Assessing hazards to aviation from sulfur dioxide emitted by explosive Icelandic eruptions

Volcanic eruptions take place in Iceland about once every 3 to 5 years. Ash emissions from these eruptions can cause significant disruption to air traffic over Europe and the North Atlantic as is evident from the 2010 eruption of Eyjafjallajökull. Sulfur dioxide (SO2) is also emitted by volcanoes, but there are no criteria to define when airspace is considered hazardous or nonhazardous. However, SO2 is a well-known ground-level pollutant that can have detrimental effects on human health. We have used the United Kingdom Met Office’s NAME (Numerical Atmospheric-dispersion Modelling Environment) model to simulate SO2 mass concentrations that could occur in European and North Atlantic airspace for a range of hypothetical explosive eruptions in Iceland with a probability to occur about once every 3 to 5 years. Model performance was evaluated for the 2010 Eyjafjallajökull summit eruption against SO2 vertical column density retrievals from the Ozone Monitoring Instrument and in situ measurements from the United Kingdom Facility for Airborne Atmospheric Measurements research aircraft. We show that at no time during the 2010 Eyjafjallajökull eruption did SO2 mass concentrations at flight altitudes violate European air quality standards. In contrast, during a hypothetical short-duration explosive eruption similar to Hekla in 2000 (emitting 0.2 Tg of SO2 within 2 h, or an average SO2 release rate 250 times that of Eyjafjallajökull 2010), simulated SO2 concentrations are greater than 1063μg/m for about 48 h in a small area of European and North Atlantic airspace. By calculating the occurrence of aircraft encounters with the volcanic plume of a short-duration eruption, we show that a 15 min or longer exposure of aircraft and passengers to concentrations ≥500μg/m has a probability of about 0.1%. Although exposure of humans to such concentrations may lead to irritations to the eyes, nose and, throat and cause increased airway resistance even in healthy individuals, the risk is very low. However, the fact that volcanic ash and sulfur species are not always collocated and that passenger comfort could be compromised might be incentives to provide real-time information on the presence or absence of volcanic SO2. Such information could aid aviation risk management during and after volcanic eruptions.


