A comparison of atmospheric dispersion model predictions with observations of SO2 and sulphate aerosol from volcanic eruptions

Authors


Abstract

[1] The UK Met Office's Numerical Atmospheric-dispersion Modeling Environment (NAME) is used both operationally and for research investigations. It has previously been used to model volcanic ash at the London Volcanic Ash Advisory Centre (VAAC), including that from the eruptions in Iceland of Eyjafjallajökull in 2010 and Grímsvötn in 2011. In this paper, the ability of NAME to model the release and dispersion of volcanic SO2, the chemical processes leading to the production of sulphate aerosol, and the subsequent dispersion of sulphate aerosol, has been investigated. Sensitivity tests were carried out to investigate the suitability of the NAME chemistry scheme for use in both the troposphere and the stratosphere. The eruptions of Sarychev in 2009, Kasatochi in 2008 and Eyjafjallajökull in 2010 were simulated and results for SO2 column density and sulphate aerosol optical depth (AOD) were compared with satellite retrievals. NAME results compare favorably with available observations in terms of both geographical distribution and magnitude for all three cases. The NAME modeled values of SO2 show a correlation of 0.8 with the corresponding observations for Sarychev. Ninety percent of modeled values of northern hemisphere averaged sulphate AOD are within a factor of 2 of those observed for Kasatochi and 71% are within a factor of 2 of those observed for Sarychev. Although significant uncertainties are present in both the model and observations, this work demonstrates that NAME's current chemistry scheme shows promise as a tool for modeling SO2 and sulphate from volcanoes.

1. Introduction

[2] Volcanic eruptions can inject large quantities of sulphur dioxide (SO2) into the upper troposphere and lower stratosphere (UTLS), affecting the global climate and leading to acid precipitation and air pollution [Grainger and Highwood, 2003; McCormick et al., 1995]. The lifetime of volcanic SO2 in the UTLS is longer than that of ash, with a mean lifetime of 9–11 days [Krotkov et al., 2010; Haywood et al., 2010]. The main mechanism for its removal is photochemical conversion to sulphuric acid through gas-phase reactions with the hydroxyl radical (OH), as described in section 2. Sulphuric acid subsequently forms sulphate aerosol, which can remain in the stratosphere for months to years, depending on SO2 injection height, total mass and dispersion pattern [Krotkov et al., 2010]. Haywood et al. [2010] derive a mean lifetime of approximately 81 days for sulphate aerosol.

[3] SO2 and the sulphuric acid formed from sulphate aerosol pose a hazard to aviation. After the eruption of Mount Pinatubo, Philippines, in 1991, airlines reported damage to airframes, increased crazing of acrylic windows, fading paint and accumulation of sulphate deposits in turbines, which can block cooling holes and lead to engine overheating [Carn et al., 2009]. Sulphur compounds also accelerate the oxidation of metals, leading to eventual corrosion of the turbine blades. This results in an increased maintenance requirement and decreases the lifetime of the engine [Fisher, 2008]. In 1992, an aircraft suffered engine power loss due to accumulated sulphate deposits in the engines. Isotopic analysis suggested that this sulphate was derived from SO2 and sulphate aerosol originating from the eruption of Mount Pinatubo the previous year [Miller and Casadevall, 2000].

[4] Volcanic SO2 and sulphate are also hazardous to human and animal health. During 1783–1784, the Laki eruption in Iceland released approximately 122 Tg SO2 into the atmosphere. Historical records suggest that 21% of the human population and 75% of livestock in Iceland died in the aftermath of this eruption, while mortality in England was 10–20% above the 51-year moving mean [Schmidt et al., 2011, and references therein]. An assessment by Schmidt et al. [2011] of the impact of a Laki-style eruption on present-day European air quality suggests that approximately 142,000 additional cardiopulmonary fatalities could occur in Europe as a result, increasing excess mortality on a scale likely to exceed that due to seasonal influenza.

[5] As volcanic aerosols can be hazardous to aviation and to human and animal health, accurate modeling of the dispersion of volcanic emissions is important. Previous comparisons of volcanic ash, SO2 and sulphate modeling with satellite, airborne and ground-based observations [Witham et al., 2007; D'Amours et al., 2010; Haywood et al., 2010; Devenish et al., 2012a; Webster et al., 2012] have shown that a range of models perform well, given the significant uncertainties present. The key uncertainty is the definition of the magnitude and height of the volcanic emissions and their variation with time. In volcanic ash modeling, the height of the ash plume can be estimated using observations from radar, satellites and pilot reports. This information can then be used to estimate the emission rate using an empirical relationship [Mastin et al., 2009]. In the case of SO2 there is no such relationship between the emission rate and plume height. The magnitude of the SO2 injection can be determined using satellite retrievals, but these have little or no vertical resolution. It is possible to use inverse modeling to estimate the vertical profile of the SO2 injection [Eckhardt et al., 2008] but, in this work, simplified vertical profiles based on observations of ash plume height have been used.

