Introduction
(a) “True”-colour VIIRS daytime image for 13:35 UTC on 16 September 2013 with the locations of
the Skarðsfjöruviti (green marker) and Mýrdalssandur (red marker) weather stations. The location of
the OPC at Maríubakki is indicated by the blue marker. The locations of the volcanoes Eyjafjallajökull and
Grímsvötn are indicated on the map by the E and G symbols respectively. (b) The recorded wind speeds at
Skarðsfjöruviti and Mýrdalssandur during 15–17 September 2013.
Iceland is one of the most active volcanic regions on Earth, with ≥ 20
eruptions per century , and explosive
eruptions can leave behind widespread ash deposits
e.g.. These
deposits are subject to intense aeolian processes: Iceland is windy, and the
lack of vegetation inhibits soil formation and particle binding, resulting in
significant remobilization events in the years following a volcanic eruption
. The eruptions of Eyjafjallajökull in 2010 and
Grímsvötn in 2011 provided a fresh source of unconsolidated ash
deposits in southern Iceland, and there have been a number of significant
resuspended ash events in the years following these eruptions
. Between 19 September 2010 and
16 February 2011 there were 12 observed resuspension episodes recorded by
PM10 counters in Drangshlidardalur (southern Iceland) of the
Eyjafjallajökull ash deposits . Following a blizzard on 6 March 2013
resuspended ash was deposited in Reykjavik, and particles were identified as
having originated from both the Eyjafjallajökull 2010 and
Grímsvötn 2011 deposits . Resuspended “ash
storms” can pose a significant hazard to the local population; decreased
visibility levels impact ground transportation and airports
and poor-air-quality episodes can
be a concern for human health e.g. and
livestock .
Following the eruption of Eyjafjallajökull in 2010, which deposited 140 ± 20 × 106 m3 of tephra in Iceland
, the Met Office in the UK has provided routine
forecasts to the Icelandic Meteorological Office (IMO) which indicate the
likely timing and location of resuspended ash clouds. Forecasts are produced
using the Lagrangian atmospheric dispersion model NAME (Numerical
Atmospheric-dispersion Modelling Environment; ), which
includes a resuspension scheme developed by .
Resuspended particles are advected by three-dimensional winds provided by the Met
Office's Numerical Weather Prediction (NWP) model and dispersed using
random-walk techniques which account for turbulent structures in the atmosphere
. Particles are removed from the
atmosphere by both dry- and wet-deposition processes
.
The emission of remobilized particles depends on the meteorological
conditions; soil moisture; terrain roughness; and characteristics of the
fallout deposit, including the size and density of particles and deposit
thickness . NAME includes a dust scheme which
explicitly models the resuspension of mineral particles; the emission rate
and the size distribution of the resuspended particles are calculated as a
function of soil moisture, vegetation fraction, clay fraction, and the wind
friction velocity . However,
information on the spatially varying surface characteristics of ash deposits
is often not available, especially when the deposits are relatively recent
. Instead
implemented a simple emission scheme in NAME for resuspended volcanic ash in
which remobilization occurs when the local wind friction velocity exceeds a
prescribed threshold and precipitation rates are low. Emission rates were
calibrated using measured PM10 data collected at multiple sites across
Iceland from two significant resuspension events during 23 May–2 July 2010
and 21 September 2010–16 February 2011, shortly after the eruption of
Eyjafjallajökull in 2010. However, in the following year the eruption of
Grímsvötn resulted in further widespread tephra deposits
, providing an additional source of remobilized
ash which is not accounted for in the calibration presented in
. It is also expected that the scaling coefficient
used to calculate emission rates of resuspended ash in the
approach will vary with time as deposits are
dispersed, eroded, and compacted.
During 16–17 September 2013 strong surface winds over tephra deposits in
southern Iceland led to the resuspension and subsequent advection of
significant quantities of volcanic ash particles. The resuspended ash cloud
was transported to the south-east over the North Atlantic Ocean and, due to
clear skies at the time, was exceptionally well observed in satellite
imagery. Here we use satellite-based measurements in combination with
radiative transfer modelling to quantify the total column mass loadings of
the resuspended ash cloud. These are then used to calibrate the emission rate
applied in the resuspension scheme in NAME. The newly calibrated scheme can
be used to provide more accurate quantitative forecasts of future events and
to assess how resuspension rates vary over time.