Introduction
Icelandic volcanism features almost all known eruption styles and types, ranging from purely effusive, ash-poor eruptions to purely explosive, ash-dominated eruptions [e.g., Thordarson and Larsen, 2007;Larsen and Eiríksson, 2008]. The 2010 explosive eruption of Eyjafjallajökull (63. 38°N, 19.36°W, 1660 m above sea level (asl)) began on 14 April 2010 and lasted 39 days resulting in severe disruption to air traffic due to the repeated presence of ash plumes in European and North Atlantic airspace [e.g., Schumann et al., 2011;Gudmundsson et al., 2012;Stevenson et al., 2012]. The eruption's impacts quickly reached global scale, affecting several industry sectors [e.g., Budd et al., 2010;Mazzocchi et al., 2010;Donovan and Oppenheimer, 2011;Harris et al., 2012] despite its low magnitude of explosivity (volcanic explosivity index of 3) and a total tephra volume of 0.18 km 3 (dense rock equivalent) . Controlled European airspace was restricted for commercial air traffic during 15-23 April 2010 followed by intermittent restrictions of parts of European airspace in the weeks thereafter. These restrictions resulted from the combination of frequent and persistent northwesterly air flow at the altitude at which significant amounts of fine-grained volcanic ash particles were injected (3-10 km) and the aviation safety protocols in place at the time (i.e., "zero tolerance") [e.g., Gudmundsson et al., 2012].
Volcanic plumes not only consist of ash but also contain species such as sulfur dioxide (SO 2 ) and hydrogen sulfide (H 2 S). Generally, SO 2 is considered hazardous to aircraft frames and engines only following its oxidation to sulfuric acid, which upon hydration forms sulfuric acid aerosol particles [Bernard and Rose, 1990], typically composed of 75 wt % H 2 SO 4 and 25 wt % H 2 O [e.g., Hamill et al., 1977]. Sulfuric acid aerosol particles can cause damage to the airframe and/or engine components as a result of sulfidation, leading to corrosion of nickel alloys in gas turbines if alkali metal salts are copresent (e.g., sea-salt, mineral dust) [Eliaz et al., 2002]. Sulfuric acid particles have the potential to cause corrosion and erosion of compressor blades and other engine components [Swadźba et al., 1993]. This may necessitate more frequent maintenance cycles for aircraft and therefore result in increased total cost of ownership.
A few aircraft encounters with volcanic plumes containing SO 2 and/or sulfuric acid aerosol have been documented [Bernard and Rose, 1990;Casadevall et al., 1996]. These incidences include reports of distress to pilots and passengers during encounters with the 2011 Grímsvötn volcanic SO 2 plume [European Space Agency, 2011] and the Sarychev 2009 plume [Guffanti et al., 2010], but there are no reported cases of damage to aircraft engines or avionics due to encounters with volcanic SO 2 plumes. In contrast to volcanic ash hazards to aviation, there are no criteria to define when airspace is considered hazardous or nonhazardous for volcanic SO 2 plumes. There is, however, International Civil Aviation Organization guidance for flight crew upon noticing the smell of sulfur in an aircraft cabin, and it is recommended that the pilot quickly contact air traffic control and the airline operations center for any information about relevant volcanic activity [International Civil Aviation Organization, 2004;International Volcanic Ash Task Force (IVATF), 2012].
At present, none of the Volcanic Ash Advisory Centres (VAACs) worldwide is required to forecast the dispersion or concentration of volcanic SO 2 or sulfuric acid aerosol, although VAACs have started to utilize satellite retrievals of volcanic SO 2 to inform their volcanic ash forecasts [Tupper et al., 2004;Brenot et al., 2014]. In future, the requirements to monitor and forecast SO 2 might change for several reasons. Carn et al. [2009] discussed that monitoring and tracking volcanic SO 2 plumes can be a useful proxy for discernable volcanic ash although ash and SO 2 are not necessarily collocated [e.g., Schumann et al., 2011;Thomas and Prata, 2011;Sears et al., 2013]. Carn et al. [2009] also discussed that cumulative effects of multiple exposure of aircraft to long-lived and aged volcanic ash and aerosol plumes may result in increased cost of ownership. Furthermore, sulfurous odors can cause distress of cabin passengers and aircrew [Guffanti et al., 2010].
Sulfur dioxide has been measured in volcanic plumes at concentrations much higher than those considered toxic to humans [e.g., Baxter, 2000;Rose et al., 2003;Hansell and Oppenheimer, 2004;Hunton et al., 2005;Voigt et al., 2014]. Many of these measurements were taken at ground level and in close proximity to the volcanic vent, but in situ measurements reported by Rose et al. [2003] suggest SO 2 concentrations of 500-1000 ppbv during an aircraft encounter with a 35 h old volcanic plume from the Icelandic Hekla eruption in February 2000 at about 1300 km distance from the source. For context, humans may perceive SO 2 at about 300 ppbv, and the World Health Organization (WHO) sets a 10 min mean ambient air quality standard of 500 μg/m 3 (about 190 ppbv at ground level) to protect public health [World Health Organization (WHO), 2014]. International Volcanic Ash Task Force, which was established following the 2010 Eyjafjallajökull eruption recommended that for volcanic SO 2 , further "work on identifying and quantifying any associated health risks to aircraft occupants […] with a view to enhancing the existing guidance […]" was needed [IVATF, 2012]. This area of research has now been taken up by the International Airways Volcano Watch Operations Group [e.g., International Airways Volcano Watch Operations Group, 2014].

10.1002/2014JD022070
We use the United Kingdom Met Office's NAME (Numerical Atmospheric-dispersion Modelling Environment) model [Jones et al., 2007] to simulate SO 2 mass concentrations for a range of hypothetical explosive Icelandic eruptions. Our aim is to inform the aviation industry and regulators on the likely SO 2 mass concentrations that aircraft could encounter over Europe and the North Atlantic. The eruption case studies span eruption magnitudes, durations, and SO 2 fluxes that are representative of past explosive eruptions in Iceland with a probability to occur about once every 3 to 5 years. The eruption source parameters and the probability are derived from the record of volcanism in Iceland over the last 1150 years [Thordarson and Larsen, 2007] and further supported by the Holocene eruption record in Iceland [Larsen and Eiríksson, 2008;Óladóttir et al., 2008;Thordarson and Höskuldsson, 2008;Óladóttir et al., 2011].