[6] The aim of this work is to model the release and dispersion of SO2, and chemical processes leading to the production and fate of sulphate, from multiple volcanoes using the Numerical Atmospheric-dispersion Modeling Environment (NAME). Two recent eruptions that released large quantities of SO2 into the UTLS were Sarychev in June 2009 and Kasatochi in August 2008. The Sarychev volcano is located in the Kuril Islands northeast of Japan. Between 12 and 17 June 2009, Sarychev erupted explosively and injected an estimated 1.2 Tg SO2 into the UTLS [Haywood et al., 2010]. The Kasatochi volcano, located in the Aleutians southwest of Alaska, injected an estimated 1.3 Tg SO2 into the UTLS on 7 and 8 August 2008 [D'Amours et al., 2010]. The dispersion and conversion to sulphate of SO2 during these eruptions, and during the 2010 eruption of Eyjafjallajökull in Iceland, are simulated. The model results from NAME for each volcanic release for both SO2 and sulphate aerosol are compared with satellite retrievals.

2. The NAME Model and Methodology

[7] The NAME model [Jones et al., 2007] was originally developed by the UK Met Office to simulate the dispersion of nuclear material and is currently used for a wide range of applications, including emergency response to industrial incidents [Webster et al., 2007; Ryall and Maryon, 1998] and outbreaks of infectious diseases affecting livestock [Gloster et al., 2007], air quality [Redington et al., 2009; Witham and Manning, 2007], long-range transport [Ryall et al., 2002] and inversion modeling [Manning et al., 2011; Reimann et al., 2005]. NAME is also used to model the dispersion of volcanic ash to support the operations of the London Volcanic Ash Advisory Centre (VAAC) and was used during the eruptions in Iceland of Eyjafjallajökull in 2010 and Grímsvötn in 2004 and 2011.

[8] NAME is a Lagrangian atmospheric dispersion model in which particles representing pollutants are released into a three-dimensional model domain. There is a great deal of flexibility available in the specification of the pollution source in NAME, both spatially and temporally. In addition, a complex source may be built up by combining a number of simpler sources. This flexibility enables the different eruptions studied here to be represented as accurately as the available data allow. Each particle can represent multiple pollutants. Particles are advected every model time step according to the mean resolved wind flow, as well as by meander – unresolved motions smaller than the model resolved wind. Particles are also advected by local atmospheric turbulence, with the unresolved motions modeled using a random walk scheme. Analysis three-dimensional meteorology is taken from the UK Met Office global numerical weather prediction model, the Unified Model [Davies et al., 2005]. The meteorological fields from the global Unified Model used by NAME have a temporal resolution of three hours and are interpolated to the specific time and location of each particle. The spatial resolutions of the meteorological fields used in the Sarychev, Kasatochi, and Eyjafjallajökull simulations vary between cases and are described later. Meteorological variables of major concern in this study include the three-dimensional wind fields that drive the advection of the particles, the cloud fraction and cloud water which partly determine the pathways of oxidation of SO2 to sulphate, and the precipitation rate which affects wet deposition.

[9] Pollutant masses on each particle evolve due to wet and dry deposition and chemical processes, which depend on the pollutants involved and the meteorological conditions. The dry deposition scheme uses a resistance analogy parameterization to calculate a pollutant dependent deposition velocity, vd, which is applied to particles within the boundary layer [Webster and Thompson, 2012]. The flux, F, of pollutant to the ground by dry deposition is given by

display math

where C is the concentration of pollutant above ground [Maryon et al., 1999]. The wet deposition scheme, applied below the cloud top height, uses scavenging coefficients to model the washout and rainout of pollutants as

display math

where t is time and Λ is the scavenging coefficient. The scavenging coefficient is dependent on the precipitation rate and whether precipitation is convective or dynamic [Maryon et al., 1999].

[10] SO2 is converted to sulphate in the atmosphere by reactions involving OH, hydrogen peroxide (H2O2) and ozone (O3). NAME is able to model O3 either as climatological background fields, on background particles, or on particles local to the pollutant plume, depending on the nature of the system to be modeled. Although modeling ozone explicitly is more accurate, this must be traded-off against increased model run-time and computing resources. In this study the model setup uses monthly background fields of H2O2 and O3 as initial states, read from fields produced by the global chemistry model STOCHEM [Collins et al., 1997] in order to maximize efficiency. Concentrations of these photo-oxidant species are passed to the chemistry module together with the concentrations of released pollutants (in this case SO2) and any secondary pollutants (in this case sulphate) which are carried by the particles. Concentrations of the species carried by the particles are calculated on a 3-dimensional grid by adding up the total mass of each species at each time step in every grid box. The concentrations of the pollutants carried on both fields and particles are modified as chemical processes are simulated and the resulting concentrations of pollutants and their secondary products in each grid box are converted back to masses and re-assigned to the particles in proportion with the original mass of the primary pollutants. The concentrations of the photo-oxidant species OH and hydroperoxy radical (HO2) are calculated anew every time step [Redington and Derwent, 2002]. Near to the poles the lat-long grid boxes are small and therefore contain fewer particles than elsewhere, so it is much harder to obtain statistically reliable concentrations. This problem could be moved away from the poles by changing the coordinate system to use, for example, a lat-long system with a displaced pole for the chemistry grid. A truly global solution would require a more complex solution, perhaps with different coordinate systems patched together. However, we have not investigated this here and have simply restricted attention to regions away from the poles.