Dust and volcanic ash may be detected by satellite instruments sensitive to
either solar or thermal radiation. Infrared (IR) detection of ash clouds and
retrieval of ash cloud properties have been described by, for example,
, , , and .
used IR bands 31 and 32 of the Moderate Resolution
Imaging Spectroradiometer (MODIS) to detect and quantify a sandstorm in
China, and the solar channels of MODIS are routinely used to produce aerosol charts
. We analyse data from the Visible Infrared Imaging
Radiometer Suite (VIIRS) on board the Suomi National Polar-orbiting
Partnership (Suomi NPP) satellite. The brightness temperature difference
(BTD) between VIIRS bands M15 and M16, BTDV=BTM15-BTM16, can be used to
detect volcanic ash using an approach similar to that applied to MODIS bands
31 and 32 . The BTDV signal
depends on a number of factors including the properties of the ash particles
(their size and shape), the altitude of the ash cloud, and the temperature of
the Earth's surface . Dispersed ash following the eruption
of a volcano often resides at high altitudes in the atmosphere, giving a
negative BTDV signal, compared to ice clouds, which give positive BTDV
values. In this study we explore how to identify low-altitude resuspended ash
clouds using the split-window method.
The manuscript is organized as follows. In Sect. observations
from the event are presented: meteorological, particulate air concentrations
from an optical particle counter (OPC), and satellite imagery. In
Sect. the radiative transfer and dispersion modelling is
described. In Sect. we calibrate the emission rate in the
resuspension scheme in NAME with the satellite-retrieved total column mass
loadings and quantify the total mass of ash resuspended during 16–17
September 2013. We discuss the results in Sect. , before
the conclusions are presented in Sect. .
Modelling
Dispersion model forecasts
The atmospheric dispersion model NAME includes a scheme to model the
resuspension of volcanic ash . Particles are
remobilized from the surface when the local friction velocity (U*), which
characterizes the wind shear at the surface, exceeds a threshold friction
velocity (Ut*). The threshold friction velocity depends on the properties
of the particles (their size and density) and on the surface conditions – such
as soil moisture and roughness, and vegetation cover. Information on the
spatially varying characteristics of volcanic ash deposits is often
unavailable, particularly as deposits change with time due to erosion,
compaction and remobilization. found that using a
threshold friction velocity of 0.4 m s-1 was most appropriate for
modelling the resuspension of ash from deposits following the 2010 eruption
of Eyjafjallajökull in Iceland. They also note that this agrees well with
a threshold of 0.42 m s-1 identified from wind tunnel experiments
. also found that a
threshold friction velocity of 0.4 m s-1 was most appropriate when
modelling the resuspension of fallout deposits from the June 2011 Cordón
Caulle eruption in central Patagonia during October 2011. We take the
threshold friction velocity to be 0.4 m s-1 and assume that
resuspension does not occur when precipitation rates are > 0.01 mm h-1. The meteorological fields used in this study are provided
by the NAE (North Atlantic and European) configuration of the Met Office's
Unified Model (UM; ), which has a horizontal resolution
of 12 km .
We consider the deposits from the eruptions of Eyjafjallajökull in 2010 and
Grímsvötn in 2011 to be potential sources of resuspended ash. The extent
of the Eyjafjallajökull 2010 ash is based on a deposit map provided by
. In the absence of a published map of
Grímsvötn deposits we use a modelled deposit generated using NAME to
simulate the eruption of Grímsvötn in 2011, as described by
. All regions where ash has a depth > 5 mm are
considered, and the source areas used are indicated in
Fig. . Source regions are represented in NAME by a
horizontal grid with a resolution of 0.01∘ longitude and
0.01∘ latitude. The driving meteorology is considered at each grid
cell in order to determine whether particles should be resuspended.
Comparison of the location of the modelled resuspended ash cloud, represented by the blue line,
to the area identified from VIIRS retrieval data, represented by the red polygons; see Sect.
for the methodology used to define this area. The outline of the modelled plume is derived from un-calibrated 1 h
averaged total column mass loadings; values > 10-7 g m-2 are considered. The source areas are
identified by the grey areas.
