SEARCH

SEARCH BY CITATION

Keywords:

  • Eyjafjallajökull;
  • RTTOV;
  • SEVIRI;
  • satellite;
  • simulated;
  • volcanic ash

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[1] During volcanic eruptions that eject ash into the atmosphere Volcanic Ash Advisory Centers issue statements on the forecast dispersion of the ash so that the aviation industry can manage airspace to avoid aircraft encountering volcanic ash. Observations, such as those from satellites, are compared with the forecasts from an atmospheric dispersion model to assess the quality of the ash forecasts. A method has been developed to enable like-with-like comparison between satellite imagery of volcanic ash and simulated imagery using the forecast ash concentration data from an atmospheric dispersion model. The ash concentration and numerical weather prediction data are used as inputs to a radiative transfer model to simulate radiances. Simulated satellite images are created from these simulated radiances. Here, Spinning Enhanced Visible and Infrared Imager volcanic ash images based on infrared brightness temperatures for the Eyjafjallajökull eruption in 2010 are simulated. In addition to providing a useful tool for forecasters in a Volcanic Ash Advisory Center, the simulated images can be used to aid the understanding of how the ash affects the satellite imagery and also the physical properties of the ash.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[2] Forecasters operating the London Volcanic Ash Advisory Center (VAAC) service obtain their information from observations (e.g., local reports, pilot reports, and radar and satellite imagery) and an atmospheric dispersion model that forecasts the movement of ash. The forecast ash maps (from an earlier model run) are compared with observations for the same time to assess the performance of the dispersion model and, in particular, the accuracy of the inputs to the dispersion model. This is a nontrivial task because the observations and the ash forecasts are not usually directly comparable (e.g., radiances and ash concentration), and are frequently in different map projections.

[3] The method developed here uses the ash concentration data from the Numerical Atmospheric dispersion Modeling Environment (NAME) [Jones et al., 2007], as an input to a fast radiative transfer model, RTTOV [Matricardi, 2005; Saunders et al., 1999; J. Hocking et al., RTTOV v10 Users Guide, unpublished report, http://research.metoffice.gov.uk/research/interproj/nwpsaf/rtm/rtm_rttov10.html]. RTTOV is used to forward model the SEVIRI radiances and these are used to produce simulated satellite imagery that can be directly compared with the satellite imagery derived from measured SEVIRI radiances. This enables forecasters to compare like-with-like imagery and to make a rapid assessment on the quality of the dispersion model ash forecast.

[4] The eruption of Eyjafjallajökull, Iceland, in April and May 2010 has been the focus of this study due to the large impact that the eruption had on aviation. The London VAAC, run by the Met Office (UK), has the responsibility for issuing Volcanic Ash Advisory Statements for airborne ash in an area of the northeast Atlantic, and so issued many advisories as a result of this eruption. The main explosive eruption started on 14 April 2010 and continued throughout April and May with the last advisory statement being issued on 24 May 2010. During this time, the NAME model was run routinely to produce ash forecasts and forecasters used satellite imagery, among other observations, to assess the extent of ash dispersal. The prevailing wind during much of the eruption was from the northwest resulting in ash being transported toward mainland Europe causing closures of busy airspace and grounding of many aircraft at airports.

2. SEVIRI Volcanic Ash Imagery

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[5] The Spinning Enhanced Visible and Infrared Imager (SEVIRI) is an instrument with 12 visible and infrared channels on board the geosynchronous Meteosat Second Generation (MSG) satellites. Data from SEVIRI on Meteosat-9 operating at 0° longitude are received at the Met Office every 15 min and used in the satellite data processing system to produce various image products for forecasters.

[6] A volcanic ash detection product using the “reverse absorption” technique is derived using the difference between equivalent blackbody brightness temperatures (BTs) in channels centered at wavelengths of 10.8 μm and 12.0 μm (BT10.8 – BT12.0). This follows the principles described by Prata [1989] and developments using Advanced Very High Resolution Radiometer (AVHRR) data at the Met Office [Watkin, 2003]. Volcanic ash is more absorbing at 10.8 μm than at 12.0 μm while water and ice are more absorbing at 12.0 μm than at 10.8 μm. Thus, a difference in brightness temperatures between two channels centered at these two wavelengths can be used to discriminate ash from water or ice clouds, with ash producing a negative signal and water/ice producing a positive signal (e.g., Figure 1a). The strength of the resulting ash signal is dependent on the optical depth of the ash cloud as well as the physical properties of the ash particles (shape, size distribution, and refractive indices). Other factors affecting the signal include the thermal contrast between the ash cloud top and underlying surface, the presence of other absorbers (water and ice result in a positive bias) and the satellite viewing angle (with increased path length through the ash cloud leading to stronger negative signals [Gu et al., 2005]). An important point to note is that a stronger ash signal in BT10.8 – BT12.0 imagery does not necessary equate to a greater concentration of ash, but may indicate the ash particles are smaller (among other factors such as height of ash and water vapor content); this will be illustrated in section 7. Also, negative signals can occur in the absence of ash particles, e.g., due to low-level inversions and desert dust. For further discussion on the known limitations of this detection method see Prata et al. [2001] and Pavolonis et al. [2006].

image

Figure 1. Measured and simulated (a–d) SEVIRI BT10.8 – BT12.0 and (e–h) dust RGB for 6 May 2010 at 1200 UTC. Images derived from the measured radiances (Figures 1a and 1e), simulated images with no cloud and no ash (Figures 1b and 1f), simulated images with cloud and no ash (Figures 1c and 1g), and simulated images with both cloud and ash (andesite, NAME size distribution, and NAME ash concentration) (Figures 1d and 1h).

