Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115, B09202, doi:10.

1029/2009JB007176, 2010

A model for wet aggregation of ash particles in volcanic plumes


and clouds:
2. Model application
A. Folch,1 A. Costa,2,3 A. Durant,4,5 and G. Macedonio2
Received 30 November 2009; revised 18 March 2010; accepted 15 April 2010; published 1 September 2010.

[1] The occurrence of particle aggregation has a dramatic effect on the transport and
sedimentation of volcanic ash. The aggregation process is complex and can occur under
different conditions and in multiple regions of the plume and in the ash cloud. In the
companion paper, Costa et al. develop an aggregation model based on a fractal relationship
to describe the rate particles are incorporated into ash aggregates. The model includes the
effects of both magmatic and atmospheric water present in the volcanic cloud and
demonstrates that the rate of aggregation depends on the characteristics of the initial
particle size distribution. The aggregation model includes two parameters, the fractal
exponent Df, which describes the efficiency of the aggregation process, and the aggregate
settling velocity correction factor ye, which influences the distance at which distal mass
deposition maxima form. Both parameters are adjusted using features of the observed
deposits. Here this aggregation model is implemented in the FALL3D volcanic ash
transport model and applied to the 18 May 1980 Mount St. Helens and the 17–18
September 1992 Crater Peak eruptions. For both eruptions, the optimized values for Df
(2.96–3.00) and ye (0.27–0.33) indicate that the ash aggregates had a bulk density
of 700–800 kg m−3. The model provides a higher degree of agreement than previous
fully empirical aggregation models and successfully reproduces the depositional
characteristics of the deposits investigated over a large range of scales, including
the position and thickness of the secondary maxima.
Citation: Folch, A., A. Costa, A. Durant, and G. Macedonio (2010), A model for wet aggregation of ash particles in volcanic
plumes and clouds: 2. Model application, J. Geophys. Res., 115, B09202, doi:10.1029/2009JB007176.

1. Introduction transport, dispersion and fallout is a matter of interest for


scientists and civil authorities. Volcanic ash transport and
[2] Sustained explosive volcanic eruptions generate col-
dispersal models (VATDM) are routinely used to forecast the
umns that inject a mixture of hot gases, particles and aerosol
trajectories of ash clouds, airborne particle loadings and the
into the atmosphere forming volcanic clouds. Ash particles
spatial distribution of ash fallout [Folch et al., 2008a; Scollo
in volcanic clouds are transported by a series of mechanisms
et al., 2009].
including advection by regional winds, diffusion by atmo-
[3] Fine ash particles form aggregates both in the eruption
spheric turbulence, and deposition by gravitational settling
column and the ash cloud during dispersion and are respon-
[Costa et al., 2006]. Volcanic ash clouds present hazards to
sible for the formation of distal mass deposition maxima [e.g.,
modern aircraft for many days after an eruptive event and
Carey and Sigurdsson, 1982; Durant et al., 2009]. Ash
often force rerouting to prevent in‐flight encounters. Tephra aggregates are larger and can have a lower bulk density than
fallout can also cause damage to human settlements and
the primary particles from which they are formed. This dra-
infrastructure at short to medium distances from the volcano
matically changes the sedimentation dynamics and reduces
[Blong, 1984; Casadevall, 1994; Casadevall et al., 1996; the atmospheric residence time of the very fine ash fraction.
Baxter, 1999; Pomonis et al., 1999; Spence et al., 2005;
In the majority of cases, VATDM forecasts deviate substan-
Macedonio et al., 2008; Costa et al., 2009]. It follows that ash
tially from observations because ash particle aggregation is
not taken into account [Sparks et al., 1997]. While particle
1
Barcelona Supercomputing Center, Barcelona, Spain. aggregation has been linked to enhanced fine particle settling
2
Istituto Nazionale di Geofisica e Vulcanologia, Naples, Italy. and the formation of distal mass deposition maxima, there
3
Department of Earth Sciences, University of Bristol, Bristol, UK. remains a low level of understanding on the physics of the
4
Department of Geography, University of Cambridge, Cambridge, UK.
5
Department of Geological and Mining Engineering and Sciences,
process. The interactions of surface liquid layers, electro-
Michigan Technological University, Houghton, Michigan, USA. static forces, differences in particle settling velocity, and
formation of hydrometeors have been proposed as contrib-
Copyright 2010 by the American Geophysical Union. uting mechanisms [Sparks et al., 1997].
0148‐0227/10/2009JB007176

B09202 1 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

[4] VATDM would need to include a full aggregation [7] 1. The total grain size distribution (TGSD) is dis-
model that accounts for interaction between all particle size cretized in nc particle classes (j = 1,…,nc). Only classes
classes. However, this approach is challenging to implement having diameters between dmin and dmax participate in
and calibrate, and becomes strictly prohibitive from a com- aggregation. Typically, dmin is the minimum considered
putational point of view because the number of particle size size and dmax falls in the submillimetric range (commonly
classes rapidly increases as aggregate growth occurs. The few F = 2) [Cornell et al., 1983; Schumacher and Schmincke,
attempts to include aggregation in VATDM [e.g., Cornell 1995; Textor et al., 2006a, 2006b].
et al., 1983; Armienti et al., 1988; Bonadonna et al., 2002] [8] 2. Particles aggregate only in an aggregation class
are based on the a priori assumption that a certain mass characterized by a diameter dA with dA > dmax.
fraction of fine classes fall as a single aggregate class. The [9] 3. The number of primary particles of diameter dj 2
aggregate settling velocity and aggregate fraction are typi- (dmin, dmax) in an aggregate of diameter dA is given by the
cally derived from field observations and without theoretical fractal relationship:
justification. For example, Cornell et al. [1983] analyze the  Df
Campanian Ignimbrite (Y5 ash) deposit and determined that dA
Nj ¼ kf ð1Þ
50% of the 63–44 mm ash, 75% of the 44–31 mm ash and dj
100% of the less than 31 mm ash could be treated as aggre-
gated particles with a diameter of 200 mm and a density of where kf is a fractal prefactor (kf ≈ 1) and Df is the fractal
200 kg m−3. From a modeling point of view, this very sim- exponent (Df ≤ 3).
plified approach is attractive because adds no extra compu- [10] 4. The number of particles of class j per unit volume
tational cost (only the input bulk granulometry must be that aggregate during a time interval Dt is
modified). However, such a simplified approach is not based
Dntot Nj
on variables directly related to the physics of the aggregation Dnj  P ðk ¼ kmin ; ::; kmax Þ ð2Þ
process and the physical variables leading these processes k Nk
(e.g., grain size distribution, quantity of atmospheric and
magmatic water available) can change from eruption to where Dntot is the total particle decay per unit volume during
eruption and/or in different regions of a volcanic cloud. As the the interval Dt:
fraction of aggregating mass is set a priori and fixed ad hoc to 
23=D
some values, Cornell et al. [1983] parameterization does not Dntot ¼ m AB n2tot þ AS 3=Df ntot f þ