[11] Aqueous phase reactions take place if the grid box contains a nonzero cloud fraction and cloud water, and dominate sulphate production in cloudy conditions. Gas phase reactions are simulated throughout the atmosphere and dominate in the stratosphere where there is little cloud present. The gas phase reaction involving sulphate is

display math

where M denotes ambient air. This reaction is dependent on temperature and pressure, and controls the rate of sulphate production as HSO3 is rapidly oxidized to HSO4. The gas to liquid phase equilibria, in which SO2 dissolves and dissociates, are

display math

[12] The aqueous phase reactions relevant to the production of sulphate are

display math

[13] There are two pathways for the oxidation of SO2 in the aqueous phase, involving either H2O2 or O3 in solution. The reaction with H2O2 is very rapid and the available H2O2 is quickly used up. The reaction with O3 is limited by the pH – as more H2SO4 is produced, the acidity of the cloud increases and the reaction rate slows. NAME uses the same chemistry scheme throughout the atmosphere and the creation of a separate stratospheric chemistry scheme is a potential area for further work.

[14] For comparison with satellite retrievals, modeled SO2 concentrations were converted to Dobson Units (DU) where

display math

and

R

= universal gas constant = 8.31 J K−1 mol−1

T

= standard temperature = 273 K

P

= standard pressure = 1.013 × 105 Pa

Mr

= relative molecular mass of SO2 = 64 g mol−1.

[15] Additionally, sulphate concentrations were converted to aerosol optical depth (AOD) at a wavelength of 750 nm, defined as

display math

where

ke

= specific extinction coefficient (m2 g−1)

ρ

= sulphate aerosol density (g m−3)

z

= height (m).

[16] The specific extinction coefficient was taken to be 1.8 m2 g−1 at 750 nm, which Haywood et al. [2010] have shown to be an appropriate value for stratospheric aerosol from Sarychev.

[17] Observations from three different satellite instruments have been used for comparison with the modeled results. The Infrared Atmospheric Sounding Interferometer (IASI) is onboard the European polar-orbiting MetOp-A satellite and has a nadir spatial resolution of 12 km. It was designed for meteorological applications, including estimating and monitoring trace gases on a global scale, and is sensitive to SO2 from the middle troposphere to the lower stratosphere [Thomas and Prata, 2011]. Retrieval of SO2 from the 2007 eruption at Jebel at Tair using IASI is described by Clarisse et al. [2008]. IASI has a detection limit of 0.3 to 1.0 DU SO2 in the UTLS, with a higher limit above and below (L. Clarisse, personal communication, 2011), and sensitivity is further limited by the presence of water vapor. The Canadian Optical Spectrograph and Infrared Imaging System (OSIRIS) [Llewellyn et al., 2004] is onboard the Swedish Odin satellite [Murtagh et al., 2002]. The limb scattered sunlight measurements made by OSIRIS can be used to retrieve vertical profiles of stratospheric aerosol extinction at 750 nm from approximately 7 km to 35 km in altitude, which can be numerically integrated to obtain the AOD [Bourassa et al., 2007, 2010]. The Ozone Monitoring Instrument (OMI) is onboard the Earth Observing System (EOS) Aura satellite and measures backscattered radiation in ultraviolet and visible wavelengths. It has a nadir spatial resolution of 13 km by 24 km and is most sensitive to SO2 above clouds and snow or ice and less sensitive below cloud. The overall uncertainty in SO2 measurements above 5 km asl is around 20%, and retrieval of SO2 from volcanic eruptions using OMI is described by Yang et al. [2007].

3. Sarychev

3.1. Model Setup

[18] The Sarychev volcano is located at 48.1°N, 153.2°E. An estimated total of 1.2 Tg SO2 was released at a constant rate of 0.025 Tg hr−1 over 15 and 16 June 2009, as derived from IASI retrievals by Haywood et al. [2010]. The release of a small quantity of SO2 (less than 0.1 Tg [Rix et al., 2009]) between 11 and 14 June was not included in the simulation. Model particles were released between 11 km and 15 km asl to accommodate the injection heights of up to 13.7 km reported by Tokyo VAAC. The model domain used was the Northern Hemisphere from 0° to 85°N, to an altitude of 30 km. Meteorological data with a horizontal resolution of 0.56° (longitude) by 0.38° (latitude) and 33 vertical levels up to 20 km asl were used. A chemistry grid with a horizontal resolution of 1° by 1° and a vertical resolution of 500 m was used. Results were output on a 1° by 1° horizontal grid.

[19] To investigate the sensitivity of the modeled results to the chemistry fields, the model was also run using concentrations of O3 [Coheur et al., 2005], HO2 and OH [Wennberg et al., 1995] measured in the stratosphere as an alternative to the default concentrations from STOCHEM. The times and locations of these measurements are different from those of the eruptions studied and only a single measured concentration was used for each species. The measured concentrations were used above an estimated tropopause height of 12 km asl. Below this altitude, the fields from STOCHEM were used. A total of four simulations using stratospheric values were carried out. One simulation was run using the stratospheric value for O3 while the default values were used for HO2 and OH. The effects of using stratospheric values for HO2 and OH were also investigated in the same way. A fourth simulation in which stratospheric values were used for all three species simultaneously was also performed.

