The eruption of the Eyjafjallajökull volcano, Iceland, in April and May 2010 caused unprecedented disruptions of European air traffic showing that timely monitoring of volcanic ash and SO2 dispersion as well as the corresponding plume heights are important for aviation safety. This paper describes the observations of SO2 and BrO columns in the eruption plume and the determination of the SO2 plume height using the GOME-2 satellite instrument. During the eruptive period in May 2010, SO2 total columns of up to ∼20 DU and BrO columns of ∼7.7 × 1013 molec/cm2 were detected. The BrO/SO2 ratio estimated from the GOME-2 observations of the Eyjafjallajökull eruption varies from 1.1 × 10−4 to 2.1 × 10−4. The SO2 plume heights estimated from the GOME-2 observations on 5 May range from 8–13 km and mostly agree within 1–3 km with visual observations, radar data and modeling results. Furthermore, the GOME-2 SO2 observations are compared with in situ measurements of the DLR Falcon aircraft on 17 and 18 May 2010 and with Brewer instruments at Valentia, Ireland and Hohenpeissenberg, Germany. The SO2 columns derived from the Falcon profile measurements range from 0.6–4.7 DU and the comparison with the GOME-2 measurements shows a good agreement, mainly within 1 DU. The Brewer observations at Hohenpeissenberg also agree well with the GOME-2 measurements with a daily average SO2 column of ∼1.3 DU during the overpass of the SO2 cloud on 18 May, whereas the Brewer instrument at Valentia shows up to 50% higher SO2 columns (∼8 DU) on 11 May.
 The eruption of Eyjafjallajökull volcano (63.63°N, 19.62 °W, 1666 m a.s.l.), southern Iceland, from 14 April to 23 May 2010 caused unprecedented and widespread disruption of European air traffic. The phreatomagmatic eruption produced high amounts of fine-grained ash [Petersen, 2010] that were transported by the prevailing winds directly toward Europe.
 Prior to this eruption only 3 minor eruptions occurred at Eyjafjallajökull during historical times, the last eruption taking place in 1821 (Global Volcanism Program, information on Eyjafjallajökull, 2011, www.volcano.si.edu/world/volcano.cfm?vnum=1702-02=). The eruption in spring 2010 started with a small flank eruption on 20 March. On 14 April an eruption began from the summit of Eyjafjallajökull producing a 5–9 km high eruption column [Gudmundsson et al., 2010]. The prevailing winds carried the eruption plume to the south and southeast toward Europe, resulting in an uninterrupted closure of most of the northern European air space from 15 April to 21 April. The restrictions were only lifted through the introduction of new preliminary guidelines concerning the allowed concentration of volcanic ash along flight corridors [International Civil Aviation Organization, 2010].
 After April 17 the eruptive intensity dropped by an order of a magnitude and the ash production was reduced. On 5 May the intensity of the eruption amplified again with increased ash and sulfur dioxide emissions and eruption columns of 5–10 km height [Stohl et al., 2011]. This again led to a disruption of aviation in parts of Europe, e.g., over Ireland and Scotland on 4 and 5 May, in Spain, northern Italy, Austria and southern Germany on 9 May and over Ireland and the UK on 16 and 17 May.
 The eruption of Eyjafjallajökull is probably one of the best documented volcanic eruptions of all times as it was closely monitored by many different instruments, from ground stations throughout Europe [e.g., Ansmann et al., 2010; Flentje et al., 2010; Gasteiger et al., 2011], to observations from aircraft [Schumann et al., 2011; Heue et al., 2011] and satellite observations [e.g., Thomas and Prata, 2011]. This paper focuses on satellite observations of volcanic gases, SO2 and BrO, in the Eyjafjallajökull eruption cloud and the estimation of the SO2 plume height using the GOME-2 instrument [Callies et al., 2000; Munro et al., 2006] onboard the MetOp-A satellite. Sulfur dioxide plumes from volcanic eruptions have been monitored by satellite instruments since the Total Ozone Monitoring Spectrometer (TOMS) [Heath et al., 1975] first detected volcanic SO2 from the eruption of El Chichon in 1982 [Krueger, 1983]. Due to the limited sensitivity of TOMS for SO2, the detection was restricted to large SO2 amounts after volcanic eruptions. With the launch of the Global Ozone Monitoring Experiment (GOME) onboard ERS-2 [Burrows et al., 1999] in 1995, the detection limit for SO2 has been greatly improved [Eisinger and Burrows, 1998; Thomas et al., 2005]. However, this instrument has a relatively poor spatial coverage and needs three days for the acquisition of a contiguous global map at the equator. The SCIAMACHY instrument [Bovensmann et al., 1999] onboard ENVISAT launched in 2002 has a better spatial resolution but it needs six days for providing global coverage. The successor of the GOME instrument (GOME-2) onboard MetOp-A launched in 2006 and OMI (Ozone Monitoring Instrument) [Levelt et al., 2006] onboard EOS-Aura launched in 2004 make it possible to monitor volcanic SO2 emissions on a global scale and on a daily basis [Carn et al., 2008; Loyola et al., 2008; Rix et al., 2009]. Recently, the first detection of BrO in a volcanic plume from satellite has been reported using the GOME-2 instrument [Theys et al., 2009].
