This paper presents a new version of the EMEP MSC-W model called eEMEP
developed for transportation and dispersion of volcanic emissions, both gases
and ash. EMEP MSC-W is usually applied to study problems with air pollution
and aerosol transport and requires some adaptation to treat volcanic eruption
sources and effluent dispersion. The operational set-up of model simulations
in case of a volcanic eruption is described. Important choices have to be
made to achieve CPU efficiency so that emergency situations can be tackled in
time, answering relevant questions of ash advisory authorities. An efficient
model needs to balance the complexity of the model and resolution. We have
investigated here a meteorological uncertainty component of the volcanic
cloud forecast by using a consistent ensemble meteorological dataset (GLAMEPS
forecast) at three resolutions for the case of SO
The European Monitoring and Evaluation Programme model developed at the Meteorological Synthesizing Centre – West (EMEP MSC-W) has been expanded to handle ash forecasting for the Norwegian Meteorological Institute. Historically, the EMEP MSC-W Eulerian model has been used to deal with problems concerning acidifying substance deposition, and long-range transport of tropospheric ozone and particles (Simpson et al., 2012). The EMEP MSC-W model is already in use in a forecasting mode as one of the ensemble members of the MACC/CAMS daily ensemble production system for regional air quality forecasting (Marécal et al., 2015). This paper will present the developments of the EMEP MSC-W model that allow the model to describe transport of both gaseous and ash emissions from a volcanic eruption in both a forecast and hindcast setting; this version of the model is called the emergency EMEP (eEMEP) model.
The volcanic emission and transport of SO
There are different approaches for volcanic ash transport and dispersion models (VATDMs). Eulerian models such as the eEMEP model are computationally more demanding compared to Lagrangian models, which most Volcanic Ash Advisory Centres (VAAC) use, e.g. NAME (Jones et al., 2007) at the London VAAC, or HYSPLIT (Draxler and Hess, 1997) at the Washington and Anchorage VAAC. Other well-known Lagrangian models used for ash dispersion are FLEXPART (Stohl et al., 2005) and PUFF (Searcy et al., 1998); the latter is also used as backup by the Washington and Anchorage VAAC. Some Eulerian models used for ash dispersion are MOCAGE (Josse et al., 2004) used at VAAC Toulouse, Fall3d (Folch et al., 2009), and Ash3d (Schwaiger et al., 2012). The Eulerian models calculate the advection of ash at every grid point, and emissions are instantaneously mixed within the grid box. In particular, peak concentrations are dependent on the grid resolution. Lagrangian models release tracers and calculate their trajectories; the mass loadings and concentrations are calculated from the number density of multiple releases of these tracers. This can lead to an uncertainty in regions with low particle concentrations, but the output resolution for Lagrangian models is independent of the resolution of the input data and can therefore be indefinitely high.
For all models, in addition to uncertainties caused by numerical diffusion and advection, uncertainties in the ash dispersion forecasting can also be due to imperfections of the meteorological driver. Initial conditions can only be set with a certain degree of accuracy when starting a numerical weather prediction model (Palmer, 2000; Iversen et al., 2011). The initial errors may amplify during the forecast and can result in forecast inaccuracies. In addition to these initial condition errors, there are uncertainties due to how the dynamics and physics are represented in the numerical weather prediction model (NWP). Ensemble forecasting was established in weather forecasting to estimate associated uncertainties by producing probability forecasts of the state of the atmosphere on the basis of multiple similar forecast runs with perturbed initial conditions or different model parameterizations. Since 1992 ensemble forecasts have been operational at both the National Meteorological Centre (NMC) (Toth and Kalany, 1993) and the European Centre for Medium-Range Weather Forecasts (ECMWF) (Palmer et al., 1993). Ensemble modelling has undergone large developments in recent years. In this study the eEMEP model will be run on state-of-the-art ensemble meteorology data at three different resolutions to see the different spreads in dispersion.