Download figure to PowerPoint

[7] The reverse absorption method has been further developed to incorporate information from additional channels [e.g., Pavolonis et al., 2006; Francis et al., 2012] to increase the probability of detecting ash clouds and reduce spurious signals in the absence of ash. While the use of additional data from infrared channels on SEVIRI has been developed at the Met Office for use in the London VAAC [Francis et al., 2012], it was decided that for simplicity and ease of understanding the simple reverse absorption (BT10.8 – BT12.0) imagery would be studied here. In the future, other simulated volcanic ash imagery could be produced.

[8] Red-green-blue (RGB) images combining information from different infrared channels are also produced. During the Eyjafjallajökull eruption in 2010 it was found that animations of the so-called dust RGB were useful in tracking volcanic ash over a number of days. The dust RGB is produced by assigning BT12.0 – BT10.8 values to the red channel, BT10.8 – BT8.7 values to the green channel and BT10.8 values to the blue channel. Volcanic ash tends to produce an orange or pink color (e.g., Figure 1e). The dust RGB images are best presented as an animation to enable forecasters to use pattern recognition to track the ash signal from source. As with the BT10.8 – BT12.0 imagery, the signal is highly variable and so careful interpretation is required. The factors affecting the BT10.8 – BT12.0 signal also affect the dust RGB imagery. In addition, sulfur dioxide (SO2) absorbs at 8–9 μm affecting the 8.7 μm brightness temperature when the SO2 cloud is above the bulk of tropospheric water vapor. As SO2 in the upper troposphere or lower stratosphere increases, BT10.8 – BT8.7 values increase so increasing the green content of the dust RGB image and this could result in a volcanic cloud that contains ash and SO2 appearing more yellow/orange than pink. Thomas and Prata [2011] discuss the detection and collocation of ash and SO2 in the Eyjafjallajökull volcanic cloud.

[9] The work described below aims to simulate the BT10.8 – BT12.0 and dust RGB imagery. In addition to providing a tool for forecasters to use to assess the performance of NAME by comparing the simulated and measured imagery, the simulations provide a means to understand the factors affecting both types of images.

3. Physical Properties of Volcanic Ash

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[10] To simulate the effect of volcanic ash on SEVIRI infrared BTs, first the physical properties of the ash must be estimated so that the absorption and scattering properties can be computed using Mie theory. Mie theory assumes that the ash particles are spherical. Volcanic ash particles are generally quite angular in shape [Riley et al., 2003] and therefore the assumption of spherical particles is probably not accurate. However, with the lack of quantitative information on ash particle shape and the added complexity of the scattering theory it was decided that as a first approximation Mie theory would be adequate.

[11] The size distribution used in this study is that used at the volcanic source in the atmospheric dispersion model, NAME. NAME uses a size distribution described in terms of cumulative fraction of mass as shown in Table 1. It is based on work by Hobbs et al. [1991] who measured ash particle sizes from the eruptions of Mount Redoubt (1990), Mount St Helens (1980) and St Augustine (1976). The representativeness of the particle size distribution used in the NAME model for the eruption of Eyjafjallajökull is the subject of discussion and it is one of the factors assessed in the sensitivity study later in this paper. B. Johnson et al. (In situ observations of volcanic ash clouds from the FAAM aircraft during the eruption of Eyjafjallajökull in 2010, submitted to Journal of Geophysical Research, 2011) show measurements of the aerosol size distribution observed during the UK Facility for Airborne Atmospheric Measurements (FAAM) flights. The measured distributions show the aerosol mass to be dominated by the coarse mode with a peak at around 4 μm (equivalent sphere diameter) on all FAAM flights and a rapid falloff in mass of particles >10 μm. For simplicity the distribution used at the source in NAME is assumed (with the peak of the aerosol mass being at around 20 μm diameter). This distribution's particles are too large for this eruption, either because the source size distribution is wrong or because evolution is neglected here; rather than attempting a more precise estimate of the evolved size distribution some sensitivity tests are performed on the size distribution (section 7.2).

Table 1. Numerical Atmospheric Dispersion Modeling Environment (NAME) Size Distribution of Ash Particles in Terms of Cumulative Fraction of Mass
Diameter (μm)Cumulative Fraction of Mass
0.10.0
0.30.001
1.00.006
3.00.056
10.00.256
30.00.956
100.01.0

[12] The size distribution in terms of fraction of mass in each size bin (Figure 2a) was converted to number density of particles for use in the Mie calculations. The density of the solid ash particles is assumed to be 2300 kg/m3 (as used in NAME; this falls within observed values for erupted material [e.g., Sparks et al., 1997]). Again, this is another variable that is unknown. The conversion to number density is performed by assuming that within each size bin the logarithm of particle diameter is uniformly distributed in terms of mass. The number density of particles in an infinitesimal size range [d, d + Δd] is calculated for a concentration of 1 kg/m3 using the mass fraction in the relevant bin, the bin width, particle diameter d and ash density. The number density per micron diameter is then calculated by dividing by Δd in microns. For plotting in Figure 2b and for use in the calculation of optical properties (section 4), the number density is normalized to a total number concentration of 1 cm−3 by dividing by the total number density.

image

Figure 2. NAME size distribution (solid line) in terms of (a) fraction of mass in each size bin and (b) number density normalized to 1 cm−3. Inset shows the fraction of mass for diameters in the range 0.1–1.0 μm. Dashed and dotted lines show alternative size distributions created by multiplying the particle diameter by a factor (0.25, 0.5, 2, or 4). The alternative size distributions have the same total mass as the original NAME size distribution (solid line).

Download figure to PowerPoint

[13] The exact composition, and therefore the refractive indices, of the ash from Eyjafjallajökull was not known at the time of the eruption. This will be the case with all future eruptions from Iceland (or elsewhere) and so it is important to gain an understanding of the sensitivity of the simulated imagery to the choice of refractive indices. With this in mind a series of simulations using each set of refractive indices in turn was performed and the results compared with the measured brightness temperatures.