3.2. Results

[20] Figure 1 shows the 24-h time-averaged NAME modeled SO2 column density between 0 and 30 km asl and the IASI SO2 retrievals. The IASI retrievals are taken throughout the depth of the atmosphere and are averages of the AM and PM swathes. The modeled results show good agreement with satellite retrievals in terms of both geographical distribution and magnitude, despite the differences in vertical- and time-averaging. In general, NAME modeled SO2 plumes cover a larger area than those observed. This can in part be explained by the fact that IASI has a detection limit of 0.3 DU to 1.0 DU in the UTLS, so that no SO2 levels below this are present in the observations. Additionally, Devenish et al. [2012b] show that the subgrid diffusion processes used in NAME can increase the spread of modeled plumes over and above that observed. Although NAME captures most of the SO2 transport well, the region surrounding Alaska is poorly modeled toward the end of June. Possible causes of this include inaccuracies in the SO2 release rate and height, and in the chemistry scheme, as discussed later.

Figure 1.

(left) The 24-h time-averaged NAME modeled SO2 column density in Dobson Units (DU) for 0 to 30 km asl and (right) daily average IASI SO2 retrievals in DU for the Sarychev eruption.

[21] Figure 2 shows the mean stratospheric longitudinally averaged sulphate AOD at 750 nm for the weeks following the eruption of Sarychev. The NAME modeled sulphate AODs are initially higher than those observed by OSIRIS by more than a factor of two in places in the early weeks. However, modeled AODs also decrease more rapidly and are lower than observed at later times. By the end of August, modeled latitudinal average AODs do not exceed 0.07 while those observed remain above 0.08 at high latitudes. The north-south gradient is well captured by NAME with the highest sulphate levels occurring at latitudes similar to and greater than that of the eruption, although there is also significant transport almost as far south as the Equator.

Figure 2.

Time versus latitude plot of the mean longitudinally averaged sulphate AOD at 750 nm modeled by NAME and observed by OSIRIS for the Sarychev eruption. The latitude and time of the eruption is marked with a cross.

[22] Figure 3a shows the time evolution of SO2 and stratospheric sulphate aerosol after the eruption of Sarychev, averaged over the Northern Hemisphere, for both NAME and observations by IASI and OSIRIS. Figure 3b shows the effect of using stratospheric values for O3, HO2 and OH on the time evolution of modeled SO2 and stratospheric sulphate aerosol. The shaded regions represent the range of model outcomes using different stratospheric concentrations of photo-oxidant species, while the line within each shaded region shows the result of using the STOCHEM fields. The upper boundary of the SO2 results and the lower boundary of the sulphate aerosol results are from model runs using observed stratospheric values for HO2 [Wennberg et al., 1995], with OH and O3 determined as in the default NAME chemistry scheme. The lower boundary of the SO2 results and the upper boundary of the sulphate aerosol results are from model runs using observed stratospheric values for O3 [Coheur et al., 2005], with OH and HO2 determined as in the default NAME chemistry scheme. Model runs using observed stratospheric values for OH result in more SO2 and less sulphate aerosol than is obtained using the default NAME chemistry, while runs using observed stratospheric values for all three photo-oxidant species (O3, HO2 and OH) give less SO2 and more sulphate aerosol. This latter case is most similar to the results obtained by manipulating the stratospheric values for O3 only, suggesting that the chemistry scheme is most sensitive to changes in O3 concentration.

Figure 3.

Time evolution of SO2 and stratospheric sulphate AOD (×10) at 750 nm averaged over the Northern Hemisphere between 0°N and 85°N for the Sarychev eruption. (a) SO2 is measured by IASI between 0 km and 30 km and sulphate is measured by OSIRIS from around 14 km upwards. (b) The shaded regions represent the range of model outcomes using different stratospheric concentrations of photo-oxidant species, while the line within each shaded region shows the result using the STOCHEM fields.

[23] The peak SO2 value around 16 June is well modeled by NAME, which is expected because the SO2 source strength is derived from IASI retrievals, although NAME modeled SO2 levels persist for longer than those observed. A contributing factor to this could be the detection limit of IASI of 0.3 to 1 DU, which would cause the SO2 retrievals to appear low. However, the fact that SO2 levels do not persist to the same extent in previous simulations of this eruption using the HadGEM2 version of the Met Office climate model [Haywood et al., 2010], which uses a different chemistry scheme, indicates that there is room for improvement in the NAME chemistry scheme. In common with the results of Haywood et al. [2010], the overall production of sulphate aerosol is more rapid in the NAME modeled results than in the observations, leading to higher sulphate aerosol AODs at earlier times. The observed sulphate AOD peaks on approximately 4 August while the NAME modeled sulphate AOD peaks earlier on approximately 15 July. This discrepancy could be caused in part by the fact that NAME does not adequately represent the nucleation of new particles in the stratosphere. In this study, when SO2 is oxidized into sulphate aerosol, it is immediately assumed to reside in the optically active accumulation mode, leading to an immediate enhancement of the AOD. As concluded by Haywood et al. [2010], representation of nucleation appears to be a critical process in getting the timing of the sulphate AOD correct in the stratosphere. The overall loss of sulphate is better modeled by NAME than that of SO2, although it is more rapid than is observed. 71% of the hemispheric mean sulphate AODs modeled using the default NAME chemistry are within a factor of 2 of the corresponding observations and 100% are within a factor of 3.