Where resuspension occurs, model particles are released with a uniform
distribution between 0 and 10 m above the ground and are assigned a density of
2300 kg m-3. Their size distribution depends on the source: the
particle size distribution (PSD) of the Eyjafjallajökull 2010 ash is
based on measurements of samples collected from deposits on 15 April 2010
, whilst the PSD of the Grímsvötn ash is
based on samples collected from deposits on 22 May 2011
. To be able to compare the modelled ash cloud to the
OPC measurements and the satellite retrievals, we only model particles with
diameters between 1 and 10 µm . The
rate at which particles are remobilized is proportional to the cube of the
excess friction velocity:
F=K(U*-Ut*)3,
where K is a dimensional constant used as a scaling coefficient. Without
calibration K is set to 1 g s-1, the emission rate increases as U*
increases, and modelled air concentrations indicate areas of high and low
concentrations, but the results are not quantitative. A discussion on the
uncertainties associated with the model set-up, the source areas,
precipitation thresholds, and source mixing by previous remobilization can be
found in .
Time series of 1 h averaged air concentrations of resuspended ash derived from OPC count
data and the un-calibrated model output, compared to the NAE precipitation rate and friction velocity (U*) at Maríubakki.
Figure shows the time series of calculated air concentrations
from OPC count data at Maríubakki for the period 9 September to 2 October
2013. Modelled air concentrations using the un-calibrated emission rate
(K=1 g s-1) are compared. The modelled peak concentration is at 19:00 UTC on 16 September 2013, ∼ 19 h earlier than the recorded peak
concentration by the OPC at 09:00 UTC on 17 September 2013. A possible
explanation for this time lag between the modelled and observed peaks could
be that resuspension is suppressed in NAME when precipitation rates are
> 0.01 mm h-1. This approach does not account for the time required
to wet the deposit and prevent resuspension, and to dry the deposit before
resuspension can restart. However, in comparing the particle concentrations from
the OPC count data and the un-calibrated model output to the NAE
precipitation rates and local friction velocity at Maríubakki, we show
that there was no precipitation in Maríubakki during 15–17 September,
indicating that during the 24 h prior to the modelled peak concentration
the deposit was dry (Fig. a and b).
Therefore it is unlikely that the offset in the modelled and observed peak
concentrations can be ascribed to the lack of parameterization for a
drying-out process in NAME. Figure c
and d show that the peak in the OPC data does not
correspond well with the peak in the modelled friction velocity (U*). This
suggests that a significant fraction of the resuspended ash particles
detected by the OPC at Maríubakki must have been transported into the
area from surrounding deposits. Comparing dispersed model output with data
collected at a single point location is challenging and non-ideal for a model
calibration e.g.. Possible explanations for the
offset in the observed and modelled peak air concentrations could be that
the NWP model did not accurately represent the local topography, leading to
errors in the modelled wind vectors, or uncertainty in the modelled
precipitation. It could also be associated with uncertainty in the defined
source areas or uncertainty associated with the OPC data.
The modelled location of the resuspended ash during 16–17 September 2013
at the times corresponding to the VIIRS data is shown by the blue lines in
Fig. . The edge of the ash cloud is identified as
1 h averaged mass loadings > 1 × 10-7 g m-2, with this
threshold taken as a pragmatic plotting choice as the emission rate is
un-calibrated. Figure shows that ash is resuspended
from both the Eyjafjallajökull and Grímsvötn deposits and
transported to the south-east over the North Atlantic on 16 September and
then to the south-west as the wind changes direction on 17 September
(Fig. ). Both the location and timing of the modelled ash
cloud agree well with the VIIRS daytime RGB composites (cf.
Fig. ). Figure shows the maximum
height of the modelled ash cloud and indicates that ash resided at low levels
in the atmosphere, < 1600 on 16 September and < 2000 m a.s.l. on 17
September. This suggests that the ash cloud was trapped below the temperature
inversion, at ∼ 1500 m (Fig. ).
The maximum height of the modelled ash cloud using NAME at the times corresponding to the
satellite retrievals. The locations of the volcanoes Eyjafjallajökull and Grímsvötn are
indicated on the map by the E and G symbols respectively.
(a) The simulated brightness temperature difference between VIIRS bands M15 and M16, BTDV,
for a 1 km thick ash layer as a function of ash layer top altitude. (b) The difference
BTDwmod-BTDw=0.0mod as a function of ash cloud top height. The curves represent varying ash
mass loading (g m-2) and are given in the legend.
Brightness temperature difference signal
The brightness temperature difference between VIIRS bands M15 and M16,
BTDV=BTM15-BTM16, can be used to detect volcanic ash. To determine
the expected BTDV signal for the altitudes at which the resuspended ash
cloud resided during 16–17 September 2013, radiative transfer
calculations were carried out for a number of ash cloud top heights.