The aim of this paper is to present the new developments and applications of
the eEMEP model for describing the dispersion of volcanic emissions in the
atmosphere. Both volcanic eruption examples with SO
The standard EMEP MSC-W model is described in Simpson et al. (2012) and updates are in addition presented in the yearly EMEP reports as well as the updated model code (EMEP Status Report, 2016). The most important aspects of the standard model for volcanic emission dispersion are briefly described, while new added components for the eEMEP are presented in more detail, as well as how the model handles the source term and the operational set-up.
Volcanic emissions are transported from the source by winds and lost due to
several processes in the atmosphere. The advection scheme has a numerical
solution based on Bott's scheme (1989a, b), with the fourth-order scheme in
the horizontal direction and a second-order version applied on the variable
grid distances over the vertical direction.
Time steps used in the advection scheme are dependent on grid resolution.
Winds and other meteorological parameters needed are given as input and the
EMEP MSC-W model is adapted to run with output from several numerical weather
prediction models. Horizontal resolution follows the meteorological driver,
and model simulations with resolutions from very fine (few kilometres) to low
resolutions of
To improve the EMEP MSC-W model capabilities to model dispersion of volcanic emissions, the model was further developed in several components such that an efficient and flexible model framework was finally available for operations at the Norwegian Meteorological Institute. This emergency model is simplified in parts with respect to the original EMEP MSC-W model to be computationally more efficient.
On a day-to-day basis the eEMEP model uses ECMWF forecast meteorology,
pre-processed for the CAMS 50 chemical weather forecasting at a resolution of
A specific volcano source module reads in volcanic emission parameters from a
file containing ash flux (kg s
To perform quick simulations of the dispersion of volcanic emissions,
sophisticated chemistry and trace species emissions are computationally too
demanding and the eEMEP model has been configured such that they are
excluded. For volcanic eruptions with SO
Apart from the wind advection of volcanic ash and wet scavenging as described
in the standard EMEP MSC-W, an important process for the simulation of
volcanic ash is the gravitational settling. In the standard EMEP MSC-W,
sedimentation and dry depositions of the different pollutants are only
calculated in the lowest model layer. Fine ash is large enough to have an
effect from gravitational sedimentation and is emitted higher into the
atmosphere compared to other coarse aerosol such as sea salt and desert dust.
A module that calculates gravitational settling at all vertical levels for
ash particles is implemented. The assumed terminal fall speed
The eEMEP model runs operationally every day at the Norwegian Meteorological Institute, for dispersion scenarios of volcanic emissions as defined in Mastin et al. (2009) for four selected volcanoes in the region of interest. If an increased risk of an eruption is given for any volcano, one or several of the default volcanoes are replaced with the volcano at risk of eruption. Meteorological input data are available every day before 08:30 and 20:30 UTC and forecasts starting from 00:00 and 12:00 UTC are run from these respectively. A standard eEMEP model simulation takes less than half an hour, making forecasts from 00:00 and 12:00 UTC available before 09:00 and 21:00 UTC.
In case of a real volcanic eruption, several simulations are used and started
as shown in Fig. 1. The purpose of the everyday initial forecast with a
default volcanic source is to provide a conservative first estimate of the
dispersion. However, because of the high uncertainty in source intensity and
vertical profile as well as ash size distribution, the resulting
concentrations are very uncertain. Thus, as soon as possible, source receptor
model simulations, with a unit emission (1 kg s
Sequence of model simulations started at the Norwegian Meteorological Institute in the case of a volcanic eruption. The single black arrows indicate 48 h forecast simulations. The thick striped arrows represent the multiple model simulations started for the inversion algorithm to retrieve an improved emissions estimate using satellite data. Dashed lines represent spin-up model simulations with emissions estimates found by the inversion algorithm. These model simulations are continued as forecasts (single black arrows). The chronological order of simulations starts from the top, so new forecast results are available every 12 h. The inversion simulations are restarted every 24 h.