discern between any different aggregation kernel or process. 24=D
þ ADS 4=Df ntot f Dt ð3Þ
[5] In the companion paper, Costa et al. [2010] develop
a simplified aggregation model which gives a theoretical
justification for aggregation in VATDM and reaches a com- where  is the solid volume fraction, am is a mean (size class
promise between the unaffordable all size class approach and averaged) sticking efficiency, and ntot is the total number of
the simplest classical strategy based on predefined percen- particles per unit volume that can potentially participate in
tages. In this paper we implement the Costa et al. [2010] the aggregation process. The expression above comes from
aggregation model in the FALL3D ash transport model integrating the coagulation kernel over all particle sizes, and
[Costa et al., 2006; Folch et al., 2009] and validate the result involves the product of the (averaged) sticking efficiency
against observations from two well‐studied recent eruptions times the collision frequency function which accounts for
where aggregation played a relevant role: the 18 May 1980 Brownian motion (AB), laminar and turbulent fluid shear (AS),
Mount St. Helens (MSH1980) and the 16–17 September and differential sedimentation (ADS):
1992 Crater Peak, Mount Spurr (CP1992) eruptions. The
manuscript is arranged as follows. First, we summarize the 4kb T
AB ¼ 
aggregation model presented by Costa et al. [2010] and 3a
describe the details of its implementation in the FALL3D 2GS 3
AS ¼ ð4Þ
model. Second, we use data from MSH1980 [Durant et al., 3
2009] and CP1992 [Durant and Rose, 2009] to calibrate two ðm  a Þg4
adjustable parameters of the aggregation model. The calibration ADS ¼ 
48a
is performed through a best fit with the observed deposits. The
model is also validated through a semiquantitative compari- where kb = 1.38 × 10−23 m2 kg s−2 K is the Boltzmann con-
son of the simulation results with weather radar and satellite stant, T is the absolute temperature, ma is the dynamic vis-
observations. Finally, we present a general discussion on cosity of air, GS is the fluid shear, x = djv−1/D
j
f
is the particle
the role of plume dynamics and magmatic water content on diameter to volume fractal relationship, vj is the particle
the formation of aggregates, summarize the main limita- volume (pd3j /6 for spheres), rm is the mean (class averaged)
tions of the aggregation model, and discuss further theoret- particle density, and ra is the air density.
ical and experimental developments necessary for improving [11] 5. The total number of available particles per unit
the model. volume ntot is evaluated at each point from the j class
concentration:
2. Aggregation Model Overview X 6Cj
ntot ¼ 3
ð j ¼ jmin ; ::; jmax Þ ð5Þ
[6] We begin by summarizing the characteristics of the j j dj
aggregation model used in the current study, which is fully
derived by Costa et al. [2010]:

2 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

   
[12] 6. The class‐averaged sticking efficiency am is 273:16 273:16
computed as log ei ¼ 9:097  1  3:566 log
T T
P P  
fi fj ij T
i j
m ¼ P P ð6Þ þ 0:876 1  þ log ð6:1071Þ ð9Þ
273:16
i j fi fj

where fk is the class mass fraction, aij is the sticking effi- where T is given in K and ew and ei in hPa.
ciency between the classes i and j, and the sums on i and j [14] 8. In presence of a pure ice phase (i.e., when ew > e >
are performed over the aggregating classes between kmin and ei or e > ew and T < Tf = 273 K), it is assumed that [Field
kmax. Although the model formulation is general, we con- et al., 2006]
sider to a first approximation aggregation influenced only by
m ¼ 0:09 ð10Þ
the presence of water. Dry aggregation is mainly driven by
electrostatic forces that are much weaker than capillary forces
by several orders of magnitude [Schumacher and Schmincke, This value of am should be considered as an upper value as it
1995]. There are also disaggregation processes (e.g., colli- is assumed that ash particles stick as ice particles. Although
sions that destroy or split aggregates) but without any ash particle surfaces have high rugosity and low sticking
research results in this area, for sake of simplicity, here efficiency, the ash may be encased in ice and then the
we neglect them. As the model calculation of am depends on assumption is valid.
the presence of liquid water or ice, it is necessary to discrim- [15] 9. In presence of a liquid phase (i.e., for e > ew and T >
inate proportions of the different water phases, the stability of Tm = 273 K), the model considers
which depend on the water vapor partial pressure in air e. 1
[13] 7. The water vapor partial pressure in air e depends ij ¼  q ð11Þ
on pressure P and mixing ratio w (ratio of the mass of water 1 þ Stij =Stcr
vapor to the mass of dry air):
where Stcr = 1.3 is the critical Stokes number, q = 0.8 is a
w constant, and Stij is the Stokes number based on the binder
e¼ P ð7Þ
w þ 0:622 liquid (water) viscosity:
 
where 0.622 is the ratio of specific gas constants of water 8m Vsi  Vsj  di dj
vapor to dry air (pressure here is expressed in Pa). If e Stij ¼ ð12Þ
9l di þ dj
exceeds the saturation vapor pressure over liquid water ew
(e > ew), supersaturation is reached and water vapor con- where Vsi and Vsj are the settling velocity of particle classes i
denses to a stable liquid phase. Freezing of water in the and j.
volcanic ash does not occur at the same temperature of the [16] 10. In our model, aggregation requires the presence
pure water [Lane and Gilbert, 1992; Durant et al., 2008]. of water either in the liquid or solid phase. This implies, for
Actually it is a stochastic process that occurs over a tem- a given ambient pressure and temperature, that wet aggre-
perature range of Tf < 250–260 K through heterogeneous gation will occur only if a minimum amount of water vapor
nucleation initiated by volcanic ash, whereas melting occurs exists (a minimum mixing ratio). In volcanic clouds, the
always around 273 K [Durant et al., 2008]. Here for sake of total water vapor content consists of both atmospheric water
computational simplicity we set Tf = Tm = 273 K. Model vapor and the magmatic water vapor transported by the
results were not very sensitive to changes of this temperature plume. Here we account for both contributions and the
of about ten Kelvin (trials were performed setting Tf = Tm). atmospheric water vapor is given at each computational
However, the effect of choosing Tf lower than Tm can have a node by a meteorological model. The magmatic water vapor
more significant effect inside the eruption column where the is introduced in the source term and transported as an addi-
plume rising becomes colder. During the settling from the tional noninteracting class of particles with a diameter 1 mm,
umbrella region (of high eruption columns) the ice contained assumed to be representative of ice crystals and water droplets
in the particles melts always around Tm = 273 K. To better (future model developments will account for a more dynamic
account for the effect of a difference between the freezing size distribution).
and melting temperature of water it would be necessary to [17] 11. The settling velocity of an aggregate of given size
develop a more sophisticated model of the eruption column. is less than the settling velocity of a “compact” particle of
Furthermore, if temperature falls within the ice phase sta- the same size, which has a density determined by taking the
bility field and e exceeds the saturation vapor pressure over average density of the aggregating particles [Costa et al.,
ice ei (e > ei), ice particles will grow directly from the vapor 2010]. The settling velocity of aggregates VsA is computed
phase through deposition. The values of ew and ei depend on as
temperature, and are computed as [Buck, 1981; Murphy and
Koop, 2005] VsA ¼ e VsC ð13Þ
 
T  273:16
ew ¼ 6:112 exp 17:67 ð8Þ where VsC is the settling velocity of a compact particle and
T  29:65 ye is a correction factor to be determined empirically, which
accounts for effects of aggregate porosity and permeability.

3 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Figure 1. Skew‐T Log‐P diagrams at Spokane, 18 May 1980 at 1200 UTC. (top) Data from the atmo-
spheric sounding. (bottom) Interpolation of WRF inner nest at Spokane. Blue lines are temperature. Wind
speed and direction are also shown.