Figure a shows calculated BTDV for a 1 km thick ash cloud
with varying ash mass loadings and ash cloud top heights ranging from 0.5 to
10.0 km. For the ash cloud with a maximum altitude of 0.5 km the ash
concentration was doubled to preserve constant mass. In addition, a
simulation with all the ash in a 10 cm thick layer on the surface was
included. The assumption of an ash layer with a thickness of 1.0 km is based
on the plume heights predicted using NAME (Fig. ). Ash
particles were assumed to have a log-normal size distribution with effective
radius re= 2.0 µm and geometric standard deviation
σ=2.0, and nadir-viewing geometry was adopted. For an ash cloud at 8 km the BTDV decreases
from about 0.5 to -9 K when the ash mass loading increases from 0 to 0.01 g m-2. Further increasing
the mass loading increases the BTDV until the
signal in the two channels saturate (BTDV of about 1.5 K). This bowl-shaped
behaviour is qualitatively similar to the behaviour shown in Fig. 2 of
and Fig. 2 of for ash clouds at higher
altitudes. Figure a further shows that, for ash cloud top
heights above 2.0 km, BTDV is negative for mass loadings less than 0.02 g m-2.
Contrarily, BTDV>0.0 when the top of the ash cloud is between
0.5 and 2.0 km and mass loadings are ≥0.02 g m-2. As 16–17
September 2013 resuspended ash cloud top is between 1 and 2 km, a positive BTDV
signal is therefore to be expected for volcanic ash, as seen in
Fig. S6.
The absorption of radiation by atmospheric water vapour is larger at
12.0 µm than at 11.0 µm. Hence, the presence of water vapour may
reduce the volcanic ash BTDV signal. To remove the water vapour
contribution to the BTDV signal, both empirical
and model-based correction procedures have been
developed. present the following correction procedure
for water vapour absorption:
BTDVc=BTDV-BTDw,
where BTDw is the BTD with water vapour and without ash:
BTDV=T15-T16,BTDw=T15m-T16m.
Here T15,16 are the measured brightness temperatures in VIIRS bands M15
and M16 respectively, and T15,16m are the modelled brightness
temperatures including only water vapour. Such a correction procedure assumes
that radiation from the water vapour is independent from the radiation from
the ash cloud. This assumption may be tested by simulating BTDV for
various ash cloud heights and ash mass loadings with (BTDwmod) and
without (BTDw=0.0mod) water vapour. The BTDw=0.0mod then
resembles BTDVc in Eq. (), whilst BTDwmod resembles
BTDV. In view of Eq. (), BTDw-BTDw=0.0 should then be
constant.
VIIRS data used in this study. The study area is limited to the area delimited by
54–65∘ N, 12–26∘ W. VIIRS data were ordered from http://sips.ssec.wisc.edu.
Date
Time (UTC)
BTDmin
BTDmax
BTM15min
Total ash mass
(start of scan)
(K)
(K)
(K)
(Gg)
16/09/2013
02:06
0.0
0.8
272.0
17.78
16/09/2013
03:42
0.0
0.5
275.0
11.80
16/09/2013
05:24
0.0
0.8
270.0
17.05
16/09/2013
12:00
0.0
0.8
270.0
19.52
16/09/2013
13:36
-0.1
0.45
270.0
14.63
16/09/2013
15:18
-0.1
1.0
270.0
24.89
17/09/2013
03:24
-0.1
0.7
275.0
26.58
17/09/2013
05:06
-0.1
0.8
275.0
8.75
17/09/2013
11:42
0.3
1.0
275.0
13.76
17/09/2013
13:18
0.0
0.5
275.0
8.67
17/09/2013
15:00
0.0
1.0
275.0
8.05
Figure b shows the difference
BTDwmod-BTDw=0.0mod for various ash mass loadings as a function
of ash cloud top height. Above an ash cloud top altitude of ∼ 5.0 km the
difference becomes constant for all mass loadings. However, the magnitude of
the difference decreases with increasing mass loading. Below 5.0 km,
BTDwmod-BTDw=0.0mod becomes smaller than the constant value
above 5.0 km. The deviation from the constant value increases with increasing
ash cloud mass loading. Most of the water vapour is located in the lower
troposphere. For an ash cloud above 5.0 km the radiation emitted by the water
vapour must traverse the ash cloud similarly to the radiation emitted by the
Earth's surface. It will contribute to BTDV in an additive manner; cf.