The EMEP MSC-W model results have been compared to model results from other
dispersion models and observations in several studies. In particular,
Steensen et al. (2016) compare model simulations for the Barðarbunga
eruption to satellite and ground observations of SO
List and names of model simulations used in this paper.
In meteorology, ensemble forecasts consist of several almost identical
simulations to quantify the uncertainty of a forecast. Large spread between
the ensemble members caused by a large difference between possible future
scenarios indicates a high uncertainty in the forecast. Combining the eEMEP
model with ensemble forecasts would create an opportunity for quantifying the
uncertainty in the eEMEP ash/SO
Here we investigate how the different resolutions of the ensemble forecasts
affect the spread of the volcanic plume for SO
The eEMEP model is run on meteorological ensemble forecast data from the Grand Limited Area Ensemble Prediction Systems (GLAMEPS) for the Barðarbunga eruption case. The starting dates, 3 to 5 September 2014, from which the respective 48 h forecasts are launched, correspond to the first part of the Barðarbunga volcanic eruption.
Significant amounts of SO
GLAMEPS aims to account for all the major sources of weather forecast inaccuracy by including both the differences due to model parameter uncertainty and initial state perturbations (Iversen et al., 2011). GLAMEPS ensemble forecast is produced at ECMWF, and in 2014 the ensemble consisted of 50 members from both the HIRLAM (High Resolution Limited Area Model) and ALADIN (Aire Limitée Adaptation Dynamique Développement International) models. To include uncertainty in the forecast, members are perturbed both in the initial field and on the model domain border. The perturbations are from the EuroTEPS (European Targeted Ensemble Prediction system), a version of the global ECMWF EPS, with higher resolution on a smaller European domain (Frogner and Iversen, 2011).
This study will only use the 24 HIRLAM (High Resolution Limited Area Model)
perturbation members of the ensemble (not the control member). The 24 HIRLAM
members are split between two different cloud physics parameterisations.
HirEPS_S members use the STRACO scheme (Sass et al., 1999; Undén et al.,
2002) for stratiform, convective cloud and precipitation; HirEPS_K members
use the Kain–Fritsch schemes for deep cumulus (Kain and Frisch, 1990; Kain,
2004; Calvo, 2007) and Rasch and Kristjansson (1998) for stratiform clouds
and precipitation (Ivarsson, 2007). To also include the uncertainty in the
forecast caused by the start time of the forecast, members are divided into
two groups with two different forecast start times. Six members of the
HirEPS_S and six of the HirEPS_K start the forecast at 00:00 and
12:00 UTC, and the remaining 12 ensemble members start the forecast at 06:00
and 18:00 UTC. All of the members are perturbed by using EuroTEPS. Each
member has a forecast time of 72 h. The original resolution is
The GLAMEPS data have been downloaded from ECMWF for the period from 3 to 5
September 2014, corresponding to the first phase of the Barðarbunga
volcanic eruption. Each member is used as input data for the eEMEP model to
run 48 h forecasts starting from 00:00 UTC from each of the 3 days by using
the 18 and 00:00 UTC meteorological forecasts. That means that for half of
the members, the forecast is 6 h old (the forecasted started at 18:00 UTC).
The relatively short forecast of 48 h is chosen due to the large
uncertainties related to the emission term when running a forecast of
volcanic emission (VAAC London only issues a maximum 24 h forecast).
Furthermore, running the full 72 h forecast is not feasible due to the
different start times of the forecasts (18:00 and 00:00 UTC). In contrast to
what is possibly done for a real case, all the forecasts are started from a
model state with no volcanic SO
Three different horizontal resolutions are generated as input: the original
high resolution of
Map of the number of ensemble members that locally exceeds a 5 DU
SO
The spread in the ensemble forecast of SO
The same as Fig. 2, counting ensemble members locally exceeding a 50 DU limit.