Model Setup and Definition of the Eruption Case Studies
The NAME model is a Lagrangian dispersion model [Jones et al., 2007], which is used for a range of applications such as air quality modeling and forecasting and emergency response work predicting the dispersion of hazardous nuclear, volcanic, chemical, or biological material [e.g., Webster et al., 2006;Derwent et al., 2007;Redington et al., 2009;Leadbetter and Hort, 2011;Leadbetter et al., 2012]. NAME is also used by the London VAAC to forecast volcanic ash dispersion operationally Witham et al., 2012]. Previously, Heard et al. [2012] used NAME to simulate SO 2 and sulfuric acid aerosol mass concentrations from the eruptions of Kasatochi in 2008, Sarychev in 2009, and Eyjafjallajökull in 2010. These authors found good agreement between satellite-retrieved and model-simulated SO 2 column densities and sulfuric acid aerosol optical depths, which gives confidence in the model's ability to capture not only atmospheric dispersion of volcanic gases and aerosol particles but also their atmospheric column densities.
To provide atmospheric SO 2 mass concentration estimates for a range of explosive volcanic eruptions in Iceland, we simulate (1) Eyjafjallajökull 2010, which is representative of a long-duration explosive eruption with sustained activity for 39 days releasing 0.39 Tg of SO 2 in total, (2) a hypothetical short-duration (2 h) explosive volcanic eruption releasing 0.2 Tg of SO 2 by mimicking the eruption of Hekla in 2000 in Iceland, and (3) a hypothetical intermediate-duration eruption releasing a total of 0.1 Tg of SO 2 over 3 days by mimicking the eruption of Grímsvötn in 2004 (Table 1). The SO 2 release rates for the long-duration and intermediate-duration eruptions vary over the course of the eruption with average hourly release rates of 0.0004 Tg and 0.0018 Tg (Table 1; see also Tables S1 and S2 in the supporting information). For the short-duration eruption, the hourly SO 2 release rate is 0.1 Tg. The SO 2 mass fluxes for the long-duration and the intermediate-duration eruption scenarios are given in Tables S1 and S2 in the supporting information.
Journal of Geophysical Research: Atmospheres

10.1002/2014JD022070
The eruption scenarios we define are generic but representative of small to moderate magnitude explosive eruptions in Iceland [Thordarson and Larsen, 2007;Óladóttir et al., 2008;Thordarson and Höskuldsson, 2008;Óladóttir et al., 2011]. The scenarios are not tied to a specific volcano or a specific volcanic system in Iceland because the exact location, magnitude, style, and composition of the next eruption in Iceland are not foreseeable. For example, the next Hekla eruption may be of a smaller or larger magnitude than simulated here and could produce magma of basaltic, intermediate or silicic composition. Since the total mass of SO 2 released is related to magma composition and eruption size (i.e., magma volume), the SO 2 mass flux to the atmosphere may be significantly higher or lower than simulated in our study.
To simulate a long-duration eruption, we use daily varying SO 2 mass fluxes derived from the petrological analysis of the sulfur content in melt inclusions and quenched eruption products for the 2010 summit eruption at Eyjafjallajökull . The SO 2 mass fluxes are used together with daily plume height data obtained from radar measurements in Keflavik, Iceland [Arason et al., 2011;Gudmundsson et al., 2012], as inputs for NAME ( Figure 1). The daily maximum plume height represents the top height of the SO 2 release, and in the model, SO 2 is released uniformly between the Eyjafjallajökull volcano summit (1666 m asl) and the plume top. To account for in-plume scavenging of SO 2 as observed during the early phase of Eyjafjallajökull 2010, we reduced the SO 2 mass flux used in the model to 0.1% of the petrological flux estimate (blue stars, Figure 1). Figure 1 additionally shows SO 2 mass fluxes derived by Flemming and Inness [2013] using inverse modeling and GOME-2 (Global Ozone Monitoring Experiment-2) satellite retrievals for the period 1 to 19 May 2010. For the period 6 to 12 May 2010 the SO 2 flux derived using GOME-2 is up to a factor of 5.9 greater than the SO 2 mass flux we derived using the petrological method ( Figure 1). These differences demonstrate that satellite-derived fluxes and petrological fluxes can differ substantially, which contributes directly to uncertainties in the model to data comparison. Eyjafjallajökull eruption derived using the petrological method  and daily maximum plume height (solid grey line and axis labels on the right, in kilometer above sea level) used in the model to simulate the long-duration eruption scenario. Blue stars indicate that in the model, the SO 2 mass flux was reduced to 0.1% of the petrological flux estimate in order to account for in-plume scavenging of sulfur species as observed during the early phases of Eyjafjallajökull 2010. Orange squares are SO 2 mass fluxes derived using GOME-2 retrievals in combination with inverse modeling [Flemming and Inness, 2013].