Eq. (). For an ash cloud below 5.0 km some of the water vapour
will be above and some below the ash cloud. Radiation emitted by the water
vapour above the ash cloud does not interact with the ash cloud, hence
BTDwmod-BTDw=0.0mod decreases. For thick ash clouds the water
vapour below the ash cloud does not contribute to the signal at the top of
the atmosphere.
The 16–17 September 2013 resuspended ash cloud had a top height of about
1.0 km (Fig. ). As is evident from
Fig. and the discussion above, any water vapour correction
for an ash cloud at this altitude is not straightforward. Thus, no water
vapour correction was applied before ash pixel identification. Rather, a
customized ash detection scheme was applied; see the next section. For the ash
mass loading retrieval the absorption of water vapour was included in the
look-up-table (LUT) calculations using area-averaged ECMWF water vapour profiles;
see Sect. and Figs. S1–S4. It is noted that
the presence of ice may give a positive BTD see for
example. However, due to the ambient temperatures and the
origin of the resuspended ash we rule out the presence of ice for the case
studied here.
Ash pixel detection
Identification of ash pixels can normally be achieved by searching for pixels
with BTDV<Tlimit, where Tlimit is zero. However, this limit
assumes that the ash resides at high altitudes, such that the ash cloud
temperature is sufficiently different from the surface temperature
. The resuspended ash cloud during 16–17 September
2013 is easily identified in the RGB composites (Fig. ). By
comparing the RGB composites with the BTDV in Fig. S6, the resuspended ash
cloud can be clearly identified in both the daytime and night-time images.
However, due to the altitude of the resuspended ash cloud during this event,
BTDV>0.0 (see Sect. and Fig. S6) and the
normal threshold for identifying ash pixels cannot be applied. Instead
pixels are identified as containing ash if
(BTDV>BTDmin)∧(BTDV<BTDmax)∧(BT15>BT15min).
The values for BTDmin, BTDmax, and BT15min are manually
selected upon inspection for each scene and listed in
Table . The BTD of the pixels identified as containing
ash by this procedure is shown in Fig. . Through
visual inspection of both the daytime (Fig. ) and
night-time images (Figs. S5 and S6) areas considered to contain
ash are then defined by polygons, as shown in Figs. , S5,
and S6 in an attempt to remove the obviously wrongly
classified pixels. As discussed in the above section, the BTD signal
depends on the atmospheric water vapour content and the resuspended ash height,
and it requires cloud-free pixels. In addition the optical properties of the
underlying surface must be accounted for. The detection method has potential
for application in other cases but must be adapted to the situation being
studied.
VIIRS BTDs for pixels identified as resuspended ash, 16–17 September 2013.
Retrieval of ash properties and radiative transfer modelling
From the satellite measurements the ash mass loading may be retrieved.
Assuming spherical ash particles the mass loading, Ml (g m-2), is
given by
Ml=ρΔzc∫0∞43πr3n(r)dr,
where ρ is the density of the ash particles, Δzc is the ash
cloud thickness, and n(r) is the ash particle number density distribution.
Assuming a log-normal size distribution,
n(r)=N02π1ln(S)1rexp[-(lnr-lnr0)22ln2(S)],
where N0 is the total number of particles per unit volume, S is the
geometric standard deviation, and r0 is the geometric mean radius, the
mass loading simplifies to
Ml=ρΔzc43πN0re3exp(-62ln2S),
where re is the ash particle effective radius:
re=∫0∞πr3n(r)dr∫0∞πr2n(r)dr.
It is noted that for the log-normal size distribution r0 is related to
re by
re=r0exp(52ln2S).
It is common to assume values for S and ρ. For the case studied here,
Δzc is approximately known from temperature profiles and dispersion
model calculations. Thus we have
Ml=Ml(N0,re).
The VIIRS infrared measurements provide BTs. The
brightness temperature is a function of the state of the atmosphere and the
underlying surface. This relationship is described by the radiative transfer
equation. The state of the atmosphere is described by the temperature
profile, the density profiles of relevant trace gases (for example H2O),
liquid water and ice cloud particle densities, and ash cloud particle
densities. For infrared radiative transfer the temperature and emissivity of
the underlying surface is also needed. In addition knowledge about the
absorption and scattering across sections of the atmospheric constituents are
required. For example the ash cloud optical depth τa is given by
τa(λ)=Δzc∫Qext(λ,r)πr2n(r)dr,
where Qext(λ,r) is the ash cloud extinction efficiency as a
function of wavelength λ and radius r, and a vertically homogeneous
ash cloud is assumed.