Compared to the two higher resolution forecasts, the low_res forecasts have
a large area where 20 or more members agree and have column loads over the
5 DU threshold after 48 h of dispersion. The mid_res and high_res
simulations show however a larger spread between the members with a bigger
area with only one or a few members above the low threshold. This larger
spread among the higher resolution forecast is also shown in Fig. 3. The
high_res forecast have members that exceed the 50 DU threshold far away
from the source, as seen over the coast of northern Russia in the first
forecast (3 September 00:00 UTC
The difference in the spread is also seen to be weather dependent, especially when using a low threshold. Figure 4 shows the 5 DU contour line for the forecasts corresponding to Fig. 2, for four of the ensemble members. Each of the four members represents one of the perturbed members from the two different model parameterisations and starting times. For the first forecast started there are large differences between the members for areas where they have VCDs above 5 DU. In the second forecast started, the differences between the members are smaller, while the last forecast from 5 September at 00:00 UTC shows that, although the members all have plumes with VCDs over 5 DU going south from Iceland, they have quite different positions, indicating a different position of the low pressure system.
To further investigate the differences in the three resolutions, Table 2
shows the area summed up at time step
Figure 5 shows the frequency above 10 DU of low and high resolution averaged
over the hours from 08:00 to 16:00 UTC on 5 September, for the forecast
starting on 4 September at 00:00 UTC. The model results are compared to OMI
(Ozone Monitoring Instrument) satellite observations from overpasses during
the same time. Retrievals are described in Theys et al. (2015) and have an
assumed plume height of 7 km, which is higher than the actual plume height,
and as a consequence the retrievals have overly low values. Even though the
column burdens from the OMI and the model results are not easily comparable
(see the discussion in Steensen et al., 2016), the patterns should be
similar. The satellite has high VCD vaules going south from Iceland in a thin
filament. Even though the total amount of area, where ensemble members show
SO
Total area A where SO
The higher resolution ensemble members show higher concentrations further
away from the volcanic eruption site in narrower SO
Even though a high resolution is desirable, the computational efficiency is
important in an emergency forecast environment. For this study, the highest
resolution runs use over 13 times more computational time than the lowest
resolution runs, while the mid_res simulations use only 5 times more. To run
a total ensemble forecast with high resolution for volcanic eruptions may
therefore not be feasible. From a pragmatic point of view, ensemble forecasts
for volcanic emissions are most valuable in situations where the weather
forecast is uncertain. Thus an alternative would be to launch ensemble
forecasts only in unstable weather situations (as predicted by the ensemble
weather prediction models). As in this study, to exclusively look at the
spread due to the uncertainty in the weather forecast, the same source term
should be used in all the members. Therefore the model simulations used as
input for the inversion calculations will only be driven by the deterministic
meteorology. This study indicates that less information is lost between
high_res and mid_res than going from the mid_res to low_res resolutions,
suggesting that resolutions around
The 5 DU contour lines for four exemplary members after 48 h of
forecast in the low, mid, and high resolution ensembles, in the left, middle,
and right columns respectively, for start time 00:00 UTC on 3 September
OMI retrieval of SO
The eEMEP model with improved ash modelling capabilities as described above
is tested here for the Eyjafjallajökull eruption in 2010. For this
purpose the model is run with the emission term from Stohl et al. (2011), an
emission term constrained by satellite observations through an inversion
routine. The ash is distributed over nine size bins with characteristic sizes
of 4, 6, 8, 10, 12, 14, 16, 18, and 25
Part of the validation has been done in the scope of the Norwegian ash
project and shall not be repeated here in all detail. We compared initially
ash dispersion from this eruption calculated with eEMEP and FLEXPART model
results as well as the Norwegian Meteorological Institute version of NAME
model SNAP (Saltbones et al., 1994), and found very similar ash plumes in all
three models (Norwegian ash project, 2014). Figure 6 shows results where all
three model results are compared to satellite ash retrievals from SEVIRI
(Spinning Enhanced Visible and Infrared Imager) and IASI (Infrared
Atmospheric Sounding Interferometer) available from
Mean ash column burdens from 08:00 to 09:00 UTC on 16 April for SEVIRI and IASI satellite ash retrievals, and eEMEP, SNAP, and FLEXPART model simulations.
