Você está na página 1de 20

This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy

Atmospheric Environment 48 (2012) 165e183

Contents lists available at ScienceDirect

Atmospheric Environment
journal homepage: www.elsevier.com/locate/atmosenv

Validation of the FALL3D ash dispersion model using observations of the 2010
Eyjafjallajökull volcanic ash clouds
A. Folch a, *, A. Costa b, c, S. Basart a
a
Barcelona Supercomputing Center e Centro Nacional de Supercomputación, Jordi Girona 29, 08034 Barcelona, Spain
b
Environmental Systems Science Centre, University of Reading, Reading RG6 6AL, UK
c
Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Vesuviano, Via Diocleziano 328, Napoli, Italy

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

Article history: During AprileMay 2010 volcanic ash clouds from the Icelandic Eyjafjallajökull volcano reached Europe
Received 1 February 2011 causing an unprecedented disruption of the EUR/NAT region airspace. Civil aviation authorities banned
Received in revised form all flight operations because of the threat posed by volcanic ash to modern turbine aircraft. New
22 June 2011
quantitative airborne ash mass concentration thresholds, still under discussion, were adopted for
Accepted 24 June 2011
discerning regions contaminated by ash. This has implications for ash dispersal models routinely used to
forecast the evolution of ash clouds. In this new context, quantitative model validation and assessment of
Keywords:
the accuracies of current state-of-the-art models is of paramount importance. The passage of volcanic
Volcanic ash dispersion
Numerical model
ash clouds over central Europe, a territory hosting a dense network of meteorological and air quality
Model validation observatories, generated a quantity of observations unusual for volcanic clouds. From the ground, the
2010 Eyjafjallajökull eruption cloud was observed by aerosol lidars, lidar ceilometers, sun photometers, other remote-sensing instru-
ments and in-situ collectors. From the air, sondes and multiple aircraft measurements also took
extremely valuable in-situ and remote-sensing measurements. These measurements constitute an
excellent database for model validation. Here we validate the FALL3D ash dispersal model by comparing
model results with ground and airplane-based measurements obtained during the initial 14e23 April
2010 Eyjafjallajökull explosive phase. We run the model at high spatial resolution using as input hourly-
averaged observed heights of the eruption column and the total grain size distribution reconstructed
from field observations. Model results are then compared against remote ground-based and in-situ
aircraft-based measurements, including lidar ceilometers from the German Meteorological Service,
aerosol lidars and sun photometers from EARLINET and AERONET networks, and flight missions of the
German DLR Falcon aircraft. We find good quantitative agreement, with an error similar to the spread in
the observations (however depending on the method used to estimate mass eruption rate) for both
airborne and ground mass concentration. Such verification results help us understand and constrain the
accuracy and reliability of ash transport models and it is of enormous relevance for designing future
operational mitigation strategies at Volcanic Ash Advisory Centers.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction episode, lasting from 15 to 21 April, was caused by the arrival and
residence at low tropospheric levels of volcanic ash and aerosols in
The AprileMay 2010 explosive eruption of Eyjafjallajökull the United Kingdom (U.K.) and North and Central Europe. Because
volcano in Iceland triggered an unprecedented disruption of the volcanic ash poses a threat to modern high-bypass turbine
European-North Atlantic (EUR/NAT) region airspace. The resulting aircrafts (Casadevall et al., 1996) and following the International
socioeconomic impact was enormous: 25 countries affected, over Civil Aviation Organization (ICAO) regulations (ICAO, 2000),
4 million passengers were stranded due to cancellation or delay of European civil aviation regulators banned all flight operations on
over 100,000 flights, and direct costs to airlines were measured in 15 April. On 20 April new guidelines based on some safe ash
billions of Euros (Oxford-Economics, 2010). The main disruptive concentration thresholds were adopted. The adoption of the new
guidelines, which substituted the previous “zero ash tolerance”
criterion, consented to resume operations in large areas previously
* Corresponding author. banned. Other sporadic disruptions of the European airspace
E-mail address: arnau.folch@bsc.es (A. Folch). occurred until mid-May but caused less impact. At the end of the

1352-2310/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.atmosenv.2011.06.072
Author's personal copy

166 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

crisis, quantitative criteria defined by the U.K. Civil Aviation across Europe. Moreover, sondes and aircraft measurements gave
Authority (CAA) and EUROCONTROL discerned zones of low (ash additional extremely valuable in-situ measurements.
concentration between 0.2 and 2 mg m3), medium (between 2 Here we compare some of these measurements made using
and 4 mg m3) and high (>4 mg m3) contamination. Whether different techniques with FALL3D (Costa et al., 2006; Folch et al.,
these quantitative criteria (or other similar) will eventually 2009) dispersal model simulations focusing on the 14e23 April
become regulatory and alternative to the classical “avoid any ash” 2010 main eruption phase. The manuscript is arranged as follows.
approach is a matter of ongoing debate. In any case, the Eyjafjal- Firstly, we briefly summarize the eruption chronology and the cloud
lajökull event dramatically demonstrated the limitations of the dispersal characteristics during 14e23 April 2010. Secondly, we
precautionary “zero ash tolerance” criterion when applied to long- overview the FALL3D model and discuss the model set up for this
lasting eruptions affecting large areas with dense air traffic and particular simulation, including the meteorological dataset used, ash
hosting main international hubs, such as the case of central granulometry and aggregation, source term evaluation, and model
Europe. grid resolution. Thirdly, we perform a sensitivity analysis to estimate
Several Volcanic Ash Transport and Dispersal Models (VATDM) the effect of the model input uncertainties on the results. Model
are routinely used by Volcanic Ash Advisory Centers (VAACs) to results are then compared against remote ground-based and in-situ
forecast ash cloud location and issue periodic Volcanic Ash Advi- aircraft-based measurements, including lidar ceilometer from the
sories. Other VATDM are used to describe ash ground deposits German Meteorological Service, aerosol lidars and sun photometers
especially in highly urbanized areas near volcanoes (e.g. Catania, from EARLINET and AERONET networks, and flight missions of the
Italy, Scollo et al., 2009). The London VAAC, responsible for the German DLR Falcon aircraft. In general, such an opportunity to vali-
Icelandic volcanoes, uses the U.K. Met Office’s Lagrangian Numer- date models versus a wide range of multi-technique observations has
ical Atmospheric-dispersion Modeling Environment (NAME) model no precedent. Finally, we conclude with a general discussion on the
(Ryall and Maryon, 1998; Jones et al., 2007). The adjacent Toulouse results and on the implications for operational VATDM. Comparison
and Montreal VAACs use MOCAGE (Peuch et al., 1999) and MLDP0 results help to constrain VATDM accuracy and reliability and it is of
(D’Amours et al., 2010) respectively. During the Eyjafjallajökull enormous relevance for informing the design of future operational
crisis, other models different from those used at VAACs were also mitigation strategies at VAACs.
run to support national decision makers in different countries,
including EURAD (Ackermann et al., 1998), FALL3D (Costa et al., 2. Eruption chronology and ash cloud trajectory
2006; Folch et al., 2009), FLEXPART (Stohl et al., 1998), HYSPLIT
(Draxler and Hess, 1998), MCCM (Grell et al., 2000), REMOTE After several months of seismic activity and crustal deformation
(Langmann et al., 2008), and CMAQ (Byun and Ching, 1999). The Eyjafjallajökull volcano (19.62 W, 63.63 N, 1666 m above sea level,
main difficulties to model transport of volcanic ash are given by the a.s.l.) erupted on 20 March 2010, producing fire fountains and basaltic
quantification of the source term (mass eruption rate, column lava flows at the East flank of the volcano (Sigmundsson et al., 2010).
height, total grain size distribution), the possible occurrence of ash On 14 April a new intrusion of trachy-andesitic magma lead to
aggregation, and the properties of ash particles themselves that can a submittal activity. Melting of the volcano summit caldera ice cap
vary notably in shape and size (from sub-microns to millimeters). caused phreatomagmatic explosions, generation of relatively fine-
In AprileMay 2010 the aviation industry and regulators reques- grained ash, and eruption columns rising up to 9 km a.s.l. The
ted the U.K. Met Office to supplement the London VAAC standard subsequent ash clouds were transported toward central Europe by the
ICAO binary forecast (ash/no ash) with a forecast of ash concentra- prevailing high and mid-troposphere north-westerly winds. On 18
tions according to the new criteria. The eventual adoption by ICAO of April phreatomagmatism waned and the eruption activity shifted to
quantitative safety thresholds of ash concentration would have pulsating water-rich ash-poor columns of much lower height. This
important implications for modeling strategies. If models are situation went on for about two weeks until explosive activity
required to quantify concentration rather than just predict ash cloud increased again on 5 May producing a sustained magmatic activity
location it is essential to constrain the source term (column height that lasted for another three weeks. Both the 14e18 April and the
and mass eruption rate) and to quantitatively validate VATDM. Some 5e20 May episodes caused significant air traffic disruption in Europe.
validations of VATDM are typically done by comparing with ground Fig. 1 shows the synoptic meteorological situation during 14e22
deposit and satellite measurements. Comparison with deposits is April. In opposition to what typically occurs during the winter, on
the traditional way volcanologists test VATDM but becomes chal- 14e15 April there was a high located south of Iceland and another
lenging or even impossible for the finest fraction of the cloud (made north of Scandinavia (Petersen, 2010). This situation favored north-
of micron-size particles) reaching very large distances. Multi- westerly tropospheric winds that transported ash over the Faroe
channel satellite imagery and ash detection algorithms (e.g. Prata, Islands and Norway during 15 April early morning and later on over
2008) are typically used for validating VATDM qualitatively South England and Denmark (Fig. 1aed). Airspace was closed in
through simultaneous agreement of modeled and observed top- several countries including Norway, Sweden, Finland, United
view cloud position and shape (problems with this method arise Kingdom, Belgium, Netherlands and Ireland. The cloud penetrated
when meteorological clouds obscure the view). A more quantitative German airspace during the early morning of the 16 April and then
validation with satellite retrievals is also possible (e.g. Corradini moved toward central Europe during 16 and 17 April when it
et al., 2011) but is less common. In contrast, VATDM comparisons became partly mixed with the atmospheric boundary layer.
with ground-based remote and aircraft in-situ measurements are German airspace was closed on afternoon 16 April. When the ash
very scarce in the literature (see e.g. Wang et al., 2008; Dacre et al., cloud reached the Alps range it was deflected E-W roughly and
2011). circulated clockwise around a high centered over the English
The passage and the residence of Eyjafjallajökull volcanic clouds channel to re-enter the continent a couple of days later (see
over a densely monitored territory generated an unprecedented streamlines on Fig. 1cee). The formation of a trough and a pro-
quantity of data. The cloud was observed by aerosol lidars (lidars longed calm wind situation was the cause of ash residing in the
originally deployed for cloud aerosol studies), lidar ceilometers lower troposphere over Europe (Fig. 1eef, 18e19 April). Because of
(single wavelength lidars originally deployed for cloud base height the mixing of fresh and recirculated ash fluxes, heterogeneous
monitoring), sun photometers and other remote-sensing instru- multilayer sub-clouds (detected also by different aircraft observa-
ments used at several meteorological and air quality observatories tions) were formed. The situation changed dramatically late on the
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 167