If we adopt best-guess values for the parameters listed in
Table , the brightness temperature becomes a function
of N0 and re:
BT=BT(N0,re).
For the ash mass loading estimate we thus tabulate BTi as a function of
N0 and re for i= M15 and M16. The tabulated values are then used to
retrieve N0 and re from measured BTM15 and BTM16, and finally
the mass loading is calculated using Eq. ().
The retrieval of N0 and re is done using the Bayesian method described
by . The cost function
J(x)=(x-xb)TB-1(x-xb)+(yob-y(x))TR-1(yob-y(x))
is
minimized using the Levenberg–Marquardt method. Here x is the
atmospheric state vector consisting of the two elements (N0, re),
y(x) is the brightness temperature calculated by the forward
model for the atmospheric state x, and yob is the observed
brightness temperatures of VIIRS bands M15 and M16. The prior estimate xb is
set to N0=106 and re=1.0 µm. The background error
covariance matrix is assumed to be diagonal with elements
σN02=(1012)2 and σre2=( 10 µm)2. The
latter value is adopted from . The diagonal elements of
B are large, implying that the background state only provides a weak
constraint on the retrieved values. The error covariance matrix R is
also assumed to be diagonal. Its diagonal elements, σi2, are the
combined variance of the observational and forward-model variances. The
observational variances are σM152=(0.0028K)2 and
σM162=(0.0036K)2, and the forward-model variance is taken as
σFM2=(1.0K)2. This gives σi2=(1.0K)2.
Assumed parameters and values
used for the ash cloud retrieval. “ECMWF average” means the
parameter is calculated from ECMWF analysis data averaged
over the region for 16–17 September 2013. See text for more
details.
Parameter
Value
Comment
ρ
2300 (kg m-3)
Ash particle density
Tc
ECMWF average
Temperature of ash cloud top (Fig. S4)
Δzc
1000.0 m
Ash cloud thickness
n(r)
Log-normal
Particle number density distribution
S
2.0
Geometric standard deviation
Ts
ECMWF average
Surface temperature (Fig. S3)
ϵ
Seawater
Emissivity of surface
T(z)
ECMWF average
Temperature profile (Fig. S4)
Qext
Andesite
Ash type
ρH2O(z)
ECMWF average
Water vapour profile (Fig. S2)
The uvspec tool from the libRadtran radiative transfer package (; ;
www.libradtran.org) was used as the forward model
to calculate VIIRS brightness temperatures for bands M15 and M16. A
plane-parallel atmosphere was assumed, and the discrete-ordinate method was
used to solve the radiative transfer equation with 16 streams
. The ambient atmosphere profiles of
temperature, pressure, and water vapour were taken from the averaged ECMWF
profiles as described in Sect. (Figs. S1–S4). The surface was assumed to be seawater with wavelength emissivity
taken from http://www.icess.ucsb.edu/modis/EMIS/html/seawater.html. For
the gas absorption the REPTRAN parameterization was used
. The resuspended ash was included as a plane-parallel
layer. The ash particles were taken to be of andesite composition, and the
refractive index was adopted from . The ash particles were
assumed to be spherical in shape, and their optical properties were calculated
using Mie theory. It is noted that porosity and non-sphericity of the ash
particles may affect the electromagnetic IR radiation measured by VIIRS
. The uvspec model is computationally too slow to be used
online in the retrieval; therefore LUTs were calculated as a
function of N0 and re for surface temperatures between 280 and 284 K.
Figure shows the retrieved ash mass loading of the
resuspended ash cloud for the areas identified as containing ash, and
Table gives the retrieved mass of ash in the atmosphere
for each overpass. The location of the ash cloud agrees well with the
forecasts using NAME (cf. Fig. ). Quantifying the
uncertainty on satellite retrievals of volcanic ash is non-trivial, and it
includes uncertainties in the retrieval and uncertainties in the assumed
parameters, such as the refractive index and particle size distribution
. Based on the work by and in
addition considering the uncertainty due to particle shape
, we assign an uncertainty of ±50 % to the total mass
retrieved for each image.
Quantifying the total mass of ash resuspended
Here we determine the scaling coefficient (K) for the emission rate (F,
Eq. ) in the resuspension scheme in NAME. As we have data from
only one OPC instrument, we are unable to perform a robust calibration with
surface PM10 data. Instead we perform a calibration using the total
column mass loadings of the remobilized ash cloud retrieved from VIIRS.