Apart from the horizontal dispersion, the vertical placement of the transported ash may have important consequences for impact assessments, both for air quality and air traffic perturbations. Meteorological processes such as subsidence and frontal lifting may alter the initial vertical distribution of ash. In addition, ash removal and settling may alter the vertical distribution. Although several observational sets are available for the Eyjafjallajökull eruption, to test here the treatment of gravitational settling for ash particles in the eEMEP model, model results with and without gravitational settling of ash included are compared to lidar observations of the ash layer.
Lidar observations provide a vertical location of aerosol. The European
Aerosol Research Lidar Network (EARLINET) consisted at the time of the
Eyjafjallajökull eruption of 27 aerosol stations over Europe. On 15
April, an alert was given to start continuous measurements providing, if
weather conditions permitted, an hourly vertical coverage of the ash cloud
over Europe (Pappalardo et al., 2013), documented as a consolidated dataset
which we use here. Ash is detected as a significant aerosol backscatter
signal, linked to the Iceland eruption through backward trajectory analysis.
Only backscatter profiles with a relative statistical error from signal
detection of less than 50 % are used to retain a reliable aerosol mask.
The vertical resolution in the dataset ranges between 60 and 180 m for the
different stations. The dataset includes the identified top and bottom of the
ash layer, as well as the centre of mass, the altitude where most of the
aerosol load is located. Identified ash layers where other aerosol sources
are also found from e.g. continental aerosol are classified as mixed layers.
These mixed layers are also given with the maximum and minimum observed
height and centre of mass. Observed planetary boundary layer (PBL) height is
also included in the database. The six lidar stations used here are situated
in central Europe (see Fig. 7), covering coastal stations and inland and
mountain regions. Weather conditions at the lidar stations, and sometimes
technical issues, made it difficult to continuously produce observations. For
example, frequent low clouds over Cabauw prevented most lidar retrievals
there. Observations at Neuchatel are also limited to the first episode in
April. Altogether, the ash layer was observed over a long period over central
Europe during the Eyjfjallajökull eruption and, as the ash has been
transported over a long distance, the effect of gravitational settling may be
visible for the fine ash particles, making this dataset the best available at
the time for our purpose. Webley et al. (2012) found by studying model
results from WRF-Chem that ash particles larger than 62.5
Map of EARLINET lidar measurement sites used in the study.
Figure 8 shows the model concentrations for the simulation with gravitational settling over the entire Eyjafjallajökull period along with observed height of the ash layer and height of the mixed aerosol layer at the EARLINET stations. Although the mixed layers may be weighted with the other aerosol, they are plotted here also. Figure 9 concentrates on the centre of mass comparison (without the mixed layers) for both model simulations with and without gravitational settling.
Height–time profiles of ash concentrations from the eEMEP model, including gravitational settling, at the six EARLINET lidar stations (see Fig. 6) in the April–May 2010 episode (contour graph in the background). Lidar-detected upper and lower heights of the ash layer are presented as grey dots. The lidar-retrieved centre of mass for ash is plotted as black dots. For mixed layers where ash is identified with continental aerosol, the height of the layer is presented as light pink dots, and centres of mass are red dots. The height of the planetary boundary layer is shown in violet. Due to weather conditions and technical difficulties, the lidar measurements are not a continuous series.
Modelled and observed centres of mass for ash at the lidar stations.