Fig. 1. Meteorological synoptic situation over Europe during 14 (a), 15 (b), 16 (c), 17 (d), 18 (e), 19 (f), 20 (g), 21 (h), 22 (i), and 23 (j) April 2010 at 00 UTC. Plots show streamlines and
contours of wind velocity (in m s1) at 5500 m above ground level (500 hPa circa). Meteorological data has been interpolated from ERA-Interim reanalysis at 1.5 horizontal
resolution.
Author's personal copy

168 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

Fig. 1. (continued).

20 April and through the 21 April onwards when strong north- shape of eruption column. The eruption column height and shape
westerly winds finally cleared the remaining suspended ash from and the mass distribution within the column (source term) can be
the airspace (Fig. 1hei). described using either a 1D Buoyant Plume Theory (BPT) model
based on that of Bursik (2001) or an empirical relationship based on
3. Model simulation of ash cloud during 14e23 April Suzuki (1983). The BPT option can provide the MER and the vertical
distribution of mass given a column height. The empirical rela-
3.1. The FALL3D model tionship option requires an independent assessment of MER, e.g. by
using the MER versus column height fits (Sparks et al., 1997; Mastin
FALL3D (Costa et al., 2006; Folch et al., 2009) is an Eulerian et al., 2009). Primary FALL3D model outputs are the time-
model for transport and deposition of volcanic ash particles. The dependent deposit load and airborne mass concentration for each
model solves a set of advection-diffusion-sedimentation equations particle class. However, other physical and optical properties of ash
(one equation for each particle class) on a structured terrain clouds (e.g. particle number concentration, optical depth and
following grid using a second-order Finite Differences explicit extinction coefficients at 0.5 mm) are also computed as explained in
scheme. Each particle class is characterized by particle diameter, Corradini et al. (2011).
density and shape. The model deals simultaneously with a wide FALL3D has already been validated against several tephra
spectrum of particle sizes (from lapilli to fine ash) and trace gas deposit data and space-based ash cloud observations from different
components (e.g. H2O). Particle aggregation can be accounted for eruptions (Costa et al., 2006; Macedonio et al., 2008; Folch et al.,
using the wet aggregation model described in Costa et al. (2010) but 2009; Scollo et al., 2008, 2009; Corradini et al., 2011). Here we
the model has no wet deposition mechanisms yet. The FALL3D validate the model against several types of ground-based and
model follows an off-line strategy, i.e., independent global or space-based observations. To this purpose we run a high-resolution
mesoscale meteorological models provide the meteorological fields simulation using, as the only constrained inputs, the meteorological
a priori. The main volcanological model inputs are mass eruption fields, the total grain size distribution (TGSD), and the observed
rate (MER), total grain size distribution (TGSD), and height and column heights.
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 169

3.2. Meteorological data Φ

As meteorological input data for FALL3D we use ECMWF ERA-


Interim reanalysis obtained from the ECMWF data server. ERA-
Interim reanalysis archive contains four times daily data at a 1.5
horizontal resolution and at 37 pressure levels, from 1000 to 1 hPa.
Meteorological fields are interpolated to the FALL3D computational
mesh on an hourly basis interval using a linear interpolation in time
and bilinear in space. For this simulation we fixed the FALL3D grid
resolution to 0.25  0.25 in the horizontal and to 100 m in the
vertical (the resulting mesh is of 241  141  100 nodal points). To
run the dispersal model on a grid finer than that of the meteoro-
logical model does not improve the advection term but reduces
numerical diffusion and improves model accuracy. Note that the
resolution is much finer than the resolution typically used in
operational mode, when model execution time is a priority.