4 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Table 1. Total Grain Size Distribution Used in the MSH1980 and mixture properties at the vent (exit temperature, velocity,
Simulationsa water content and bulk granulometry). BPT also calculates
d r Weight
temperature along the plume. The main volcanological inputs
F (mm) (kg m−3) Percent Comments necessary for running FALL3D are the MER, or alternatively
the column height (H), and the TGSD. In order to determine
−2 4000 914 1 pumice and lithic average
−1 2000 1824 1 pumice and lithic average
the necessary meteorological inputs, FALL3D was coupled
0 1000 2200 3 pumice and lithic average to the Weather Research and Forecasting (WRF) model
1 500 1800 6 pumice, lithic and crystal average [Michalakes et al., 2004] using an off‐line approach. WRF
2 250 2600 13 pumice, lithic and crystal average is a fully compressible, Eulerian, nonhydrostatic mesoscale
3 125 2100 1 pumice meteorological model that solves the equations of atmospheric
3 125 2800 5 lithic and crystal average
4 62 2350 3 pumice motion. The model was configured to integrate the primi-
4 62 2900 7 lithic and crystal average tive equations using the Advanced Research WRF (ARW)
5 31 2600 13 pumice dynamics solver [Skamarock et al., 2005] in a high‐resolution
5 31 3500 1 crystals nested domain. After running WRF, meteorological variables
6 15 2620 41 average of all minor components were interpolated from the WRF mesh to the terrain‐following
water 1 1000 5
FALL3D grid at user‐defined time resolution.
a
Granulometric distribution assumes 12 classes ranging from F = −2 to [20] We implemented the aggregation model in the par-
F = 6 plus a 5% of magmatic water. allel version of FALL3D. Parallelization operates at two
levels; one for particle classes (a group of processors is
Accounting only for settling in the Stokes laminar regime, assigned to a single particle class) and another for the
the above expression reduces to domain (each processor of a group working with a given
class, is assigned to a part of the computational domain).
ðm  a Þg 2 Whenever meteorological variables change, the collision
VsA ¼ e dA ð14Þ
18a frequency functions (4) and the averaged sticking efficiency
(6) are updated at each spatial point. At each time step Dt, it is
[18] In summary, given (1) the concentration of each necessary to modify the FALL3D source/sink term of each
particle class at a point and time instant, (2) the class particle class Sj:
properties (dj, rj, fj, Vsj and x), and (3) the meteorological
variables (w, T, ma, GS and ra), equations (1) to (14) give the dj j
Sj ¼ Sjo þ Dnj ð15Þ
particle class decay per unit volume and time and the 6
aggregate settling velocity. The aggregation model has two
parameters that need to be calibrated, the fractal exponent Df where Dnj is given by (2) and Sjo is the original source term.
(equation (1)) and the aggregate settling velocity correction From a computational point of view, the main consequences
factor ye (equation (13)). The first parameter gives a mea- of including the aggregation model are (1) introduction of a
sure of how effective the aggregation process is (the value is nonlinear source term as Sj depends on concentration and
proportional to the number of primary particles in a given (2) coupling of the different granulometric classes. In the
aggregate). The second parameter accounts for the settling parallel version it implies a considerable increase of the
velocity of aggregates and affects the position of the sec- computational time because each group of processors assigned
ondary maximum. Lower values of ye result in smaller set- to a given class must now communicate and exchange values
tling velocities which increases the distance from the volcano of concentration at each time step.
to the distal mass deposition maximum resulting from [21] The aggregation model has two parameters that need
deposition of aggregates. to be estimated, the fractal exponent Df (equation (1)) and the
aggregate settling velocity correction factor ye (equation (13)).
Due to the lack of experimental data of volcanic ash aggre-
3. Modeling System gation, we estimate them through a best fit procedure with the
observed deposits. In order to select the best values we
[19] The modeling system combines a meteorological
minimized the following deviation of regression [Costa et al.,
model, a volcanic plume model and a tephra dispersal model.
2009]:
FALL3D [Costa et al., 2006; Folch et al., 2008b, 2009] is
an Eulerian VATDM that can be used from local to mesoscale. X
N
The model solves the advection‐diffusion‐sedimentation D2 ¼ wi ðYci  Yoi Þ2 ð16Þ
equation with turbulent diffusion based on gradient transport i¼1
theory, several semiempirical particle terminal velocity para-
meterizations, and a time‐dependent three‐dimensional wind and the normalized mean bias (NMB):
field furnished by global or mesoscale meteorological models. PN
ðYci  Yoi Þ
The source term is described in terms of the Buoyant Plume NMBð%Þ ¼ 100  i¼1
PN ð17Þ
Theory (BPT) [Bursik, 2001]. We apply BPT as described by i¼1 Yoi
Costa et al. [2006] but estimate the radial entrainment
parameter [Bursik, 2001] as a function of the local Richardson where N is the number of points (119 in the MSH1980 case
number [Carazzo et al., 2006, 2008], whereas the wind and 23 in the CP1992 case), Yci and Yoi denote the computed
entrainment parameter [Bursik, 2001] is set to 0.34. Moreover, and the observed ground load, respectively, at each point
an iterative procedure allows us to estimate the mass eruption i, and wi are weighting factors. The value of the weighting
rate (MER) for a given plume height, ambient conditions, factors depends upon the distribution of the errors [Aitken,

5 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

blast. Durant et al. [2009] provide an overview of the eruptive


chronology. A number of coignimbrite columns were gen-
erated to the north of the volcano [Sparks et al., 1986] and
the associated ash clouds were transported eastward by the
dominant regional winds. Approximately 30 min after the
initial event the eruption generated a sustained Plinian col-
umn that reached a height in excess of 30 km [Holasek and
Self, 1995]. For the following 8 hours the eruption column
height fluctuated between approximately 13 and 19 km. The
climactic phase occurred midafternoon and was dominated
by coignimbrite activity and concluded with a late Plinian
phase until 18 May 1730 LT (19 May 0030 UTC). The
eruption waned until 2230 LT (19 May 0530 UTC) with
pulses of eruptive activity that generated weak collapsing
columns ( 2 to 4 km height) and small pyroclastic flows. The
MSH1980 ash fall deposit blanked a vast area with ash fall
reported as far south as Oklahoma. The ashfall blanket over
Washington State featured a distal mass deposition maxi-
Figure 2. Eruption column height (km above sea level) mum, the so‐called “Ritzville bulge,” at about 325 km from
used in the simulations for each eruption phase (black line, the volcano [Sarna‐Wojcicki et al., 1981]. There were
see Table 2 for details). Heights are obtained from averaging abundant observations of ash aggregate fallout from the
radar measurements (filled circles) [Harris et al., 1981]. For eruption cloud that coincided with the distal mass deposition
comparison, empty circles show also maximum cloud maximum. Additional observations, such as mammatus on
heights inferred from GOES satellite images [Holasek and the base of the MSH1980 cloud, indicate that cloud micro-
Self, 1995]. physics played a major role in the dynamics of the distal ash
cloud and ash sedimentation [Durant et al., 2009].
1935]. If all measurements have an equal uncertainty the
weighting factors are uniform (in this study we used w−1 i = 4.1. Model Initialization
PN 2
k¼1 Y ok ). If the observations have different uncertainty [23] In order to simulate the event, we first ran the WRF
distributions proportional weighting factors (here w−1 i = model [Michalakes et al., 2004] using a nesting procedure
2 −1 with nudging to generate a 4D meteorological data set from
NY ) or statistical
qoiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi weighting factors (here w i =
PN 2
18 May 0000 UTC to 20 May 0000 UTC. The inner WRF
( N k¼1 Yok )Yoi) have to be used. In practice, uniform model nest had horizontal spatial resolution of 6 km and
weighting factors tend also to give more relevance to the 40 vertical layers (top pressure 10 hPa, about 31 km). The
points with large values of Yoi, i.e., favor the considered model was initialized with the NCEP‐DOE Reanalysis‐
proximal points and those of the secondary maxima. In 2 data at 2.5° resolution, allowing 15 h for model heat up.
contrast, proportional weighting factors tend to favor the Figure 1 compares the Skew‐T Log‐P diagrams of an
weight of the points with smaller values of Yoi, i.e., the more atmospheric sounding at Spokane on 18 May 1200 UTC
distal points or those at the margins of the deposit. The best and the WRF model results at the same locality and time.
fit procedure consists of finding the (Df,ye) couple which Temperature profiles and wind fields above the planetary
minimizes the D2 function with the weighting factors that boundary layer (PBL), where most of the ash transport
produce the minimum NMB. In all cases we found PNthat the occurred, show good agreement. We also compared WRF
best weighting factor was the uniform, i.e., w−1
i =
2
k¼1 Yok.
modeled and observed soundings at Boise, Quillayute, and
Salem (not shown in Figure 1), which all demonstrated a
high degree of agreement.
4. Test Case I: May 1980 Mount St. Helens
[24] The WRF meteorological variables were then inter-
Eruption polated within the FALL3D mesh with a time resolution of
[22] The 18 May 1980 eruption of Mount St. Helens 30 min. The FALL3D computational domain spans from
(MSH1980) volcano (46.20°N, 122.18°W) began at 0832 LT 45°N to 50°N in latitude and from 123°W to 114°W in lon-
(U.S. Pacific Daylight Time; 1532 UTC) as the flank of the gitude, with horizontal and vertical spatial resolutions of
volcano collapsed and initiated a northward directed lateral 5 and 1 km, respectively. For the MSH1980 simulations we

Table 2. Time Interval of Each Eruptive Phase and Top of the Eruption Column Assumed for the 18–19 May MSH1980 Simulationsa
Maximum Height (km)
Interval Time UTC Time LT Duration (hours) Above Vent Level Above Sea Level Eruptive Phase
1 1540–1550 0840–0850 0.16 29.5 32.0 lateral blast and associated explosions
2 1550–2320 0850–1620 7.3 13.5 16.0 sustained Plinian
3 2320–0030 1620–1730 1.16 17.0 19.5 late Plinian pulse
4 0030–0200 1730–1900 1.5 4.5 7.0 late ash flow, eruption waning
5 0200–0500 1900–2200 3.0 2.5 5.0 posteruption, weak ash‐rich plume
a
Vent elevation is 2500 m.