Green and blue dots represent the centre of ash mass, computed from the
entire model column, for simulations with and without gravitational settling,
shown where ash concentrations were larger than 0.1
Ash was first detected at the Hamburg station during the morning of 16 April,
48 h after the start of the eruption. Ash was also observed early at the
other stations, and while the timing of the observations matches well at
Hamburg and Leipzig, at Neuchatel ash is observed before the model has
transported ash to this station. At Cabauw, the first part of the ash plume
is not covered by the lidar because no measurements are available, while the
second part shows similar simulated and observed levels of maximum
concentrations. Even though a lidar does not measure concentrations, it is
possible to retrieve these using mass-to-extinction coefficients. Ansmann et
al. (2011) and Wiegner et al. (2012) estimated maximum ash concentrations of
around 1100
From 2 May the model results show small ash concentrations at the lidar
stations, due to small ash emissions after 29 April. On 5–6 May ash is
observed lower down in the atmosphere compared to simulated ash at Hamburg;
however, a layer where ash is mixed with other aerosols is detected at higher
altitudes more similar to where the model has ash. More ash was then emitted
on 5 May (Stohl et al., 2011), but southerly winds transported the ash over
Spain and the Atlantic Ocean. Not until the night of 16/17 May are weather
conditions favourable again for transport of ash to central Europe. No
measurements are available for this time at Neuchatel. The other stations
have observations of the ash layer at similar altitudes to the model. Ash
concentrations estimated over Cabauw on 17 May are around
500
To show more broadly the impact of the gravitational settling processes on the vertical profile of ash, Fig. 8 shows all calculated centres of mass for ash in the model simulation with and without gravitational settling. The rather small displacement between the two model simulations implies that not gravitation, but rather weather and emission height, are the main drivers of the ash layer height. This is especially visible in the simultaneous rapid decrease in the centre of mass height for the first plume (17–18 April) in both simulations. On some occasions there are larger differences between the two model simulations, specifically in the beginning of May during a period with smaller concentrations. Unstable north-westerly winds at this time can cause the small differences in height distribution of ash to grow over time due to different wind directions in the column.
In order to compare to the observed values more properly, a centre of mass above the observed PBL is calculated for the two model simulations (only for the cases when an observed height is available); see Fig. 8. Model centres of mass are generally lower than observed altitudes for both model simulations, indicating that the model simulations have too much descent of the ash layer e.g. around 18 April, independent of inclusion or not of gravitational settling. Figure 10 show scatterplots where the observed ash centre of mass height (not including the mixed layers) is plotted against the model with and without gravitational settling at the stations. As discussed above, some measured and modelled values are unrealistically high; therefore, only values below 8 km are taken into account for correlation calculations. The scatterplot confirms that observed heights are generally higher than model calculations. At Palaiseau and Hamburg, model height descends faster than observed on 20 April, causing the low correlation at these stations. Neuchatel generally exhibits a higher observed centre of mass, explaining possibly a slightly higher correlation for the model simulation with no gravitational settling. Except for Neuchatel and Hamburg, however, the model simulation with gravitational settling exhibits a slightly higher correlation with lidar-retrieved height data compared to the model simulation without gravitation.
Scatterplots for observed versus simulated centres of ash mass with (magenta) and without (orange) gravitational settling. Data correspond to Fig. 8 using model and observed values under 8 km but above the PBL. Correlation between observed and model values is given in the upper left corner.
A new model version of the standard EMEP MSC-W model has been developed, aimed at modelling dispersion of volcanic emission, called the eEMEP model. Changes with respect to the standard model are a simplified gas chemistry; a modification of the aerosol part to handle ash particles in different size classes; the description of gravitational settling of ash particles; a volcanic source module which has a default source term and can be altered to include improved source estimates; an increase in vertical levels to increase the model top and vertical resolution; the possibility to run as an ensemble model based on ensemble meteorological forecasts; a formal procedure for an operational use of the model in an emergency case; and an inversion algorithm coupled to the model, using satellite data to retrieve an improved source estimate (Steensen et al., 2017). With this model version we document here selected important aspects of the volcanic gas and ash dispersion simulation.
We have first studied the impact of ensemble meteorological input fields of
different resolutions on the dispersion of volcanic emissions from Iceland.
Compared to Lagrangian VATDMs, Eulerian models such as the eEMEP model have
inherent numerical diffusion dependent on the grid size. eEMEP model
simulations thus have to have a sufficiently high resolution, especially when
peak concentrations shall be predicted, for example for the purpose of
establishing flight restriction zones. High resolution simulations are
however computationally demanding, while obtaining results quickly is
critical in situations with volcanic eruptions. How to best use CPU resources
for transport of volcanic emission is studied here by looking at the change
in spread between ensemble model simulations at three different resolutions.