10.1002/2014JD022070
For all simulations reanalyzed meteorological fields (ERA-interim) from the European Centre for Medium-Range Weather Forecasts [Dee et al., 2011] are used for the period 14 April 2010 to 30 May 2010. We used the same meteorological data for all eruption scenarios, and each eruption commenced midday on 14 April 2010 because under these meteorological conditions SO 2 will be dispersed into European and North Atlantic airspace [e.g., Petersen et al., 2012]. All simulations are run accounting for oxidation of SO 2 as well as dry and wet removal of SO 2 and its oxidation products [Webster andThomson, 2011, 2014]. Details of the NAME chemistry scheme are described in Heard et al. [2012] and Redington et al. [2009]. Briefly, we simulate the gas-phase reaction of SO 2 with the hydroxyl radical (OH), which is the dominant sink of volcanic SO 2 in the free troposphere. Aqueous-phase oxidation of SO 2 by hydrogen peroxide and ozone occurs in grid boxes where both liquid water and cloud fraction are nonzero [Redington et al., 2009;Heard et al., 2012]. Mass concentrations of SO 2 are output as 15 min means on a regular longitude-latitude grid of 0.5°by 0.5°.
We apply a similar procedure as the London VAAC applies for volcanic ash to our analysis of SO 2 mass concentrations at flight altitudes in European and North Atlantic airspace by outputting mass concentrations for 2500 ft deep layers [see also Webster et al., 2012;Witham et al., 2012]. Flight levels (FLs) are barometric pressures expressed as a nominal altitude in 100 s of feet. For the analysis, we combined the 2500 ft deep output to three FL ranges by finding the maximum mass concentration in the following FL ranges: FL000-FL200, FL200-FL350, and FL350-FL550. The cruising altitude of commercial aircraft on an intercontinental flight lies between 20,000 ft and 35,000 ft, equivalent to FL200-FL350. The cruising altitude on transatlantic flights typically lies between 35,000 ft and 45,000 ft (FL350-FL450). FL000-FL200 is relevant for low-altitude cruising and for takeoff and landing.

Ozone Monitoring Instrument Satellite Retrievals and In Situ Measurements of SO 2
The retrieval of SO 2 vertical columns densities (VCDs, in Dobson units, DU) from Ozone Monitoring Instrument (OMI) is achieved by applying Differential Optical Absorption Spectroscopy (DOAS) [Platt and Stutz, 2008] to the ultraviolet measured spectra in the 312-326 nm wavelength range. An empirical offset Map showing the five representative flight paths (black lines) that are used to calculate the probability of 15 min or longer periods that an aircraft could be exposed to volcanic sulfur dioxide. These paths are defined so that they pass through the volcanic plume of each eruption scenario.

10.1002/2014JD022070
correction is then applied to account for possible biases and is estimated over clean regions. The obtained quantity is converted into an SO 2 VCD using an air mass factor that accounts for changes in measurement sensitivity due to observation geometry, ozone column, clouds, and surface reflectivity. The SO 2 VCD is estimated with the assumption of a plume height at 7 km above sea level and is provided along with the column operator, which accounts for the altitude-dependent sensitivity of the OMI measurements. Figure S1 shows that the SO 2 VCDs retrieved using the DOAS algorithm are in close agreement with those retrieved using the operational OMI algorithm described in Krotkov et al. [2006].
To compare simulated SO 2 VCDs with those retrieved by OMI, vertical model output resolutions of 100 m in the first 1 km and 500 m between 1 km and 16 km have been used. In order to directly compare model output and satellite retrievals, the OMI column operators (accounting for instrument sensitivity in the vertical) have to be applied to the model profiles by taking each OMI pixel and finding the corresponding model profile which is coincident in both time (within the model 15 min mean time window) and space (within a model grid box). The model profile is then interpolated to the OMI vertical retrieval grid, resulting in a simulated VCD that can be directly compared to OMI.

Calculation of Aircraft Encounters With Volcanic SO 2 Along Flight Paths
To calculate the probability of 15 min or longer periods that an aircraft could be exposed to SO 2 , we define five flight paths that are representative of actual flight paths along great circles in North Atlantic and European airspace ( Figure 2). These flight paths are defined so that they pass through the volcanic plume of each eruption scenario. For simplicity, we assume that the aircraft flies with a constant ground speed of 700 km/h along these flight paths. The probability of aircraft transiting through or aircraft occupants being exposed to volcanic SO 2 concentrations ≥ X i for 15 min or longer is given as the mean probability (in percent) for the five flight paths with the sample space defined by the analysis period considered for each eruption scenario (Table 1).