6 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Figure 3. (left) Values of D2 (uniform weighting factors) for the MSH1980 simulations as a function of
ye and for the different values of Df. (right) A zoom around the minimum. Best compromise between D2
and NMB minimizations was obtained for Df = 2.99 and ye = 0.29.

ran FALL3D and considered 12 particle classes ranging from spond to averaged values given by Harris et al. [1981] and
F = −2 (4 mm) to F = 6 (15 mm) and assumed 5 wt % of Holasek and Self [1995] for five distinct eruptive phases: the
magmatic water. The 12‐class TGSD, reported in Table 1, initial lateral blast and associated explosions, the sustained
was obtained by simplifying the original granulometry given Plinian column, the late Plinian pulse, the waning phase, and
by Durant et al. [2009]. The number of particle classes were the final posteruption ash‐rich weak plumes. Figure 2
reduced for computational reasons and the corresponding compares the column heights used in the simulations (the
class densities were computed as a weighted average of corresponding MER are calculated using the BPT) with
densities of pumices, lithics and crystals. In line with the radar measurements. For comparison, upper values inferred
experimental results reported by Durant et al. [2009], from GOES satellite images are also shown. The column
sphericity of ash particles was fixed at a mean value of model was initialized using results from BPT, assuming an
0.92. The diameter of the aggregate class was set to F = 2 exit mixture temperature of To = 1173 K and exit velocities
(250 mm) based on the observation that most of the particles of 250, 150, 150, 100, and 100 m s−1 for the five distinct
smaller than 64 mm fell out as aggregates of 250–500 mm phases, respectively. The corresponding MER (dependent
[Sorem, 1982; Carey and Sigurdsson, 1982; Durant et al., on the wind strength) in the five phases are about 2 × 109,
2009]. 9 × 106, 2 × 107, 3 × 105 and 5 × 104 kg s−1, respectively.
[25] The simulations considered a time‐dependent erup- Under these conditions, BPT predicts a total erupted mass
tion column height as reported in Table 2. Heights corre- of 5.4 × 1011 kg, in good agreement with the 3–8 × 1011 kg

Figure 4. (left) MSH1980 simulated column mass (vertical integration of concentration) in t km−2 at
18 May 2340 UTC (1640 LT). Run using the optimal values Df = 2.99 and ye = 0.29. (right) Meteorological
radar reflectivity at 19 May 0040 UTC (1740 LT) (modified from Harris et al. [1981]). Dark grey shading
corresponds to level 2 reflectivity of Harris et al. [1981]; the light grey shading corresponds to a weaker
reflectivity. Although these quantities are not directly comparable, qualitative information is provided on
the cloud extent and density.

7 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Figure 5. Comparison between MSH1980 simulations without aggregation, aggregation considering the
Cornell et al. [1983] model, and the proposed aggregation model with Df = 2.99 and ye = 0.29. (left)
Final deposit (contours in kg m−2). (right) Scatterplot comparing observed versus computed deposit at
119 localities. Solid line indicates the perfect fit. Upper and lower dashed lines show values of 5 (over-
estimation) and 1/5 (underestimation) times the ideal case (1/1; solid line).

8 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Table 3. Value of D2 (Considering Uniform wi) and NMB for the within the range Df = 2.96–2.99 and ye = 0.28–0.30, and the
MSH1980 Runs Considering No Aggregation, the Cornell et al. absolute minimum occurs for Df = 2.99 and ye = 0.29, with
[1983] Model, and the Proposed Aggregation Model With the D2 = 0.286 and NMB = −15.59%. As is common for ash
Optimal Values of Df = 2.99 and ye = 0.29a fallout simulation problems [e.g., Bonadonna et al., 2005;
D2 NMB R2
Pfeiffer et al., 2005; Costa et al., 2006, 2009], the model
Run (Uniform wi) (%) (Correlation) T Test is only able to capture the correct order of magnitude (for
instance, in this case 88% of the points are comprised between
No aggregation 0.798 −82.04 0.624 1.5 · 10−5 1/5 and 5 times the observed values). Besides the comparison
Cornell model 0.289 −24.36 0.781 0.205
Aggregation model 0.286 −15.59 0.823 0.599 with the observed deposits, in Figure 4 the reliability of the
a
model for the cloud transport in the atmosphere was verified
Correlation and T test values are referred to log(observed deposits) through comparison of the model results for MSH1980 cloud
versus log(calculated deposits).
at 18 May 2340 UTC (1640 LT) with the radar reflectivity
measurements for the same time. Radar reflectivity is pro-
reported by Rose and Durant [2009] or the 4.9–5.5 × 1011 kg portional to particle size and number concentration in the
estimated by Sarna‐Wojcicki et al. [1981]. The column cloud. In the simulation, regions of high integrated particle
model also furnishes the distribution of mass along the plume mass correlate with regions of observed high radar reflectivity.
(source term So). Of special interest is the region of enhanced distal radar
[26] Horizontal diffusion in FALL3D was calculated reflectivity centered over Ritzville which corresponds to high
using the Regional Atmospheric Modeling System param- simulated particle loading, implying that particle size and
eterization [Pielke et al., 1992; Folch et al., 2009] with Cs = mass is increased in the cloud through aggregation. Addi-
0.2 and Prandtl number Pr = 1. The particle terminal tional valuable validation could be obtained comparing the
velocities were computed using the Ganser model [Ganser, simulated grain size distributions with those actually depos-
1993; Costa et al., 2006]. ited. However, in the framework of our model, this is not
[27] The emission source for this eruption was dynamic, computationally possible as it would be necessary to store, at
for example, as described by Sparks et al. [1986], the initial each computational cell, information on all the particle classes
formation of the umbrella cloud occurred at approximately that contribute to form aggregates.
0840 LT, 8–10 km north of the summit. Then the plume
center moved northeast at a velocity of about 15 m s−1, 4.3. Comparison With Other Models
moving the center to 27 km north of the summit at 0900 LT. [30] In this section we compare the results obtained from
In order to match the simulations obtained using the PUFF the FALL3D model simulations using the aggregation
model [Searcy et al., 1998] with satellite imagery, Fero et al. model described above, with results given using the original
[2008] shifted the vent position 27 km northward for the granulometry reported in Table 1, i.e., without considering
initial phase. In order to reproduce the fallout deposits, aggregation, and results furnished by the simple aggregation
Armienti et al. [1988] shifted the emission source approxi- model suggested by Cornell et al. [1983]. As we mentioned
mately 30 km north of the vent position for the duration of the above, Cornell et al. [1983] present an empirically based
simulation. However, the simulated deposits by Armienti aggregation model that assumes aggregates consist of 50% of
et al. [1988] are systematically 10–20 km southern than the 63–44 mm ash, 75% of the 44–31 mm ash and 100% of the
the observed deposits, especially in the distal region, which finer than 31 mm ash form aggregates with a diameter of 200–
may reflect systematic error in the determination of the wind 300 mm. Deposit granulometry for the simulation with
field. Since in this work we are not interested in reproducing the Cornell et al. [1983] was obtained applying the rules reported
proximal part of the fallout deposit (within 60 km), and in order above to the MSH1980 12‐class TGSD reported in Table 1.
to eliminate a systematic bias due to the virtual position of the Aggregation is assumed to occur instantaneously in the
vent and probable error in the wind directions, the location of eruptive column which modifies the TGSD. In order to esti-
emission source was centered 50 km north of the summit mate the density of the aggregates in this case, the density of
(46.64°N, 122.20°W) for the entire duration of the eruption. the aggregate class was varied from ra = 200 kg m−3 to ra =
600 kg m−3. The optimal value for the aggregate density was
4.2. Optimization of the Aggregation Model ra = 400 kg m−3.
Parameters [31] Figure 5 compares results from simulations assuming
[28] Here we use the well‐studied MSH1980 deposit to no aggregation, aggregation following the model presented
assess the optimal values for Df and ye through comparison by Cornell et al. [1983], and aggregation following the model
of the simulated surface loading with measured values at presented by Costa et al. [2010] with the optimal values of
119 localities (in the analysis we excluded the location Df = 2.99 and ye = 0.29. In Table 3 the values of D2 and
(48.390°N, 115.560°W) which presented a value we con- NMB are reported for the three cases. Under the assumption
sidered as an outlier). These correspond to the points given of no aggregation FALL3D is not able to reproduce the
by Durant et al. [2009] located at distances larger than secondary maximum and underestimates the deposit almost
60 km from the volcano (twice the maximum column height). systematically (most points in the log‐log plot lay beyond
On the basis of the parametric study presented in the com- the 1/5 line). In contrast, both aggregation models reproduced
panion paper, Costa et al. [2010] ran a set of simulations the secondary maximum and gave similar D2 values (0.286
varying Df from 2.8 to 3.0 and ye from 0.10 to 0.50 at step of for Costa et al. [2010] and 0.288 for Cornell et al. [1983]).
0.01 for both parameters. However, the aggregation model presented here provides
[29] Figure 3 plots the MSH1980 D2 function for the an improved NMB over the Cornell et al. [1983] model
range of Df and ye considered. The optimal values are (−15.59% versus −24.36%). Comparing Log(observed