[14] Refractive indices for volcanic dust [Volz, 1973] (and tabulated in WCRP [1986]), andesite ash [Pollack et al., 1973], Oregon obsidian (rhyolite) ash [Pollack et al., 1973] and desert dust with 0.9% hematite [Balkanski et al., 2007] are considered in this work (imaginary refractive indices are shown in Figure 3a). The refractive indices for volcanic dust [Volz, 1973] were used in the original calculation of the volcanic ash aerosol coefficients file for RTTOV (Radiative Transfer for TOVS) (M. Matricardi, personal communication, 2010) and so are included in this study. Andesite refractive indices [Pollack et al., 1973] are commonly used in infrared retrievals of ash characteristics [e.g., Gangale et al., 2010; Pavolonis et al., 2006; Prata and Grant, 2001].

image

Figure 3. (a) Imaginary refractive indices for andesite ash, Oregon obsidian ash, volcanic dust, and desert dust with 0.9% hematite. (b) Mass absorption coefficient of volcanic ash calculated using Mie theory with a NAME size distribution and the four sets of refractive indices shown in Figure 3a. Vertical dotted lines mark the central wavelengths of the SEVIRI channels.

Download figure to PowerPoint

[15] The four sets of imaginary refractive indices all exhibit the important “reverse absorption” feature [Prata, 1989] between the 10.8 μm and 12.0 μm wavelengths commonly used for volcanic ash detection with higher imaginary refractive indices at 10.8 μm than at 12.0 μm (Figure 3a). However, the magnitude of the refractive indices varies greatly so affecting the strength of the BT10.8 – BT12.0 signal.

[16] Studies of the ash from the Eyjafjallajökull eruption have shown that there were two distinct types of ash ejected. During the flank eruption (20 March 2010–12 April 2010) at Fimmvörðuháls the ejected material was alkali-olivine basalt (∼47.7% silicate oxide (SiO2)). Whereas during the main summit eruption (14 April 2010 onward) the ejected material was trachyandesite (56.7–59.6% SiO2) (N. Óskarsson, unpublished data, 2010, available from the Institute of Earth Sciences, Nordic Volcanological Center, http://www.earthice.hi.is/page/IES-EY-CEMCOM). This would indicate that refractive indices for andesite may be the most appropriate to use for the main eruption studied here.

4. Optical Properties of Volcanic Ash

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[17] The Edwards-Slingo radiation code [Edwards and Slingo, 1996] was used to compute the aerosol optical properties. The Edwards-Slingo code uses Mie theory and has great flexibility allowing the spectrally dependent information (number of bands, central wavelengths, absorbing species, size distribution and refractive indices) to be read in from files and used in the scatter preprocessing code to create an output file containing mass absorption, mass scattering coefficients and asymmetry parameter at each central wavelength specified.

[18] The aerosol optical properties are written to a RTTOV coefficients file in the form of absorption and scattering cross sections and backscatter coefficient at each SEVIRI infrared central wavelength [Matricardi, 2005]. The absorption and scattering cross sections are calculated from the outputs from the Edwards-Slingo code by dividing the mass absorption or scattering coefficient (in m−1) by the number density (in m−3). The backscatter coefficient (b) is the integral of the backward facing hemispheric values of the phase function; it is approximated using the asymmetry parameter (g) by Wiscombe and Grams [1976]:

  • display math

[19] Marshall et al. [1995] suggest a more complex relationship between b and g that depends on the imaginary part of the refractive index and the width of the aerosol size distribution. However, the simple relationship in equation (1) is used in this paper.

[20] The mass absorption coefficients derived using the NAME size distribution and the four sets of refractive indices via the Edwards-Slingo scatter code are shown in Figure 3b. The mass absorption coefficients derived from all four sets of refractive indices show the same general trend of decreasing between 10.8 and 12.0 μm; thus indicating that the brightness temperature between channels centered at these wavelengths (BT10.8 – BT12.0) would be negative. However, the gradient of the decrease in mass absorption coefficients varies between the different sets of refractive indices. Andesite and desert dust have the largest gradients implying that a stronger BT10.8 – BT12.0 signal could be expected with these refractive indices than with those for obsidian or volcanic dust.

5. NAME Ash Concentration

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[21] The Numerical Atmospheric dispersion Modeling Environment (NAME) is the Met Office's atmospheric dispersion model [Jones et al., 2007]. It is a Lagrangian particle model that calculates the dispersion of material by tracking model particles through the modeled atmosphere. The source strength is calculated according to the empirical relationship between the height of the eruption column above the volcano summit and the mass emission rate [Mastin et al., 2009]. The NAME data used in this study was output as a 6 h average air concentration of ash for a layer in the atmosphere defined by flight levels on a latitude-longitude grid. A flight level (FL) is a standard nominal altitude of an aircraft, in hundreds of feet, calculated from the standard pressure datum of 1013.25 hPa (it is not necessarily the same as the aircraft's true altitude either above mean sea level or above ground level). The NAME output vertical levels range from FL000 to FL550 at FL025 vertical resolution. The near source fallout, due both to large grain sizes and aggregation of small grains, is assumed to be 95%, so that only 5% of the mass is assumed to be represented in the size range of the NAME size distribution (0.1 – 100 μm). Estimates of the distal fine ash fraction (i.e., the dispersed ash cloud away from the source) lie in the range between about 0.05% and 10% as found in various case studies, for example, during the recent Eyjafjallajökull eruption [Dacre et al., 2011] and for some previous eruptions [Rose and Mayberry, 2000]. NAME III version 6.0 was used to generate the data in this study. For a description of the set up of NAME during and since the Eyjafjallajökull eruption (see Webster et al. [2012]).

[22] RTTOV requires the number density of aerosols as input, so it is necessary to convert the ash concentration from NAME to number density using the size distribution information. The number density is then interpolated on to the Numerical Weather Prediction (NWP) model grid and assigned to model vertical levels based on a conversion of the flight levels to pressure levels using a standard atmosphere.