Model Evaluation for the 2010 Eyjafjallajökull Eruption
To evaluate the ability of NAME to simulate SO 2 VCDs and the dispersion of SO 2 , we compare simulated SO 2 VCDs for the long-duration Eyjafjallajökull 2010 eruption with satellite retrievals from OMI. Additionally, we compare simulated mass mixing ratios of SO 2 to in situ measurements from the United Kingdom's Facility for Airborne Atmospheric Measurements (FAAM) research aircraft Turnbull et al., 2012]. Figure 3 compares the spatial distribution of simulated and measured SO 2 VCDs for 4 days during Eyjafjallajökull 2010 applying the column operators that assume a plume height of 7 km, which best reflects the conditions during the eruption. Qualitatively, the presence or absence of volcanic SO 2 can be captured using NAME as evident from comparing retrieved and simulated SO 2 VCDs. However, NAME underpredicts maximum SO 2 VCDs compared to OMI with a normalized mean bias of À13.6% (Figures 3 and 4a). The relative frequency distributions of all simulated and retrieved SO 2 VCDs greater than 0.5 DU (Figures 4b and  4c) and the relative frequency the bias (NAME minus OMI) further demonstrate that NAME most frequently underpredicts SO 2 VCDs by up to 3 DU compared to OMI (Figure 4d).
The daily maximum SO 2 VCDs retrieved by OMI during April and May 2010 (Figure 4a) are in good agreement with those reported by Rix et al. [2012] using GOME-2. For context, a SO 2 VCD of 10 DU converts to mixing ratios on the order of 300 ppbv when assuming a plume thickness of 1 km.   (Figure 5a). Spatially, Figure 4. Comparison of (a) maximum SO 2 vertical column densities (VCDs in Dobson units, DU) greater than 0.5 DU retrieved by the Ozone Monitoring Instrument (OMI) with maximum NAME VCDs for the period 14 April to 30 May 2010 of the Eyjafjallajökull eruption. Also shown is the normalized mean bias (NMB in percent). For the comparison, the maximum was located anywhere in a given scene (i.e., maximum VCDs shown may not be spatially collocated in NAME and OMI scenes). The relative frequency distribution of all gridded SO 2 VCDs greater than 0.5 are shown (b) for NAME and (c) for OMI. and (d) The relative frequency distribution of the bias (NAME minus OMI) in SO 2 VCDs. May 2010 (grey) with that of simulated SO 2 mixing ratios, which have been sampled along the FAAM flight paths for the same period (colored lines). The colored lines show simulated mixing ratios obtained using different levels of model sampling tolerance with regard to the flight track (blue = exact location in terms of longitude, latitude, and height above sea level; green = maximum SO 2 mixing ratio found within ±2.5°longitude/latitude; brown = maximum SO 2 mixing ratio found within ±2 vertical output levels (±1 km); orange = maximum SO 2 mixing ratio found within ±2.5°longitude/latitude and ±2 vertical output levels).

Journal of Geophysical Research: Atmospheres
10.1002/2014JD022070 simulated SO 2 mixing ratios are in good agreement with FAAM measurements throughout most vertical levels although there are mismatches, which are particularly pronounced for mixing ratios greater than 10 ppbv (Figure 5b). Comparing FAAM aircraft in situ measurements for the period 14-18 May 2010 to the simulated SO 2 mixing ratios, it is apparent that maximum SO 2 mixing ratios are not captured by the model, although in situ measurements of SO 2 mixing ratios greater than 40 ppbv represent less than 2% of all measurements (Figure 5c). Mixing ratios below 40 ppbv are captured by the model if up to ±2.5°tolerance in terms of longitude and latitude and/or ±2 vertical levels (±1 km) tolerance is used with respect to the location of the in situ measurements (Figure 5c, orange and green lines).