9 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Figure 6. Skew‐T Log‐P diagrams (top) for atmospheric sounding and (bottom) WRF interpolations at
Anchorage on 17 September 1992 at 1200 UTC. Blue lines are temperature. Wind speed and direction are
also shown.

10 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Table 4. TGSD Used in the CP1992 Simulationsa rological model from 16 September 1992 at 0000 UTC to
F d (mm) r (kg m )−3
Weight Percent
19 September 1992 at 0000 UTC using a nesting procedure.
The inner model nest had an horizontal spatial resolution of
−2 4000 900 4.5 4 km and 38 vertical layers (top pressure 50 hPa, ∼20 km).
−1 2000 1800 4.5 WRF was initialized with the NCEP‐DOE Reanalysis‐2 data
0 1000 1800 5.4
1 500 1800 2.4 at 2.5° resolution, leaving 24h for model heat up. Comparison
2 250 2650 2.0 with soundings shows again good agreement for temperature
3 125 2650 15.8 and wind field, specially above the PBL. For illustrative
4 64 2650 23.5 purposes, Figure 6 shows Skew‐T Log‐P diagrams for
5 31 2650 16.8
6 15 2650 10.1 atmospheric sounding and WRF results at Anchorage during
7 8 2650 12.0 17 September at 1200 UTC. The FALL3D computational
water – 1000 3.0 domain for the CP1992 simulations spans from 59°N to 63° N
a
Granulometric distribution assumes 10 classes ranging from F = −2 to
in latitude and from 153°W to 144°W in longitude, with
F = 7. A 3% of magmatic water is used. Grain size data reported by horizontal and vertical spatial resolutions of 4 and 0.5km,
Durant and Rose [2009] were fitted assuming a bi‐Gaussian distribu- respectively. Meteorological variables were interpolated
tion. Since particle densities and spericities are not available, they are hourly from the WRF mesh to the FALL3D grid.
chosen like the average values measured for the TGSD of MSH1980. [34] We ran the CP1992 simulations with 10 aggregating
Sphericity is assumed equal to 1.
classes ranging from F = −2 (4 mm) to F = 7 (8 mm) and
assumed a contribution of 3 wt % magmatic water (see
deposits) versus Log(computed deposits) for the three cases Table 4). The 10‐class TGSD granulometry was derived
(i.e., (1) the aggregation model considered here, (2) Cornell et from the distribution reported by Durant and Rose [2009].
al. [1983] model, and (3) no aggregation), we have a corre- However, for this eruption, component analysis for densities
lation coefficients equal to (1) 0.823, (2) 0.781, and (3) 0.624, and particle shapes were not available, and granulometry
respectively. Results from the Student T test comparison of appears very coarse depleted due to the small number of sam-
observed and simulated deposits also confirmed the best per- pling points available. Class particle densities were assumed
formance of the aggregation model considered here. Although similar to the average values of the TGSD for MSH1980.
the aggregation assumption introduced by Cornell et al. [1983] [35] The eruption duration and the variation column height
works satisfactorily around the secondary maximum (see with time are given in Table 5. The corresponding MER,
points close to 10 kg m−2 in the log‐log plot of Figure 5), calculated using the BPT model, for the three considered
it tends to overestimate the deposit at distances larger than phases are about 3.2 × 106, 4.8 × 106 and 7.7 × 106 kg s−1,
400 km (see the points close to 0.1 kg m−2 in the log‐log plot of respectively. The predicted total erupted mass is 7.9 ×
Figure 5). In contrast, the proposed aggregation model is able 1010 kg. For comparison, McGimsey et al. [2002] estimated
to reproduce the deposit at any distance with a high degree an erupted volume (DRE) of 1.5 × 107 m3, which for a rock
of agreement, even beyond the secondary maximum because density of r = 2600 kg m−3 gives 3.9 × 1010 kg. Horizontal
90–99% of aggregating particles belong to the 4F and 5F and vertical diffusion, and terminal velocity model were set
particle size classes which is generally larger than the values as for the MSH1980 case.
assumed by Cornell et al. [1983] (50% and 75%, respectively).
As a consequence a larger amount of mass can fall at shorter 5.2. Optimization of Aggregation Model Parameters
distance. [36] As for the previous test case, for CP1992 we per-
formed several series of simulations in order to find the
optimal values of Df and ye in the ranges 2.8–3.0 for Df and
5. Test Case II: September 1992 Crater Peak 0.1–0.5 for ye (see Figure 7). Simulations were compared
Eruption with surface loading measurements [Durant and Rose,
[32] Crater Peak volcano on Mount Spurr, Alaska (61.30°N, 2009]. There were 24 sampling locations distributed 5–
152.25°W), erupted on 27 June, 18 August, and 17 September 361 km from the vent. In our analysis we used 23 points, as
1992 and generated a series of short‐lived but violent sub‐ we excluded the location (61.197°N, 149.845°W) which
Plinian plumes that penetrated the stratosphere and produced presented a value we considered as an outlier. Figure 7 plots
ashfall at distances >1000 km downwind [McGimsey et al., the D2 function for the range of Df and ye considered.
2002] and covered most of the populated regions of south Minimum values lay in the ranges Df = 2.98–3.0 and ye =
central Alaska [Neal et al., 1995]. Pyroclastic flows occurred 0.27–0.33. The D2 function presents an absolute minimum
in both the August and September 1992 eruptions, but were for Df = 3.0 and ye = 0.29, with D2 = 0.636 and NMB =
small in scale. The distal ash fall deposits for both the August −63.57%. Both D2 and NMB values indicate the fit is of
and September 1992 eruptions had distal mass deposition lower quality than the MSH1980 case, which is probably
maxima at approximately 200–300 km from the volcano. due to the lower sampling density. Sparse sampling can
Although the sampling density was lower than the case of the result in the reconstruction of an apparent TGSD which
MSH1980 eruption, we used the available data for the Crater lacks representation of certain undersampled particle size
Peak September event (CP1992) to perform a further test for classes [Bonadonna and Houghton, 2005].
the proposed aggregation model. [37] The simulated deposits and comparison between
observed versus computed deposit surface loadings are
5.1. Model Initialization shown in Figure 8. Figure 8 shows FALL3D results without
[33] As in the MSH1980 case, 4‐D meteorological condi- an aggregation model applied and using the aggregation
tions were obtained by running the WRF mesoscale meteo- model described by Cornell et al. [1983]. It is worth to

