1 FALL3D: A Computational Model for Trans-

2 port and Deposition of Volcanic Ash

3 A. Folch a,1, A. Costa a G. Macedonio a

4
a Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Vesuviano, Napoli,
5 Italy.

6 Abstract

7 FALL3D is a 3-D time-dependent Eulerian model for the transport and deposition of
8 volcanic ash. The model solves the advection-diffusion-sedimentation (ADS) equa-
9 tion on a structured terrain-following grid using a second-order Finite Differences
10 (FD) explicit scheme. Different parameterizations for the eddy diffusivity tensor
11 and for the particle terminal settling velocities can be used. The code, written
12 in FORTRAN 90, is available in both serial and parallel versions for Windows and
13 Unix/Linux/Mac X Operating Systems (OS). A series of pre- and post-process util-
14 ity programs and OS-dependent scripts to launch them are also included in the
15 FALL3D distribution package. Although the model has been designed to forecast
16 volcanic ash concentration in the atmosphere and ash loading at ground, it can also
17 be used to model the transport of any kind of airborne solid particles. The model
18 inputs are meteorological data, topography, grain-size distribution, shape and den-
19 sity of particles, and mass rate of particle injected into the atmosphere. Optionally,
20 FALL3D can be coupled with the output of the meteorological processor CALMET, a
21 diagnostic model which generates 3-D time-dependent zero-divergence wind fields
22 from mesoscale forecasts incorporating local terrain effects. The FALL3D model can
23 be a tool for short-term ash deposition forecasting and for volcanic fallout hazard
24 assessment. As an example, an application to the 22 July 1998 Etna eruption is also
25 presented.

26 Key words: volcanic ash, fallout, computational model, FORTRAN code

Code is available from server at http://www.iamg.org/CGEditor/index.htm

Preprint submitted to C&G 29 November 2008

27 1 Introduction

28 Explosive volcanic eruptions can inject into the atmosphere large

29 amounts of pyroclasts. These particles, globally known as tephra,
30 can be classified according to their diameter d as blocks (d >
31 64 mm), lapilli (2 mm < d < 64 mm), coarse ash (64 m < d <
32 2 mm), and fine ash (d < 64 m). Blocks and larger lapilli follow
33 ballistic and non-ballistic trajectories and fall rapidly close to the
34 volcano. In contrast, very fine ashes can remain entrapped in the
35 atmosphere for months to years, and may affect the global cli-
36 mate in the case of large eruptions. Particles having sizes between
37 these two end-members remain airborne from hours to days and
38 can cover wide areas downwind. Such volcanic fallout entails a
39 serious threat to aircraft safety and can create many undesirable
40 effects to the communities located around the volcano.
41 The assessment of volcanic fallout hazard is an important scien-
42 tific, economic, and political issue, especially in densely populated
43 areas. Effective tools for forecasting ash transport and for strate-
44 gic land use planning can contribute to hazard assessment and
45 risk mitigation of the potentially affected regions. From a scien-
46 tific point of view, considerable progress has been made during
47 the last two decades through the use of increasingly powerful com-
48 putational models and capabilities. Nowadays, models are used
49 to quantify hazard scenarios and/or to give short-term forecasts
50 during emergency situations (Folch et al., 2008).
51 Volcanic ash fallout models can be grouped within two main cat-
52 egories: particle-tracking models (PTM) and advection-diffusion
53 models (ADM). PTM are Eulerian or Lagrangian models devoted
54 to track the position of a volcanic cloud. They are mainly used by
55 the Volcanic Ash Advisory Centers (VAACs) for aviation-safety
56 purposes. Examples of PTM are CANERM (DAmours, 1998), PUFF
57 (Searcy et al., 1998), VAFSTAD (Heffter and Stunder, 1993), or
58 VOL-CALPUFF (Barsotti et al., 2008). On the other hand, ADM
59 are Eulerian models based on the ADS equation. Simplest ADM
60 assume that volcanic ash dispersal is described as a quasi-steady

2
61 and quasi 2-D process in order to derive semi-analytical solutions
62 of the ADS equation. They can only forecast ash accumulation on
63 the ground and are mainly used for civil protection purposes such
64 as giving public warnings or territorial planning through devel-
65 opment of hazard maps. They cannot be used on large domains
66 or for very transient episodes, where their basic assumptions are
67 no longer valid. Moreover, semi-analytical models cannot be used
68 to predict the variations of ash concentration in the air. Exam-
69 ples of these models are HAZMAP (Barberi et al., 1990; Macedonio
70 et al., 2005), ASHFALL (Hurst, 1994), and TEPHRA (Connor et al.,
71 2001).
72 FALL3D (Costa et al., 2006) is a 3-D time-dependent Eulerian
73 model which circumvents many of the simplifications behind the
74 simplified ADM. The model can be used to forecast both par-
75 ticle concentration in the atmosphere (i.e. ash cloud evolution)
76 and particle loading at ground level. The main inconvenience of
77 such generalization is the increase in the computing times, espe-
78 cially for large domains and long time intervals. Here we present
79 an improved version of the original FALL3D model (Costa et al.,
80 2006) together with a new parallelized code version which dras-
81 tically reduces the computational times. It is worth highlighting
82 that the applications of FALL3D are not constrained to volcanic
83 ash transport. In fact, the model can also be used to simulate
84 the transport of industrial airborne particles at short to medium
85 scales (from few hundreds of metres to few hundreds of kilome-
86 tres). This manuscript is arranged as follows. Firstly we present
87 the governing equations and parameterizations used by FALL3D,
88 and overview the solving algorithm and the code parallelization
89 strategy. Secondly, we describe the program I/O files and the as-
90 sociated utility programs. We conclude presenting an application
91 example, included in the FALL3D package, to the 22 July 1998
92 Mt. Etna eruption.