4. Kasatochi

4.1. Model Setup

[24] The Kasatochi volcano is located at 52.2°N, 175.5°W and was reported by the Alaska Volcano Observatory (AVO) to have erupted for almost 20 h beginning at around 2000 UTC on 7 August 2008 [D'Amours et al., 2010]. Estimates of the total SO2 released during this eruption vary greatly [Kristiansen et al., 2010] and the AVO reported that the eruption was characterized by three distinct explosions, detected by the seismic network at 2201 UTC, 0150 UTC and 0435 UTC. The first two explosions were ash-poor but gas-rich, while the third was rich in both ash and gas. All three explosions reached 14 to 15 km asl (Alaska Volcano Observatory, available at http://www.avo.alaska.edu/activity/Kasatochi08/Kasatochi2008PLW.php (accessed 1 December 2011)). For this simulation, total of 1.3 Tg SO2, derived from AURA/OMI data by S. Carn (NASA OMI image available at http://earthobservatory.nasa.gov/IOTD/view.php?id=8998 (accessed 29 June 2011)), was released at a constant rate of 0.065 Tg hr−1 from 2000 UTC on 7 August to 1600 UTC on 8 August 2008. Two different vertical source profiles were tested. In profile 1, 10% of the model particles and SO2 mass were emitted between 5.5 km and 7.5 km asl, 50% between 7.5 km and 12 km asl and the remaining 40% between 12 km and 14 km asl (R. D'Amours, personal communication, 2010). In profile 2, 40% of the model particles and SO2 mass were emitted between 4 km and 8 km asl and the remaining 60% between 11 km and 17 km asl to replicate the source derived by Kristiansen et al. [2010] using inverse transport modeling. The model domain, meteorological data resolution, chemistry grid and output grid are the same as those used for the Sarychev simulation.

4.2. Results

[25] Figure 4 shows the NAME modeled SO2 column densities and retrievals from the AURA/OMI satellite. The left column shows modeled results using the vertical source profile 1 and the central column shows those using profile 2. The right column shows the corresponding satellite retrievals. The results using profile 1 and profile 2 are similar, although profile 1 gives better results for the Eastern part of the plume while profile 2 gives better results for the Western part. This suggests that the plume is being advected in different directions at different heights. There is also good agreement between both sets of modeled results and the satellite retrievals but, as before, the results are not directly comparable due to differences in spatial- and time-averaging. In particular, the first three OMI images are from single swath measurements while the last four are obtained by combining swaths together. The NAME modeled SO2 column densities are similar in magnitude to those observed by the AURA/OMI satellite; this is to be expected because the SO2 source strength was derived from OMI data. As with the Sarychev simulation, the NAME modeled SO2 plume shows greater dispersion and covers a larger area than that observed. In addition to the reasons given previously, such as excessive sub-grid diffusion and inaccuracies in the chemistry scheme, the use of a simplified, constant emission rate also contributes to these discrepancies.

Figure 4.

Natural logarithm of SO2 column density in Dobson Units (left and middle) as modeled by NAME using two different vertical source profiles and (right) observed by AURA/OMI (NASA OMI image courtesy Simon Carn, Joint Center for Earth Systems Technology (JCET), University of Maryland Baltimore County (UMBC), available at http://earthobservatory.nasa.gov/IOTD/view.php?id = 8998 (accessed 20 July 2011)) for the Kasatochi eruption. NAME results are time averages over 1 h for the first 3 plots and 24 h for the remaining 4 plots to approximate the time periods of the observations. The OMI images are composites of 1 or more swaths. The beginning of each time interval corresponds to the east-most swath while the end of each time interval corresponds to the west-most swath.

[26] Figure 5 shows the mean stratospheric longitudinally averaged sulphate AOD for the weeks following the eruption of Kasatochi. As is the case in the Sarychev simulation, the NAME modeled sulphate AODs are initially higher than those observed by OSIRIS. In addition, the NAME modeled sulphate AOD is not transported as far south as is observed, particularly in the case of profile 1 in which there is less SO2 released at higher altitudes, and neither source profile is able to reproduce transport into the equatorial circulation system as is observed by OSIRIS. It is possible that the use of a simplified, constant emission rate is a contributing factor to this discrepancy. However, it may also be the case that some of the sulphate aerosol in the satellite retrievals is from a source other than Kasatochi, such as the Kilauea volcano located at 19.4°N, 155.3°W which continued to erupt throughout 2008 (Hawaiian Volcano Observatory, available at http://hvo.wr.usgs.gov/kilauea/timeline/ (accessed 12 December 2011)).