Sulfur Dioxide Mass Concentrations at Flight Altitude and Threshold Exceedances
Based on the UK Daily Air Quality Index defined by the Department for Environment, Food and Rural Affairs (DEFRA), 15 min mean ground-level SO 2 concentrations greater than 265 μg/m 3 would result in a public health warning being issued [Connolly et al., 2013]. The WHO health protection guideline for SO 2 mass concentrations is set at 500 μg/m 3 for an averaging period of  14,189 10 min [WHO, 2014]. On a 15 min mean basis, SO 2 concentrations greater than 531 μg/m 3 and 1063 μg/m 3 are considered "high" and "very high" pollution by DEFRA. Figure 6 shows the spatial distribution of maximum 15 min mean SO 2 mass concentrations simulated at any time for each eruption case study in the three flight altitude ranges (see section 2 and Table 1 for details). For the long-duration Eyjafjallajökull 2010 eruption, we find that maximum SO 2 mass concentrations do not  (Figure 6a), except in a small region around Iceland. In contrast, during a short-duration explosive eruption emitting 0.2 Tg of SO 2 over a 2 h period (Table 1), maximum SO 2 mass concentrations exceed 1063 μg/m 3 in FL200-FL350 and FL350-FL550 (Figure 6c). The 1063 μg/m 3 threshold is exceeded as far downwind of the volcanic vent as~3100 km for a period of just over 2 days (Figure 7c). Overall, during a short-duration explosive eruption, volcanic SO 2 (with concentrations > 0.1 μg/m 3 ) is present in a relatively small area of airspace of about 0.6 million km 2 out of about 100 × 10 6 km 2 of total area considered in the model domain. For this eruption scenario, SO 2 concentrations exceed 265 μg/m 3 for at least 3 days in FL200-FL350 and FL350-FL550, which based on the UK DEFRA air quality legislation would result in a public health warning being issued if exceeded at ground level [Connolly et al., 2013]. For the intermediate-duration eruption (emitting a total of~0.1 Tg of SO 2 over 3 days), SO 2 mass concentrations are greater than 531 μg/m 3 in FL200-FL350 up to about 1900 km from the volcanic vent, and the 265 μg/m 3 threshold is exceeded as far downwind as 2900 km during the first 2 days of such an eruption (Figures 6b and 7c).   Figure 2) for (a) the long-duration eruption, (b) the intermediate-duration eruption, and (c) the short-duration eruption scenario. The sample space used to calculate the probabilities is defined by the analysis period considered for each eruption scenario (Table 1).

10.1002/2014JD022070
The SO 2 concentrations simulated for each eruption scenario can be compared with in situ measurements made during the Hekla eruption in the year 2000. Rose et al. [2003] measured SO 2 volume mixing ratios up to 1000 ppbv at 11.3 km altitude (using the U.S. 1976 standard atmospheric profile, 1000 ppbv is equivalent to about 770 μg/m 3 at about 11 km), which is of the same order as the maximum SO 2 mass concentrations of about 1050 μg/m 3 that we find in the 36 h old plume in FL350-FL550 for the short-duration eruption.

Probability of 15 min or Longer Exposure to SO 2 Along Flight Paths
For the long-duration Eyjafjallajökull 2010 eruption, we find that long continuous encounters of high SO 2 concentrations are unlikely. We find that the maximum concentration that an aircraft would encounter for 15 min or longer is 120 μg/m 3 (see section 2 and Figures 2 and 8). Figure 8 shows that there is a~40% probability of exposure for 15 min or longer to concentrations ≥0.01 μg/m 3 and of about 5% of exposure to concentrations ≥5 μg/m 3 along the flight paths in FL000-FL200 (Figure 8a). For the intermediate-duration eruption, there is a 0.7% probability of exposure for 15 min or longer to concentrations ≥100 μg/m 3 in FL200-FL350, and the probability of encountering concentrations ≥500 μg/m 3 is about 0.05% in FL200-FL350 ( Figure 8b). For the short-duration eruption scenario, we find that SO 2 mass concentrations ≥500 μg/m 3 could be encountered for 15 min or longer in FL200-FL350 or FL350-FL550, although the probability of such an encounter is just under 0.1% (Figure 8c). The probability of encountering concentrations ≥ 1000 μg/m 3 for 15 min or longer in FL350-FL550 is about 0.01%.
A study on SO 2 infiltration into aircraft cabins commissioned by the Department for Transport in the United Kingdom suggests that SO 2 concentrations inside a narrow body aircraft (e.g., Boeing 757-300 or Airbus 320) will reach about 80% of the ambient atmospheric concentrations within less than 3 min upon encounter [Chitty and Shipp, 2013]. This is valid for ambient volume mixing ratios that exceed 200 ppbv for SO 2 (equivalent to~182 μg/m 3 at 10 km altitude), and the clearance time required for in-cabin concentrations to return to "zero" is about 12 min [Chitty and Shipp, 2013]. Although the probability of encountering SO 2 concentrations greater than 500 μg/m 3 for 15 min or longer is lower than 0.1% for the short-duration eruption scenario, such an encounter would violate public health protection legislation set for ground-level pollution. This may be relevant even more if the cabin air clearance time is taken into account.