3.3. Reconstructed TGSD, ash aggregation and effective TGSD

TGSD depends on eruption style and intensity, and may vary


from one eruption phase to another as actually occurred during the Fig. 2. TGSD reconstructed by Bonadonna et al. (2011a) for the 4e8 May period (black bars)
Eyjafjallajökull eruption. Data from samples collected by scientists and effective TGSD (i.e. particle percentages remaining after aggregation) used in the
of the Institute of Earth Sciences, University of Iceland, during 15, simulations (gray bars). The lower axis shows the particle class or bin diameter (in mm) and
the upper axis indicates the corresponding F value, defined as d (in mm) ¼ 2  pow (F).
17, and 18 April 2010 at distances varying between 20 and 55 km
The class of aggregates in the effective TGSD is at 250 mm (F ¼ 2). The distribution contains
from the volcano (http://www.earthice.hi.is/page/ies_EYJO2010_ 14 classes at equally-spaced F values. However, as we do not mean to model proximal
Grain) revealed a relatively fine-grained tephra as expected distances (teens to few hundreds of kms) in detail, all particles above F ¼ 2 (250 mm in
because of the phreatomagmatic activity occurring initially. diameter) were grouped into a single class. Thus, our simulations consider in practice 10
Unfortunately, data available from the first days are too scarce in particle classes (bins) varying in diameter from F ¼ 1 (d ¼ 500 mm) to F ¼ 10 (d ¼ 1 mm).
Density and sphericity were set to typical values of 2700 kg m3 and 0.9 respectively.
order to reconstruct the TGSD during the initial 14e18 April
phreatomagmatic period considered here.
Bonadonna et al. (2011a) reconstructed the TGSD from in-situ
than 5 microns (typical effective classes used for LIDAR retrieval) is
sampling carried out from 4 to 8 May 2010 at distances ranging
less than 3.5%.
from 2 to 56 km from the vent, applying a Voronoi tessellation
technique, both with and without the integration of information
derived from satellite (MSG-SEVIRI) for the fine ash fractions. The 3.4. Source term: column height and mass eruption rate
fraction of fine ash (diameter < 64 mm or F > 4) is about 35%. Here
we assume that the TGSD during the simulated period is similar to The source term in VATDM is defined by eruption column
the that reconstructed by Bonadonna et al. (2011a) using Voronoi height, MER, vertical distribution of mass and TGSD (discussed in
tessellation technique to field observations without information Section 3.3). Quantifying the source term is a bottleneck of
derived from satellite (Fig. 2). It is worth noting that the stronger VATDM although it is essential if models have to compute
phreatomagmatic activity during the 14e18 April period could have airborne ash concentration. Because column height is the easiest
produced a higher amount of fine ash, although it is still very parameter to constrain in real-time (e.g. by using radar, visual
reasonable to assume that TGSDs in the two periods were similar. observations or ground-based lidars), MER is typically inferred
The wet aggregation model implemented in FALL3D (Costa et al., from height either using empirical or semi-empirical relation-
2010; Folch et al., 2010) requires a reliable estimation of the total ships (e.g. Wilson and Walker, 1987; Sparks et al., 1997; Mastin
amount of water (magmatic plus meteoric) and needs a very fine et al., 2009) or running 1D radial-averaged BPT models (e.g.
computational grid (few kms) to compute the wet aggregation Woods, 1995; Bursik, 2001).
processes occurring in the plume. Because the FALL3D model can The C-band weather radar of the Icelandic Meteorological Office
adopt a uniform horizontal grid resolution only, the use of the wet (IMO) observed the Eyjafjallajökull volcanic plume during the
aggregation model for such large-scale domains (60  35 in our whole eruption period (Petersen, 2010). The radar, located 150 km
case) is prohibitive from a computational point of view. To from the volcano, provided 5-min plume altitudes with a minimum
circumvent this limitation and, simultaneously, to estimate the detection height of 2.5 km a.s.l. (z0.8 km a.g.l.) because of
effect of fine ash wet aggregation, in our simulations we considered mountain ranges and curvature of the Earth. Here we use hourly-
an effective TGSD that accounts for wet aggregation. Such an averaged values of the heights provided by IMO (http://andvari.
effective TGSD has been obtained running FALL3D with the wet vedur.is/warason/radar/), interpolating linearly to cover observa-
aggregation model on a small-scale domain (2  2 ) considering tion holes, and assumed a minimum altitude of 2.5 km (a.s.l.) when
the reconstructed TGSD (Bonadonna et al., 2011a) and assuming the plume was not detected by radar (Fig. 3a).
fully saturated water conditions. The use of an effective fine ash Given hourly values of column height, we used both a BPT model
depleted TGSD to model long-range transport can be justified (Bursik, 2001; Carazzo et al., 2006, 2008) and the Suzuki (1983)
because most of ash aggregates from Eyjafjallajökull eruption had mass distribution with A ¼ 4 and l ¼ 1 (Pfeiffer et al., 2005)
a quite large settling velocity (typically >1 m s1, see Taddeucci et combined with the empirical relationship given in Mastin et al.
al., (2011); Bonadonna et al., (2011a)) therefore falling mainly in the (2009) to estimate the MER. Empirical relationships hold better
proximal region. The resulting effective TGSD is shown in Fig. 2 for strong sustained vertical columns and can capture the correct
(gray bars). We estimate that more than 20% of the fine ash order of magnitude. However, they are not suitable to describe
(d  64 mm) aggregated and the fraction of particle classes finer weak bent-over plumes where plume dynamics is strongly affected
Author's personal copy

170 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

a b

Fig. 3. Source term characterization. a) Hourly-averaged plume height (km a.s.l.) as measured by the IMO weather radar. Detection limit of the weather radar is 2.5 km (a.s.l.).
b) MER versus time predicted by the BPT model (solid line) and by the empirical fit given by Mastin et al. (2009) (dashed line). The BPT plume model is similar to Bursik (2001) with
radial entrainment coefficient a calculated in accord to Carazzo et al. (2006) and the cross-wind entrainment coefficient b estimated on the basis of the local Richardson number
(0.06  b  0.2). Total erupted masses for the 14e23 April 2010 period are of 1.9  1011 kg and 2  1010 kg respectively.

by the intensity of the surrounding variable wind. BPT approach 4.1. Analysis of uncertainty on the simulation results
also has limitations mainly due to the steady-state assumption and
the poorly constrained cross-wind entrainment effects, but, to date, Before proceeding to compare simulations with observations, it is
is the only alternative to more sophisticated 3D plume models (e.g. necessary to estimate how the typical uncertainties in model inputs
ATHAM, see Herzog and Graf, 2010) that are not operationally can affect the predictions of the model. Rather than an exhaustive
viable. Ambient conditions, necessary to solve the BPT equations, sensitivity analysis, beyond the scope of this paper, our purpose is
were supplied interpolating the ERA-Interim wind field and to perform different simulations varying relevant inputs (column
temperature profiles to the vent coordinates. Conditions at the vent height, MER and vertical mass distribution inside the eruption
were set to 65 m s1 for the exit velocity of the mixture (M. Ripepe, column) within the typical uncertainty. This allows us to assess the
personal communication 2010), 2% (in weight) for the magmatic spread of forecasts with respect to a reference forecast, i.e. to asso-
water content, and 1030  C for the exit temperature of the mixture ciate an uncertainty to the reference simulation used later to
(a value typical of this magmatic composition). compare with the observations. We consider:
BPT model and empirical MERs are shown in Fig. 3b. According
to the BPT model, the total erupted mass during the 10-day simu- i) Column height uncertainty. The IMO C-band weather radar has
lated period is 1.9  1011 kg, almost one order of magnitude larger a minimum detection height of 2.5 km a.s.l. (due to the
than that of the empirical relationship proposed by Mastin et al. mountain ranges and curvature of the Earth) and an error
(2009) (2.0  1010 kg). Such a difference is not surprising for associated with the radar measurements (due to scanning
sustaining weak columns because cross-wind effects cause large angles) that, at certain altitude, can be up to 1 km. Although
variabilities in the MER versus height relationship. The BPT model reasonably constrained, this error can affect the estimation of
needs larger MER because of the stronger entrainment due to cross- MER and the dispersion of ash in the case of a strong wind
wind effects. As we will show later, when we compare with shear. Here we accounted for a mean column height
observations BPT model seems to overestimate the total mass up to measurement error of DH ¼ 500 m and run simulations
a factor 3e4, whereas the Mastin et al. (2009) relationship strongly considering H  DH and H þ DH, where H is the column
underestimate the total mass (up to an order of magnitude). height of the reference run.
However, all observations are bound within these two cases. ii) MER and column mass distribution uncertainty. As discussed in
Section 3.4, the BPT model and the fit proposed by Mastin
4. Model results and validation et al. (2009) give upper and lower bounds for the MER. In
addition, we consider also the effect of different vertical mass
Several reasons preclude a direct quantitative validation of distributions, namely given by distribution predicted by BPT
VATDM. First of all, VATDM give mass concentration, a variable hard (Bursik, 2001), Suzuki distribution (Suzuki, 1983; Pfeiffer
to measure directly. Comparison involves model derived variables et al., 2005), and point source concentrated at the top of the
(e.g. Aerosol Optical Depth (AOD) estimated from model concen- column.
tration) and retrievals from observations. These, in turn, are also
based on assumptions that have an associated uncertainty. On the The reference case assumes column heights as given by IMO
other hand, it is often difficult to discriminate in the observed weather radar, MER and column mass distribution estimated through
signals the effect of other atmospheric components from that of the BPT model, and the effective TGSD presented in Section 3.3. Fig. 4
ash. Notwithstanding these limitations, we perform here semi- gives a snapshot of the FALL3D ash cloud simulation for the reference
quantitative and quantitative comparisons with data obtained case showing the ash mass load from 14 to 21 April (an animated gif
using several techniques. file showing the ash mass load from 14 to 22 April is available as online
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 171

Fig. 4. Composite image of FALL3D results for the reference run at different time slices. Contours are ash column load in ton km2 or g m2. Assuming a homogeneous ash cloud
1 km thick these contours would correspond to mass concentrations expressed in mg m3. An animated gif file showing the ash mass load from 14 to 22 April is available as online
material.
Author's personal copy

172 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

Table 1
Sensitivity analysis at 4 different points of the domain. (1) plume height as given by the IMO radar. (2) MER estimated by means of the Buoyant Plume Theory (BPT). (3) MER
estimated as in Mastin et al. (2009). Meaning of used symbols is the following: H column height of the reference run; DH ¼ 500 m; C ¼ Concentration; Cref ¼ Concentration of
the reference run; CL ¼ Column load; CLref ¼ Column load of the reference run.