 Satellite-based observations provide valuable information for detecting and tracking eruption plumes and, therefore, reduce the risk of aircraft encounter with hazardous volcanic clouds [Prata, 2008; Carn et al., 2009]. Although volcanic ash is the main hazard to aviation, SO2 has proven to be an excellent tracer for volcanic eruption clouds, especially if ash detection techniques fail. SO2 is released during volcanic eruptions and is a robust marker for the onset of volcanic activity. The Volcanic Ash Advisory Centers (VAACs), which are the official organizations to issue alerts to airlines and air traffic control centers of the potential danger of volcanic eruption plumes, make use of satellite observations as an essential information source for issuing alerts [Van Geffen et al., 2007]. Additionally, modeling of the SO2 and ash dispersion is an important tool to ensure aviation safety after volcanic eruptions, as satellite and ground-based observations can only provide snapshots of the situation at the time of observation. Moreover, satellite observations can play an important role in validating and correcting model simulations and the integration of satellite data into the models can improve the forecast results.
 Further, the correct determination of the volcanic plume height is a central issue for aviation safety, as well as for air quality and climate studies. For aviation safety, information on the plume height is important to avoid flying through volcanic plumes. As the injection height also determines the further distribution of the volcanic cloud, it is an important input parameter for modeling and forecasting the future development of the plume. The altitude of the volcanic SO2 also plays an important role in determining the climatic effects of an eruption, as sulfate aerosols forming from SO2 can persist in the stratosphere for a long time and lower the global surface temperature, as occurred after the Pinatubo eruption in 1991 [McCormick et al., 1995]. The plume height has traditionally been inferred from direct observations at the volcano or through inverse trajectory modeling [Eckhardt et al., 2008; Kristiansen et al., 2010]. Recently, the direct retrieval of the SO2 plume height from observations of the OMI and GOME-2 instruments has been demonstrated [Yang et al., 2009; Nowlan et al., 2011].
 In this paper we focus on satellite observations of SO2 and BrO inside the Eyjafjallajökull eruption cloud using the GOME-2 instrument. We present first the observations of SO2 and BrO from the GOME-2 instrument for the eruptive period starting on 5 May. The emissions are analyzed and the BrO/SO2 ratio is estimated and compared with the ratio estimated from aircraft observations of the Eyjafjallajökull plume [Heue et al., 2011]. In the second part a new method for the fast estimation of the SO2 plume height from GOME-2 data is described and applied to the Eyjafjallajökull eruption cloud on 5 May 2010. The retrieved plume heights are compared with observations at the volcano, with results of dispersion modeling of the ash plume [Stohl et al., 2011] and with the SO2 plume simulated by the Integrated forecasting system (IFS) of ECMWF (European Centre for Medium-Range Weather Forecasts), as part of the Monitoring Atmospheric Composition and Climate (MACC) project [Hollingsworth et al., 2008]. Afterwards, we present a validation of the GOME-2 SO2 observations with in situ SO2 measurements from the DLR Falcon research aircraft in the Eyjafjallajökull plume over the North Sea on 17 May 2010 and over Germany on 18 May 2010 [Schumann et al., 2011]. Finally, a comparison of the GOME-2 and model SO2 data with Brewer measurements is conducted for the Brewer stations of Valentia, Ireland and Hohenpeissenberg, Germany.
2. Operational SO2 and BrO Retrieval From GOME-2 Measurements
2.1. GOME-2 Instrument and Data Processing
 The second Global Ozone Monitoring Experiment (GOME-2) onboard the MetOp-A satellite is the successor of the GOME instrument on the ERS-2 satellite [Callies et al., 2000]. GOME-2 is a nadir-scanning UV-VIS spectrometer that measures the backscattered radiation from the earth-atmosphere system with a spectral coverage of 240–790 nm and a spectral (FWHM) resolution between 0.26 nm and 0.51 nm. In addition, a direct sun spectrum is recorded once a day. The default ground pixel size is 80 km × 40 km (across-track x along-track) covering a swath width of 1920 km, which allows global coverage at the equator in about 1.5 days. Owing to an optimized movement of the scan mirror, the ground pixel size remains nearly constant over the full scan.