[23] The performance of NAME is affected by the quality of the model itself, the NWP data used in NAME and the accuracy of the source term. The source term includes the volcano's location; source geometry; eruption date, time and duration; upper height of the eruption plume (this determines the emission rate), vertical ash distribution and the particle size distribution. Much of this information is often limited and has large margins of error. It usually comes from reports from pilots, radar or local observers, but can be unknown. During the eruption of Eyjafjallajökull the best information available at the time was used to define the source term. After the event NAME was rerun using redefined source terms at certain times, including:

1. Changes in the eruption plume height. Following the end of the eruption detailed information on the observed plume height from radar [Arason et al., 2011], from pilot reports and from Icelandic coastguard observations was gathered and studied. A revised plume height time profile was produced by Webster at al. [2011] and is shown by the solid line in Figure 4.

image

Figure 4. The plume height time profile (above sea level) determined during the event (dashed line) and determined postevent (solid line).

Download figure to PowerPoint

2. Using a different relationship between plume height and mass of ash, as given by Mastin et al. [2009].

3. Using a vertical resolution of 25 flight levels (2,500 feet).

4. Using analysis NWP data (interspersed with short-term forecast NWP data) as opposed to all forecast NWP data.

[24] The rerun data (hereafter called “revised height run”) are used in this study, unless otherwise stated, since they are thought to provide the best available forecast of the dispersion of the ash.

6. Simulation of SEVIRI Imagery

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[25] The fast radiative transfer model RTTOV is used to forward model the SEVIRI radiances using NWP and NAME data. RTTOV is a very fast radiative transfer model for nadir viewing passive infrared and microwave satellite radiometers, spectrometers and interferometers. RTTOV-10 was used in this study, but RTTOV-9 onward includes the facility to simulate cloudy and aerosol multiscattered radiances. The details of the RTTOV aerosol scattering and absorption scheme are given in the RTTOV-9 science and validation plan (NWP SAF website, http://research.metoffice.gov.uk/research/interproj/nwpsaf/rtm/) and also by Matricardi [2005] which describes the scheme implemented in a precursor version of RTTOV. The aerosol multiple scattering is parameterized by scaling the layer optical depth by a factor derived by including the backward scattering in the emission of a layer and in the transmission between levels (a scaling approximation). The scaling approximation introduces small errors (typically <0.2 K) into the simulated radiances but preserves the spectral variation between the channels [Matricardi, 2005], thus introducing very little error into the BT10.8 – BT12.0 values. RTTOV includes by default eleven aerosol components from the database of optical properties generated using the Lorentz-Mie theory assuming these particles have a spherical shape and one is specifically for volcanic aerosols. However, for this work the RTTOV scattering coefficients were updated to include more representative optical properties for the volcanic aerosol type as described in section 4.

[26] NWP analyses from the Met Office's Global Model version of the Unified Model [Cullen, 1993] are used as input to RTTOV in this study. These data include profiles of temperature, humidity and pressure and model cloud fields. In addition the ash number density in units of cm−3 are assigned to the aerosol variable and used as an input to the RTTOV state vector. Switches to control the output of RTTOV result in simulated radiances, converted to brightness temperatures, of (1) clear sky, (2) clear sky and clouds, (3) clear sky and ash, or (4) clear sky, clouds, and ash. Ultimately, it is the cloud plus aerosol affected brightness temperatures that are required. However, it is interesting to study the other situations in order to understand how the individual components to the satellite observed radiance affect the signal.

[27] The simulated brightness temperatures from RTTOV allow the calculation of the brightness temperature difference (BTD) between the 10.8 and 12.0 μm channels. Images of the BTD are generated using the same software that produces the measured BT10.8 – BT12.0 imagery. The dust RGB image is produced by combining the simulated brightness temperatures of the 8.7, 10.8 and 12.0 μm channels, again using the same software that produces the measured dust RGB image. Figure 1 shows an example of how the cloud and aerosol components affect the clear sky simulated imagery. The clear sky simulation (Figures 1b and 1f), without cloud or aerosol, shows how the surface features and humidity and temperature variations in the atmosphere affect the BT10.8 –BT12.0 and dust RGB images. Add in cloud (Figures 1c and 1g), and more structure can be seen: dark red in the RGB due to thick, ice clouds causing positive BTDs in the BT10.8 – BT12.0 image, and green/brown/orange in the dust RGB due to lower-level water clouds resulting in some negative BTDs in the BT10.8 – BT12.0 image. Add in the ash aerosol (Figures 1d and 1h) and the ash cloud (in this case) is instantly obvious due to the yellow to red colors in the BT10.8 – BT12.0 and pink/orange color in the dust RGB.

7. Sensitivity of Imagery to Ash Properties

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[28] In order to gain some understanding of the effect of the different ash properties on the simulated signal a series of simulations was carried out using twelve cases, see Table 2. The cases were selected so that there was a reasonable spread of dates, but mainly on the basis that there was a significant amount of ash to study. The imagery was simulated for these twelve cases using each of the four sets of refractive indices, a variety of size distributions (all based on the NAME size distribution) and different ash concentrations (all based on the NAME ash concentrations). The resulting simulated images were visually compared with the measured images to assess the optimum set of ash properties. The images were also automatically compared with each other to assess the optimum set of ash properties (Figure 5). This was done by:

Table 2. Optimum Set (Based on Median Values) of Volcanic Ash Properties of Those Studied (Refractive Indices, Size Distribution and Ash Concentration) for Simulated BT10.8 – BT12.0 Images Compared With Measured BT10.8 – BT12.0 Images for Twelve Casesa
CaseVariable Particle PropertiesbFixed Particle Propertiesc
DateTime (UTC)Concentration FactorRefractive IndicesSize Distribution FactorConcentration Factor
  • a

    The size distribution factor is multiplied by the particle diameter in the NAME size distribution to obtain an adjusted NAME size distribution. The concentration factor is multiplied by the NAME ash concentration to obtain a revised ash concentration.