Column height H(1) H þ DH H  DH H H


(2)
MER estimation model BPT BPT BPT BPT Mastin(3)

Column vertical mass distribution BPT BPT BPT Point Point


16APR 15UTC Leipzig C/Cref at 3 km 1.0 0.86 1.71 0.84 0.05
CL/CLref 1.0 0.58 1.43 0.90 0.05
17APR 06 UTC Paris C/Cref at 2 km 1.0 0.34 2.92 2.18 0.12
CL/CLref 1.0 0.80 2.17 1.41 0.07
18APR 15 UTC Chilbolton C/Cref at 3 km 1.0 0.62 1.14 0.63 0.04
CL/CLref 1.0 0.54 1.50 0.95 0.06
19APR 12UTC Lille C/Cref at 2 km 1.0 1.22 0.53 0.49 0.03
CL/CLref 1.0 2.99 0.50 2.07 0.12

material). A first qualitative inspection reveals that the simulation height and vertical mass distribution affect the column loadings and
reproduces the main features of the ash cloud evolution described in concentration values by a factor of 2e3. In contrast, as expected,
Section 2. A more detailed comparison between FALL3D ash mass variations in concentration due to different estimations of the MER
column load and processed MSG satellite SEVIRI images for the 16 differ by a factor 10 or even larger.
April, reported in Fig. 5, shows a qualitative agreement. An overview of
the sensitivity of results to input uncertainty is shown in Table 1, where 4.2. Comparison with the DWD lidar-ceilometer network
airborne concentration and column load obtained for the different
runs at four different selected locations, are compared with results The German Meteorological Service (DWD) manages a network
obtained for the reference case. Typically, uncertainties in column of lidar ceilometer originally designed for cloud base height

Fig. 5. Qualitative comparison between FALL3D ash mass column load (in ton km2 or g m2) for the reference run and processed MSG Spinning Enhanced Visible Infra-red Imager
(SEVIRI) images (Pavolonis, 2010). Results over Germany at 16 Apr 06 UTC (top) and 18 UTC (bottom). The satellite-based volcanic ash retrievals were developed in preparation for
the next generation of Geostationary Operational Environmental Satellite (GOES-R) where a retrieval scheme developed for cirrus clouds (Heidinger and Pavolonis, 2009; Heidinger
et al., 2010) was adapted to volcanic ash clouds. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 173

monitoring. The network consists of Jenoptik CHM15K low-power 10.4 E, 25 m a.s.l.) and Hohenpeissenberg (South Germany, 47.8 N,
instruments providing aerosol backscatter intensity profiles at 11.0 E, 985 m a.s.l.) with FALL3D simulated extinction coefficients (the
a single wavelength of 1064 nm (Flentje et al., 2010a). Ceilometers simulated contour of 0.1 km1 relates roughly with a measured
are simple backscatter lidars (one wavelength) that do not resolve scatter intensity of 200 in arbitrary units). Comparison permits
depolarization and hence do not distinguish particle types. a validation of the model cloud arrival times, cloud altitudes and
However, if they are calibrated with co-located measurements can thicknesses. For both locations simulated cloud arrival times differ
be used to estimate, with large uncertainties, particle extinction from observations by less than 1 h and top layer altitudes differ by less
coefficients and mass concentrations (Flentje et al., 2010a; Heese than 1 km (consistent with Dacre et al. (2011)). On the 17 April, at
et al., 2010) that can be used for a quantitative model validation. Hohenpeissenberg at 3 km altitude level, the ash layer was first
The arrival of the ash cloud front over Germany during 16 April was observed at 06 UTC, in excellent agreement with the simulations
registered by all DWD network stations as a layer at about 6e7 km (Fig. 6c). Typically simulated layer thicknesses are thicker than the
altitude that gradually descended with time to 2e4 km. The cloud was observed one (2e3 times thicker). This is due to a few reasons. First,
observed as a dense layer of w1 km thick. According to Emeis et al. the low observation threshold is not well defined, second, the real
(2011) this apparent sinking motion can be attributed to the slanted layers are sometimes thinner than the vertical computational reso-
orientation of the layer (probably reflecting, in turn, a progressive lution (100 m), third, it can reflect numerical diffusion effects inherent
decrease of the eruption column height). The negative (downwards) of Eulerian models, forth, it can be due to an incorrect vertical
wind vertical component and the particle sedimentation also favored distribution of source. The thin ash layers visible by ceilometer on 18
the sinking motion to some extent but were too weak (z0.01 m s1 April (z00 UTC) at 3e4 km and on 19 April afternoon are also well
and <0.01 m s1 respectively) to be the only causes. On the 17 April captured by the model at right times and locations (Fig. 6b). The
only south German stations observed the layer, which entrained and model predicted also a peak on 20 April at noon but this is not visible
mixed with the convective planetary boundary layer from noon from ceilometer, clearly obscured by low clouds at that time.
onwards (Flentje et al., 2010a). This phenomenon is described in detail Using co-located sun photometer aerosol optical depth and
in Flentje et al. (2010b) at the Hohenpeissenberg Meteorological nephelometer measurements, Flentje et al. (2010a,b) estimate
Observatory (47.8 N, 11.0 E, 985 m a.s.l.). extinction coefficient profiles from ceilometer backscatter intensity at
Fig. 6 compares semi-quantitatively time series of measured 1064 nm. This allows for a more quantitative comparison with the
intensity backscatter profiles at Dornick (North Germany, 54.2 N, FALL3D results, shown in Fig. 7. Inferred extinction coefficient peak

Fig. 6. Time series of ceilometer intensity backscatter profiles (left, in arbitrary units) and FALL3D reference run computed extinction coefficient at 0.5 mm (right, in km1). Top:
Dornick (North Germany) from 15 to 18 April 2010. Measured signals above 5 km were identified as meteorological clouds. Middle: Hohenpeissenberg (South Germany) from 16 to
21 April 2010. Bottom: zoom at Hohenpeissenberg on 17 April. The seemly intermittence of the observed layer is due to low clouds that prevented the observation of the ash layer
above. Ceilometers profiles modified from Flentje et al. (2010a,b).
Author's personal copy

174 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

Fig. 7. Extinction coefficient profiles inferred by Flentje et al. (2010a) from ceilometer backscatter intensity at 1064 nm (left) and computed by FALL3D at 532 nm (right, solid line:
BPT model, dashed line: Mastin et al. (2009)). Observations at Hohenpeissenberg on 17 April at 07:30 (black), 08:30 (blue), 09:30 (green), and 10:30 (yellow) UTC. (For interpretation
of the references to color in this figure legend, the reader is referred to the web version of this article.)