 The operational GOME-2 SO2 and BrO column products are provided by the German Aerospace Center (DLR) in the framework of EUMETSAT's Satellite Application Facility on Ozone and Atmospheric Chemistry Monitoring (O3M-SAF). The first step in the processing chain is the production of calibrated and geolocated level 1 radiances (level 0-to-1 processing). Level 1 products are generated operationally in the Core Ground Segment (CGS) at EUMETSAT headquarters in Darmstadt, Germany. The calibrated and geolocated level-1 radiances are delivered to DLR and other users via the EUMETCast broadcast system approximately 1:45 h after sensing. The operational level 1-to-2 retrieval of the SO2 and BrO column products is performed with the UPAS (Universal Processor for Atmospheric Spectrometers) system, a new generation system for the processing of operational trace gas and cloud property products in near-real time and off-line [Livschitz and Loyola, 2003]. The resulting GOME-2 level 2 products are disseminated via EUMETCast, WMO/GTS and the Internet. In total, less than 15 min are required for level 1 data reception, processing with UPAS and level 2 data dissemination. End users receive the GOME-2 level 2 near-real time products in less than two hours after sensing with a committed service 24 h a day and 365 days of the year. Further, offline and reprocessed GOME-2 level 2 products are provided by DLR on an orbital basis. These can be ordered via the EUMETSAT product navigator (http://navigator.eumetsat.int) or DLR EOWEB systems (http://eoweb.dlr.de). The German National Remote Sensing Data Library [Heinen et al., 2009] stores the GOME-2 SO2 data for enabling long-term monitoring and data reprocessing. Imagery of SO2, other trace gases and cloud properties derived from GOME-2 are freely available at http://atmos.caf.dlr.de/gome2.
2.2. SO2 Retrieval Algorithm
 SO2 columns are retrieved with the GOME Data Processor (GDP) version 4.4 [Loyola et al., 2011a; Valks et al., 2011] from GOME-2 UV backscatter measurements of sunlight using the well established Differential Optical Absorption Spectroscopy (DOAS) method [Platt, 1994]. The DOAS slant column fit of SO2 is performed in the UV wavelength range between 315 and 326 nm [Thomas et al., 2005]. The SO2 cross-section included in the fit is the SCIAMACHY Flight Model (SCIA FM) cross-section from Bogumil et al.  reconvolved with the GOME-2 slit function [Siddans et al., 2006]. For the DOAS fit, the 203 K cross-section is used and the retrieved slant columns are corrected a posteriori for the temperature dependence of the cross-sections, according to the assumed SO2 plume height [Valks et al., 2011]. Further, cross-sections of the interfering trace gases ozone and NO2 are included in the fit. Best results in this wavelength region are obtained using the Malicet et al.  ozone cross-sections at two temperatures (218 K and 243 K). For NO2, the GOME-2 Flight Model/CATGAS cross-section is used at 241 K [Gür et al., 2005].
 The Ring effect, caused by inelastic rotational Raman scattering, is responsible for the filling-in of the solar and absorption features in the earthshine spectra. This is treated as an additional absorber by means of two additive Ring pseudo absorption cross-sections generated after Vountas et al. . This accounts also for the telluric contribution (molecular Ring effect). Intensity offset effects, that may be induced by residual stray-light or remaining calibration issues in the GOME-2 level-1 product are known to be possible sources of bias in DOAS retrievals of minor trace gas species; to partly correct for possible offset the inverse of the earthshine spectrum is fitted as another effective cross-section. A cubic polynomial closure term is added that accounts for broadband effects: molecular scattering, aerosol scattering and absorption and reflection from the Earth's surface. A solar spectrum that is recorded by GOME-2 once a day is used as reference spectrum.
Figure 1 (left) shows an example of the DOAS SO2 fit for a GOME-2 measurement within the Eyjafjallajökull eruption plume on May 14, 2010. The measurement selected is showing a high SO2 slant column density of ∼22 Dobson Units (1 DU = 2.687 × 1016 molec/cm2) and the strong SO2 absorption features are clearly visible in the fit.
 In the wavelength range 315–326 nm, used for the retrieval of SO2, a strong interference of the SO2 and ozone absorption signals can be observed, because the absorption cross-sections of the two gases show similar structures. As the ozone columns are typically much larger than the SO2 columns, the DOAS fit can produce “negative” SO2 slant column values for high ozone amounts. Therefore, an interference correction needs to be applied to the SO2 slant column values [Rix, 2011; Valks et al., 2011]. The empirical correction is calculated based on two years of GOME-2 SO2 and O3 data.
 The second component in the retrieval is the conversion of the corrected SO2 slant column density (S) into the vertical column density (V) using the air mass factor M: M = S/V. The air mass factor depends on the vertical SO2 profile in the atmosphere and a set of forward model parameters, including the GOME-2 viewing geometry, surface albedo, clouds and aerosols. For the air mass factor calculations in the retrieval algorithm of the GDP 4.4, a volcanic SO2 profile is assumed, defined via a central plume height with a Gaussian SO2 distribution (FWHM = 1.5 km) around that central height. The air mass factors and vertical SO2 column are computed for three different assumed SO2 plume heights: 2.5 km, 6 km and 15 km above ground level. The lowest height represents passive degassing of low volcanoes, the second height effusive volcanic eruptions or passive degassing of high volcanoes and the third height explosive eruptions, which can reach stratospheric heights. In this study, a plume height of 6 km has been used for the air mass factor calculations of the Eyjafjallajökull eruption (see also section 2.3).