  • b

    Allowing ash concentration, refractive indies, and size distribution to vary.

  • c

    Only allowing ash concentration to vary (andesite refractive indices and size distribution factor of 0.25).

15 April 201018002.0Andesite0.252.0
17 April 201012001.0Andesite0.251.0
6 May 201012001.0Volcanic Dust0.50.5
7 May 201012001.0Obsidian0.50.5
8 May 201012002.0Obsidian0.251.0
9 May 201012001.0Andesite0.251.0
11 May 201012002.0Andesite0.252.0
13 May 201012002.0Andesite0.252.0
14 May 201012002.0Andesite0.252.0
15 May 201012002.0Andesite0.252.0
16 May 201012002.0Andesite0.252.0
17 May 201012002.0Andesite0.252.0
image

Figure 5. Graphs of the median simulated brightness temperature difference (BTD) – median satellite BTD in the area of the ash plume (as described in the text) for twelve cases (all at 1200 UTC except 1800 UTC on 15 April 2010). (a–c) Results for BT10.8 – BT8.7 and (d–f) results for BT10.8 – BT12.0. The simulations were performed using different sets of ash properties: andesite refractive indices, NAME ash concentrations and size distributions derived by multiplying the particle diameter of the NAME size distribution by the factor shown on x axis (Figures 5a and 5d); NAME ash concentrations, NAME size distribution and different refractive indices (AN, andesite; OB, obsidian; VD, volcanic dust; DD, desert dust) (Figures 5b and 5e); NAME size distribution, andesite refractive indices and NAME ash concentration multiplied by the factor shown on the x axis (Figures 5c and 5f).

Download figure to PowerPoint

1. Isolating the area of interest (i.e., the area that (1) contained forecast ash from NAME, (2) had negative BT10.8 – BT12.0 values in the measured image, and (3) had differences between measured and simulated clear sky BT10.8 – BT12.0 values of less than −0.2 K). This resulted in pixels 0 – 3700 km from Eyjafjallajökull being selected over the twelve cases, with an average distance of 970 km.

2. Computing the difference between the median simulated BT10.8 – BT12.0 and median satellite BT10.8 – BT12.0 in the area of interest. This was repeated for the BT10.8 – BT8.7, since this difference is used in the dust RGB.

3. Plotting the differences in the median for each case and each set of ash properties (Figure 5).

4. Finding the set of ash properties for which the absolute difference in the median values is a minimum; these are listed in Table 2.

[29] The BT10.8 – BT8.7 values used in the dust RGB images and displayed in Figures 5a–5c are, in general, too low in the simulated imagery due to too little absorption at 8.7 μm (simulated dust RGB plumes too pink). The RTTOV simulations do not take account of absorption by sulfur dioxide (SO2), which is a strong absorber at this wavelength. Indeed, the 8.7 μm channel is frequently used to detect volcanic SO2 [Watson et al., 2004] and so any judgments on the most appropriate set of refractive indices should not be based on information using the 8.7 μm channel. The dust RGB images are used in this paper, and operationally at the Met Office, because the satellite derived dust RGB images have been found to be so useful in the tracking of volcanic ash and thus the simulated dust RGB images are a useful tool to monitor the forecast position of the plume, but perhaps less so the physical properties of the ash. In sections 7.17.3, we focus on the BT10.8 – BT12.0 results, illustrated with results for 7 May 2010.

7.1. Refractive Indices

[30] As described earlier, refractive indices for volcanic dust [Volz, 1973], andesite ash [Pollack et al., 1973], obsidian (rhyolite) ash [Pollack et al., 1973] and desert dust with 0.9% hematite [Balkanski et al., 2007] were considered here. Figure 6 shows examples of simulated BT10.8 – BT12.0 and dust RGB images for 7 May 2010 using each of the four sets of refractive indices. Andesite and desert dust result in stronger negative signals in the BT10.8 – BT12.0 images than obsidian or volcanic dust, as would be expected from the steeper gradient between the 10.8 and 12.0 μm channels in the mass absorption graph (Figure 3b). In this case the stronger negative signals from the andesite and desert dust refractive indices give a closer match to the strong negative signal observed. The color of the ash plume in the dust RGB varies considerably with different refractive indices, mainly due to the differences at 8.7 μm. The andesite and volcanic dust result in pink plumes, while the obsidian and desert dust result in plumes that are orange due to greater absorption at 8.7 μm. The ash plume in the satellite dust RGB is pink near the volcano becoming orange further away. Overall, visual examination results in the conclusion that the andesite or desert dust refractive indices produce reasonable simulations for this case.

image

Figure 6. (a, f) Measured and (b–e and g–j) simulated BT10.8 – BT12.0 (top) and dust RGB (bottom) for 7 May 2010 at 1200 UTC. The simulations were performed for ash with a NAME size distribution and NAME forecast ash concentrations with different refractive indices: andesite (Figures 6b and 6g), obsidian (Figures 6c and 6h), volcanic dust (Figures 6d and 6i), desert dust and (Figures 6e and 6j). See Figure 1 for BT10.8 – BT12.0 key.

Download figure to PowerPoint

[31] Figure 5e is a graph showing the results of the objective method for assessing the refractive indices. The choice of refractive indices does not have a large effect for most of the cases. Overall, using refractive indices for andesite and desert dust result in simulated median BT10.8 – BT12.0 values that are closer to the measured median BT10.8 – BT12.0 values. However, for 6, 7 and 8 May 2010 using refractive indices for volcanic dust and obsidian result in a signal closer to that observed when the size distribution and ash concentration are also allowed to vary (Table 2).