values on 17 April morning are 5(2)  104 m1 at heights between with the PBL on afternoon of 18 April. This is quite consistent with the
2.7 and 3 km a.s.l., whereas the reference run simulation predicts model results although the simulated cloud is always thicker than
values of 2.5  104 m1 at heights between 3 and 3.6 km. Peak observed. Simulated peak concentration values for the reference run
differences are within a factor of two, the simulated maximum is are of about 1500 mg m3 at 2 km height on 17 April 06 UTC.
w0.5 km higher but it can be observed that the model is more Fig. 9 shows a similar semi-quantitative comparison with
diffusive (simulated profile width is about thrice larger). A very good measurements from a lidar deployed at the Science and Technology
agreement is obtained for mass concentration as peak values inferred Facilities Council (STFC) facility at Chilbolton Observatory (51.14 N,
from measurements are 600(400) mg m3 whereas the model 1.42 W, http://www.chilbolton.rl.ac.uk/) during 16, 18, and 20 April
predicts a decrease from 875 (07:30 UTC) to 750 mg m3 (10:30 UTC). 2010. At Chilbolton, the model predicts the cloud to arrive around
midnight of 16 April at 3e4 km height. This early arrival was not
4.3. Comparison with ground-based aerosol lidars observed by the lidar, although was predicted even by few other
models used to simulated ash cloud transport (Dacre et al., 2011). Likely
Europe hosts a large number of ground-based lidars designed for it reflects cumulative errors in the driving NWP wind fields (Dacre
meteorological and aerosol observations. Lidar networks like the et al., 2011), including the ERA reanalysis used here. At noon of 16
European Aerosol Research LIdar NETwork (EARLINET; www.earlinet. April, a layer of ash was detected descending from 2.5 to about 1 km
org) allow scientists to study continental and long-term processes not height in 3 h (from 12:00 to 15:00 UTC). This decrease is in good
evident from single-point lidar measurements. Lidars provide infor- agreement with the model, that in addition predicts layer mass
mation about vertical structure of troposphere aerosol layers. concentrations of 400 mg m3 for the reference run. On 18 April the
Advanced multi-wavelength lidars that measure polarization can be model predicts a concentration peak at 2.5 km from 14:00 to 17:00 UTC
combined with retrieval algorithms to give information about optical (Fig. 9b). This was actually observed by the lidar a couple of hours later
(e.g. extinction, backscatter, AOD) and microphysical aerosol proper- (18:00 UTC) at a lower altitude (w1.8e2 km). On 20 April the model
ties (e.g. size or shape). This provides an excellent framework for predicts a third flux of ash crossing over Great Britain from north to
model validation. We have compared the FALL3D model results with south (see Fig. 4) and arriving at Chilbolton early morning. Layer height
measurements from ground-based lidars at Belsk (Poland), Cabauw is predicted by model to fall from about 4 to 1 km height in about 8 h
(Netherlands), Cardington and Chilbolton (U.K.), Leipzig and Munich and to achieve a maximum concentration of 900 mg m3 at 07:00 UTC
(Germany) and Palaiseau (France). For some of these lidars only approximately. This series of event was clearly observed by the lidar,
a semi-quantitative comparison is possible while for others, like the although during the first 3 h of 20 April clouds partly obscured the lidar
multi-wavelength Munich and Leipzig Raman lidars, a more quanti- observations. Also in this case, the agreement between model and
tative validation can be carried out. observation is satisfactory with time differences of 1e2 h and
Fig. 8 shows a comparison with measurements of the SIRTA lidar discrepancies in altitude below 0.5 km. Similar results have been found
located at Palaiseau, 20 km south of Paris (49 N, 2 E, http://sirta.ipsl. for other lidars all over Europe (not all results are shown here).
polytechnique.fr/), during 17 and 18 April 2010, when the ash cloud Ansmann et al. (2010) reported Raman-lidar observations at
was stagnant over northern France (see Fig. 4). Observations indicate Leipzig (51.4 N, 12.4 E) and Munich-Maisach (48.2 N, 11.3 E)
a thin layer descending from 3.5 to 2 km altitude during the morning during 16e17 April 2010, including inferred vertical concentration
and afternoon of 17 April and remaining at this height until it mixes and extinction profiles at different time slices. Fig. 10 compares
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 175

Fig. 8. Comparison between 532 nm linear polarization measured at Paris during 17e18 April 2010 (left) and computed FALL3D mass concentration for the reference run (right).

these profiles with the FALL3D results obtained for the BPT refer- Additionally, direct-sun AOD processing now includes the Spectral
ence run and the run having the MER calculated in accord to Mastin Deconvolution Algorithm (SDA) described in O’Neill et al. (2003).
et al. (2009). At Leipzig, time-averaged concentration (from 14:15 This algorithm yields fine (sub-micron) and coarse (super-micron)
to 15:30 UTC on 16 April) from the lidar reached a peak of roughly AOD at a standard wavelength of 500 nm from which the Fine Mode
850 mg m3 at 3.3 km height and the ash layer was observed to be Fraction (FMF) to total AOD can be computed.
about 1.5 km thick. In this case the simulations show concentration Here we use AOD and AODcoarse at 500 nm from the SDA
peaks of 3400 and 100 mg m3 respectively, i.e. 4 times higher and retrievals at several AERONET sites for further quantitative model
8.5 times lower than the observation. Layer thicknesses are validation. Two limitations should be pointed out about AERONET
generally overestimated by a factor 2 for the reference run, whereas data. First, direct-sun radiance measurements are possible during
the run having the MER calculated in accord to Mastin et al. (2009) daylight only and can be obscured by clouds, although this should
systematically underestimate concentration. Results for the refer- be corrected in Level 1.5 and 2.0 data. Second, if no information on
ence run are in a much better agreement with observations at particle type is available, total AOD does not allow for the distinc-
Munich on 17 April, where differences with observations are tion between ash and other aerosol. This can mislead the effect of
smaller in both peak intensity (about 950 and 1000 mg m3 for ash if the background contribution is not well constrained. We
model and observation respectively) and altitude (Fig. 10b). Layer compare with AODcoarse to eliminate the contribution of anthro-
thickness are overestimated by a factor 2 also in this case, reflecting pogenic aerosols, mainly found in the finer fractions. Notwith-
the fact that BPT model tends to overestimate MER. standing these limitations, the comparison allows us to test the
ability of the model to predict cloud arrival times and AOD
4.4. Comparison with AERONET sun photometers modulation.
Fig. 11 shows the simulated ash AOD over Central Europe at
AErosol RObotic NETwork (AERONET, Holben et al., 1998) is different time slices. According to the model (reference run simu-
a federation of ground-based remote-sensing aerosol networks lation), large values (AOD > 1.5) are predicted only during 15 April
devoted to assess aerosol optical properties and validate satellite and 16 April morning over Denmark and Germany, although the
retrievals of aerosol optical properties. AERONET provides globally 18e19 April pulse also causes values of AOD z 1.5 over Scotland.
distributed observations of columnar AOD in diverse aerosol During most of the episode, low to moderate ash AOD values
regimes and at different wavelengths computed for three data (0.1e0.75) are dominant over the continent. Fig. 12 compares time
quality levels: Level 1.0 (unscreened), Level 1.5 (cloud-screened; series of sun-photometer derived AOD and AODcoarse with FALL3D
Smirnov et al., 2000), and Level 2.0 (cloud-screened and quality- ash AOD predictions at 4 different localities: Cabauw, Helgoland,
assured, longer delay). AOD is calculated from sun-photometer Lille, and Munich. Results at other locations (e.g. Belsk, Brussels,
measurements of the direct (collimated) solar radiation. Chilbolton, or Lägren, not shown here) present a similar agreement
Author's personal copy

176 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

Fig. 9. Comparison between backscatter range corrected signal measured at Chilbolton (UK) during 16, 18, and 20 April 2010 (left) and computed FALL3D mass concentration for the
reference run (right).

a b

Fig. 10. Comparison between mass concentrations estimated by Ansmann et al. (2010) from Raman-lidar observations (open circles) and FALL3D results for the reference run (solid
line) and computing the MER as in Mastin et al. (2009) (dashed line). a) Leipzig 16 April 2010. Observations are averaged from 14:15 to 15:30 UTC. Model results are plotted at 15:00
UTC. b) Munich-Maisach 17 April 2010. Observations are averaged from 01:40 to 02:30 UTC. Model results are plotted at 02:00 UTC.
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 177

Fig. 11. FALL3D ash AOD (532 nm) over central and north Europe at different time slices for the reference run. The localities reported in Fig. 12 are indicated.