Discussion
A handful of other studies and aviation safety protocols have suggested that more research on volcanic SO 2 may be useful for assessing aviation hazards [Carn et al., 2009;Bonadonna et al., 2012Bonadonna et al., , 2014IVATF, 2012;Rix et al., 2012]. However, data on likely SO 2 concentrations at flight altitude for Icelandic eruptions are scarce and there are no hazard assessments available.
For the Eyjafjallajökull eruption in 2010, we show that the spatial distribution of simulated SO 2 VCDs compare qualitatively well with those retrieved by OMI (Figures 3 and 4). The comparison gives confidence in the suitability of the London VAAC NAME dispersion model as a tool to monitor and to predict the presence or absence of volcanic SO 2 , which may be a useful quantity to aid aviation risk management. However, simulating SO 2 concentrations at exact point locations or integrating concentrations along flight paths (i.e., dosage) is more challenging and subject to large uncertainties, which is a similar issue in quantitative volcanic ash concentration forecasting [e.g., Rauthe-Schöch et al., 2012;Webster et al., 2012].
In order to quantitatively predict SO 2 mass concentrations, several sources of uncertainty have to be taken into account such as the accuracy of the eruption source terms (plume height and SO 2 mass flux), meteorological fields, chemical conversion rates, gas and aerosol removal rates, as well as the limitations of the vertical, horizontal, and temporal model and output resolution ( Figure 5) and the representation of in-plume processes near the source in the model. For our model simulation of the 2010 Eyjafjallajökull eruption, we have accounted for in-plume scavenging of SO 2 during the early phases of the eruption in a very simplistic way by reducing the SO 2 mass flux to the atmosphere to 0.1%. Uncertainties in meteorological fields used in the model appear to be minor given the reasonable representation of the spatial occurrence of the volcanic plume compared to in situ measurements and OMI satellite retrievals. However, in general, the farther away from the volcanic vent, the greater the errors on plume concentrations and plume location tend to become in models (see Boichu et al. [2013] for a detailed discussion). In terms of SO 2 mass fluxes, we Journal of Geophysical Research: Atmospheres 10.1002/2014JD022070 estimate an uncertainty on the petrological SO 2 mass fluxes of about ±20%. When converting satellite SO 2 VCDs to mass fluxes, uncertainty arises from the assumptions made about the chemical conversion rates of SO 2 to form aerosol particles [e.g., Carn et al., 2013;Flemming and Inness, 2013].
There are no aviation safety regulations for volcanic SO 2 ; therefore, we have used air quality and health protection guidelines defined for ground-level pollution in order to assess the hazards due to volcanic SO 2 at flight altitude. Based on this approach, we conclude that at no time during the long-duration Eyjafjallajökull eruption in 2010 did 15 min mean SO 2 mass concentrations in the far field (≥1000 km away from the volcanic source) significantly exceed air quality standards set to protect public health even under the assumption that NAME underpredicts maximum SO 2 concentrations. However, in a small area (less than 200 km 2 ) over and around Iceland, SO 2 concentrations exceeded 265 μg/m 3 on several days in our model (Figures 6a and 7a), which based on the UK DEFRA air quality legislation would result in a public health warning being issued [Connolly et al., 2013]. In contrast, during a short-duration explosive eruption with an eruptive phase lasting 2 h (Table 1, with a SO 2 release rate 250 times the average release rate of the long-duration Eyjafjallajökull 2010 eruption scenario), SO 2 mass concentrations greater than 1063 μg/m 3 occur for about 48 h up to about 3500 km away from the volcanic vent ( Figure 7c). Concentrations greater than 1063 μg/m 3 are considered "very high pollution" by DEFRA if exceeded at ground level [Connolly et al., 2013]. For the short-duration eruption scenario, SO 2 concentrations farther than~1500 km from the volcanic vent reach up to~3000 μg/m 3 in both FL200-FL350 and FL350-FL550 (Figure 6c).
On the basis that in-cabin concentrations reach 80% of the ambient atmospheric concentrations within three minutes and about 12 min of cabin air clearance time [Chitty and Shipp, 2013], our results suggest that passengers in an aircraft encountering volcanic SO 2 from a short-duration eruption as simulated here could be exposed to concentrations greater than 500 μg/m 3 for 15 min or longer. Although exposure of humans to SO 2 concentrations greater than ≥500 μg/m 3 for 10 min or longer may lead to irritations to the eyes, nose and throat, and cause increased airway resistance even in healthy individuals [Baxter, 2000;World Health Organization, 2000;EPA, 2008;International Volcanic Health Hazard Network (IVHHN), 2014;WHO, 2014], the probability of such a plume encounter in North Atlantic airspace is as low as 0.1% (Figure 8c). Most likely, such an encounter would result in passenger comfort being compromised (rather than in detrimental health effects) since individuals may perceive sulfur odors at these concentrations.