3
93 2 FALL3D Ash Transport Model

94 2.1 Governing equation

95 The main factors controlling atmospheric transport of ash are

96 wind advection, turbulent diffusion, and gravitational settling
97 of particles. Neglecting particle-particle interaction effects (col-
98 lisions, aggregation, etc.), the Eulerian form of the continuity
99 equation written in a generalized coordinate system (X, Y, Z) is
100 (Costa et al., 2006):
C C C C Vsj
+ VX + VY + (VZ Vsj ) = C V + C
t X Y Z Z
(1)
C/ C/ C/
+ KX + K
Y + KZ + S
101 X X Y Y Z Z

102 where C is the scaled averaged concentration, V = (VX , VY , VZ )

103 is the scaled wind speed, KX , KY and KZ are the diagonal terms
104 of the scaled eddy diffusivity tensor, is the scaled atmospheric
105 density, and S is the scaled source term. FALL3D solves Eq. (1) for
106 each particle class j using a terrain-following coordinate system
107 (x = X, y = Y, z Z). The scaling factors for this particular
108 case are given in Table 1. The generic particle class j is defined
109 by a triplet of values characterizing each particle (dp , p, Fp), that
110 are, respectively, diameter, density, and a shape factor. For dp we
111 use the equivalent diameter d, which is the diameter of a sphere of
112 equivalent volume. For the shape factor Fp we choose the spheric-
113 ity , which is the ratio of the surface area of a sphere with diam-
114 eter d to the surface area of the particle. In our approximation,
115 each triplet (d, p, ) is sufficient to define the settling velocity. As
116 mentioned above, the model assumes no particle-particle interac-
117 tion and that particles settle down at their terminal velocities.
118 Moreover we assume a negligible effect of Earths curvature. This
119 assumption constrains the applicability of the model to domains
120 smaller than few hundreds of kilometres.

4
 Parameter Scaling
Coordinates X=x Y =y Z = z h(x, y)
Velocities VX = vx VY = vy VZ = vz J 1 Vsj = vsj J 1
Diffusion Coefficients KX = Kx KY = Ky KZ = Kz J 2
Concentration C = cJ
Density = J
Source Term S = SJ
Table 1
Scaling factors for a terrain-following coordinate system (x = X, y = Y, z Z).
(x, y, z) are the Cartesian coordinates, h is the topographic relief, and J is the
determinant of the Jacobian of the coordinate system transformation.

121 2.2 Eddy Diffusivity Tensor

122 FALL3D admits either constant or variable values for vertical and
123 horizontal diffusion. For the more realistic case of variable diffu-
124 sion, the vertical Kz and the horizontal Kh = Kx = Ky compo-
125 nents of the eddy diffusivity tensor are estimated depending on
126 meteorological variables.

127 (1) Vertical component Kz . Inside the Atmospheric Boundary

128 Layer (ABL), FALL3D evaluates Kz as:

h z 1

z
! !

u z
1 1 + 9.2 h/L 0 stable



h L h!

Kz = (2)
z h z 1/2
!

u z 1 1 13 h/L 0 unstable



129 h Lh

130 where is the von Karman constant ( = 0.4), u is the

131 friction velocity, h is the ABL height, and L is the Monin-
132 Obukhov length (see Costa et al., 2006). The expression above
133 comes from an extension of the Monin-Obukhov similarity
134 theory to the entire ABL (Ulke, 2000). On the other hand,
135 above the ABL (z/h > 1), Kz is considered a function of the
136 local vertical wind gradient, a characteristic length scale lc ,
137 and a stability function Fc which depends on the Richardson

5
138 number Ri:
2  uz 
 
 
Kz = lc   Fc (Ri) (3)
139 z 

140 For lc and Fc , FALL3D adopts the relationship used by the

141 CAM3 model (Collins et al., 2004):
!1
1 1
lc = + (4)
142 z c
1


stable (Ri > 0)



1 + 10Ri(1 + 8Ri)

Fc (Ri) = (5)
1 18Ri unstable (Ri < 0)



143

144 where c is the so-called asymptotic length scale (c 30m).

145 (2) Horizontal components Kh. A large eddy parameterization as
146 that used by the RAMS model (Pielke et al., 1992) is used
147 for evaluating Kh :
v
u !2 !2 !2
u
2u vx vy vx vy
Kh = 3 max km ; (CSH ) t

+ + 2 + (6)

148
y x x y

149 where km = 0.0754/3, = xy, x and y are the
150 horizontal grid spacings, and CSH is a constant ranging from
151 0.135 to 0.32.

