 AD eruption of Vesuvius, Italy, which is of the scale of the reference eruptive scenario for the
Accepted 2 November 2009 emergency planning, at Vesuvius. Applying a novel least-squares method, the bulk grain-size distribution,
Available online xxxx
the total mass, and the eruption column height were obtained by fitting the computed ground load and
granulometries with the observed ones. The analysis of the effect of three different weighting factors in the
Keywords:
tephra fallout
minimization procedure was also performed. Results showed that the statistical weighting factor produced
Pollena eruption the minimum bias. The best correlation between calculated and measured deposit was found, even though
HAZMAP model the quantity of the input data was not very high, as it commonly occurs for several ancient eruptions. Model
Vesuvius results were also in agreement with estimations provided by other independent methods.
inverse problem © 2009 Elsevier B.V. All rights reserved.

1. Introduction the estimation of crucial eruptive parameters, such as total erupted

Volcanic eruptions can produce several hazardous phenomena,
such as tephra fallout, pyroclastic density currents, lavas and lahars. In
particular tephra fallout can be responsible for severe damages to
buildings, infrastructures, viability, agriculture, livestock and humans
(Casadevall, 1994). Moreover, airborne ash may significantly affect air
traffic.
The analysis of tephra deposits allows for estimating some im-
portant parameters characterizing the eruption, such as the erupted
mass, the column height and the bulk grain-size distribution. However
the preservation of tephra deposits is often incomplete, especially for
ancient eruptions, because the deposits can be buried or eroded. For
these reasons, it is very useful to develop procedures that allow to
estimate reliably the parameters listed above, by using a minimum
amount of data on the thickness and the grain-size distribution of
the deposits. Computational models based on an analytical solu-
tion of a simplified advection–diffusion–sedimentation (ADS) equa-
tion for volcanic tephra (Pfeiffer et al., 2005; Macedonio et al., 2005;
Bonadonna et al., 2005) can be used for this purpose for their
simplicity of the physical parameterization which allows high
computation speed. These advantages allow for the solution of the
inverse problem, which consists in a best-fit with field data, by
carrying out thousands of runs in a relatively short time even on an
ordinary personal computer. Solving the inverse problem allows for
⁎ Corresponding author. Istituto Nazionale di Geofisica e Vulcanologia, Sezione
“Osservatorio Vesuviano”, Via Diocleziano, 328, Napoli, Italy.
E-mail address: bonasia@ov.ingv.it (R. Bonasia).
1
Also at Department of Earth Sciences, University of Bristol, UK.
mass and column height. In order to obtain the solution of the inverse
problem, besides these parameters other quantities need to be esti-
mated such as the bulk granulometry, the mass distribution inside
the column (e.g., column shape parameters), the wind profile (which
is needed for ancient eruptions), and the turbulence diffusion coef-
ficient. This implies that there are several free parameters needed to
be determined. Hence for constraining all these parameters it would
be ideal to know deposit thicknesses, together with grain-size and
component distributions, in a large number of well distributed sites
(Pfeiffer and Costa, 2004a). Unfortunately it is not always possible to
have all this amount of information, especially for ancient eruptions.
Therefore, a typical problem in volcanology consists of the solution of
an inverse problem using a relatively small amount of data (Pfeiffer
and Costa, 2004b).
Independent analytical methods can help for estimating a reliable
range of values for column height (Carey and Sparks, 1986), wind
intensity and total mass (Pyle, 1989). In addition, in order to reduce
the number of degrees of freedom the bulk granulometry can be
reconstructed independently from individual analyses of field data.
For example, Carey and Sigurdsson (1982) calculated the total grain-
size distribution of tephra fall deposits of the May 18, 1980, eruption
of Mount St. Helens volcano as a weighted average of the measured
grain-sizes, while, more recently, in Bonadonna and Houghton (2005)
total grain-size distribution of the Ruapehu deposits was calculated by
means of the Voronoi Tessellation method.
Here we propose a new procedure that allows for improving the
fitting results when only a few points are available. The procedure is
based on the assumption that quantities such as i) column height,
ii) column shape, iii) wind profile and iv) diffusion coefficient do not

0377-0273/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.jvolgeores.2009.11.009

Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
 ARTICLE IN PRESS
2 R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx
depend on the granulometry classes, i.e. on the settling velocity classes.
This is true because, according to the model of Woods (1995), pyro-
clastic particles smaller than 2 mm in diameter, which constitute the
great part of the particles distributed throughout the plume, are both in
thermal and mechanical equilibrium into the column and can be
considered as uniformly distributed inside the column itself. Therefore,
in a first step we estimate these parameters, that we consider inde-
pendent among them, using a restricted range of settling velocity classes
having the baricenter of the corresponding deposit inside the “convex
hull” delimited by the sampled sites. Then, in a second step, we extend
the inversion to all the other classes by maintaining fixed the
parameters determined in the first step. In this way, the first step
allows us to estimate the column height, the wind velocity and the
diffusion coefficient using a reliable range of settling velocity classes,
while the second step allows for estimating the bulk grain-size
distribution and the total erupted mass.
This method is applied to reconstruct and analyze the tephra
deposit and the bulk grain-size distribution of the sub-Plinian 472 AD
eruption of Vesuvius, Italy. This eruption is chosen for two main
reasons. Firstly because it is of the same scale of the reference eruption
in case of renewal activity of Vesuvius (Macedonio et al., 1990,2008),
and secondly because, since it was previously studied using classi-
cal methods (Rosi and Santacroce, 1983; Rolandi et al., 2004; Sulpizio
et al., 2005), it is possible to carry out a comparative study.
Section 2 describes briefly the ADS model for volcanic ash dis-
persal, HAZMAP (Macedonio et al., 2005). Section 3 describes the new
best-fit inversion procedure, based on the least-squares function,
applied for the determination of the input parameters required by
HAZMAP. Then, in Section 4 we apply the method to reconstruct the
deposits produced during the 472 AD (Pollena) eruption using the
best-fit method, with the application of three different weighting
factors. Finally we compare model results with the values calculated
by Sulpizio et al. (2005) by using independent methods, based on the
analysis of field data.
2. The advection–diffusion–sedimentation model for volcanic
ash dispersal
For this study we used the HAZMAP model (Macedonio et al.,
2005). One of the assumptions of the model is that beyond a certain
distance from the eruption column, the dispersion and sedimenta-
tion of volcanic particles are governed mainly by wind transport,
turbulent diffusion and their terminal velocity (Armienti et al., 1988;
Costa et al., 2006; Folch et al., 2009). Particles with the same set-
tling velocity are commonly grouped into families or classes that
have no interactions among them. Moreover, a constant and horizon-
tally uniform wind field, and a negligible vertical diffusion coeffi-
cient, are assumed.
Under these hypotheses the concentration of particles of class j is
described through the mass conservation equation (Armienti et al.,
1988; Pfeiffer et al., 2005; Macedonio et al., 2005):
∂Cj ∂Cj ∂Cj ∂Vsj Cj ∂2 Cj ∂2 Cj
+ Ux + Uy − = Kx + Ky
∂t ∂x ∂y ∂z ∂x2 ∂y2
where Cj denotes concentration of particles class j, t is time, (Ux, Uy)
are the components of the wind velocity vector, Kx, Ky are the tur-
bulent diffusion coefficients (K = Kx = Ky, i.e. isotropic horizontal dif-
fusion is assumed), and Vsj and Sj stand for the terminal settling
velocity and source term for class j respectively.
The ADS Eq. (1) has to be resolved numerically, but, under addi-
tional hypothesis, it can be simplified to derive an analytical solution
(Macedonio et al., 2005). Dividing the vertical computational domain
in Nlayers layers among which Nsources point sources are distributed
and in which settling and wind velocity are assumed constant, the
total mass on the ground, MG, can be computed as the sum of the
Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
contributions from each of the point sources distributed above the
vent and from each particle settling velocity class:
" #
Nvs Nsources Mi fj ðx−xGij Þ2 + ðy−yGij Þ2
MG ðx; yÞ = ∑ ∑ 2
exp − ð2Þ
j=1 i=1 2πσGij 2σG2i j
→
where →xGij = → 2
x0i + ∑k U k Δtkj and σGij = 2K ∑k Δtkj are the centre and
the variance of the Gaussian respectively (Δtkj = (zk −zk-1)/Vskj is the
time for a particle to cross the layer k), Nsources indicates the number of
source points, Nvs is the total number of settling velocity classes, Mi is the
total mass emitted from the point source in the layer i (∑i Mi =Mtot,
with Mtot total mass injected into the system), and fj is the fraction
of that mass belonging to the settling velocity class j (∑j fj = 1).
Following Macedonio et al. (2005) and Pfeiffer et al. (2005), the source
term was described using a modified parameterization proposed by
Suzuki (1983):
n z h z ioλ
Sðx; y; z; tÞ = S0 1− exp A −1
H H ð3Þ
× δðt−t0 Þδðx−x0 Þδðy−y0 Þ
where S0 is the normalisation constant, x0, y0 are the coordinates of the
vent, H is the column height and A and λ are two empirical parameters
introduced in Pfeiffer et al. (2005). In this parameterization the value of
the parameter A describes the vertical position of the maximum con-
centration relative to the maximum column height, whereas the para-
meter λ is a measure of how closely the total mass is concentrated
around the maximum.
One of the limitations of the model used for our simulations
derives from the parameterization of the eruption column that is
described as a vertical line. This simplification is only valid far from
the vent, where the wind field is not perturbed by the ascending
column and the radius of the eruption column can be neglected (i.e. at
a distance of few kilometers). Therefore, the use of the model is
limited to mid-distal sites. Another main limitation is due to the
assumption of a horizontally uniform wind field that typically holds
up to distances of about 100 km (Pfeiffer et al., 2005). Moreover,
another reason for restricting the application of this kind of models
within a distance of about 100 km is due to the fact that we do not
account for aggregation of fine ashes (Carey and Sigurdsson, 1982;
Cornell et al., 1983; Durant et al., 2009). So the range of applicability of
the model goes from few km to few hundreds of km.
3. Best-fit inversion method
Input parameters required by HAZMAP can be obtained by best-fit
with field data. Fitting is performed using a least-squares method
comparing measured and calculated deposit thickness and grain-
sizes. Because of the inherent limitation of the model to medium-
range and distal parts of a deposit, only available data from 5 km of
distance from the eruptive vent were considered in this study. The
model parameters were defined by minimising the function (e.g.
+ Sj ð1Þ Pfeiffer et al., 2005; Costa et al., 2009):
2 1 N 2
χ = ∑ w ½Y −Ymod;i  ð4Þ
N−p i = 1 i obs;i
where wi are weighting factors, N is the number of observed data, p is
the number of free parameters, Yobs,i denote the observed ground load
(kg/m2) and Ymod,i the values predicted by the model. The choice of
the weighting factors, wi, in this equation, depends upon the dis-
tribution of the errors (Costa et al., 2009). If all the measurements
have equal uncertainty, uniform weighting factors (wi = 1) have to
be used. If the observations have different uncertainty distributions,
proportional weighting factors (wi = 1/Y2obs,i) or statistical weighting
 ARTICLE IN PRESS