7.2. Size Distribution

[32] Throughout this work, the NAME ash size distribution has been used since the aim is to simulate the satellite imagery using the forecast NAME ash dispersion. To understand the sensitivity to the size distribution and to gain an insight into the applicability of the NAME size distribution for this volcanic event, the simulations were rerun with the size distribution shifted to larger and smaller particles sizes. In each case the total mass of ash particles was kept constant. Thus, as the sizes increased the number density decreased and vice versa (Figure 2).

[33] As the particle size distribution decreases, with the total mass kept constant, the BT10.8 – BT12.0 ash signal increases (i.e., stronger negative values), as shown in Figure 7 and in the graph in Figure 5d. As the particle size increases, the BT10.8 – BT12.0 ash signal decreases and the plume becomes more difficult to identify. This illustrates the large effects that ash particle size has on the signal. Thus, an ash cloud containing predominantly large particles may go undetected using this imagery. Based on the comparison of the twelve cases here (Table 2), the NAME size distribution with particle diameters shifted by a factor of 0.25 was found to give the best (in terms of median values) simulations overall of the distributions studied, i.e., smaller particles with a peak in the size distribution in terms of mass at approximately 5 μm particle diameter. This is a similar to the peak particle diameter of 4 μm observed during FAAM flights (Johnson et al., submitted manuscript, 2011). Here, we have not considered the variance or “shape” of the size distribution; this has been kept constant (maintaining the “shape” of the NAME size distribution). This is an important factor and one worth considering in a future study.

image

Figure 7. Simulated BT10.8 – BT12.0 imagery for 1200 UTC on 7 May 2010 using NAME ash concentrations, refractive indices for andesite, and variations of the NAME size distribution: (a) diameters × 0.5, (b) diameters × 1, and (c) diameters × 2. See Figure 1 for BT10.8 – BT12.0 key.

Download figure to PowerPoint

7.3. Ash Concentration

[34] The ash concentration received from the NAME model was halved and doubled to look at the effect on the signal (Figure 8). As would be expected, as the concentration of ash increased the BT10.8 – BT12.0 ash signal increased and a decrease in concentration results in a decreased BT10.8 – BT12.0 ash signal. While a strong ash signal may indicate a high concentration of ash, as noted in section 2, it may also indicate that the particles are small (as shown in section 7.2) and thus a direct correlation between ash signal and ash concentration cannot be made without considering other factors. Figure 5f shows that for 6 and 7 May, the ash concentration from NAME result in the simulated signal closest to that observed with refractive indices of andesite and NAME size distribution, while for the other ten cases doubling the NAME ash concentration improves the match to the observed BT10.8 – BT12.0 signal. However, if the ash particle properties are fixed to andesite and a size distribution with a peak at 5 μm (size distribution factor of 0.25), the study indicates that the NAME ash concentration should be reduced for 6 and 7 May and increased for seven of the twelve cases to better match the observed BT10.8 – BT12.0 image (see Table 2).

image

Figure 8. Simulated BT10.8 – BT12.0 imagery for 1200 UTC on 7 May 2010 using NAME size distribution, refractive indices for andesite, and variations of NAME ash concentration: (a) concentration × 0.5, (b) concentration × 1, and (c) concentration × 2. See Figure 1 for BT10.8 – BT12.0 key.

Download figure to PowerPoint

8. Individual Cases

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

8.1. 15 April 2010

[35] The eruption began on 20 March 2010 at Fimmvörðuháls, part of the Eyjafjallajökull volcanic system. The eruption was mainly Hawaiian eruptive style with fire fountains and continued until 12 April. After a brief pause in activity, a new set of craters opened up on 14 April under the volcano's ice covered central summit caldera (Institute of Earth Sciences Daily Reports, unpublished data, 2010, http://www2.hi.is/page/ies_EYJO_compiled). This was the start of the main explosive eruption. By early morning an ash plume was observed rising up to 9 km above the volcano. The explosive eruption continued on 15 April with the ash plume rising to 6 km (as measured by radar). Ash reached mainland Europe causing closure of airspace across much of northern Europe, including the UK, Norway and Sweden.

[36] The BT10.8 – BT12.0 and dust RGB images for 1800 UTC on 15 April 2010 (Figures 9a and 9e) show a strongly positive BTD signal indicative of the presence of ice emanating from Iceland (white in BT10.8 – BT12.0 image and black in dust RGB). The main explosive eruption occurred under a glacier and therefore it released a lot of water vapor that mixed with the ash and volcanic gases and rose in the plume. Moreover, the large quantity of water melting into the crater interacts explosively with the magma to create small ash particles. Water vapor would have also been entrained from the surrounding atmosphere as the plume rose. The ice crystals in the ice cloud may have encased ash particles and so the presence of ash in this cloud cannot be ruled out [Rose et al., 1995]. An ash signal can be seen between Scotland and Norway (58 – 60 °N). It is not a strong or extensive signal, possibly due to masking by the ice cloud and the presence of a large quantity of (frozen) water droplets in the ash cloud.

image

Figure 9. (a, e) SEVIRI measured and simulated BT10.8 – BT12.0 (top) and dust RGB (bottom) images for 1800 UTC on 15 April 2010. (b, f) Simulated images using the original height run data and NAME size distribution. (c, g) Simulated images using the revised height run data and NAME size distribution. (d, h) Simulated images using the revised height run ash concentration × 2 and NAME size distribution with diameters × 0.25. All simulations use refractive indices for andesite. See Figure 1 for BT10.8 – BT12.0 key.