Figure 5.

Time versus latitude plot of the mean longitudinally averaged sulphate AOD at 750 nm (top) modeled by NAME using two different vertical source profiles and (bottom) observed by OSIRIS for the Kasatochi eruption. The latitude and time of the eruption is marked with a cross.

[27] Figure 6 shows the time evolution of SO2 and sulphate after the eruption of Kasatochi, averaged over the Northern Hemisphere. Although the same amount of SO2 was released when testing each vertical source profile, as can be seen from the identical peak SO2 values, profile 2 results in greater and more long-lived sulphate aerosol levels than profile 1. This is expected because of the greater release of SO2 at higher altitudes in profile 2, which will in turn result in a greater quantity of sulphate aerosol residing in the stratosphere. The stratified nature of the atmosphere and limited vertical transport and wet deposition enable the sulphate aerosol to remain in the atmosphere for longer, and build up to greater levels, when profile 2 is used. As with the Sarychev simulation, the NAME modeled sulphate AOD (using profile 1) peaks earlier than is observed, although the effect is less pronounced in this case. A contributing factor in this discrepancy may be the lack of representation of the sulphate aerosol nucleation mode, as discussed earlier. The modeled sulphate AOD (using profile 1) peaks on approximately 9 September while the observed sulphate AOD reaches a peak on approximately 18 September. 90% of the modeled hemispheric mean sulphate AODs are within a factor of 2 of the corresponding observations when using profile 1.

Figure 6.

Time evolution of NAME modeled SO2 and modeled and observed stratospheric sulphate AOD at 750 nm averaged over the Northern Hemisphere between 0°N and 85°N for the Kasatochi eruption. Modeled results are shown for two different vertical source profiles.

5. Eyjafjallajökull

5.1. Model Setup

[28] The Eyjafjallajökull volcano is located at 63.63°N, 19.62°W and erupted over a period of six weeks during April and May 2010. While determining the SO2 emissions early in the eruption period was hampered by the presence of cloud, during 7 and 8 May the conditions were sufficiently cloud-free for IASI to make a reasonably continuous set of coherent measurements. The SO2 release rates estimated from IASI retrievals are given in Table A1 in Appendix A. As there were no observations of the vertical SO2 source profile available, particles were released uniformly between 1666 m asl (the summit height) and 10000 m asl (the maximum height of the ash plume estimated using observations from radar, pilot reports and the Icelandic coastguard). The model domain used was the Northern Hemisphere from 26°N to 85°N. Meteorological data with a horizontal resolution of 0.35° (longitude) by 0.23° (latitude) and 59 vertical levels up to 20 km asl were used. The chemistry grid has the same vertical resolution as was used for the Sarychev simulation and a horizontal resolution of 0.5° by 0.5°. Results were output on a 0.5° by 0.5° horizontal grid.

5.2. Results

[29] Figure 7 shows NAME modeled SO2 column densities and the corresponding IASI retrievals for two days part way through the six week eruption period. There is a good agreement between the modeled results and the IASI retrievals in terms of both SO2 column density and plume location. However, as was the case for Sarychev, this is to be expected because the SO2 source strength is derived from IASI retrievals. Near to the volcano, the satellite does not accurately detect SO2 due to its low altitude and masking by steam and volcanic ash, making comparison with observations difficult in this region.

Figure 7.

The (left) 12-h time-averaged NAME modeled SO2 column density and (right) IASI retrievals for the Eyjafjallajökull eruption, both in DU.

6. Discussion

[30] In general the results from NAME compare well with observations, demonstrating the model's ability to simulate long range transport and chemical processes. Time-averaged NAME modeled SO2 column densities and sulphate AODs are within a factor of three of those observed by satellite. However, both estimates have significant uncertainties.

[31] Uncertainties exist in the SO2 source strengths, which are derived from satellite retrievals. The SO2 signal can be masked by clouds, as well as by steam and ash from the volcano, and observations are only available as daily ‘snapshots’ taken on each pass of the satellite. Detection of SO2 by IASI will depend on the brightness temperature difference caused by the SO2, which will be greatest when the SO2 is at a similar altitude to the tropopause. In addition, sensitivity to SO2 is dependent on the altitude of the SO2, which could lead to underestimation of SO2 levels close to the volcano summit. This is particularly important in the case of Eyjafjallajökull where the SO2 is at a lower altitude and there is more uncertainty in the overlying temperature and humidity fields. The six week duration of this eruption, combined with the lifetime of SO2, further increase the difficulty in estimating the source strength as the relationship between the SO2 present in the atmosphere and the release rate is more complex than for shorter eruptions. The uncertainties in observations should also be considered when comparing observations with the modeled results. In the case of Eyjafjallajökull, the lower SO2 levels involved compared to those from Sarychev and Kasatochi could make small discrepancies between the model and observations more apparent, while measurement of small values of SO2 can result in large relative errors in the observed values.