(generally there are a few periods of good agreement and others several flights from 19 April to 18 May equipped with a Doppler wind
that show poor agreement). lidar and in-situ aerosol instrumentations like optical particle
counter spectrometers and trace gas instruments (Schumann et al.,
4.5. Comparison with in-situ aircraft particle counter 2011). The first Falcon flight was performed on 19 April as a request
to support airspace re-opening decisions after four days of airspace
Several measurement flights were carried out during the 14e23 shut down. The 19 April flight path crossed over Munich, Leipzig,
April week including, among others, the German DLR Falcon 20 Cabauw, and Stuttgart in coincidence with ground-based lidars.
(http://www.dlr.de/), the U.K. NERC Dornier (http://arsf.nerc.ac.uk/) Schumann et al. (2011) derived ash mass concentrations at different
and FAAM BAe164 (www.faam.ac.uk/), the Swiss METAIR DIMO flight legs from measured ash particle number size distributions.
(www.metair.ch), or the French SAFIR ATR (www.safire.fr). In-situ Fig. 13 shows modeled airborne ash concentration at different
and remote airborne measurements played a key role in support- flight levels on 19 April at the time of the Falcon flight (17:00 UTC).
ing decisions of agencies and operators. Different height-dependent structures are visible reflecting the
Here we compare FALL3D model results with in-situ measure- mixing of different ash pulses and the recirculation of ash sus-
ments from the German DLR Falcon 20. The Falcon aircraft performed pended in the lower troposphere for several days. Actually, the 19
Author's personal copy

178 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

Fig. 12. Comparison between FALL3D ash AOD, level 1.5 aerosol AOD (circles), and AODcoarse (solid points) at 500 nm measured by different AERONET sun photometers (Cabauw,
Helgoland, Lille and Munich). Results shown for the entire period of simulation. Note that all the particle classes considered in the TGSD belong to the aerosol coarse mode (>1 mm).
Contribution to this mode is expected to be mainly from natural aerosols (e.g. sea salt, mineral dust, volcanic ash). Two FALL3D simulations are plotted, BPT reference run (solid line)
and with the MER as in Mastin et al. (2009) (dashed line).

April flight Lidar observations revealed multiple ash layers at presence of two layers at roughly 2 and 4 km height with concen-
3.7e5.8 km altitudes that typically were 0.5e2 km thick and tration maxima of circa 180 mg m3. The reference run (mass over-
100e300 km wide (Schumann et al., 2011). Fig. 14 compares estimation) captures only a single peak of 340 mg m3 at 3 km above
different estimations of concentration with the FALL3D model the ground. For the other 3 legs, model results for the two simula-
results. At Leipzig during 19 April 1510 UTC measurements show the tions shown in Fig. 14 cover the measured concentration values.
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 179

Fig. 13. FALL3D (reference run) ash concentration contours (mg m3) over Germany on 19 April 17:00 UTC, in coincidence with the time of the DLR Falcon 20 flight. Results are
shown at four different flight levels of FL050, FL100, FL150 and FL200, roughly corresponding to altitudes of 5, 10, 15, and 20 thousand feet respectively.

4.6. Comparison with in-situ ground-based measurements For comparison, Fig. 15 shows FALL3D model predicted PM5 ground
concentrations at this locality for the reference run and for the run
Ground-based particle counters for boundary layer aerosol having the MER calculated in accord to Mastin et al. (2009). The
studies measure size-resolved particle number concentrations. This simulated concentration, for the reference run, increases along 17
kind of data may provide an additional method for validating April up to 68 mg m3, in agreement with observations.
models. Some of them registered in some locations a sharp increase
in number concentration of particles larger than 1 mm in coinci- 5. Summary and discussion
dence with the overpass of the ash cloud. For example, Flentje et al.
(2010b) report that at Hohenpeissenberg (Germany) PM5 particle We performed several semi-quantitative and quantitative
mass concentrations increased from 20 mg m3 to 40 mg m3 (with comparisons between FALL3D VATDM and independent
peaks of 50 mg m3) on 17 April and remained elevated till 23 April. measurements and observations of the Eyjafjallajökull volcanic
Author's personal copy

180 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

Fig. 14. Comparison between FALL3D vertical concentration profiles (above ground level, a.g.l.) and concentration values reported in Schumann et al. (2011) (Table 3) at different
flight legs during 19 and 22 April Falcon flights. FALL3D results are shown for the reference run (solid line) and computing the MER as in Mastin et al. (2009) (dashed line). These can
be considered as upper and lower bounds for the erupted mass and hence for the concentration downwind. The two circles in the Stuttgart, Munich, and Skagerrak legs indicate
maximum and minimum concentration values estimated by Schumann et al. (2011).

ash clouds obtained using different techniques during 14e23 effects and ii) a distribution (Suzuki, 1983; Pfeiffer et al., 2005)
April 2010. FALL3D model simulations used ERA-Interim mete- combined with the empirical relationship suggested by Mastin
orological data at 1.5 resolution (off-line coupling strategy), an et al. (2009) to estimate the MER from column height. We
effective TGSD derived from field observations (Bonadonna have found that, for this particular eruption (characterized by
et al., 2011a) and simulations on a small-scale that account for a low to moderate eruption intensity and a very oscillating
wet aggregation, and hourly-averaged values of eruption column height), these two strategies provide an upper and
column heights as measured by the IMO weather radar. Source a lower bound for the MER and hence for the downwind ash
term (MER and vertical distribution of mass) has been computed concentration. In particular, model results were compared with:
from observed heights using both i) a BPT model that accounts i) DWD lidar ceilometer, ii) ground-based lidars (from EARLINET
for wind induced plume bent-over and cross-wind entrainment and others), iii) sun photometers from AERONET, iv) DLR Falcon
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 181

during the Eyjafjallajökull eruption are the subject of ongoing


research. However, preliminary modeling of aggregation process
of Eyjafjallajökull eruption indicates that more than 20% of the
fine ash aggregated and the fraction of particle classes finer than
5 microns (the effective classes used for LIDAR retrieval) is less
than 3.5%. For comparison, Dacre et al. (2011) estimated that
around 3% of the total emitted mass was transported as far as
these sites by small ash particles (although they used a very
different TGSD). However they estimated MER through Mastin
et al. (2009) and they assumed that this fraction consists of all
particles  100 microns diameter, although it is necessary to
consider that only particle finer than 5 microns are typically used
for LIDAR retrieval.
The adoption of quantitative ash concentration safe criteria
has important implications on mitigation strategies. From
a modeling point of view, a key point to investigate is whether
VATDM can predict airborne concentrations with a sufficient
accuracy. In this context, an adequate estimation of the source
term is a pivotal issue because model input uncertainties are
a major source of inaccuracy. For the case of the Eyjafjallajökull
Fig. 15. PM5 ground concentration (in mg m3) predicted by FALL3D at Hohen-
peissenberg (South Germany) from 16 to 23 April 2010. Two cases, BPT reference volcanic ash clouds simulations considered here, where input
run (line with circles) and with the MER as in Mastin et al. (2009) (line with cross), parameters were much more constrained than during the time of
are plotted. an emergency, we found that differences between the concen-
tration computed by the model and the peak concentration
inferred from observations are typically overestimated within
a factor 2e3 (up tp 4e5 in a few cases) for the reference run,
20 aircraft particle counter and, v) a ground-based particle either airborne and at ground. However, it must be stressed that,
counter located at Hohenpeissenberg (Germany). Main findings at present, to achieve such agreement in operational mode is
can be summarized as follows: unlikely because model inputs are typically less constrained. For
example, TGSD is very often not available during an eruption
 Observed and computed cloud arrival times agree with time because difficulties in sampling and time needed for data pro-
differences of few hours (typically lower than 1e2 h) at all cessing. Column height measurements may not be available or if
analyzed localities (except for the UK on the 16 April). so, be available at a low frequency (e.g. daily values), especially for
 Discrepancies between observed and computed cloud alti- poorly monitored volcanoes. Finally, techniques to measure MER
tudes are below 1 km (0.5 km in most cases). The simulated on real or nearly real-time are not well consolidated yet, so that
cloud is always much thicker than the observed (by a factor of this key model input has to be inferred from BPT models or
2e3 at least). As explained, this can be partially explained by empirical fits from the observed column heights. In the case of
the numerical diffusion of Eulerian models, by poor quantifi- weak plumes, like Eyjafjallajökull eruption, BPT models tend to
cation of the detection threshold, and by the poor vertical overestimated whereas empirical fits tend to underestimate
resolution of NWP. In fact, the vertical resolution of the concentration values.
underlying NWP data (not the interpolated resolution) can This study demonstrates that current state-of-the-art VATDMs,
only capture the gradients/features in the flow over a number like FALL3D, can model far-range dispersion and sedimentation of
of grid lengths. volcanic ash clouds satisfactorily when inputs are well con-
 Measurements of ash mass concentration are always bounded strained. Strategies for improving model accuracy during emer-
by the two cases characterized by different estimations of the gencies, such as real-time or nearly real-time data assimilation or
MER, i.e. through BPT theory (Bursik, 2001; Carazzo et al., 2008) ensemble forecast techniques, were discussed recently at the “Ash
and using an empirical relationship (Mastin et al., 2009) dispersal forecast and civil aviation workshop” (Bonadonna et al.,
respectively. In general, total mass difference between the BPT 2011b) and reader are encouraged to read the final report at
and the empirical model is almost an order of magnitude. http://www.unige.ch/sciences/terre/mineral/CERG/Workshop.
However, the reference run (MER estimated by means of a BPT html.
model) shows, in general, a better agreement with the obser-
vations, typically within a factor 2e3 and up to 4e5 maximum. Acknowledgments
The tendency to overestimate concentration reflects the fact
that, for such low columns, cross-wind entrainment is not This work has been partially funded by the Spanish CGL2009-
properly estimated. 10244 ATMOST project. AF is grateful to the Ramón y Cajal scientific
 The simulations (for the reference case) captures AOD varia- program. We acknowledge the constructive reviews of two anon-
tions and differences with AERONET AODcoarse measurements ymous referees which have improved this manuscript notably. We
are lower than 0.2. also thank S. Karlsdóttir and G.N. Petersen from IMO for providing
radar eruption column heights. C. Bonadonna from University of
Results show that is crucial to improve methods for estimation Geneva provided granulometric data and suggestions for
of MER (for weak eruptions), estimation of TGSD, and aggregation improving the paper. A. Ansman from Leibniz Institute for Tropo-
effects. The use of an effective TGSD (combining Bonadonna et al. spheric Research and U. Schumann from DLR kindly provided us
(2011a) analysis with preliminary modeling of aggregation) data from Leipzig and Munich lidars and Falcon 20 flight respec-
allowed us to model long-range transport. Further studies more tively. We thank M. Pavolonis from NOAA for the satellite images.
focussed on the quantification of aggregation on the ash transport We also wish to acknowledge the British Atmospheric Data Centre,
Author's personal copy