Download figure to PowerPoint

[37] When the explosive eruption occurred on 14 April there were few observations with which to set the source term for the NAME model. Initially, the plume top height was estimated to be 11 km on 14 April and 8 km by midday on 15 April (dotted line in Figure 4). NAME was run, for research purposes, with the plume top heights close to these original estimates (hereafter called “original height run”). Postevent analysis of data from radar revealed that the plume top height was 8–9 km on 14 April, reducing to 5–6 km on 15 April [Arason et al., 2011] (solid line in Figure 4). NAME was re-run [Webster et al., 2012] with plume top height at 9 km on 14 April reducing to 6 km on 15 April (revised height run). NAME ash concentrations from both of these runs have been used to simulate SEVIRI imagery (Figure 9).

[38] The original height run simulated images (Figures 9b and 9f) show strong ash signals with the ash cloud extending from Iceland to the Norwegian coast, in the same location as the observed ice and ash clouds (Figures 9a and 9e). The simulated images then show the ash cloud being advected across the North Sea over England and Wales. The strength of the simulated ash signal is too strong implying that the ash concentrations are too high (probably due to too high an ash plume at source, but possibly also due to other errors, e.g., in the assumed plume-rise versus source-strength relationship or in the near source fallout fraction) or that the ash particles are too small. The simulated images using the NAME ash concentrations generated using the revised plume height (reduced from 11 km to 9 km) (Figures 9c and 9g) do not show an ash signal at all implying that the ash concentrations are too low or the ash particles are too large. The images were simulated again using the revised height run data, a size distribution with small particles (diameters × 0.25) and double the NAME ash concentration (Figures 9d and 9h); this produced an ash signal with a strength that better matched the observed signal, but the location is still not correct. Dacre et al. [2011] used lidar observations of ash at Chilbolton, southern UK, to indicate that the location error is due to a timing error in the NAME output. The timing error, caused by cumulative errors in the driving NWP wind fields used in NAME, is of the order of 12–15 h (or a 200 km positional error). This example shows the large effect that a change in the ash plume height in the source term for NAME can have on the signal in the simulated imagery and how the simulated imagery may be used to identify an error in the NAME source assumptions.

8.2. 9 May 2010

[39] Explosive activity increased late on 5 May and continued in a strong explosive phase on 6 May, decreasing slightly on 7 and 8 May. On 6 May the plume reached altitudes of 4–9 km and by 9 May the plume height was reported to be 4–5 km, occasionally up to 6 km (Institute of Earth Sciences Daily Reports, unpublished data, 2010, http://www2.hi.is/page/ies_EYJO_compiled). By 9 May ash from the previous four days was circulating over the northern Atlantic Ocean, with the oldest ash (3–4 days old according to NAME simulations (not shown)) still producing a strong signal in the center of the circulation (Figure 10). The satellite imagery shows a narrow ash cloud to the south of Iceland, bending around to the west at approximately 52°N, 18°W, then swinging north and curving back toward Iceland and into the center of the circulation.

image

Figure 10. (a, d) SEVIRI measured and simulated BT10.8 – BT12.0 (top) and dust RGB (bottom) images for 1200 UTC on 9 May 2010. (b, e) Simulated images using NAME size distribution, refractive indices for andesite and NAME ash concentration. (c, f) Simulated images using NAME size distribution with diameters × 0.25, refractive indices for andesite and NAME ash concentration. See Figure 1 for BT10.8 – BT12.0 key.

Download figure to PowerPoint

[40] Images from two simulations are shown in Figure 10, one with the standard setup (andesite refractive indices, NAME size distribution and NAME ash concentration) and one with the diameters in the particle size distribution multiplied by a factor of 0.25. The simulation with the standard setup does not show a strong ash signal (Figures 10b and 10e). There is a weak signal to the southwest of Iceland for the older ash, but the ash cloud to the south of Iceland is barely detectable. By decreasing the size of the particles a strong ash signal results in a good match to the measured satellite imagery. The revised setup (Figures 10c and 10f) results in a strong ash signal to the south of Iceland and a strong ash signal to the west of Iceland and in the center of the circulation. There is too little ash at around 53°N, 25°W. This would need a modification of the source term in NAME to produce a pulse of ash from the volcano at around 0000 UTC on 8 May (using the NAME age of ash simulation, not shown). Otherwise, the location of the ash in the simulation and thus in the NAME output is in good agreement to the ash detected in the satellite imagery.

8.3. 17 May 2010

[41] During the four days prior to the 17 May, the plume height was 6–7 km, occasionally reaching 8–9 km with a slightly lower plume height on 14 May of around 5 km, as measured by radar (Institute of Earth Sciences Daily Reports, unpublished data, 2010 http://www2.hi.is/page/ies_EYJO_compiled). The ash drifted in a southeasterly direction and by 1200 UTC on 17 May it could be detected in the satellite imagery to the east and southeast of Iceland and over the North Sea close to the east coast of the UK (Figure 11).

image

Figure 11. (a, d) SEVIRI measured and simulated BT10.8 – BT12.0 (top) and dust RGB (bottom) images for 1200 UTC on 17 May 2010. (b, e) Simulated images using NAME size distribution, refractive indices for andesite and NAME ash concentration. (c, f) Simulated images using NAME size distribution with diameters × 0.5, refractive indices for andesite and NAME ash concentration. See Figure 1 for BT10.8 – BT12.0 key.

Download figure to PowerPoint

[42] The simulated satellite imagery using the standard setup (andesite refractive indices, NAME size distribution and NAME ash concentration) shows a strong ash signal over Iceland and a weaker signal to the southeast of Iceland over the Faeroe Islands toward the Shetland Islands (Figures 11b and 11e). When the simulation is repeated with smaller particles (diameters multiplied by a factor of 0.5) a strong ash signal over a larger area is revealed (Figures 11c and 11f). This is a better match to the measured satellite imagery in terms of location of ash and strength of signal. However, the location of the ash signal in the simulation needs to be in a more meridional direction from the east of Iceland rather than southeast from the volcano and needs to extend further south over the North Sea approaching the Norfolk coast in the UK.