[32] Due to lack of observations of SO2 plume rise heights, these were inferred from those observed for volcanic ash. In reality, however, these heights may be different and this assumption could introduce significant errors in the transport of both SO2 and sulphate aerosol [Prata and Kerkmann, 2007]. This is particularly true for prolonged eruptions such as that of Eyjafjallajökull, during which the ash plume rise height varied significantly over the course of the six weeks. Plume height can dramatically affect dispersion, with material moving in one direction at one height and in a different direction at another height, amplifying small errors in plume height into larger errors in distribution In the case of Eyjafjallajökull, the SO2 release was simplified by assuming a constant height throughout the eruption, although the release rate was varied according to observations. In the cases of both Sarychev and Kasatochi, SO2 was assumed to be released at a constant rate and between fixed heights for the durations of these much shorter eruptions. It is possible that varying both the SO2 release rate and height with time would give results that compare more favorably with observations; this should be investigated in future simulations. Although NAME output has been used for inversion modeling using surface observations [Manning et al., 2011; Reimann et al., 2005], the capability of estimating height and time varying sources using satellite retrievals is still under development.

[33] The cut-off of the model domain at 85°N resulted in some loss of both SO2 and sulphate from the simulation because model particles are lost when they leave the model domain, and is a source of inaccuracy in the results. In addition, the model output gives a time average over the preceding 1, 12 or 24 h while the observations are either snapshots at a single time or averages of two snapshots taken 12 h apart. This, combined with different vertical extents and averaging for the models and observations, is expected to result in differences when comparing the two.

[34] The chemistry scheme used throughout the vertical domain was designed to model tropospheric chemistry and, as a result, chemistry in the stratosphere is not optimally represented. The sensitivity tests carried out show that increasing O3 increases the production of sulphate aerosol while increasing OH or HO2 decreases production of sulphate aerosol. Additionally, volcanic sulphate aerosol can lead to O3 depletion by providing a surface for chlorine-liberating reactions [Krotkov et al., 2010] which are not included in the simulations used here. Using decreased O3 concentrations may give a sulphate aerosol production rate more comparable to that observed when a large proportion of the total SO2 is released into the stratosphere. More accurate modeling of stratospheric O3, as well as a more interactive chemistry scheme that includes the effects of oxidant depletion, would benefit future simulations. However, this must be traded off against increased run-time and computing resources.

[35] The loss of sulphate aerosol from the NAME model simulations also proceeds at a greater rate than observed. This could in part be caused by the loss of particles from the model domain poleward of 85°N but could also result from the slowing or premature termination of sulphate production, so that the sulphate that is lost through wet and dry deposition is not replenished. The modeled rate of sulphate loss is higher when NAME produces greater amounts of sulphate; this is expected because both wet and dry deposition rates are proportional to sulphate concentration.

[36] NAME takes the cloud fraction solely from the meteorological data and does not explicitly model water vapor resulting from the eruption, due to either melting snow and ice or emission of water vapor by the volcano. It is possible that this water vapor may be present in the Unified Model as a result of the satellite data assimilation process [Rawlins et al., 2007; Hilton et al., 2009]. However, the presence of volcanic ash may cause the satellite data to be rejected by quality control. The absence of water vapor would prevent the aqueous chemistry involved in the conversion of SO2 to sulphate from taking place. As the Sarychev and Kasatochi eruptions released SO2 mainly into the stratosphere where there is little moisture, gas phase chemistry is dominant. However, water vapor associated with the eruptions, as well as that entering the rising plume by entrainment, would also enable aqueous phase reactions to take place. As this water vapor is not directly included in the simulation these aqueous phase reactions, and the resulting initial pulse of sulphate production, may not be reproduced. Lack of water vapor would also affect wet deposition at low altitudes close to the summit, where sulphate aerosol could act as cloud condensation nuclei and be absorbed into the cloud droplets as they form. This would cause modeled pollutants to remain in the atmosphere for longer than they would otherwise.

[37] NAME results compare favorably with available observations in terms of both geographical distribution and magnitude. Table 1 shows the Pearson correlation and root-mean square error for the comparison of NAME modeled SO2 and sulphate aerosol with observations. The comparison is carried out using mean hemispheric values at the times when nonzero observations are available. IASI retrievals are used for comparison with modeled SO2 while OSIRIS retrievals are used for comparison with modeled sulphate aerosol. The low Pearson correlation coefficient for sulphate AOD from Sarychev is in part caused by the premature peak and decline of modeled sulphate aerosol. NAME modeled SO2 column densities compare well with the corresponding observations above the IASI detection limit for Sarychev, with a Pearson correlation of 0.8, but the SO2 persists for longer in NAME. 90% of modeled values of northern hemisphere averaged sulphate AOD are within a factor of 2 of those observed for Kasatochi and 71% are within a factor of 2 of those observed for Sarychev, despite significant uncertainties in both the model and observations. The use of a chemistry scheme designed for use in the troposphere is a likely cause of some of the discrepancies between modeled and observed sulphate AODs, as is the absence of directly modeled water vapor associated with the volcanic eruptions, and the consequential reduced aqueous processes. In addition, NAME does not presently include any microphysical aerosol processes. This omission is a further source of inaccuracy but, as is the case with oxidant depletion, its inclusion must be traded-off against increased run-time and computing resources.