R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx
factors (wi = 1/Yobs,i) should be used (Aitken, 1935). Generally, the
best weighting factor is the one that yields a random residual plot.
In order to well constrain the input parameters, it is necessary to
know thickness and grain-size values in a large number, N, of strati-
graphic sections, such as N ≫p. For example, Pfeiffer and Costa (2004a)
showed that a good reliable inversion is given by at least 40–50 well
distributed stratigraphic sections and at least 3–5 sections with
individual granulometry and particle component distributions. Howev-
er, especially for ancient eruptions, finding outcrops with a good ex-
posure is often very arduous.
The risk of finding parameters not physically correct occurs
especially when we deal with portions of the deposit, which represent
the tail of the distribution relative to a settling velocity class. In fact, as
explained in the previous section, in order to compute the total mass
on the ground, the vertical domain is divided in layers among which
the point sources are distributed. Under the effect of wind,
atmospheric turbulence and gravity, it is assumed that the Gaussian
distribution of particles spreads and settles into the underlying layer,
while its centre horizontally translates, leading to a final position of
the mass distribution centre at the ground. The final position of the
Gaussian centres may fall in areas where we do not have information
about the actual deposits.
In order to avoid these problems, in the scientific literature, other
techniques for the calculation of the particles spectrum, in terms of
the total grain-size distribution, were adopted, allowing the reduction
of the degrees of freedom. In this work we propose a new procedure
for fitting also the bulk settling velocity distributions, even when only
few stratigraphic sections are available. For this reason, we assume
that the best-fitted parameters such as column height, column shape
coefficients, wind velocity and diffusion coefficient, do not vary with
the choice of the grain-size classes and their number. Thus, between
all the possible choices, we focus on a restricted number of grain-size
classes, for which the inverse problem results are better constrained.
Therefore, in the first step, the program identifies the area enclosed by
the measured ground sections, that, in the case of the eruption studied
here, is delimited by the “convex hull”, shown in Fig. 1. The calculation
of the total mass at the ground is performed by summing the con-
tribution of each settling velocity class that leads to the Gaussian
centre placed inside the “convex hull” (Fig. 1). It is worth noting that
this procedure reduces the number of settling velocity classes, and
allows for estimating the total mass relative to these classes only. In
order to obtain the total mass over a wider range of velocity classes, in
a second step, we repeat the inversions fixing the values of column
height, wind direction, wind speed and diffusion coefficients, as they
Fig. 1. Location map of the stratigraphic sections and deposit Gaussian centres. White
star: 472 AD eruption vent; dots: stratigraphic spections measured by Sulpizio et al.
(2005); black stars: sections considered in this work; black line represents the “convex
hull” of the considered stratigraphic sections; squares: Gaussian centres that fall inside
the “convex hull”; crosses: Gaussian centres that fall outside the “convex hull”.
(Modified after Sulpizio et al., 2005).
Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
3
were obtained in the previous step, and extend the computation of
the bulk particle spectrum also to the particles classes previously
neglected, thus, leading to the final estimation of the effective total
mass and of the total bulk velocity class distribution.
4. Application to the 472 AD (Pollena) eruption of Vesuvius
4.1. Characteristics of the eruption
The most recent study of this eruption is due to Sulpizio et al.
(2005) who presented new stratigraphic data coupled with dispersal,
grain-size, componentry, density/vesicularity and SEM data of the
products. In their work more than 40 stratigraphic sections were
documented on the slopes of the volcano and its surroundings (Fig. 1).
The stratigraphic data presented by the authors, showed significant
differences with respect to previous works (see Rosi and Santacroce,
1983; Arnò et al., 1987; Joron et al., 1987), especially in the estimation
of the volume of fall layers and dispersal area of pyroclastic surge
deposits.
The composite stratigraphic section of the Pollena eruption, as it
has been reconstructed by Sulpizio et al. (2005) is shown in Fig. 2. The
continuous succession of fall beds characterized by variable grain-size,
named L1 to L7, constitutes Phase I of the eruption. The L1 bed is the
first deposit of the Pollena eruption. It is a massive, well-sorted, lapilli
sized fall deposit, characterized by highly vescicular juvenile frag-
ments and scarce accidental lithic clasts. The abundance of accidental
lithic fragments sharply increases in L2 and L3 beds which merge
in medial and distal sites to form a single layer. According to Sulpizio
et al. (2005) the presence of these beds suggests an extensive cra-
terization, confirmed by the lithology of the lithic component, which
mainly consists of lava fragments. The L2–L3 bed is overlained by the
L4 bed. Grain-size in this bed oscillates upwards and contains a
smaller amount of accidental lithics with respect to the preceding L2–
L3 bed. A low accidental lithic/juvenile fragments ratio characterizes
the L5 and L6 beds, both massive, well-sorted lapilli units, comprising
dark gray scoriae. Phase I is closed by the emplacement of L7 bed,
which shows a limited dispersal area and is represented by a massive,
well-sorted lapilli unit, rich in accidental lithics, mainly hydrothermal
lava fragments. Phase II is characterized by the alternation of PDC
and fall deposits, the latter indicated as L8 and L9. The L8 fallout
bed is characterized by the presence of deep-seated accidental lithic
fragments of syenites, marbles, cumulates and skarns. Phase II is
closed by the deposition of L9 fallout bed, which is a massive, well-
sorted lapilli unit, comprising dark coloured scoriae and minor
accidental lithics, mainly lava fragments. Phase III is dominated by
PDC deposition.
4.2. Ranges of the physical parameters of the eruption and input data
As mentioned in Section 2, the computational model needs field data
located relatively far from the eruptive vent (≈5 km) and fallout beds
that can be stratigraphically correlated. We selected five, out of the eight
fallout units recognized by Sulpizio et al. (2005), i.e. L1, L2–L3, L4, L6, L8,
that can be recognized in the 14 considered stratigraphic sections (see
Fig. 1). For the deposit thicknesses, values relative to the considered
fallout beds measured in Sulpizio et al. (2005) were used, whereas for
the deposit densities we used the average value of 900 kg/m3 as
reported in Cioni et al. (2003), for the “Pollena” eruption. Units L5, L7
and L9 were not taken into account because they do not outcrop in all
the considered stratigraphic sections.
Data relative to component composition and grain-size distribu-
tion are a crucial requirement for the model. Four independent com-
ponents were used, pumice, lithics, crystals and obsidian. For lithics,
crystals and obsidian, grain-size independent constant densities of
2500, 2650 and 2570 kg/m3, respectively, were used, whereas the
density of pumice was taken as grain-size dependent using the values
 ARTICLE IN PRESS
4 R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx

Fig. 2. Composite stratigraphic section of the Pollena eruption (modified after Sulpizio et al., 2005). Thicknesses are not to scale.

published in Mele (2003), measured in a sample of the flow deposit Suzuki (1983) as modified by Pfeiffer et al. (2005), reported in Eq. (3).
that is related to the total column collapse occurred during the sus- Theoretical and empirical observations on buoyant plumes (e.g.
tained phase that generated the L8 fallout bed. Morton et al., 1956; Sparks, 1986) show that the ratio HB/HT between
Juvenile fragment density values, used in this work, are reported in
Table 1. The settling velocity distributions were obtained by means of
Table 1
an HAZMAP utility that transforms particle size and density distribu-
Grain-size dependent density of juvenile pumice used in the model.
tions into the particle settling velocity spectrum, by means of the
terminal settling velocity model reported in Pfeiffer et al. (2005), Φ Density
which assumes a constant particles shape factor (F = 0.7). b−1.0 1615 kg/m3
A unidirectional wind profile (Cornell et al., 1983; Carey and − 0.5 1726 kg/m3
0.0 1830 kg/m3
Sparks, 1986) was applied for the reconstruction of the eruption
0.5 1936 kg/m3
deposit. Maximum velocity at tropopause level was found by best-fit 1.0 2045 kg/m3
with field data. The direction was also obtained by best-fit. 1.5 2152 kg/m3
The ranges used in the first step of the procedure for the search of 2.0 2260 kg/m3
the best-fit parameters are reported in Table 2. For what concerns the 2.5 2367 kg/m3
3.0 2475 kg/m3
shape of the eruption column, it assumed the empirical formula of

Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
 ARTICLE IN PRESS
R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx 5

Table 2
Parameter ranges and steps used in the first step best-fit procedure.

Column height Wind direction at 11 km Wind speed at 11 km Diffusion coefficient

H = 5–25 (km) θ = 20°–40° w = 10–90 (m/s) K = 1000–10,000 (m2/s)

ΔH = 0.5 (km) Δθ = 2° Δw = 2 (m/s) ΔK = 500 (m2/s)

the height of neutral buoyancy of the plume, HB, and the maximum
height, HT, is usually around 3/4. We tried to account for this by fixing
A = 4 and λ = 1 in all the calculations.
4.2.1. Results obtained using different weighting factors
Best-fit between measured and calculated data is obtained by
minimising the χ2 function reported in Eq. (4).
We performed a series of inversion runs applying the three types
of weighting factors discussed in Section 3. As mentioned above, the
first step inversions, restricted to the area delimited by the “convex
hull” of the considered stratigraphic sections, returned the values of
those parameters not dependent on the granulometry classes, i.e.
column height, wind velocity, wind direction, and diffusion coeffi-
cients, and gave the mass resulting from the sum of the contribution of
each velocity class that has the centre of mass of the deposit inside the
“convex hull”. Results of these inversions are reported in Table 3.
Results for total mass obtained with the different weighting factors,
cannot be compared among each other, because they are calculated
over different ranges of settling velocity classes. However results of
column height, wind direction, wind speed and diffusion coefficients
are comparable between the different best-fit methods, since they
are not directly dependent on the granulometry. As it may be ob-
served from the results, wind directions vary between the fallout
beds. They range from 20° to 30° using the first best-fit method
(uniform weighting factor), from 20° to 44° using the second method
(proportional weighting factor), and from 20° to 34° using the third
method (statistical weighting factor). This indicates that the wind
profile was not unidirectional, as we assumed, but likely it changed
with height confirming the not straight isopachs shape reconstructed
for each fallout bed by hand in Sulpizio et al. (2005). The variations
in the deposits' axis directions could also be attributed to the em-
placement of the different fallout beds in different periods. The
uncertainty on the value of the wind intensity is very large, being in
the range between 26 and 88 m/s, depending on the fallout bed. This
is due to the interdependence of column height and diffusion
coefficient on the wind speed calculation. However considering the
average of the values obtained between the fallout bed, it can be seen
that they lie in the range of average wind intensities at the tropopause
level, typical of the Naples area (Costa et al., 2009). In particular the
average wind intensity at the tropopause level furnished by the first
two methods is about 40 m/s whereas the third method gives about
60 m/s.
Column height values, calculated for each fallout bed, are different
depending on the used best-fit method. Column heights closer to the
values calculated by Sulpizio et al. (2005), by means of the Carey and
Sparks (1986) method, are those calculated with a uniform weighting
factor (χ2-Method-1).
The second step inversion runs led to the reconstruction of the
total bulk settling velocity distribution and to the estimate of the
total mass. Results of these inversions are summarized in Table 4