152 2.3 Settling velocity models

153 There are several semi-empirical parameterizations for the parti-

154 cle settling velocity vs if one assumes that particles settle down
155 at their terminal velocity:
v
u 4g (p
a ) d
u
vs = t
(7)
156 3Cda

157 where a and p denote air and particle density, respectively, d is

158 the particle equivalent diameter, and Cd is the drag coefficient. Cd

6
159 depends on the Reynolds number, Re = dvs/a (a = a /a is the
160 kinematic viscosity of air, a the dynamic viscosity). In FALL3D
161 several options are possible for estimating settling velocity, such
162 as:

163 (1) ARASTOOPOUR model (Arastoopour et al., 1982):

24

(1 + 0.15Re0.687) Re 103



Cd = Re

(8)
0.44 Re > 103


164

165 valid for spherical particles only.

166 (2) GANSER model (Ganser, 1993):
24 n 0.4305K2
1 + 0.1118 [Re (K1K2)]0.6567 +
o
Cd = 3305 (9)
ReK1 1+
167 ReK1 K2

where K1 = 3/(1 + 2 0.5), K2 = 101.84148(Log)

0.5743
168 are two
169 shape factors, and is the particle sphericity ( = 1 for
170 spheres).
171 (3) WILSON model (Walker et al., 1971; Wilson and Huang, 1979)
172 using the interpolation suggested by Pfeiffer et al. (2005):
24 0.828
q
+ 2 1.07 Re 102



Re




Cd = 1 1 Cd |Re=10 (103 Re) 102 Re 103 (10)


2

900
Re 103

1


173

174 where = (b + c)/2a is the particle aspect ratio (a b c

175 denote the particle semi-axes).
176 (4) DELLINO model (Dellino et al., 2005):
a  0.5206
vs = 1.2605 Ar 1.6 (11)
177
d
178 where Ar = gd3 (p a )a /2a is the Archimedes number,
179 g the gravity acceleration, and is a particle shape factor
180 (sphericity to circularity ratio).

7
181 Since for FALL3D the primary particle shape factor is the spheric-
182 ity , for sake of simplicity, it calculates in (10) and in (11)
183 approximating particles as prolate ellipsoids.

184 2.4 Meteorological variables

185 FALL3D reads time-dependent meteorological data (wind field, air

186 temperature, Monin-Obukhov length L, friction velocity u, and
187 ABL height h) and topography from a database created by an ex-
188 ternal utility program (SETDBS). There are two ways to generate
189 this database. The simplest option consists of using a horizontally
190 uniform wind derived from a vertical profile, typically obtained
191 from vertical sounding measurements. The second choice (CALMET
192 option) is more elaborate and uses data derived from the out-
193 put of the meteorological diagnostic model CALMET (Scire et al.,
194 2000). The second option is used for assimilating and interpolat-
195 ing short-term forecasts (or re-analysis) from Mesoscale Meteo-
196 rological Prognostic Models (MMPM). Assimilating topographic
197 information (roughness and terrain heights e) and the MMPM
198 output on a coarse mesh, CALMET generates a zero-divergence
199 wind field V = (VX , VY , VZ ) on a finer grid using a terrain fol-
200 lowing coordinate system X = x, Y = y, Z = z e(x, y). In
201 a first step, the initial guess wind field (in our case that given
202 by a MMPM) is adjusted for (i) kinematic terrain effects (lifting
203 and acceleration of the airflow over terrain obstacles), (ii) ther-
204 modynamically generated slope flows, and (iii) blocking effects.
205 After a divergence-minimisation procedure, a Step 1 mass con-
206 sistent wind field (VX1, VY 1 , Vz1) is obtained. In a further stage
207 meteorological observations are added to the Step 1 field and
208 an objective analysis procedure gives a second intermediate field
209 (VX2, VY 2 , VZ2). The scheme is designed so that observations are
210 used to correct the Step 1 wind field within a user-specified ra-
211 dius of influence, whereas it remains unchanged in subregions
212 where observations are unavailable. Finally, a new divergence-
213 minimisation procedure is applied iteratively to (VX2, VY 2, VZ2)

8
214 until the inequality:

215 V < (12)

216 is satisfied ( is a user-defined bound). The final products of

217 CALMET are a zero-divergence wind field consistent with the ob-
218 servations (or pseudo-observations), a finer scale interpolated
219 temperature field and all the quantities needed in the parameter-
220 isation of the eddy diffusivity tensor such as the Monin-Obukhov
221 length L, the friction velocity u , and the atmospheric boundary
222 layer height h (see section 2.2). The approximation of a zero-
223 divergence wind field (12) is fully adequate at heights close to
224 one kilometre (Dutton and Fichtl, 1969) although it is commonly
225 extended up to few kilometres (Sang et al., 1999; Park and Kim,
226 1999). In a further generalization of FALL3D, for applications that
227 cover large domains (several hundreds of kilometres), this limi-
228 tation can be overcome through a direct coupling of the model
229 with the meteorological fields furnished by MMPM, which typi-
230 cally involve horizontal spatial resolutions of about 5-10 km and
231 variable vertical grid spacings from about 50 m near the surface
232 to nearly one km at the upper layers.