Table 1. Linear Pearson Correlation and Root-Mean Square Error for Modeled Hemispheric Mean SO2 and Sulphate AOD (×10) for Sarychev and for Modeled Hemispheric Mean Stratospheric Sulphate AOD (×10) for Kasatochi (Using Profile 1)
 Pearson CorrelationRoot-Mean Square Error
Sarychev SO20.830.056
Sarychev sulphate AOD (×10)0.320.014
Kasatochi sulphate AOD (×10)0.920.002

7. Conclusions

[38] In this study, a chemistry scheme primarily designed for modeling regional formation and transport of atmospheric pollutants (for example ozone, sulphate and nitrate aerosol) over Europe has been applied to modeling the transformation of sulphur dioxide to sulphate in a volcanic plume for the first time. The model has shown very encouraging results, however it is likely that some of the model chemistry parameterizations are not particularly suited to the volcanic situation and that improvements to the model's ability to predict volcanic sulphate could be achieved by developing a specific scheme. For example the use of modeled pH in the aqueous phase calculations is not likely to be at all representative of what is actually occurring in a volcanic plume, and a simpler approach may be more appropriate.

[39] This work shows that NAME is able to model the transport of SO2 and sulphate from volcanoes and that the chemistry scheme shows promise as a tool for modeling the production of sulphate aerosol in the UTLS. Modeling ability could be further improved by varying the SO2 release rate and height with time and by more accurately modeling any aqueous production of sulphate immediately after the eruption, among other things. It would also be valuable in future to modify the chemistry scheme and background fields to better simulate stratospheric processes, and to reduce computing costs by optimizing the chemistry scheme for sulphur chemistry.

Appendix A

[40] For the Eyjafjallajökull simulation, the SO2 release rates were estimated from IASI retrievals assuming a SO2 altitude of 10 km asl. The estimated release rates are shown in Table A1.

Table A1. SO2 Source Strength Based on IASI Retrievals Assuming a SO2 Altitude of 10 km asl
Release Rate (Gg h−1)Start Time (UTC)Stop Time (UTC)
0.09070567615 Apr 2010 12:0016 Apr 2010 00:00
0.03975810118 Apr 2010 12:0019 Apr 2010 00:00
0.04542060919 Apr 2010 00:0019 Apr 2010 12:00
0.07871129321 Apr 2010 00:0021 Apr 2010 12:00
0.04725120222 Apr 2010 12:0023 Apr 2010 00:00
0.0666252423 Apr 2010 00:0023 Apr 2010 12:00
0.08473590823 Apr 2010 12:0024 Apr 2010 00:00
0.10026491124 Apr 2010 00:0024 Apr 2010 12:00
0.01558515928 Apr 2010 12:0029 Apr 2010 00:00
0.09176035730 Apr 2010 12:0001 May 2010 00:00
0.09981538601 May 2010 00:0001 May 2010 12:00
0.11512482705 May 2010 12:0006 May 2010 00:00
0.56092626606 May 2010 00:0006 May 2010 12:00
1.55680795806 May 2010 12:0007 May 2010 00:00
0.24048327607 May 2010 00:0007 May 2010 12:00
0.61841075907 May 2010 12:0008 May 2010 00:00
0.17927683708 May 2010 00:0008 May 2010 12:00
1.07314337108 May 2010 12:0009 May 2010 00:00
0.58557315809 May 2010 00:0009 May 2010 12:00
0.60136504209 May 2010 12:0010 May 2010 00:00
0.29170288710 May 2010 12:0011 May 2010 00:00
0.10889716711 May 2010 00:0011 May 2010 12:00
0.12217386111 May 2010 12:0012 May 2010 00:00
0.31773649712 May 2010 00:0012 May 2010 12:00
0.6817998512 May 2010 12:0013 May 2010 00:00
0.90677069213 May 2010 00:0013 May 2010 12:00
1.1900569713 May 2010 12:0014 May 2010 00:00
2.20050792814 May 2010 00:0014 May 2010 12:00
2.25430378614 May 2010 12:0015 May 2010 00:00
3.30727966815 May 2010 00:0015 May 2010 12:00
3.52547109915 May 2010 12:0016 May 2010 00:00
2.83760807716 May 2010 00:0016 May 2010 12:00
2.85616584916 May 2010 12:0017 May 2010 00:00
2.12503651917 May 2010 00:0017 May 2010 12:00
2.41222777117 May 2010 12:0018 May 2010 00:00
2.76844613918 May 2010 00:0018 May 2010 12:00
1.28513709718 May 2010 12:0019 May 2010 00:00
0.15331593319 May 2010 00:0019 May 2010 12:00
0.01287216622 May 2010 00:0022 May 2010 12:00

Acknowledgments

[41] This work was funded by the Civil Aviation Authority. L. Clarisse is a Postdoctoral Researcher with the Belgian F. R. S.-FRNS. The authors would like to thank Réal D'Amours (Canadian Meteorological Centre) and Sarah Millington (UK Met Office) for constructive input and Simon Carn (UMBC) for making the OMI data from the Kasatochi eruption available.

Ancillary