Table 3
Results of the first step best-fit runs. Subscript 1 indicates inversion results obtained by applying a uniform weighting factor (χ2-Method-1); subscript 2: proportional weighting
factor (χ2-Method-2); subscript 3: statistical weighting factor (χ2-Method-3).

Fall beds L1 L2–L3 L4 L6 L8

Total mass (kg) M1 1.60 × 1010 2.97 × 1010 3.65 × 1010 2.42 × 1010 4.27 × 1010
M2 4.20 × 109 3.10 × 1010 8.97 × 109 6.16 × 109 4.32 × 1010
M3 8.24 × 109 2.45 × 1010 2.45 × 1010 8.85 × 109 4.34 × 1010
Column height (km) H1 14.5 15 10.5 12.5 14
H2 13.5 15 9 9 17
H3 12 11.5 7 7.5 15.5
Wind direction at 11 km θ1 30° 26° 30° 28° 20°
θ2 42° 26° 44° 36° 20°
θ3 34° 20° 34° 32° 20°
Wind speed at 11 km (m/s) v1 34 30 40 32 78
v2 32 26 52 38 64
v3 34 88 68 48 70
Diffusion coefficients (m2/s) K1 1000 2000 1000 1000 10,000
K2 1000 1500 1000 1000 10,000
K3 1000 5000 1000 1000 10,000
Velocitya class ranges (m/s) Vsett1 4.5–7.5 4.0–7.0 3.5–6.0 3.5–6.0 5.0–9.0
Vsett2 5.0–6.0 3.5–6.5 5.0–5.5 3.0–4.5 5.0–9.0
Vsett3 4.0–6.0 5.0–9.0 3.0–5.5 2.5–4.5 5.0–9.0
a
Ranges of velocity classes that lead to Gaussian centres placed inside the “convex hull”.

Table 4
Total mass values obtained by the second step best-fit runs. Subscript 1 indicates inversion results obtained by applying a uniform weighting factor (χ2-Method-1); subscript 2:
proportional weighting factor (χ2-Method-2); subscript 3: statistical weighting factor (χ2-Method-3).

Fall beds L1 L2–L3 L4 L6 L8

Total mass (kg) M1 2.30 × 1010 4.60 × 1010 9.10 × 1010 7.10 × 1010 1.04 × 1011
M2 1.60 × 1010 5.30 × 1010 4.30 × 1010 6.20 × 1010 1.16 × 1011
M3 1.90 × 1010 4.80 × 1010 8.00 × 1010 8.68 × 1010 1.10 × 1011

Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
 ARTICLE IN PRESS
6 R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx
and Fig. 4. Note that in the second step the obtained total mass
values can be compared among the different best-fit methods as
they are calculated over the same range of velocity classes, i.e. 1.0–
9.0 m/s.
From the plots of the observed tephra deposits versus the cal-
culated deposits, Fig. 3, we can see which of the three weighting
factors leads to the minimum bias in the best-fit procedure. All the
three weighting factors reproduce well the L8 fallout deposit. The first
2
Fig. 3. Log–Log plots of the observed tephra deposits (kg/m ) versus the calculated
deposits (kg/m2), for the five fallout beds, obtained by using three different weighting
factors. a): Uniform weighting factor (wi = 1); b): proportional weighting factor
(wi = 1/Y2obs,i); c): statistical weighting factor (wi = 1/Yobs,i).
Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
method (uniform weighting factor) does not work well for reprodu-
cing deposits of the L4 and L6 fallout beds. Similarly the second
method (proportional weighting factor) produces a plot with scat-
tered points, even though, also in this case it can be seen that the L8
deposit is well reproduced. The best-fit method having the minimum
bias between calculated and observed data, is the one that uses a
statistical weighting factor (χ2-Method-3) that is able to reproduce
well the deposits of all the fallout beds.
4.2.2. Total bulk settling velocity distributions
The reconstructed total bulk settling velocity distributions are shown
in Fig. 4. For comparison, in the same figure, we report also the bulk
settling velocity spectrum of the flow deposit from the total column
collapse occurred during the closing phase of the 79 AD eruption
(Macedonio et al., 1988) that can be considered as representative of
both a Plinian and a sub-Plinian eruption of Vesuvius (Cioni et al., 2008,
and data presented by Cioni et al. at the Exploris Final Workshop, Naples,
May 10–13, 2006).
In order to compare the calculated velocity distributions with the
spectrum of the 79 AD eruption, we preferred to use the median value
of the distribution. In fact, since we are comparing distributions with
different ranges of settling velocity classes (between 1 and 9 m/s for
the calculated spectra and between 0.5 and 23.5 m/s for the 79 AD
eruption), it is more suitable to compare the median value with
respect to the mean value, because it is less influenced by the tails of
the distribution.
Spectra obtained with the use of the uniform and the proportional
weighting factors (χ2-Method-1 and χ2-Method-2 in Fig. 4) are not
unimodal, with medians centred at the velocity class values of 5.5 and
5.4 m/s respectively. The distribution calculated with the statistical
weighting factor (χ2-Method-3) is fairly unimodal and its median
corresponds to the velocity class 4.5 m/s, very close to the median of
the 79 AD spectrum (4.2 m/s).
It is worth noting that the reconstructed bulk velocity distributions
are coarse and fine particles depleted. In fact, we neglect coarse
particles (with settling velocity classes greater than 9 m/s) because of
the limitations of the model to the very proximal parts of the deposits,
and we do not account for very fine ashes (with the settling velocity
class lower than 0.5 m/s) because they fall in the distal parts of the
deposit. Comparing the bulk settling velocity distribution obtained by
the χ2-Method-3, with the 79 AD spectrum (Fig. 4), we found that the
coarse population we neglected accounts for about 20% of the total
Fig. 4. Total bulk settling velocity distributions obtained by the inversion runs that uses the
three weighting factors. χ2-Method-1 uses a uniform weighting factor; χ2-Method-2 uses
a proportional weighting factor; χ2-Method-3 uses a statistical weighting factor. For
comparison the 79 AD eruption distribution is also reported. In the last case we used the
gray phase particle sizes and densities reported in Macedonio et al. (1988) for obtaining
the particle settling velocity spectrum.
 ARTICLE IN PRESS
R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx 7