11 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Table 5. Eruptive Phases and Eruption Column Height Assumed for the 17 September 1992 Mount Spurr Simulationsa
Maximum Height (km)
Phase Time (UTC) Duration (hours) Above Vent Level Above Sea Level Comments
1 0800–0830 0.5 9.4 11.6 PIREPS and NWS radar
2 0830–0900 0.5 10.9 13.1 NWS radar
3 0900–1130 2.5 12.4 14.6 NWS radar
Data from Durant and Rose [2009] based on pilot reports (PIREPS) and radar measurements. Vent elevation is 2200 m; avl denotes “above vent level.”
a

mention that the simulation without considering aggregation the parallel version of FALL3D. The aggregation model has
shows an incipient distal mass deposition maximum. This is two input parameters that need to be estimated, the fractal
not related to aggregation but caused by bias introduced exponent Df and the aggregate settling velocity correction
from the input TGSD, which is coarse depleted and is very factor ye. Both parameters were calibrated through a best fit
peaked around 3–4F. Moreover, without aggregation part procedure. For MSH1980, the best values were obtained for
of the mass associated with finer classes does not deposit the ranges Df = 2.96–2.99 and ye = 0.28–0.30, with a mini-
within the computational domain. Table 6 gives the values mum at Df = 2.99 and ye = 0.29. For CP1992, best values lay
of D2 and NMB for the three cases considered. Values are in the range Df = 2.98–3.0 and ye = 0.27–0.33, with a mini-
similar for all three cases. However, when we compare Log mum at Df = 3.0 and ye = 0.29. Results of both best fits show
(observed deposits) versus Log(computed deposits) for the similar ranges for Df and ye indicating that aggregation pro-
three cases (i.e., (1) the aggregation model considered here, cesses were similar in both eruptions. However, some dis-
(2) Cornell et al. [1983] model, and (3) no aggregation), crepancies may be due to the lower deposit sampling density
models have a correlation equal to (1) 0.817, (2) 0.837, used to generate the TGSD for the CP1992 eruption, com-
and (3) 0.615, respectively. The Student’s T test supports pared to the MSH1980 eruption. Taking the values of particle
the aggregation model considered here. In contrast to the density used in the simulations with the proposed aggregation
MSH1980 case, the fraction of aggregating particles belonging model into consideration, ye = 0.29 corresponds to an
to the classes 3–7F is lower in the proposed model than in the effective minimum aggregate density of ra ≈ 750 kg m−3,
Cornell et al. [1983] model, with a total mass of aggregates whereas best fit with Cornell et al. [1983] model needs ra ≈
of 2.2 × 1010 kg and 2.7 × 1010 kg, respectively. 400 kg m−3.
[39] We also performed a comparative study between the
proposed aggregation model and a simplified model for
6. Summary and Discussion aggregation based on field observations [Cornell et al.,
[38] Dispersion and transport of particles in volcanic 1983]. The model we proposed has a stronger theoretical
clouds generated by the MSH1980 and CP1992 eruptions physical basis and is able to reproduce similarly or even
was simulated using a modeling system comprising the better the fallout deposits produced by ash aggregation
WRF mesoscale model to provide meteorological inputs, processes. Moreover, in the framework of the new model,
BPT to describe the source term and the plume temperature, the aggregation rate depends on the amount of both mag-
and the FALL3D model to simulate the atmospheric disper- matic and atmospheric water and on the TGSD. In contrast
sion and deposition of tephra. This study applied an aggre- to previous models, the mass fraction of aggregating parti-
gation model that accounts for presence of liquid and frozen cles is now an output and changes from eruption to eruption
water (described by Costa et al. [2010]) and implemented it in (see Table 7).

Figure 7. (left) Values of D2 (uniform weighting factors) for the CP1992 simulations as a function of ye
and for the different values of Df. (right) Focus on values where ye is a minimum. Best compromise
between D2 and NMB minimizations is obtained for Df = 3.0 and ye = 0.29.

12 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Figure 8. (left) Comparison between CP1992 simulations without aggregation, aggregation consider-
ing the Cornell et al. [1983] model, and aggregation using our model with Df = 3.0 and ye = 0.29. (left)
Final deposit (contours in kg m−2). (right) Scatterplot comparing observed versus computed deposit at
23 localities. Solid line indicates the perfect fit. Upper and lower dashed lines show values of 5 (over-
estimation) and 1/5 (underestimation) times the ideal case (1/1; solid line).

13 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Table 6. Value of D2 (Considering Uniform wi) and NMB for the amount of liquid water (large coagulation kernel K). This can
CP1992 Runs Considering No Aggregation, the Cornell et al. occur either in a height window within the eruption column
[1983] Aggregation Model, and the Proposed Aggregation where liquid water can exist or, to a lesser extent, in certain
Model With the Optimal Values of Df = 3.0 and ye = 0.29a atmospheric layers. These are the conditions that likely favor
Run D2 (Uniform wi) NMB (%) R2 (Correlation) T Test
the formation of accretionary lapilli. However, their forma-
tion depends also on the ratio between the residence time of
No aggregation 0.637 −66.04 0.615 0.125 aggregating particles within the column window where liquid
Cornell model 0.593 −58.94 0.837 0.731 water can exist and the time required for aggregates to form
Aggregation model 0.636 −63.57 0.817 0.758
[Costa et al., 2010]. In addition, liquid water may again
a
Correlation and T test values are referred to log(observed deposits) appear during descent, as the cloud passes through the 0°C
versus log(calculated deposits). isotherm and ice crystals melt. Aggregation by liquid water in
Plinian and sub‐Plinian eruptions, characterized by vigorous
[40] Volcanic ash particles can aggregate in presence of plumes with large vertical velocities, is likely to be less
water or by other mechanisms (e.g., electrostatic aggrega- effective due to the lower residence times of particles in the
tion). The model proposed here considers aggregation in liquid water window. In contrast, less energetic eruptions
presence of water only, as these forces dominate over others, with shorter columns and lower ascending velocities are more
such as electrostatic aggregation [e.g., Sparks et al., 1997]. likely to meet the conditions where liquid aggregation is
Furthermore, the model discriminates between aggregation dominant, especially for water‐rich plumes or if the plume
due to liquid water, and aggregation due to the presence height coincides with the atmospheric layers with higher
of ice, which is less effective. In the case of MSH1980, the mixing ratio. For the initial phase of the MSH1980 eruption
simulated erupted mass was 54 × 1010 kg; the simulated where plume height exceeded 32 km, BPT predicts a window
deposit comprised 1.8 × 1010 kg (37.13% of total mass for liquid water between 13 and 19 km above the vent and a
erupted) of tephra in the absence of aggregation and 44 × mean ascending velocity in this region of about 190 m s−1. It
1010 kg (85.48 % of total mass erupted) when accounting for implies a residence time of about 30 s, lower than 60–100 s
aggregation using the model described by Costa et al. estimated for the aggregation by Costa et al. [2010]. In this
[2010]). For the CP1992 eruption, the simulated erupted case wet aggregation must proceed extremely rapidly before
mass was 81 × 109 kg; the simulated deposit comprised 28 × aggregation in the ice region becomes a more relevant
109 kg (36.63% of total mass erupted) of tephra in the absence mechanism. For a less vigorous column, like the CP1992
of aggregation and 42 × 109 kg (54.78% of total mass erupted) case, the window for liquid water is between 3 and 5 km
of tephra when accounting for aggregation. In both cases above the vent, the mean ascending velocity in this interval
between 50% and 130% more tephra was deposited within the is 45 m s−1 and the residence time of about 45 s. This case
model domain when aggregation was accounted for. The represents likely a regime where both mechanisms can be
missing fine fraction in each case was 63% of total mass important. Finally, for lower columns wet aggregation will
erupted in the absence of aggregation and reduced to between dominate.
15% and 45% when aggregation was included in the simu-
lations. The simulated MSH1980 deposit was far more 7. Conclusions
complete compared to CP1992, which may result from a
higher degree of aggregation in response to the different [42] This study is the first step in developing a physically
TGSD and larger eruption column height, and higher water based model for ash aggregation in volcanic plumes that
content of the MSH1980 cloud derived from magmatic includes the effects of cloud hydrometeors. The specific
volatiles. conclusions suggested by the results of the proposed model
[41] Water can exist in three different phases depending are as follows:
on temperature and pressure (height). The basal gas thrust or [43] 1. The microphysical‐based particle aggregation
jet region and the lower part of the overlaying convective model proposed here indicates that aggregation depends on
region contain vapor water only. As the plume ascends and the amount of both magmatic and atmospheric water present
cools by expansion and entrainment of ambient air at lower in the volcanic cloud, and on the characteristics of the
temperature, a point is reached where the vapor partial pres- TGSD.
sure exceeds the saturation pressure and vapor condenses [44] 2. For both eruptions studied here, the best fit values
into a stable water phase. Finally, further ascent and cooling for the fractal exponent Df and the aggregate settling velocity
allows ice to form, nucleated by volcanic ash particles. Once
at the neutral buoyancy level, ice crystals are transported with Table 7. Mass Fraction of Each Aggregating Class That Forms a
the ash cloud and, eventually, if they reach a lower atmo- 2F Aggregate According to the Cornell et al. [1983] Aggregation
spheric level where temperature raises above 0°C, will melt Model and the Proposed Model for the MSH1980 and CP1992
to form liquid water. Wet aggregation is limited to regions of Casesa
the ash cloud that are above the freezing temperature of water, F d (mm) Cornell (%) MSH1980 (%) CP1992 (%)
which is in the column or during cloud descent. Furthermore,
3 125 0 96 23
as the cloud becomes more dilute, the growth or evaporation/ 4 62 50 98 22
sublimation of hydrometeors will depend on the degree of 5 31 75 99 32
mixing and the amount of water vapor in the entrained air. 6 15 100 72 54
Aggregation will be more relevant in the proximal region, i.e., 7 8 100 – 45
in or near the column where particle concentration is high a
Values of percentages should be considered as indicative only because
(large initial number density n0) and in presence of a large strongly depend on the input TGSD.