233 2.5 Source term

234 FALL3D reads the time-dependent source term (mass released per
235 unit time at each grid point) from an external file. This file can
236 be generated by the SETSRC utility program (see section 6.3).

237 3 Equation-solving algorithm and code parallelization

238 FALL3D solves Eq. (1) on a structured grid using a FD explicit

239 algorithm. The mesh is assumed uniform along the horizontal
240 but it can vary along the vertical. The advective terms are dis-
241 cretised using a second-order Lax-Wendroff scheme whereas the

9
242 diffusive terms are evaluated using central differences (see Costa
243 et al. (2006) for details). As boundary conditions we assume zero
244 normal derivatives for outgoing fluxes and null concentrations for
245 in-going fluxes.
246 The parallel version of FALL3D is based on the Message-Passing
247 Interface (MPI) library. The parallelization of the code is done
248 at two levels, one for the particle classes and another for the
249 domain. First, the processors available are distributed among
250 groups. Each group works on one particle class or on a series
251 of particle classes. Since it is assumed that particles do not in-
252 teract, the first parallelization is straightforward. Second, if each
253 particle class has more than one processor assigned (i.e., if the
254 number of processors is a multiple of the number of classes), a sec-
255 ond parallelization is performed for the vertical layers. The first
256 parallelization (on the classes) scales linearly because broadcast
257 operations are minimum. This is not true for the second par-
258 allelization because broadcast operations among processors of a
259 group grow proportionally to the number of processors. The scal-
260 ability analysis has shown that optimal performance (best ratio
261 between CPU time and number of processors) is achieved when
262 each processor works with 3-4 vertical layers.

263 4 Program setup

264 For information regarding the installation and execution of FALL3D

265 and the utility programs please see the README Files.pdf in-
266 cluded in the distribution package. The FALL3D package comes
267 with a set of utility programs (see section 6) which can be used
268 to generate input files in the format required by FALL3D or to
269 post-process the results. The order of execution is the follow-
270 ing (see Figure 1): (i) run the program SETGRN to generate the
271 granulometry file, (ii) run the program SETDBS to generate the
272 database files containing meteorologic and topographic data, (iii)
273 run the program SETSRC to generate the source term file, (iv) run
274 FALL3D (serial or parallel version) and, finally, (v) run the pro-

10
275 gram FALL3DPOSTP to postprocess the results. Steps (i) and (iii)
276 can be avoided if the user provides the granulometry and source
files directly.
FileInp
FALL3D GRANULOMETRY SOURCE METEO BATABASE POSTPROCESS
block block block block block

SETGRN SETDBS FileDat

FileTop
FileGrn SETSRC FileDbs

FileSrc

FALL3D FileSym

FileRes FALL3DPOSTP

FileLog
FileGRD
FilePS

Fig. 1. Execution flow for FALL3D and the utility programs (see section 6). Boxes
indicate I/O files. File names are passed to programs as a call argument.
277

278 5 The I/O files

279 FALL3D and the utility programs use several files. For a detailed
280 description of the contents and formats of all files see the README Files.pdf
281 file located in the folder Documents. Names and locations of files
282 are defined by the user who can change them editing the script
283 files that launch the programs. The I/O files are:

284 The control input file FileInp. This file is composed of a se-
285 ries of blocks which define all the computational and physical
286 parameters needed by FALL3D and the utility programs. Each
287 program reads a single block of the file (blocks could, in fact,

11
288 be also in separate files, one for each program).
289 The granulometry file FileGrn. This file contains the sizes,
290 densities, shapes and distribution of particles. For some par-
291 ticular cases, it can be created by the utility program SETGRN.
292 The source file FileSrc. This file contains the definition of
293 the source term(s) for each particle class and at different time
294 slices. Source point number, position, and values (mass erup-
295 tion rate) can vary from one time slice to another. There is no
296 restriction on the number and duration of the time slices. That
297 allows the user to discretise any type of time-dependent source
298 term (time-dependent mass eruption rate, column height, etc).
299 This file can be created by the utility program SETSRC.
300 The meteorological data file FileDat. This file contains meteo-
301 rological data and is used as input for the SETDBS utility which
302 generates data needed at each node (see below). Two options
303 are available: FileDat can be i) an output of CALMET in binary
304 format or ii) a user-defined ASCII file containing the vertical
305 profile of wind and temperature at different time slices. These
306 values are typically measured through vertical soundings.
307 The topography file FileTop. This file contains the topogra-
308 phy in GRD format and it is used by the SETDBS program when
309 the option profile is used for the meteorological file FileDat.
310 When FileDat is an output of CALMET it is not necessary be-
311 cause CALMET output files already contain the topography. The
312 topography file specifies ground elevation at a regional scale
313 (i.e. in a region typically larger than the computational do-
314 main). Topography must be specified on a structured grid using
315 arbitrary (but constant) grid spacing. Discretisations along x-
316 and y-directions can be different. The only necessary require-
317 ment is that the computational domain must lay within the
318 bounds of the region where topography is specified. SETDBS
319 reads the topography file and automatically interpolates eleva-
320 tions onto the nodes of the computational grid.
321 The database file FileDbs. This is a direct access binary file
322 generated by the SETDBS utility program. It contains the to-
323 pography and meteorological data at each node of the compu-

12
324 tational grid for the run time interval.
325 The results file FileRes. This is a binary file that contains
326 the results of a FALL3D run. This file must be processed by
327 the FALL3DPOSTP utility program to produce files for the post-
328 processing phase (normally in GRD or PostScript formats).
329 The log file FileLog. This file contains information concerning
330 the run (summary of input data and memory requirements,
331 run-time error messages, CPU time, etc).
332 The symbols file FileSym. This file is optionally used by the
333 FALL3DPOSTP program for plotting symbols and legends in the
334 PostScript map files.