distribution, whereas the neglected fine particles are of the order of
30%. The latter is also in agreement with the results of Rose and
Durant (2009), who estimated that tephra generated during silicic
eruptions may contain large proportions of fine ash, up to 30–50%.
5. Discussion
Reconstruction of the deposits using the three different best-fit
methods introduced in Section 3 and assuming the 79 AD bulk grain-
size distribution gave similar results in terms of total mass. Final
results are summarized in Table 3, Table 4, and Fig. 4. The overall
quality of the fitting obtained using the first two least-squares
methods (χ2-Method-1 and χ2-Method-2) is not very good for some
deposits although it is still reasonable for the most massive fallout
bed, L8, and also for the units L2–L3. The method that gives the
minimum bias in the best-fit of all the levels is the one that uses the
statistical weighting factor (χ2-Method-3), implying a Poissonian
error distribution.
Maximum eruption column heights predicted by the model for the
fallout bed L8 range from 14 to 17 km. These values are lower than the
estimate of 20 km by Sulpizio et al. (2005). Results for column heights
have relevant differences. This is due to the fact that, as shown in
Scollo et al. (2008), this parameter could not be calibrated. In fact,
there exists a strong link between column height and wind speed,
which is intrinsic in the dispersal process and cannot be easily re-
solved by best-fit procedure.
Concerning the wind directions, it can be observed that the main
distribution axis can be defined at an easterly direction, with mode-
rate differences between each bed.
A reasonable diffusion coefficient value of 5000 m2/s is obtained
for the fallout bed L2–L3. Inversions on the other fallout beds gave
diffusion coefficients values that range from 1000 to 10,000 m2/s,
which are the lowest and the upper limit of realistic values. Likely this
reflects the difficulty of the method for reproducing the correct value
of the diffusion coefficient when only few sampling points are
available and the wind profile is not unidirectional.
Table 5 summarizes our best values for total mass and column
height and the corresponding values calculated by Sulpizio et al.
(2005) using Fierstein and Nathenson (1992) and Carey and Sparks
(1986) methods respectively. Note that total deposit values are
calculated by considering all the deposit thicknesses including the L5,
L7 and L9 fallout beds, not previously inverted. Fig. 5 shows the results
of the computed isopachs using the input parameters returned by the
χ2-Method-3 best-fit method. The simulations satisfactory reproduce
the actual deposit and are in good agreement with the isopachs drawn
by Sulpizio et al. (2005).
For comparison, values obtained by imposing the bulk settling
velocity distribution of the 79 AD eruption using the same sampled
points used for the previous simulations are also reported in
Table 5. In this case the total mass is slightly higher than the one
calculated with the reconstructed grain-size distributions, due to
the presence of fine and coarse particles in the bulk grain-size
distribution.
As we explained above, in our simulations we excluded the very
coarse and the very fine particle population because of the limitations
of the model. Being the fraction of mass associated to these particle
classes at about 50%, a rough estimation of the total mass would be of
about 9–10 × 1011 kg. Since we used as stratigraphic framework the
data reported in Sulpizio et al. (2005), we compare our total mass
results with their estimates obtained using other methods. The value
of total mass that we calculate is very close to the total mass of
12.4 × 1011 kg (corresponding to a volume of 1.38 km3) estimated by
Sulpizio et al. (2005).
6. Conclusion
We presented a new procedure for solving the inverse problem
related to the reconstruction of tephra fallout deposits of an erup-
tion for which only few sampling data are available. The procedure
is based on a semi-analytical ADS model. We simulated tephra
fallout of the sub-Plinian 472 AD eruption at Vesuvius (also called
Pollena eruption). The solution of the inverse problem obtained
through the best-fit procedures comparing model results with field
data, allowed us to estimate the total mass, the column height, and
the particles' bulk settling velocity distribution, together with the
meteorological conditions (i.e., wind intensity and turbulent
diffusion coefficient).
The inversion procedure consists of two steps: i) firstly we esti-
mated the parameters not directly dependent on granulometry (i.e.,
column height, column shape, wind profile and diffusion coefficient)
on a restricted range of granulometry classes having the baricenter of
the relative deposit distributions inside the “convex hull” delimited by
the sample sites; and ii) secondly we extended the inversion to all the
other classes by fixing the parameters determined in the first step,
allowing for the estimation of the bulk grain-size distribution and the
total erupted mass.
In the application to the Pollena eruption the fitting was satis-
factory despite only 14 sampling points were available. The best
results were obtained for the most massive fallout bed (L8). The
lowest bias in the best-fits was obtained using a statistical
weighting factor. Best-fit results of the erupted mass were also in
good agreement with estimations carried out through classical
methods by Sulpizio et al. (2005). In particular, assuming that the
mass fraction due to very coarse and very fine particles (not
considered in this study) contributes for about 50%, we estimate a
total erupted mass of about 9–10 × 1011 kg. Total bulk settling
velocity distribution obtained using the statistical weighting factor
shows a median value similar to the one measured for the 79 AD
eruption spectrum.
Acknowledgements
We thank A. Folch and S. Scollo for useful comments that improved
the clarity of the manuscript. This work was mainly supported by the
Civil Defense of Italy, DPC, and Istituto Nazionale di Geofisica e
Vulcanologia, INGV, in the framework of the project entitled “SPeeD”.