Calculated fractional bias between VIIRS-retrieved total column mass loadings and modelled
total column mass loadings where the emission rate in NAME is calibrated using the scaling coefficient (K)
derived from a peak-to-peak scaling to the VIIRS data.
Time + Date
K=1×103
K=1×104
02:06 16/09/2013
0.89
1.85
03:42 16/09/2013
1.14
1.89
05:24 16/09/2013
0.22
1.71
12:00 16/09/2013
0.40
1.75
13:36 16/09/2013
0.51
1.78
15:18 16/09/2013
0.70
1.82
03:24 17/09/2013
-0.61
1.37
05:06 17/09/2013
0.82
1.84
11:42 17/09/2013
0.40
1.75
13:18 17/09/2013
0.77
1.83
15:00 17/09/2013
1.16
1.90
Figure shows the frequency of binned total column mass
loadings from the satellite retrievals and the NAME modelled mass loadings
where K is set to unity (1 g s-1). The mode of the VIIRS mass loadings
varies with time during the event, from 10-1–100 to
100–101 g m-2; this variation includes the uncertainty
associated with the retrieval. The un-calibrated modelled total column mass
loadings have a mode at 10-4–10-3 g m-2. Considering the
difference in the mode of the VIIRS-retrieved mass loadings and the model output
at each retrieval time, this suggests that we need to apply a scaling
of between K=1×103 and 1×104 to the emission rate in the
resuspension scheme in NAME to match the observed mass loadings in the
atmosphere.
The resuspended ash mass loading retrieved from VIIRS infrared bands M15 and
M16 for the areas identifed as containing ash.
Comparing the frequency of binned total column mass loadings of the resuspended ash cloud modelled
using NAME with an un-calibrated source strength to those retrieved from VIIRS during 16–17 September 2013.
Simulated mass loadings using these calibration factors are given in
Fig. . The performance of the calibration factors are
assessed by calculating the fractional bias between the satellite-retrieved
and the modelled total column mass loadings within the polygons
(Table ). The fractional bias is a measure of the mean
bias and indicates over- or underestimation of the model output: values range
between -2 and +2, a positive value represents over-prediction of the model
with respect to the VIIRS-retrieved mass loadings, a negative value
represents under-prediction; and a value of 0 represents a perfect match. Scaling the source
strength by K=1×104 systematically overestimates mass loadings,
whereas using K=1×103 results in a better match to the
satellite retrievals. This is still the case when we consider that the
retrieved mass loadings have an uncertainty of ±50 %.
Summing the mass loadings from each VIIRS retrieval
(Table ) gives a total observed mass of remobilized ash
of 0.17 Tg. This represents contributions only from the mass in the
atmosphere at the time of each overpass and may double-count between
retrievals. Using the modelled emission rates, scaled by K=1×103, the total mass of ash remobilized from the Eyjafjallajökull 2010
and Grímsvötn 2011 deposits between 00:00 UTC on 16 September
2013 and 00:00 UTC on 18 September 2013 is ∼ 0.2 Tg.
Discussion
The total mass of ash erupted from Eyjafjallajökull in 2010 was estimated
from ground surveys and remote sensing to be 384 ± 96 Tg
. Calculated estimates using plume rise models are
also found to lie within the error bounds of this observational estimate
. Preliminary results from mapping the Grímsvötn
2011 fall deposits indicate that the bulk volume of ash from this eruption is
2–3 times larger . We estimate that
∼ 0.2 Tg of ash was remobilized during 16–17 September 2013.
Modelled 1 h averaged total column mass loadings (g m-2), where the source strength in NAME
is calibrated using the scaling coefficient determined from the peak-to-peak scaling to the satellite-retrieved
total column values (a) K=1×103 and (b) K=1×104. The locations of the volcanoes
Eyjafjallajökull and Grímsvötn are indicated on the map by the E and G symbols respectively.