14 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

correction factor ye range between Df = 2.96–3.00 and ye = jets, J. Volcanol. Geotherm. Res., 178, 94–103, doi:10.1016/j.jvolgeores.
0.27–0.33, respectively. This indicates that the settling 2008.01.002.
aggregates had a bulk density of about 700–800 kg m−3. Carey, S., and H. Sigurdsson (1982), Influence of particle aggregation on
deposition of distal tephra from the May 18, 1980, eruption of Mount
[45] 3. Particle aggregation in distal regions of volcanic St‐Helens volcano, J. Geophys. Res., 87(B8), 7061–7072, doi:10.1029/
clouds is enhanced as the cloud subsides and passes through JB087iB08p07061.
the melting level, which provides liquid water for wet Casadevall, T. (1994), The 1989–1990 eruption of Redoubt Volcano, Alaska:
impacts on aircraft operations, J. Volcanol. Geotherm. Res., 62, 301–316,
aggregation. doi:10.1016/0377-0273(94)90038-8.
[46] The introduction of the aggregation model in FALL3D Casadevall, T., P. Delos Reyes, and D. Schneider (1996), The 1991
increases the computational time by a factor of 2–4 depending Pinatubo eruptions and their effects on aircraft operations, in Fire and
Mud: Eruptions and Lahars of Mount Pinatubo, Philippines, edited by
on the cases. This is because the parallel version of the C. Newhall and R. Punongbaya, pp. 625–636, Univ. of Wash. Press,
code was not originally designed to investigate the interaction Seattle.
among classes, so that different classes were assigned to Cornell, W., S. Carey, and H. Sigurdsson (1983), Computer simulation and
transport of the Campanian Y‐5 ash, J. Volcanol. Geotherm. Res., 17,
different groups of processors which worked independently. 89–109, doi:10.1016/0377-0273(83)90063-X.
The result was a linear code scalability concerning the number Costa, A., G. Macedonio, and A. Folch (2006), A three‐dimensional Eulerian
of classes. This is certainly not the best strategy when aggre- model for transport and deposition of volcanic ashes, Earth Planet. Sci.
gation is introduced because processors assigned to different Lett., 241, 634–647, doi:10.1016/j.epsl.2005.11.019.
Costa, A., F. Dell’Erba, M. Di Vito, R. Isaia, G. Macedonio, G. Orsi, and
classes must now exchange data at each time step. It implies a T. Pfeiffer (2009), Tephra fallout hazard assessment at the Campi Flegrei
large amount of time‐consuming broadcast operations which caldera (Italy), Bull. Volcanol., 71, 259–273, doi:10.1007/s00445-008-
decrease the scalability of the code and, eventually, can even 0220-3.
Costa, A., A. Folch, and G. Macedonio (2010), A model for wet aggrega-
block it. Future lines of research will contemplate code opti- tion of ash particles in volcanic plumes and clouds: 1. Theoretical formu-
mization by changing the parallelization strategy and the lation, J. Geophys. Res., 115, B09201, doi:10.1029/2009JB007175.
inclusion of other phenomena leading to aggregation in Durant, A., and W. Rose (2009), Sedimentological constraints on
hydrometeor‐enhanced particle deposition: 1992 eruptions of Crater
absence of water. Moreover, further experimental studies Peak, Alaska, J. Volcanol. Geotherm. Res., 186, 40–59, doi:10.1016/
would be necessary to better constrain sticking efficiency of j.jvolgeores.2009.02.004.
volcanic ashes. Durant, A., R. Shaw, W. Rose, Y. Mi, and G. Ernst (2008), Ice nucleation
and overseeding of ice in volcanic clouds, J. Geophys. Res., 113,
D09206, doi:10.1029/2007JD009064.
Durant, A., W. Rose, A. Sarna‐Wojcicki, S. Carey, and A. Volentik (2009),
[47] Acknowledgments. A.F. is grateful to the Ramón y Cajal scien- Hydrometeor‐enhanced tephra sedimentation: Constraints from the
tific program. Numerical simulations have been done using the Barcelona 18 May 1980 eruption of Mount St. Helens, J. Geophys. Res., 114,
Supercomputing Center‐Centro Nacional de Supercomputación (BSC‐ B03204, doi:10.1029/2008JB005756.
CNS) MareNostrum Supercomputer. A.C. and G.M. acknowledge funding Fero, J., S. Carey, and J. Merrill (2008), Simulation of the 1980 eruption of
from the Italian Department of Civil Protection (Research Project SPeeD). Mount St. Helens using the ash‐tracking model PUFF, J. Volcanol.
A.D. acknowledges funding from the Leverhulme Foundation. We are very Geotherm. Res., 175, 355–366, doi:10.1016/j.jvolgeores.2008.03.029.
grateful to C. Bonadonna and two anonymous reviewers for their useful Field, P., A. Heymsfield, and A. Bansemer (2006), A test of ice self‐
comments that improved the clarity of the paper. collection kernel using aircraft data, J. Atmos. Sci., 63, 651–666,
doi:10.1175/JAS3653.1.
Folch, A., C. Cavazzoni, A. Costa, and G. Macedonio (2008a), An automatic
References procedure to forecast tephra fallout, J. Volcanol. Geotherm. Res., 177,
Aitken, A. (1935), On the least squares and linear combination of observa- 767–777, doi:10.1016/j.jvolgeores.2008.01.046.
tions, Proc. R. Soc. Edinburgh, 55, 42–48. Folch, A., O. Jorba, and J. Viramonte (2008b), Volcanic ash forecast—
Armienti, P., G. Macedonio, and M. Pareschi (1988), A numerical model Application to the May 2008 Chaitén eruption, Nat. Hazards Earth Syst.
for the simulation of tephra transport and deposition: Applications to Sci., 8, 927–940, doi:10.5194/nhess-8-927-2008.
May 18, 1980 Mount St. Helens eruption, J. Geophys. Res., 93, 6463– Folch, A., A. Costa, and G. Macedonio (2009), FALL3D: A computational
6476, doi:10.1029/JB093iB06p06463. model for transport and deposition of volcanic ash, Comput. Geosci., 35,
Baxter, P. (1999), Impacts of eruptions on human health, in Encyclopedia 1334–1342, doi:10.1016/j.cageo.2008.08.008.
of Volcanoes, edited by H. Sigurdsson, et al., pp. 1035–1043, Academic, Ganser, G. (1993), A rational approach to drag prediction of spherical and
San Diego, Calif. nonspherical particles, Powder Technol., 77, 143–152, doi:10.1016/
Blong, R. (1984), Volcanic Hazards: A Sourcebook on the Effects of Erup- 0032-5910(93)80051-B.
tions, Academic, North Ryde, N. S. W., Australia. Harris, D., W. Rose, R. Roe, M. Thompson, P. Lipman, and D. Mullineaux
Bonadonna, C., and B. Houghton (2005), Total grain‐size distribution and (1981), Radar observations of ash eruptions, in The 1980 Eruptions of
volume of tephra‐fall deposits, Bull. Volcanol., 67, 441–456, Mount St. Helens, Washington, edited by P. Lipman and D. Mullineaux,
doi:10.1007/s00445-004-0386-2. U.S. Geol. Surv. Prof. Pap., 1250, 323–333.
Bonadonna, C., G. Macedonio, and R. Sparks (2002), Numerical modelling Holasek, R., and S. Self (1995), GOES weather satellite observations and
of tephra fallout associated with dome collapses and Vulcanian explo- measurements of the May 18, 1980, Mount St. Helens eruption, J. Geo-
sions: application to hazard assessment on Montserrat, in The Eruption phys. Res., 100, 8469–8487, doi:10.1029/94JB03137.
of Soufriére Hills Volcano, Montserrat, From 1995 to 1999, edited by Lane, S., and J. Gilbert (1992), Electric potential gradient changes during
T. Druitt and B. Kokelaar, pp. 517–537, Geol. Soc., London. explosive activity at Sakurajima volcano, Japan, Bull. Volcanol., 54,
Bonadonna, C., C. B. Connor, B. F. Houghton, L. Connor, M. Byrne, 590–594, doi:10.1007/BF00569942.
A. Laing, and T. K. Hincks (2005), Probabilistic modeling of tephra Macedonio, G., A. Costa, and A. Folch (2008), Ash fallout scenarios at
dispersal: Hazard assessment of a multiphase rhyolitic eruption at Vesuvius: Numerical simulations and implications for hazard assessment,
Tarawera, New Zealand, J. Geophys. Res., 110, B03203, doi:10.1029/ J. Volcanol. Geotherm. Res., 178, 366–377, doi:10.1016/j.jvolgeores.
2003JB002896. 2008.08.014.
Buck, A. (1981), New equations for computing vapor pressure and McGimsey, R., C. Neal, and C. Riley (2002), Areal distribution, thickness,
enhancement factor, J. Appl. Meteorol., 20, 1527–1532, doi:10.1175/ mass, volume, and grain size of tephra‐fall deposits from the eruptions of
1520-0450(1981)020<1527:NEFCVP>2.0.CO;2. Crater Peak vent, Mt. Spurr volcano, Alaska, U.S. Geol. Surv. Open File
Bursik, M. (2001), Effect of wind on the rise height of volcanic plumes, Rep., 01‐370.
Geophys. Res. Lett., 28, 3621–3624, doi:10.1029/2001GL013393. Michalakes, J., J. Dudhia, D. Gill, T. Henderson, J. Klemp, W. Skamarock,
Carazzo, G., E. Kaminski, and S. Tait (2006), The route to self‐similarity in and W. Wang (2004), The Weather Research and Forecasting Model:
turbulent jets and plumes, J. Fluid Mech., 547, 137–148, doi:10.1017/ Software architecture and performance, paper presented at Eleventh
S002211200500683X. ECMWF Workshop on the Use of High Performance Computing in Mete-
Carazzo, G., E. Kaminski, and S. Tait (2008), On the dynamics of volcanic orology, Reading, U. K., 25–29 Oct.
columns: A comparison of field data with new model of negatively buoyant