335 6 The utility programs

336 6.1 The program SETGRN

337 The program SETGRN is an optional utility program that reads

338 the GRANULOMETRY block from the control input file (FileInp)
339 and generates the granulometry file for FALL3D. It is assumed
340 that the mass fraction of particles follows a Gaussian distribu-
341 tion in and that the density of particles varies linearly with
342 ( is related to the particle diameter, expressed in millimeters,
343 by d(mm) = 2). Note that FALL3D can deal with more general
344 granulometric distributions different from Gaussian. In this case,
345 the granulometry file FileGrn must be supplied by the user di-
346 rectly, providing sizes, densities, sphericities and distribution of
347 particles (see README files.pdf).

348 6.2 The program SETDBS

349 The program SETDBS generates a database file needed by FALL3D.

350 As input either a vertical profile (sounding) together with a to-
351 pography file (in GRD format) or an output from the meteorolog-
352 ical processor CALMET (version 6.2) can be used. The latter option

13
353 is preferable because CALMET generates a 3-D wind field that ac-
354 counts for topographic effects and computes micrometeorological
355 variables in the ABL that are needed by FALL3D to estimate the
356 eddy diffusivity tensor (see section 2.2). FALL3D and the database
357 have the same horizontal discretisation, but the number and the
358 spacing of the vertical layers can be different.

359 6.3 The program SETSRC

360 The program SETSRC is an optional utility program that reads

361 the SOURCE block from the control input file (FileInp) and gen-
362 erates a source file for FALL3D. The program admits three types
363 of source: point source (mass is released in a single source point),
364 Suzuki distribution (Suzuki, 1983; Pfeiffer et al., 2005), and a
365 buoyant plume model (Bursik, 2001). The last option is more
366 elaborate and involves the solution of the 1D radial-averaged
367 plume governing equations that describe the convective region
368 of an eruptive column. These equations are intimately coupled
369 with the wind field which, for small plumes, may cause a sub-
370 stantial plume bend-over and, as a consequence, variations of
371 plume height and of mass release location. When this option is
372 chosen, the SETSRC program reads the values of the wind profile
373 at the vent from the database file (FileDbs) and then solves the
374 plume governing equations for each time slice and particle class
375 accounting for wind effect. Note that it introduces a time depen-
376 dence in the source term due to the wind variability even if all
377 the eruptive parameters (mass eruption rate, class fraction, etc)
378 are kept constant.

379 6.4 The program FALL3DPOSTP

380 The program FALL3DPOSTP is a program that reads a FALL3D

381 output binary file, calculates some relevant quantities at selected
382 heights and times, and produces elementary maps in GRD and
383 PostScript formats. Files in GRD format can be readed directly

14
384 by several plotting programs like the commercial software GRAPHER.
385 Alternatively, the user may also generate plots using functions
386 from several free packages (e.g. gnuplot).

387 7 Application Example

388 As an example of application we consider the paroxystic phase

389 of 22 July 1998 Mt. Etna eruption. Table 2 shows the files nec-
390 essary to run the Etna 1998 application example (these file are
391 included in the folder Runs/Etna1998 of the FALL3D distribution
392 package).
393 An intense explosive eruption was generated at Mt. Etna Vor-
394 agine Crater on 22 July 1998. The eruption reached its climax at
395 about 16:40 UTC (18:40 LT) forming a column 9 km high above
396 the vent (Andronico et al., 1999; Aloisi et al., 2002).
397 Field observations used for comparison and input data were de-
398 rived from Andronico et al. (1999) (available at http://www.ct.ingv.it/Progetti/Iavcei/
399 Concerning grain size distribution we considered 15 particle classes
400 ranging in size from = 3 (d = 8 mm) to = 4 (d = 62.5 m)
401 with a = 1. For each -value we accounted for both pumice
402 and lithic components, assuming constant relative fractions of
403 0.95 and 0.05, respectively. Particle sphericity values were as-
404 sumed similar to those measured for the 2002 Etna eruption
405 (Coltelli et al., 2008), and range from 0.93 to 0.95 (see the gran-
406 ulometry file Etna1998.grn).
407 In order to reconstruct the wind field during the eruption, we
408 used radio-sounding data from University of Wyoming (http://weather.uwyo.edu/)
409 at Trapani station on 22 July 1998. Since wind data at the erup-
410 tion time are unavailable, we calculated wind speed and direction
411 interpolating values from soundings at 12:00 and 24:00 UTC. The
412 file Etna1998.profile.dat, containing vertical profiles for veloc-
413 ity and temperature, can be used as input for the SETDBS utility to
414 generate an horizontally homogeneous wind database. However,
415 due to the long distance between Trapani and Mt. Etna (220 km)
416 and to the local effects of topography, the wind profile at Trapani