Table 5
Total mass and column height values calculated using i) the χ2-Method-3 (statistical weighting factor), ii) the preset 79 AD bulk settling velocity distribution, and iii) calculated by
Sulpizio et al. (2005). TOT refers to total mass values calculated by summing the contribution of each total mass value calculated for every fallout beds, adding also deposit thickness
of the L5, L7 and L9 fallout beds, not accounted for in the first step inversion runs.

L1 L2–L3 L4 L6 L8 TOT
2 10 10 10 10 11
Total mass (kg) χ -Method-3 1.90 × 10 4.80 × 10 8.0 × 10 8.68 × 10 1.10 × 10 4.80 × 1011
79 AD 2.62 × 1010 8.24 × 1010 1.09 × 1011 1.04 × 1010 1.20 × 1011 4.84 × 1011
Sulpizio et al. (2005) 1.10 × 1010 4.50 × 1010 9.00 × 1010 3.60 × 1010 4.70 × 1011 12.42 × 1011
Column height (km) χ2-Method-3 12 11.5 7 7.5 15.5 21.5
79 AD 13 12.5 14 14 16.5 20
Sulpizio et al. (2005) 13 16 12 14 20 –

Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
 ARTICLE IN PRESS
8 R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx

Fig. 5. Isopachs obtained with HAZMAP, with the input parameters values reported in Table 3 for column height, wind velocity, wind direction, diffusion coefficient, and in Table 4 for
total mass. Best-fit method: χ2-Method-3.

Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
 ARTICLE IN PRESS
R. Bonasia et al. / Journal of Volcanology and Geothermal Research xxx (2009) xxx–xxx
References
Aitken, A., 1935. On the least squares and linear combination of observations. Proc. R.
Soc. Edinb. 55, 42–48.
Armienti, P., Macedonio, G., Pareschi, M., 1988. A numerical model for simulation of
tephra transport and deposition: application to May 18, 1980 Mount St. Helens
eruption. J. Geophys. Res. 93 (B6), 6463–6476.
Arnò, V., Principe, C., Rosi, M., Santacroce, R., Sbrana, A., Sheridan, M., 1987. Eruptive
history. Somma-Vesuvius. In: Santacroce, R. (Ed.), Quaderni della Ricerca Scientifica,
vol. 114. CNR, Roma, pp. 53–103.
Bonadonna, C., Houghton, B., 2005. Total grain-size distribution and volume of tephra-
fall deposits. Bull. Volcanol. 67, 441–456.
Bonadonna, C., Connor, C., Houghton, B., Connor, L., Byrne, M., Laing, A., Hincks, T., 2005.
Probabilistic modeling of tephra dispersal: hazard assessment of a multiphase
rhyolitic eruption at Tarawera, New Zealand. J. Geophys. Res. 110.
Carey, S., Sigurdsson, H., 1982. Influence of particle aggregation on deposition of distal
tephra from the May 18, 1980, eruption of Mount St-Helens volcano. J. Geophys.
Res. 87 (B8), 7061–7072.
Carey, S., Sparks, R., 1986. Quantitative models of the fallout and dispersal of tephra
from volcanic eruption columns. Bull. Volcanol. 48, 109–125.
Casadevall, T., 1994. Volcanic ash and aviation safety. Proceedings of the First Inter-
national Symposium on Volcanic Ash and Aviation Safety, U. S. Geological Survay
Bullettin, vol. 2047, pp. 1–6.
Cioni, R., Longo, A., Macedonio, G., Santacroce, R., Sbrana, A., 2003. Assessing pyroclastic
fall hazard through field data and numerical simulations: example from Vesuvius.
J. Volcanol. Geotherm. Res. 108, 2063–2073.
Cioni, R., Bertagnini, A., Santacroce, R., Andronico, R., 2008. Explosive activity and eruption
scenarios of Somma-Vesuvius (Italy): towards a new classification scheme. J. Volcanol.
Geotherm. Res. 178, 331–346.
Cornell, W., Carey, S., Sigurdsson, H., 1983. Computer simulation and transport of the
Campanian Y-5 ash. J. Volcanol. Geotherm. Res. 17, 89–109.
Costa, A., Macedonio, A., Folch, A., 2006. A three-dimensional Eulerian model for
transport and deposition of volcanic ashes. Earth Planet. Sci. Lett. 241, 634–647.
Costa, A., Dell'Erba, F., Di Vito, M., Isaia, R., Macedonio, G., Orsi, G., Pfeiffer, T., 2009. Tephra
fallout hazard assessment at Campi Flegrei caldera (Italy). Bull. Volcanol. 71 (3),
259–273.
Durant, A., Rose, W., Sarna-Wojcicki, A., Carey, S., Volentik, A., 2009. Hydrometeor-
enhanced tephra sedimentation: constraints from the 18 May 1980 eruption of
Mount St. Helens. J. Geophys. Res. 114 (B13).
Fierstein, J., Nathenson, M., 1992. Another look at the calculation of tephra fallout
volumes. Bull. Volcanol. 54, 156–167.
Folch, A., Costa, A., Macedonio, G., 2009. FALL3D: a computational model for transport
and deposition of volcanic ash. Comput. Geosci. 35 (6), 1334–1342.
Joron, J., Metrich, N., Rosi, M., Santacroce, R., Sbrana, A., 1987. Chemistry and petrography.
Somma-Vesuvius. In: Santacroce, R. (Ed.), Quaderni della Ricerca Scientifica, vol. 114.
CNR, Roma, pp. 105–174.
Please cite this article as: Bonasia, R., et al., Numerical inversion and analysis of tephra fallout deposits from the 472 AD sub-Plinian eruption at
Vesuvius (Italy) through a new best-fit procedure, J. Volcanol. Geotherm. Res. (2009), doi:10.1016/j.jvolgeores.2009.11.009
9
Macedonio, G., Pareschi, M., Santacroce, R., 1988. A numerical simulation of the Plinian
fall phase of 79 AD eruption of Vesuvius. J. Geophys. Res. 93, 14817–14827.
Macedonio, G., Pareschi, M., Santacroce, R., 1990. Renewal of explosive activity at Vesuvius:
models for the expected tephra fallout. J. Volcanol. Geotherm. Res. 40, 327–342.
Macedonio, G., Costa, A., Longo, A., 2005. A computer model for volcanic ash fallout and
assessment of subsequent hazard. Comput. Geosci. 31, 837–845.
Macedonio, G., Costa, A., Folch, A., 2008. Ash fallout at Vesuvius: numerical simulations
and implication for hazard assessment. J. Volcanol. Geotherm. Res. 178, 366–377.
Mele, D., 2003. Studio quantitativo per la ricostruzione dei meccanismi di trasporto e
deposizione delle eruzioni esplosive: l'eruzione di Pollena (472 AD) al Vesuvio. Ph.D.
Thesis, University of Bari, Italy.
Morton, B., Taylor, G., Turner, S., 1956. Turbulent gravitational convection from main-
tained and instantaneous sources. Proc. R. Soc. Lond. 234, 1–23.
Pfeiffer, T., Costa, A., 2004a. A numerical reconstruction of fall deposits from Agnano-
Montespina (4100 BP) Plinian eruption in the Campi Flegrei area. Tech. Rep. OV Prot.
N. 4440 (10.11.2004), Osservatorio Vesuviano-INGV, http://www.earth-prints.org/
handle/2122/2068.
Pfeiffer, T., Costa, A., 2004b. Reconstruction and analysis of a sub-Plinian fall deposit
from the Astroni volcano (ca. 4100–3800 BP) in the Campi Flegrei area, Italy. Tech.
Rep. OV Prot. N. 4440 (10.11.2004, Osservatorio Vesuviano-INGV, http://www.
earth-prints.org/handle/2122/2069.
Pfeiffer, T., Costa, A., Macedonio, G., 2005. A model for the numerical simulation of
tephra fall deposits. J. Volcanol. Geotherm. Res. 140, 237–294.
Pyle, D., 1989. The thickness, volume and grainsize of tephra fall deposits. Bull.
Volcanol. 51, 1–5.
Rolandi, G., Munno, R., Postiglione, I., 2004. The AD 472 eruption of Somma volcano.
J. Volcanol. Geotherm. Res. 129, 291–319.
Rose, W., Durant, A., 2009. Fine ash content of explosive eruptions. J. Volcanol. Geotherm.
Res. 187 (1–2), 32–39.
Rosi, M., Santacroce, R., 1983. The A.D. 472 “Pollena” eruption: volcanological and petro-
logical data from this poorly known Plinian-type event at Vesuvius. J. Volcanol.
Geotherm. Res. 17, 249–271.
Scollo, S., Tarantola, S., Bonadonna, C., Coltelli, M., Saltelli, A., 2008. Sensitivity analysis
and uncertainty estimation for tephra dispersal models. J. Geophys. Res. 113,
B06202. doi:10.1029/2006JB004864.
Sparks, R., 1986. The dimensions and dynamics of volcanic eruption columns. Bull.
Volcanol. 48 (1), 3–15.
Sulpizio, R., Mele, D., Dellino, P., La Volpe, L., 2005. A complex Subplinian-type eruption
from low-viscosity, phonolitic to tephri-phonolitic magma: the AD 472 (Pollena)
eruption of Somma-Vesuvius, (Italy). Bull. Volcanol. 67, 743–767.
Suzuki, T., 1983. A theoretical model for dispersion of tephra. In: Shimozuru, D., Yokoyama,
I. (Eds.), Volcanism: Physics and Tectonics. Terrapub, Tokyo, pp. 95–113.
Woods, A., 1995. The dynamics of explosive volcanic eruptions. Rev. Geophys. 33,
495–530.