Summary and Implications
At present, there is no defined level of susceptibility of aircraft, avionics or aircraft passengers to volcanic SO 2 , which limits the assessment of the hazards and risks. If aviation regulator and airline operators consider information on the presence or absence of SO 2 as useful for aiding aviation risk mitigation, then models like NAME could be used to provide these information. Airborne volcanic ash clearly poses a greater threat to aircraft operations [e.g., Guffanti et al., 2010] than volcanic SO 2 for those eruption scenarios we have assessed. However, the fact that volcanic ash and sulfur species are not always collocated [e.g., Schumann et al., 2011;Thomas and Prata, 2011;Sears et al., 2013] or that passenger comfort could be compromised might be further incentives to forecast SO 2 concentrations operationally in addition to volcanic ash. Probabilistic assessment of the SO 2 hazards from explosive volcanism in Iceland similar to those carried out for volcanic ash [Biass et al., 2014] would be possible if regulators and the aviation industry were to define levels of susceptibility.
Quantitative forecasting of volcanic SO 2 concentrations or their integrated quantities along flight paths (i.e., dosage) would be much more challenging than merely providing information on the presence or absence of volcanic SO 2 . This is mainly because accurate eruption source terms at high temporal resolution (SO 2 mass flux, height of the emissions, and eruption characteristics such as near-source scavenging of sulfur species) are required as model inputs. Providing near-real-time data on the SO 2 flux is challenging, and we have shown that there can be significant differences between satellite-derived source terms and those derived using petrology (see section 2 and Figure 1). These differences are mainly a result of specific eruption characteristics like scavenging of sulfur species in the volcanic plume [e.g., Sigmarsson et al., 2013], but also the fact that satellites retrieve column density instead of fluxes although techniques are developed to convert these into mass fluxes [e.g., Carn  In an operational context, assimilating ground-based, aircraft, and satellite measurements of volcanic SO 2 would help to reduce uncertainties in model forecasts of volcanic SO 2 [Brenot et al., 2013;Flemming and Inness, 2013;Boichu et al., 2014], but these techniques have not yet been fully developed for operational use.
We have assessed the SO 2 hazard for eruption case studies of explosive volcanism in Iceland that are representative of small to moderate magnitude explosive eruptions. These case studies have been defined based on the record of eruptions in Iceland in the last 1150 years [e.g., Thordarson and Larsen, 2007]. From this record, it also is known that there have been at least four long-lasting (months to years) flood lava eruptions that produced magma volumes greater than 1 km 3 . For the 1783-1784 A.D. Laki eruption, for example, the average hourly SO 2 release rate during the first 3 months [Thordarson et al., 1996;Thordarson and Self, 2003] was up to 3 orders of magnitude greater than that during Eyjafjallajökull in 2010. On the basis of our results for explosive eruptions, and previous modeling studies of a Laki-type eruption [Schmidt et al., 2010[Schmidt et al., , 2011Schmidt, 2015], sulfur species from infrequent but very large-scale flood lava eruptions like Laki may present a major challenge for aviation risk management depending on volcanic plume heights, SO 2 release rates, and the duration of volcanic activity. VANAHEIM NE/I015612/1. N.A.D.R. is funded by the NERC's National Centre for Earth Observation (NCEO). K.S.C. is a Royal Society Wolfson Merit Award holder. The data for this paper are available upon request from the corresponding author (Anja Schmidt, a. schmidt@leeds.ac.uk). The data supporting Figure 1 are available in Table S1 in the supporting information.