15
 File Description
Etna1998.inp Control input file.
Etna1998.grn Granulometry file. Not needed if SETGRN is used.
Etna1998.calmet62.dat CALMET output file.
Needed by SETDBS program if TYPEDATA=CALMET62.
Etna1998.profile.dat Vertical wind profile and temperature.
Needed by SETDBS program if TYPEDATA=PROFILE.
Etna1998.regionaltopo.grd Regional topography file in GRD format.
Needed by SETDBS program if TYPEDATA=PROFILE.
Etna1998.sym Symbols file. Optionally used by FALL3DPOSTP program.
Table 2
List of files needed to run the Etna 1998 application example.

417 may actually differ from that at Etna, especially in the lower at-
418 mospheric levels. For this reason we consider also a variable wind
419 field obtained by using CALMET. The file Etna1998.calmet62.dat
420 can be used by the SETDBS utility to generate a database having
421 a wind field which incorporates terrain effects.
422 Concerning the source term, the SETSRC utility using the buoyant
423 plume theory with a Mass Eruption Rate (MER) of 2.5106 kg/s
424 and a magma temperature of 1000 o C predicts a plume height of
425 9 km above the vent ( 12 km a.s.l.), in complete agreement
426 with the observations. The effective duration of the paroxystic
427 phase was set equal to 0.2h (12 minutes). The total mass, calcu-
428 lated as the MER times the eruption duration, is 1.8 109 kg,
429 consistent with the 1.5 109 kg estimated by Andronico et al.
430 (1999). It is worth nothing that the best-fit duration that repro-
431 duces the observed deposits ranges from about 5 to 20 minutes
432 depending on the weighting factor used in the minimization pro-
433 cedure (see e.g. Costa et al., 2008). As consequence total mass
434 estimation ranges from about 0.8 109 to 3 109 kg.
435 Figure 2 shows the deposit obtained by FALL3D using a grid with
436 51 51 18 nodes (1 km of horizontal spacing) and the wind
437 field generated by CALMET. The time evolution of the airborne
438 ash concetration at 1500 m a.s.l. is also shown in Figure 3. As-
439 suming 106 kg/m3 as a concentration threshold for flight safety,

16
440 the outer contours in the figure bound the unsafe region for the
441 fly level of 1500 m. Such information can be important for evalu-
442 ating whether landing is possible at the Catania airport.
In order to quantify the influence of the the spatial discretisa-
4190000
Taormina

0.03

0.13
4180000
0.36
83.95 0.34
20.24
25.42
14.64
Vent 0.94
0.32
0.48
Y coordinate (UTM in m)

5.15
3.50 2.68
1.47 3.21 1.55
0.26
0.27 1.68 1.48

4170000
1.52 0.99

0.44
0.49
0.11 0.50
0.44
0.32
Acireale
0.38

4160000 0.24
0.25

0.13
Catania
4150000

4140000
485000

495000

505000

515000

525000

X coordinate (UTM in m) 535000

Fig. 2. Etna 22 July 1998 deposit load at the end of the simulation (20:00h UTC).
Contours in kg/m2 . Dots show measured ground ash loads.
443

444 tion we also performed the same simulation using a finer mesh
445 (101 101 18) verifying that the effect of mesh refinement was
446 negligible.
447 We used the GANSER terminal velocity model and an eddy diffusiv-
448 ity tensor described by the similarity theory (vertical component)
449 and a constant horizontal diffusion (see section 2.2). For compar-
450 ison we carried out the same simulation using the DELLINO and
451 the WILSON terminal velocity models verifying that the choice of

17
 4190000 4190000
Taormina Taormina
17:00 UTC 18:00 UTC

4180000 4180000

Vent Vent

Y coordinate (UTM in m)

Y coordinate (UTM in m)
4170000 1E-004 4170000

Acireale Acireale
4160000 1E-005 4160000

Catania Catania
4150000 4150000
1E-006

4140000 4140000
485000

495000

505000

515000

525000

535000

485000

495000

505000

515000

525000

535000
X coordinate (UTM in m) X coordinate (UTM in m)
4190000 4190000
Taormina Taormina
19:00 UTC 20:00 UTC

4180000 4180000

Vent Vent
Y coordinate (UTM in m)

Y coordinate (UTM in m)
4170000 1E-004 4170000

Acireale Acireale
4160000 1E-005 4160000

Catania Catania
4150000 4150000
1E-006

4140000 4140000
485000

495000

505000

515000

525000

535000

485000

495000

505000

515000

525000

535000
X coordinate (UTM in m) X coordinate (UTM in m)

Fig. 3. Temporal evolution of the Etna 22 July 1998 airborne ash concentration at
1500 m a.s.l., from 17:00 UTC to 20:00h UTC. Contour levels indicate concentrations
of 104 , 105 , and 106 kg/m3 . The latter value can be considered as a concentration
threshold for air navigation safety.