The calibration applied in this study is uniquely related to the event
studied and the source areas defined, but this approach can be used to
consider how the emission rate of resuspension has varied with time since the
ash was deposited. The calibrated emission flux of K=1 × 103
is lower than the original calibration determined by
, K=1.1 × 107 (taking an emission
flux in grams), for the Eyjafjallajökull ash source in 2010. This suggests
that resuspension rates had declined by 2013, perhaps due to depletion and
compaction of the ash with time since it was deposited and/or re-growth of
vegetation. However, the retrieved mass loadings from VIIRS and the
calibrated modelled mass loadings show that the resuspended ash cloud still
contained significant quantities of ash. Dividing the calculated total mass
of ash resuspended over the emission time period (48 h), we calculate an
average emission rate of 1.04 × 103 kg s-1. This is
equivalent to the minimum calculated eruption rates of tephra from
Eyjafjallajökull 2010 using plume rise models, which range between
103 and 106 kg s-1 over the 39-day eruption
. The magnitude of the retrieved ash
mass loading in individual scenes from the VIIRS data is also comparable in
magnitude to those determined by using SEVIRI of the distal
ash cloud from the eruption of Eyjafjallajökull in 2010 over the southern
North Sea on 17 May 2010. This suggests that remobilization of ash
deposits can produce ash clouds with mass loadings equivalent to those
observed from explosive volcanic eruptions. One important distinction is that
the buoyant ash plume generated from the eruption of Eyjafjallajökull
released ash to altitudes up to 10 km a.s.l., and the resulting ash cloud was
consequently transported by upper-air winds. In contrast, resuspended ash,
remobilized from deposits, is necessarily closer to the surface, and during
16–17 September 2013 the ash was trapped below a temperature inversion
at < 2 km a.s.l., restricting further vertical dispersion. Ash sedimenting from
a low-altitude resuspended ash cloud will be deposited more quickly than ash which
is released at upper levels, as it does not have as far to fall and because
it will be rained out by precipitation from clouds formed above the ash
layer.
measured the PSD of resuspended ash deposited in
Reykjavik during 6–7 March 2013 following a significant remobilization
event of the Eyjafjallajökull 2010 and Grímsvötn 2011 deposits.
Most of the mass was contained within the 32–63 µm size fraction, and < 10 % of the total mass was
on particles with diameter < 10 µm. Here
we have considered particles with diameter ≤ 10 µm only, to be
consistent with the particle size range the satellite retrievals are most
sensitive to. No observations of the PSD of the remobilized ash cloud were
made during 16–17 September 2013. Taking the PSD from
suggests that our calculated remobilized mass of
0.2 Tg for this event may represent a fraction of the total mass actually
resuspended.
We have used the extent of tephra deposits defined immediately after the
eruptions of Eyjafjallajökull in 2010 and Grímsvötn in 2011 to
identify the potential source area from which ash can be resuspended. This
does not consider how the deposits may have been modified since they were
formed. Compaction and cementation processes increase deposit cohesion and
can reduce the emission flux of particles. Here, we have applied the same
scaling coefficient to both the Eyjafjallajökull and Grímsvötn
deposits, which could underestimate the flux from the younger
Grímsvötn deposits and overestimate the flux of particles from the
older Eyjafjallajökull deposits . Deposits are
also re-distributed as ash is resuspended, advected, and re-deposited.
Jökulhlaups (sub-glacial floods) can also transport large volumes of ash
which is then re-deposited on outwash planes (sandurs). The sandur planes
represent large areas of unstable sediments and are known to be an
additional source of remobilized particles across Iceland
.
calculated the total emission from a remobilized “dust
storm” on 25 May 2012 in Dyngjusandur, a large glacio-fluvial plain north
of Vatnajökull, to be 3.65 × 105 t (∼ 0.3 Tg). The
calculated emission is based on measurements of the horizontal extent of the
plume and visibility (weather) observations, which were validated with MODIS
satellite imagery. More recently
estimated the total mass of dust resuspended during two storms in south-west
Iceland, on 15 June 2015 and 4 August 2015, from observations of the
horizontal extent of the plume and visibility measurements to be ∼ 0.18 and ∼ 0.28 Tg respectively. These masses are comparable to the value
calculated here. The VIIRS satellite imagery of the resuspended ash cloud
during 16–17 September 2013 clearly indicates that the source of the
remobilized ash cloud is over southern Iceland, and the two distinct plumes
observed in the visible imagery (Fig. ) suggest that both
the Eyjafjallajökull and Grímsvötn deposits are the primary
sources of the remobilized ash. The good agreement between the modelled and
observed location and timing of the resuspended ash cloud gives us confidence
that our source areas are well defined (Fig. ).
However, the sandur planes on the south coast at Mýrdalssandur and
Skeiðarársandur may also have been an additional source of ash which
has not been accounted for. It is not yet understood whether the mechanism of
resuspension, and hence the rate at which particles are remobilized, from the
sandur planes differs to that from the tephra deposits. Applying the same
calibration coefficient to a larger source area, to include the sandur
plains, would increase the total modelled emission flux.