182 A. Folch et al. / Atmospheric Environment 48 (2012) 165e183

which is part of the NERC National Centre for Atmospheric Science Grell, G.A., Emeis, S., Stockwell, W.R., Schoenemeyer, T., Forkel, R., Michalakes, J.,
Knoche, R., Seidl, W., 2000. Application of a multiscale, coupled MM5/Chem-
(NCAS), for the use of lidar data from Chilbolton and Cardington. H.
istry Model to the complex terrain of the VOTALP Valley Campaign. Atmo-
Dacre from University of Reading, UK, is acknowledged for her spheric Environment 34, 1435e1453.
suggestions. Finally we warmly thank G. Macedonio from INGV, Heese, B., Flentje, H., Althausen, D., Ansmann, A., Frey, S., 2010. Ceilometer lidar
Italy for his support and suggestions. comparison: backscatter coefficient retrieval and signal-to-noise ratio deter-
mination. Atmospheric Measurement Techniques 3, 1763e1770.
Heidinger, A.K., Pavolonis, M.J., 2009. Gazing at Cirrus Clouds for 25 Years through
Appendix. Supplementary material a Split Window. Part I: Methodology. Journal of Applied Meteorology and
Climatology 48 (6), 1100e1116.
Heidinger, A.K., Pavolonis, M.J., Holz, R.E., Baum, B.A., Berthier, S., 2010.
The Supplementary data associated with this article can be A comparison of the sensitivity to cloud pressure offered by the NPOESS/VIRRS
found online, at doi:10.1016/j.atmosenv.2011.06.072. and GOES-R/ABI Infrared observations for cirrus cloud remote sensing. Journal
of Geophysical Research 115. doi:10.1029/2009JD012152.
Herzog, M., Graf, H.F., 2010. Applying the three-dimensional model ATHAM to
References volcanic plumes: dynamic of large co-ignimbrite eruptions and associated
injection heights for volcanic gases. Geophysical Research Letters 37, L19807.
Ackermann, I.J., Hass, H., Memmesheimer, M., Ebel, A., Binkowski, F.S., Shankar, U., doi:10.1029/2010GL044986.
1998. Modal aerosol dynamics model for Europe: development and first Holben, B.N., et al., 1998. AERONET e a federated instrument network and data
applications. Atmospheric Environment 32, 2981e2999. archive for aerosol characterization. Remote Sensing of Environment 66,
Ansmann, A., et al., 2010. The 16 April 2010 major volcanic ash plume over central 1e16.
Europe: EARLINET lidar and AERONET photometer observations at Leipzig and ICAO, 2000. International Civil Aviation Organisation International Handbook on
Munich, Germany. Geophysical Research Letters 37, L13810. doi:10.1029/ the International Airways Volcano Watch (IAVW) Operational Procedures and
2010GL043809. Contact List, ICAO Doc 9766-AN/968, first ed.
Bursik, M., 2001. Effect of wind on the rise height of volcanic plumes. Geophysical Jones, A.R., Thomson, D.J., Hort, M., Devenish, B., 2007. The U.K. Met Office’s next-
Research Letters 18, 3621e3624. generation atmospheric dispersion model, NAME III. In: Borrego, C.,
Bonadonna, C., Genco, R., Gouhier, M., Pistolesi, M., Alfano, F., Cioni, R., Hoskuldsson, A., Norman, A.-L. (Eds.), Air Pollution Modeling and its Application XVII
Ripepe, M., 2011a. Tephra sedimentation during the 2010 Eyjafjallajökull eruption (Proceedings of the 27th NATO/CCMS International Technical Meeting on Air
(Iceland) from deposit, radar and satellite observations. Geophysical Research Pollution Modelling and its Application). Springer, pp. 580e589.
Abstracts 13, EGU2011-3476-2, EGU General Assembly 2011. Langmann, B., Varghese, S., Marmer, E., Vignati, E., Wilson, J., Stier, P., O’Dowd, C.,
Bonadonna, C., Folch, A., Loughlin, S., Puempel, H., 2011b. Future developments in 2008. Aerosol distribution over Europe: a model evaluation study with detailed
modelling and monitoring of volcanic ash clouds: outcomes from the first aerosol microphysics. Atmospheric Chemistry and Physics 8, 1591e1607.
IAVCEI-WMO workshop on Ash Dispersal Forecast and Civil Aviation, Bulletin of doi:10.5194/acp-8-1591-2008.
Volcanology, doi:10.1007/s00445-011-0508-6. Macedonio, G., Costa, A., Folch, A., 2008. Ash fallout scenarios at Vesuvius:
Byun, D., Ching, J., 1999. Science Algorithms of the EPA Models-3 Community numerical simulations and implications for hazard assessment. Journal of
Multiscale Air Quality Modeling System, EPA/600/R-99/030. US Environmental Volcanology and Geothermal Research 178, 366e377. doi:10.1016/
Protection Agency, Office of Research and Development, Washington DC. j.jvolgeores.2008.08.014.
Carazzo, G., Kaminski, E., Tait, S., 2006. The route to self-similarity in turbulent jets Mastin, L.G., et al., 2009. A multidisciplinary effort to assign realistic source
and plumes. Journal of Fluid Mechanics 547, 137e148. parameters to models of volcanic ash-cloud transport and dispersion during
Carazzo, G., Kaminski, E., Tait, S., 2008. On the dynamics of volcanic columns: eruptions. Journal of Volcanology and Geothermal Research. doi:10.1016/
a comparison of field data with new model of negatively buoyant jets. Journal of j.jvolgeores.2009.01.008.
Volcanology and Geothermal Research 178, 94e103. O’Neill, N.T., Eck, T.F., Smirnov, A., Holben, B.N., Thulasiraman, S., 2003. Spectral
Casadevall, T.J., et al. (Eds.), 1996. The 1991 Pinatubo Eruptions and Their Effects on discrimination of coarse and fine mode optical depth. Journal of Geophysical
Aircraft Operations. Philippine Institute of Volcanology and Seismology, Quezon Research 108 4559.
City and University of Washington Press, Seattle, p. 1115. Oxford-Economics, 2010. The Economic Impacts of Air Travel Restrictions due to
Corradini, S., Merucci, L., Folch, A., 2011. Volcanic ash cloud properties: comparison Volcanic Ash, Report for Airbus.
between MODIS satellite retrievals and FALL3D transport model. IEEE Geo- Pavolonis, M.J., 2010. Advances in extracting cloud composition information
science and Remote Sensing Letters 8 (2), 242e252. from space-borne infrared radiances e a robust alternative to brightness
Costa, A., Macedonio, G., Folch, A., 2006. A three-dimensional Eulerian model for temperatures, part 1: theory. Journal of Applied Meteorology and Clima-
transport and deposition of volcanic ashes. Earth and Planetary Science Letters tology 49 (9).