The eEMEP model is run for a 48 h forecast from three start dates for the
Barðarbunga eruption period with meteorological fields from 24 HIRLAM
ensemble members originally produced for the GLAMEPS forecast. The original
The increased numerical diffusion causes a larger area (
The vertical dispersion of ash transport was studied. Gravitational settling
for ash tracers is added in the model over the entire vertical column. This
addition is evaluated by comparing a model simulation with and without
gravitational settling to observations during the 2010 Eyjafjallajökull
eruption. EARLINET ground stations measured the vertical location of the
volcanic ash layer over the eruption, providing hourly observations of the
height and centre of mass for the ash layer when the weather allowed it. The
centre of mass calculated for the two model simulations shows that
gravitational settling displaces the centre of mass closer to the ground by
up to 1 km. Besides emission height, the weather situation is found to be a
more important factor than gravitation for the height of the ash layer as
most of the vertical displacement is caused by subsidence in high pressure
systems and is similar in both model simulations. An example is a rapid
descent in ash plume height on 16 April caused by an anti-cyclone seen in
both observation and model. However, the descent in the model is quicker and
puts the ash closer to the ground compared to observations, especially at the
Hamburg and Leipzig lidar stations. A second descent in the ash layer at the
stations is seen on 20 April, and this subsidence occurs later in the
observational data at Hamburg and Palaiseau compared to the model data. The
model has a centre of ash mass height on average below the observed one,
independent of gravitational settling. Calculated correlation between
observed centres of mass height and corresponding model heights are higher in
the model simulation, with gravitational settling for four of the six
stations studied here, suggesting improved quality of the model when
including the gravitation process. The addition of gravitational settling is
found to have a relatively small influence on the vertical placement of the
ash layer and thus is responsible only for a small improvement in model
results. The model simulations presented here only include ash sizes of up to
25
Even with the included gravitational settling in the EMEP model, the assumed
density, shape, and size distribution of the ash particles bring along large
uncertainties during a forecast situation. Ash properties show large
differences in between volcanic eruptions (Vogel et al., 2016). The
Eyjafjallajökull model results presented here are initiated with a time
and height resolved emissions estimate calculated by inversion with FLEXPART
model results, constrained with satellite observations (Stohl et al., 2011),
to be used with the eEMEP model for a new volcanic eruption in an operational
set-up. Uncertainties in satellite retrievals due to meteorological clouds
that obscure ash clouds and a 0.2 g m
Although a correct model description of bulk volcanic emissions is useful, other factors such as model resolution, details of the source term, and the model set-up are seen as important for safety assessments. The developed model is capable of guiding near real time emergency assessments of the spread of high volcanic gas and aerosol concentrations.
The model code to the standard EMEP MSC-W model is
available on github:
The authors declare that they have no conflict of interest.
The authors would like to thank Fred Prata and Lieven Clarisse for SEVIRI and IASI satellite data for the model comparison and Nina I. Kristiansen for numerous inspiring discussions and technical help. We also thank Inger-Lise Frogner for the valuable help with obtaining GLAMEPS data. We thankfully acknowledge the EARLINET lidar data providers for assembling a very useful lidar dataset. This work has also received funding under the ACTRIS project from the European Union's Horizon 2020 research and innovation programme under grant agreement no. 654109. The work done for this paper is funded by the Norwegian ash project financed by the Norwegian Ministry of Transport and Communications and AVINOR. Model and support are also appreciated through the Cooperative Programme for Monitoring and Evaluation of the Long-range Transmission of Air Pollutants in Europe (no. ECE/ENV/2001/003). This work has also received support from the Research Council of Norway (Programme for Supercomputing) through CPU time granted at the supercomputers at NTNU in Trondheim. Edited by: A. Colette Reviewed by: two anonymous referees