[43] This case illustrates the sensitivity to particle size distribution. Without changing the ash concentration (keeping the mass constant), the reduction in particle sizes by a factor of 2 (and thus increase in number density) results in a much stronger ash signal revealing a greater extent of the ash cloud. Therefore, great care must be taken in drawing conclusions from the apparent lack of ash in the simulated imagery using the standard setup.

9. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[44] Simulated volcanic ash imagery can provide useful additional information for a forecaster operating the VAAC service. The simulated imagery when compared with the equivalent measured imagery can give an indication of the accuracy of the output from earlier runs of the atmospheric dispersion model, NAME. The comparison between the simulated and measured imagery can reveal the accuracy in the location of the predicted ash cloud at the observation time. A close collocation of the simulated and measured ash signals gives confidence in the NAME forecasts, while a mismatch in locations would indicate that the forecaster needs to take action to adjust the current and future volcanic ash advisories. This may be done by a forecaster directly adjusting the current VAAC ash prediction and by adjusting the source term for the next NAME run. However, a careful assessment must be made when comparing the simulated and measured volcanic ash imagery since there are limitations in what both simulated and measured volcanic ash images can reveal.

[45] Other methods of using satellite data to improve the advisories issued by VAACs include (1) the retrieval of ash cloud height, mass loading and effective particle radius that can be compared with dispersion model output, (2) performing an inversion that couples a priori source information and the output of a dispersion model with satellite data [Stohl et al., 2011], or (3) the development of assimilation of satellite data into a dispersion model.

[46] In addition to the use of the simulated imagery in an operational VAAC environment, the imagery can be used to indicate the physical properties of ash particles and aid the understanding of signals in the volcanic ash satellite imagery. Table 2 shows the optimum set of ash particle properties for the twelve cases studied. The studies here indicate that the refractive indices for andesite and that particle sizes smaller than the NAME size distribution, with the peak of the aerosol mass at 5 μm particle diameter, give the best simulations for this eruption. These simulations indicate that the ash concentration from NAME needed to increase for seven of the twelve cases when the ash particle properties are fixed (to andesite and a size distribution peaking at 5 μm), and that the concentration needed to decrease for the 6 and 7 May 2010. For the range of physical characteristics studied it was found that the simulated signal had greatest sensitivity to the particle size distribution and that overall, if the particle sizes were reduced and the NAME ash concentrations unchanged, there were reasonable matches in signal strength between the simulated and observed BT10.8 – BT12.0 images.

[47] There are several factors that affect the measured and simulated imagery that need to be taken into account when drawing conclusions from comparing the two. If a large amount of water vapor is released during the eruption, the ash may be well mixed with water droplets or can become encased in ice [Rose et al., 1995] and both may give the ash a positive bias in the BT10.8 – BT12.0 imagery. The simulated imagery will not take account of any water or ice of volcanic origin or of the ash cloud entraining moist air and ash becoming encased in ash; thus an accurate NAME forecast may result in a strong ash signal in the simulated imagery and little or no signal in the measured imagery. In this theoretical situation, there is no means to distinguish an ice-rich ash cloud from an ice cloud and so it would not be possible to draw useful conclusions from the comparison of simulated and observed imagery.

[48] The volcanic ash imagery (BT10.8 – BT12.0 and dust RGB) is insensitive to large particles and so areas of the volcanic cloud that contain a large mass of particles with diameters >5 μm [Wen and Rose, 1994] will not produce a strong ash signal (however, a thick ash cloud will be identifiable from contextual information in infrared and visible imagery assuming it's not obscured by water or ice cloud). Large particles (>100 μm) tend to fall out close to the volcano and therefore this should not be a problem for the dispersed ash cloud.

[49] The strength of the ash signal in the volcanic ash imagery depends on a large range of factors: size distribution of particles, composition of particles, number density of particles, opacity of ash cloud, height of ash cloud, temperature contrast between ash cloud top and underlying surface, satellite viewing angle and presence of other absorbers (e.g., water, ice, and SO2). Thus, the strongest ash signal in the BT10.8 – BT12.0 (or dust RGB) does not directly relate to areas of highest concentration and should not be directly compared to raw ash concentration output from NAME. The use of simulated imagery and an understanding of the sensitivities aid the comparison of observed and modeled ash.

[50] There are many possible extensions to this study to enhance the usefulness to VAAC forecasters and understanding to modellers and researchers. For example, it has been shown that there is a strong sensitivity to particle size distribution, so it may be useful to provide ash concentrations for different size distributions (e.g., concentration of large, medium, and small) in order to better represent the number density of particles in the dispersed ash cloud (as opposed to using the NAME size distribution for the source term). It would be useful for a simulation to be provided pre and postforecaster intervention so that the forecaster could check the impact of their changes. Additional studies in the sensitivity to ash cloud height and shape of size distribution would be interesting. All these studies aid understanding into the effects that these factors have on the observed and simulated satellite imagery. They can feedback information to modellers on the representativeness of the ash particles in NAME and RTTOV and the overall performance of the atmospheric dispersion model.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

[51] We acknowledge EUMETSAT for the provision of SEVIRI data via EUMETCast. The RTTOV model is developed as part of the EUMETSAT funded NWP SAF activities. We thank three reviewers; their comments helped to greatly improve the manuscript.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. SEVIRI Volcanic Ash Imagery
  5. 3. Physical Properties of Volcanic Ash
  6. 4. Optical Properties of Volcanic Ash
  7. 5. NAME Ash Concentration
  8. 6. Simulation of SEVIRI Imagery
  9. 7. Sensitivity of Imagery to Ash Properties
  10. 8. Individual Cases
  11. 9. Discussion
  12. Acknowledgments
  13. References
  14. Supporting Information
FilenameFormatSizeDescription
jgrd17643-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
jgrd17643-sup-0002-t02.txtplain text document1KTab-delimited Table 2.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.