241 (3e4), 634e647. Petersen, G.N., 2010. A short meteorological overview of the Eyjafjallajökull erup-
Costa, A., Folch, A., Macedonio, G., 2010. A model for wet aggregation of ash tion 14 Aprile23 May 2010. Weather 65 (8), 203e207.
particles in volcanic plumes and clouds: I. Theoretical formulation. Journal of Peuch, V.H., Amodei, M., Barthet, T., Cathala, M.L., Josse, B., Michou, M., Simon, P.,
Geophysical Research. doi:10.1029/2009JB007175. 1999. MOCAGE, MOdéle de Chimie Atmosphèrique à Grande Echelle. In:
Dacre, H.F., Grant, A.L.M., Hogan, R.J., Belcher, S.E., Thomson, D.J., Devenish, B.J., Proceedings of Météo-France workshop on atmospheric modelling,
Marenco, F., Haywood, J.M., Ansmann, A., Mattis, I., Clarisse, L., 2011. Evaluating pp. 33e36.
the structure and magnitude of the ash plume during the initial phase of the Pfeiffer, T., Costa, A., Macedonio, G., 2005. A model for the numerical simulation of
2010 Eyjafjallajökull eruption using lidar observations and NAME simulations. tephra fall deposits. Journal of Volcanology and Geothermal Research 140,
Journal of Geophysical Research 116, D00U03, doi:10.1029/2011JD015608. 273e294.
D’Amours, R., Malo, A., Servranckx, R., Bensimon, D., Trudel, S., Gauthier- Prata, A.J., 2008. Satellite detection of hazardous volcanic clouds and the risk to
Bilodeau, J.P., 2010. Application of the atmospheric Lagrangian particle global air traffic. Natural Hazards. doi:10.1007/s11069-008-9273-z.
dispersion model MLDP0 to the 2008 eruptions of Okmok and Kasatochi Ryall, D.B., Maryon, R.H., 1998. Validation of the UK Met. Office’s NAME model
volcanoes. Journal of Geophysical Research 115, D00L11. doi:10.1029/ against the ETEX dataset. Atmospheric Environment 32, 4265e4276.
2009JD013602. Schumann, U., Weinzierl, B., Reitebuch, O., Schlager, H., Minikin, A., Forster, C.,
Draxler, R.R., Hess, G.D., 1998. An overview of the HYSPLIT_4 modeling system of Baumann, R., Sailer, T., Graf, K., Mannstein, H., Voigt, C., Rahm, S., Simmet, R.,
trajectories, dispersion, and deposition. Australian Meteorological Magazine 47, Scheibe, M., Lichtenstern, M., Stock, P., Rüba, H., Schäuble, D., Tafferner, A.,
295e308. Rautenhaus, M., Gerz, T., Ziereis, H., Krautstrunk, M., Mallaun, C., Gayet, J.-F.,
Emeis, S., Forkel, R., Junkermann, W., Schäfer, K., Flentje, H., Gilge, S., Fricke, W., Lieke, K., Kandler, K., Ebert, M., Weinbruch, S., Stohl, A., Gasteiger, J., Gross, S.,
Wiegner, M., Freudenthaler, V., Gross, S., Ries, L., Meinhardt, F., Birmili, W., Freudenthaler, V., Wiegner, M., Ansmann, A., Tesche, M., Olafsson, H., Sturm, K.,
Münkel, C., Obleitner, F., Suppan, P., 2011. Measurement and simulation of the 2011. Airborne observations of the Eyjafjalla volcano ash cloud over Europe
16/17 April 2010 Eyjafjallajókull volcanic ash layer dispersion in the northern during air space closure in April and May 2010. Atmospheric Chemistry and
Alpine region. Atmospheric Chemistry and Physics 11, 2689e2701. doi:10.5194/ Physics 11, 2245e2279. doi:10.5194/acp-11-2245-2011.
acp-11-2689-2011. Scollo, S., Folch, A., Costa, A., 2008. A parametric and comparative study of different
Flentje, H., Heese, B., Reichardt, J., Thomas, W., 2010a. Aerosol profiling using the tephra fallout models. Journal of Volcanology and Geothermal Research 176 (2),
ceilometer network of the German Meteorological Service. Atmospheric 199e211. doi:10.1016/j.jvolgeores.2008.04.002.
Measurement Techniques Discussions 3, 3643e3673. Scollo, S., Prestifilippo, M., Spata, G., D’Agostino, M., Coltelli, M., 2009. Monitoring
Flentje, H., et al., 2010b. The Eyjafjallajökull eruption in April 2010-detection of and forecasting Etna volcanic plumes. Natural Hazards and Earth System
volcanic plume using in-situ measurements, ozone sondes and lidar-ceilometer Sciences 9, 1573e1585.
profiles. Atmospheric Chemistry and Physics 10, 10085e10092. Sigmundsson, F., et al., 2010. Intrusion triggering of the 2010 Eyjafjallajökull
Folch, A., Costa, A., Macedonio, G., 2009. FALL3D: a computational model for explosive eruption. Nature 468, 426e430. doi:10.1038/nature09558.
volcanic ash transport and deposition. Computer and Geosciences. doi:10.1016/ Smirnov, A., Holben, B.N., Eck, T.F., Dubovik, O., Slutsker, I., 2000. Cloud screening
j.cageo.2008.08.008. and quality control algorithms for the AERONET database. Remote Sensing
Folch, A., Costa, A., Durant, A., Macedonio, G., 2010. A model for wet aggregation of Environment 73, 337e349.
ash particles in volcanic plumes and clouds: II. model application. Journal of Sparks, R.S.J., Bursik, M.I., Carey, S.N., Gilbert, J.S., Graze, L.S., Sigurdsson, H.,
Geophysical Research. doi:10.1029/2009JB007176. Woods, A.W., 1997. Volcanic Plumes. J. Wiley and Sons, Chichester, pp. 574.
Author's personal copy

A. Folch et al. / Atmospheric Environment 48 (2012) 165e183 183

Stohl, A., Hittenberger, M., Wotawa, G., 1998. Validation of the Lagrangian particle by field and laboratory high-speed imaging. Geology 39 (9), 891e894;
dispersion model FLEXPART against large-scale tracer experiment data. Atmo- doi:10.1130/G32016.1.
spheric Environment 32, 4245e4264. Wang, X., et al., 2008. Volcanic dust characterization by EARLINET during Etna’s
Suzuki, T., 1983. A theoretical model for dispersion of tephra. In: Shimozuru, D., eruptions in 2001e2002. Atmospheric Environment 42, 893e905.
Yokoyama, I. (Eds.), Arc Volcanism: Physics and Tectonics. Terra Scientific Wilson, L., Walker, G.P.L., 1987. Explosive volcanic eruptions-VI. Ejecta dispersal in
Publishing Company (TERRAPUB), Tokyo, pp. 93e113. plinian eruptions: the control of eruption conditions and atmospheric proper-
Taddeucci, J., Scarlato, P., Montanaro, C., Cimarelli, C., Del Bello, E., Freda, C., ties. Geophysical Journal of the Royal Astronomical Society 89, 657e679.
Andronico, D., Gudmundsson, M.T., Dingwell, D.B., 2011. Aggregation- Woods, A., 1995. The dynamics of explosive volcanic eruptions. Reviews of
dominated ash settling from the Eyjafjallajökull volcanic cloud illuminated Geophysics 33, 495e530.

Você também pode gostar