452 the model has not a very strong effect on the simulation results,
453 in agreement with Scollo et al. (2008). It also is worth noting
454 that, in cases when the wind field comes from a single sound-
455 ing, large eddy models (see option RAMS model) can underes-
456 timate turbulent diffusion because there is no horizontal shear.
457 In fact, data best-fit suggests a constant horizontal diffusion of
458 Kh 5000 m2/s whereas the RAMS option gives a lower average
459 value (Kh 2000 m2/s). The best-fit input parameters are re-
460 ported in Table 3. Comparison between simulation and observed
461 ground load (Andronico et al., 1999) is shown in Figure 4. The
462 agreement is satisfactory for the whole range of values, which
463 comprises more than four orders of magnitude.

18
 100

10
Computed value (kg/m2)

0.1

0.01
0.01 0.1 1 10 100
Observed value (kg/m2)
Fig. 4. Comparison between computed and observed ground ash loadings. Dashed
lines indicate over- or under-estimations of 1/5 and 5 times the observed values.

464 8 Summary and discussion

465 FALL3D is a 3-D time-dependent model for the transport and de-
466 position of fine particles. The model, based on the Eulerian form
467 of the ADS equation, is coupled with meteorological data (vertical
468 sounding or forecast/re-analysis from prognostic mesoscale mod-
469 els), accounts for terrain effects, adopts similarity theory and LES
470 for quantifying the turbulent diffusion, and uses semi-empirical
471 parameterisations for the particle terminal settling velocities. The
472 source term can be described by using the buoyant plume the-
473 ory. The ADS equation is solved on a structured terrain-following
474 mesh using a second-order FD explicit algorithm. The code, writ-

19
 Parameter (unit) Value
Vent Coordinates (UTM) (500000, 4176000)
Vent Elevation (m) 2700
Exit Velocity (m/s) 100
Vent Magma Temperature (o C) 1000
MER (kg/s) 2.5 106
Column Height (m) 9000
Eruption Duration (h) 0.2
Erupted Mass (kg) 1.8 109
Meteorological Input Data http://weather.uwyo.edu/ (sounding at Trapani)

Table 3
Parameters used to simulate Etna ash dispersal from 22 July 1998. ( ) value derived
from best-fit of data.

475 ten in FORTRAN 90, is available in both serial and parallel ver-
476 sions, and distributed together with a set of pre- and post-process
477 utility programs.
478 FALL3D can be used to forecast both ash concentration in air and
479 ash load at ground. However, the model could also be used to
480 simulate the transport and deposition of airborne particles differ-
481 ent from volcanic ash. Other potential applications of the model
482 include dispersion of pollutants on complex terrains such as par-
483 ticles released during forest fires, fine dust and inert aerosols, or
484 emissions from large industrial stacks.
485 Although different applications have shown the reliability of FALL3D
486 to model volcanic ash dispersion, the model still has some limita-
487 tions to be addressed in the future. Improvements should include:

488 Use of spherical coordinates for large domains.

489 Adopt an anelastic atmosphere model. It mainly concerns the
490 preprocess treatment but would also require addition of an
491 extra term in the ADS equation.
492 Local effects of eruptive plume-atmosphere interaction for large
493 columns.
494 Particle aggregation, vapour and thermal effects.

20
495 Acknowledgments

496 This work has been supported by the MIUR-FIRB Italian project Sviluppo Nuove
497 Tecnologie per la Protezione e Difesa del Territorio dai Rischi Naturali and by the
498 Department of Civil Protection of Italy. D. Andronico, M. Coltelli and S. Scollo are
499 warmly thanked for providing some field and laboratory data. Finally, we wish to
500 thank H.C. Saunderson and an anonymous reviewer for improving the quality of
501 the manuscript.

21
502 References

503 Aloisi, M., DAgostino, M., Dean, K., Mostaccio, A., Neri, G.,
504 2002. Satellite analysis and PUFF simulation of the eruptive
505 cloud generated by the Mount Etna paroxysm of 22 july 1998.
506 J. Geophys. Res. 107 (B12), doi:10.1029/2001JB000630.
507 Andronico, D., Del Carlo, P., Coltelli, M., 1999. The 22 July 1998
508 fire fountain episode at Voragine Crater (Mt. Etna, Italy). In:
509 VMSG 1999, Annual Meeting. Volcanic and Magmatic Studies
510 Group, Birmingham, UK.
511 Arastoopour, H., Wang, C., Weil, S., 1982. Particle-particle in-
512 teraction force in a diluite gas-solid system. Chem. Eng. Sci.
513 37 (9), 13791386.
514 Barberi, F., Macedonio, G., Pareschi, M., Santacroce, R., 1990.
515 Mapping the tephra fallout risk: an example from Vesuvius
516 (Italy). Nature 344, 142144.
517 Barsotti, S., Neri, A., Scire, J., 2008. The VOL-CALPUFF model
518 for atmospheric ash dispersal. I. Approach and physical formu-
519 lation. J. Geophys. Res.(in press).
520 Bursik, M., 2001. Effect of wind on the rise height of volcanic
521 plumes. Geophys. Res. Lett. 18, 36213624.
522 Collins, W., Rasch, P., Boville, B., Hack, J., McCaa, J.,
523 Williamson, D., Kiehl, J., Briegleb, B., 2004. Description of
524 the NCAR Community Atmosphere Model (CAM 3.0). Tech-
525 nical Report NCAR/TN-464+STR, National Center for Atmo-
526 spheric Research, Boulder, Colorado.
527 Coltelli, M., Miraglia, L., Scollo, S., 2008. Characterization of
528 shape and terminal velocity of tephra particles erupted dur-
529 ing the 2002 eruption of Etna volcano, italy. Bull. Volcanol.(in
530 press).
531 Connor, C., Hill, B., Winfrey, B., Franklin, M., La Femina, P.,
532 2001. Estimation of volcanic hazards from tephra fallout. Nat-
533 ural Hazards Review 2, 3342.
534 Costa, A., DellErba, F., Di Vito, M., Isaia, R., Macedonio, G.,
535 Orsi, G., Pfeiffer, T., 2008. Assessment of the volcanic ash
536 loading hazard from phlegrean fields caldera (italy). Bull. Vol-