15 of 16
B09202 FOLCH ET AL.: ASH AGGREGATION MODEL IN VOLCANIC PLUMES B09202

Murphy, D., and T. Koop (2005), Review of the vapour pressures of ice Skamarock, W., J. Klemp, J. Dudhia, D. Gill, D. Barker, W. Wang, and
and supercooled water for atmospheric applications, Q. J. R. Meteorol. J. Powers (2005), A description of the Advanced Research WRF Version 2,
Soc., 131, 1539–1565, doi:10.1256/qj.04.94. Tech. Rep. NCAR/TN‐468+STR, Natl. Cent. for Atmos. Res., Boulder,
Neal, C., R. McGimsey, C. Gardner, M. Harbin, and C. Nye (1995), Colo. (Available at http://www.wrf‐model.org)
Tephra‐Fall Deposits from the 1992 eruptions of Crater Peak, Mount Sorem, K. (1982), Volcanic ash clusters: Tephra rafts and scavengers,
Spurr Volcano, Alaska: A Preliminary Report on Distribution, Stratigraphy, J. Volcanol. Geotherm. Res., 13, 63–71, doi:10.1016/0377-0273(82)
and Composition, U.S. Geol. Surv. Bull., 2139, 65–79. 90019-1.
Pfeiffer, T., A. Costa, and G. Macedonio (2005), A model for the numerical Sparks, R., J. Moore, and C. Rice (1986), The initial giant umbrella cloud of
simulation of tephra fall deposits, J. Volcanol. Geotherm. Res., 140, 273– the May 18th, 1980, explosive eruption of Mount St. Helens, J. Volcanol.
294, doi:10.1016/j.jvolgeores.2004.09.001. Geotherm. Res., 28, 257–274, doi:10.1016/0377-0273(86)90026-0.
Pielke, R., W. Cotton, R. Walko, C. Tremback, M. Nicholls, M. Moran, Sparks, R., M. Bursik, S. Carey, J. Gilbert, L. Glaze, H. Sigurdsson, and
D. Wesley, T. Lee, and J. Copeland (1992), A comprehensive meteo- A. Woods (1997), Volcanic Plumes, John Wiley, Chichester, U. K.
rological modeling system‐RAMS, Meteorol. Atmos. Phys., 49, 69–91, Spence, R., I. Kelman, P. Baxter, G. Zuccaro, and S. Petrazzuoli (2005),
doi:10.1007/BF01025401. Residential building and occupant vulnerability to tephra fall, Nat. Hazards
Pomonis, A., R. Spence, and P. Baxter (1999), Risk assessment of residen- Earth Syst. Sci., 5, 477–494, doi:10.5194/nhess-5-477-2005.
tial buildings for an eruption of Furnas Volcano, Sao Miguel, the Azores, Textor, C., H. Graf, M. Herzog, J. Oberhuber, W. Rose, and G. Ernst
J. Volcanol. Geotherm. Res., 92, 107–131, doi:10.1016/S0377-0273(99) (2006a), Volcanic particle aggregation in explosive eruption columns.
00071-2. Part I: Parameterization of the microphysics of hydrometeors and ash,
Rose, W., and A. Durant (2009), Fine ash content of explosive eruptions, J. Volcanol. Geotherm. Res., 150, 359–377, doi:10.1016/j.jvolgeores.
J. Volcanol. Geotherm. Res., 186, 32–39, doi:10.1016/j.jvolgeores. 2005.09.007.
2009.01.010. Textor, C., H. Graf, M. Herzog, J. Oberhuber, W. Rose, and G. Ernst
Sarna‐Wojcicki, A., S. Shipley, D. Waitt, D. Dzurisin, S. Wood, P. Lipman, (2006b), Volcanic particle aggregation in explosive eruption columns.
and D. Mullineaux (1981), Areal distribution, thickness, mass, volume, Part II: Numerical experiments, J. Volcanol. Geotherm. Res., 150,
and grain size of air‐fall ash from the six major eruptions of 1980, in 378–394, doi:10.1016/j.jvolgeores.2005.09.008.
The 1980 Eruptions of Mount St. Helens, Washington, edited by P. Lipman
and D. Mullineaux, U.S. Geol. Surv. Prof. Pap., 1250, 577–600.
A. Costa and G. Macedonio, Istituto Nazionale di Geofisica e
Schumacher, R., and H. Schmincke (1995), Models for the origin of accre- Vulcanologia, Sezione di Napoli, Via Diocleziano 328, I‐80124 Naples,
tionary lapilli, Bull. Volcanol., 56, 626–639. Italy. (costa@ov.ingv.it; macedon@ov.ingv.it)
Scollo, S., M. Prestifilippo, G. Spata, M. D’Agostino, and M. Coltelli
A. Durant, Department of Geography, University of Cambridge,
(2009), Monitoring and forecasting Etna volcanic plumes, Nat. Hazards Downing Place, Cambridge, CB2 3EN, UK.
Earth Syst. Sci., 9, 1573–1585, doi:10.5194/nhess-9-1573-2009. A. Folch, Earth Sciences Department, Barcelona Supercomputing Center,
Searcy, C., K. Dean, and W. Stringer (1998), Puff: A high‐resolution vol-
Centro Nacional de Supercomputación, Jordi Girona 29, E‐08034
canic ash tracking model, J. Volcanol. Geotherm. Res., 80, 1–16, Barcelona, Spain. (afolch@bsc.es)
doi:10.1016/S0377-0273(97)00037-1.

16 of 16

You might also like