22
537 canol.(in press).
538 Costa, A., Macedonio, G., Folch, A., 2006. A three-dimensional
539 Eulerian model for transport and deposition of volcanic ashes.
540 Earth Planet. Sci. Lett. 241, 634647.
541 DAmours, R., 1998. Modeling the ETEX plume dispersion with
542 the Canadian emergency response model. Atmos. Environ.
543 32 (24), 43354341.
544 Dellino, P., Mele, D., Bonasia, R., Braia, L., La Volpe, R., 2005.
545 The analysis of the influence of pumice shape on its terminal
546 velocity. Geophys. Res. Lett. 32 (L21306).
547 Dutton, J., Fichtl, G., 1969. Approximate equations of motion
548 for gases and liquids. J. Atm. Sci. 26, 241254.
549 Folch, A., Cavazzoni, C., Costa, A., Macedonio, G., 2008. An
550 automatic procedure to forecast tephra fallout. J. Volcanol.
551 Geotherm. Res.(in press).
552 Ganser, G., 1993. A rational approach to drag prediction of spher-
553 ical and nonspherical particles. Powder Technol. 77, 143152.
554 Heffter, J., Stunder, B., 1993. Volcanic Ash Forecast Transport
555 and Dispersion (Vaftad) Model. Weather and Forecasting 8,
556 533541.
557 Hurst, A., 1994. ASHFALL - A computer program for estimat-
558 ing volcanic ash fallout. Report and users guide. Science Re-
559 port 94/23, 22p., Institute of Geological and Nuclear Sciences,
560 Wellington, New Zealand.
561 Macedonio, G., Costa, A., Longo, A., 2005. A computer model
562 for volcanic ash fallout and assessment of subsequent hazard.
563 Computers & Geosciences 31, 837845.
564 Park, S., Kim, C., 1999. A numerical model for the simulation
565 of so2 concentrations in the Kyongin region, Korea. Atmos.
566 Environ. 33, 31193132.
567 Pfeiffer, T., Costa, A., Macedonio, G., 2005. A model for
568 the numerical simulation of tephra fall deposits. J. Volcanol.
569 Geotherm. Res. 140, 273294.
570 Pielke, R., Cotton, W., Walko, R., Tremback, C., Nicholls, M.,
571 Moran, M., Wesley, D., Lee, T., Copeland, J., 1992. A compre-
572 hensive meteorological modeling system-RAMS. Meteor. At-

23
573 mos. Phys. 49, 6991.
574 Sang, J., Lin, G., Zhang, B., 1999. Numerical modeling for emer-
575 gency response of nuclear accident. J. Wind Eng. Ind. Aerod.
576 81, 221235.
577 Scire, J., Robe, F., M.E., F., Yamartino, R., 2000. A Users Guide
578 for the CALMET Meteorological Model. Tech. Rep. Version 5,
579 Earth Tech, Inc., 196 Baker Avenue, Concord, MA 01742.
580 Scollo, S., Folch, A., Costa, A., 2008. A parametric and com-
581 parative study of different tephra fallout models. J. Volcanol.
582 Geotherm. Res.(in press).
583 Searcy, C., Dean, K., Stringer, W., 1998. Puff: A high-resolution
584 volcanic ash tracking model. J. Volcanol. Geotherm. Res. 80,
585 116.
586 Suzuki, T., 1983. A theoretical model for dispersion of tephra. In:
587 Shimozuru, D., Yokoyama, I. (Eds.), Arc Volcanism: Physics
588 and Tectonics. Terra Scientific Publishing Company (TERRA-
589 PUB), Tokyo, pp. 93113.
590 Ulke, A., 2000. New turbulent parameterization for a disper-
591 sion model in atmospheric boundary layer. Atmos. Environ.
592 34, 10291042.
593 Walker, G., Wilson, L., Bowell, E., 1971. Explosive volcanic erup-
594 tions I. Rate of fall of pyroclasts. Geophys. J. Roy. Astron. Soc.
595 22, 377383.
596 Wilson, L., Huang, T., 1979. The influence of shape on the atmo-
597 spheric settling velocity of volcanic ash particles. Earth Planet.
598 Sci. Lett. 44, 311324.

24