 The air mass factors are calculated with the radiative transfer model LIDORT 3.3 [Spurr, 2008] at 315 nm. The cloud information is obtained from the OCRA and ROCINN algorithms [Loyola et al., 2007]. OCRA provides the cloud fraction from the GOME-2 broadband polarization measurements and ROCINN retrieves the cloud top height (pressure) and cloud albedo from GOME-2 measurements in the oxygen A-band around 760 nm. Sun-glint and clouds are properly discriminated in the retrieval [Loyola et al., 2011a]. The air mass factor for cloudy conditions is calculated using the independent pixel approximation (IPA) for clouds and a Lambertian reflecting boundary cloud model.
2.3. Uncertainty Analysis and SO2 Detection Limit
 The error in the SO2 total vertical column (σV) is mainly a function of errors in the slant column density (S), which are to a large extent due to instrument noise, and a function of errors in the air mass factor (M):
Here we focus on the random error in the slant column density, which determines the SO2 detection limit, and the uncertainty due to the unknown SO2 plume height, which is the main error source in the air mass factor in this case.
 To estimate the random error of the GOME-2 SO2 vertical columns, the retrieval statistics for a pristine ocean region over the Pacific, from 10°N–10°S and 170°E–120°W, have been examined. In this oceanic region only SO2 background concentrations below about 50 pmol/mol are expected [Seinfeld and Pandis, 2006]. The GOME-2 SO2 values have a statistical distribution with a Gaussian shape around zero due to instrument measurement noise. The standard deviation of the SO2 total columns, which can be derived from this distribution, provides an estimate of the random error in the slant and vertical column densities (as volcanic SO2 clouds can occasionally occur in this area due to volcanic eruptions, all measurements containing SO2 values larger than two times the standard deviation are excluded from the analysis).
 For individual GOME-2 observations, the random error in the vertical column due to the instrument measurement noise has been estimated to be 0.44 DU for the Eyjafjallajökull eruption in May 2010 (Note that the measurement noise has been increasing significantly since the beginning of GOME-2 operations in 2007 due to instrument degradation [Lang et al., 2009; Dikty et al., 2010]). By averaging over a 1° × 1° region (about 3–5 GOME-2 measurements) the precision can be significantly improved and the standard deviation is reduced to 0.24 DU. The detection limit is typically considered twice the standard deviation.
 It has been found that the main contribution to systematical errors in the retrieval is an unknown SO2 plume height, which leads to errors due to the differences between assumed and real atmospheric temperature and due to the height dependence of the air mass factor [Rix, 2011]. One can assume that for the Eyjafjallajökull eruption the plume height of 6 km (as used in the GOME-2 SO2 retrieval algorithm for this study) is within ±4 km of the actual plume height, as the volcano has a summit height of ∼1700 m and eruptions of this magnitude typically do not reach heights significantly higher than 10 km. This results in an error in the temperature of ∼20 K causing an error in the vertical column due to the temperature dependence of the SO2 cross-section of about 10%. The error caused by the plume height dependence of the air mass factor ranges from approximately +50% for an actual plume height of 2 km (i.e., an overestimation of the plume height by 4 km) to approximately −15% for an actual plume height of 10 km (i.e., an underestimation of the plume height by 4 km). Other contributions to the systematical error are the ozone interference and the Ring effect, which are typically below 5% for moderate SO2 loadings as observed for Eyjafjallajökull. This results in a total systematic error in the GOME-2 SO2 columns of 10–60%. Assuming a total random error for moderate SO2 loadings occurring during the Eyjafjallajökull eruption of 5–20% (see above), the overall error in the vertical column density (including random and systematic contributions) is estimated to be in the 20–70% range. A detailed discussion on the error sources in the GOME-2 SO2 retrieval can be found in Rix .
Figure 1 (right) shows a BrO fit for the GOME-2 measurement within the Eyjafjallajökull eruption plume on May 14. As the BrO slant column density is closer to the detection limit (∼5 × 1013 molec/cm2), the quality of the fit is limited compared to that of SO2. A detailed description of the BrO retrieval from GOME-2 observations can be found in Theys et al. [2009, 2011].
3. GOME-2 Observations of SO2 and BrO
 During the early eruptive episode until 4 May, GOME-2 and other satellite instruments detected only small amounts of sulfur dioxide that were mostly located close to the volcano. We will focus on the eruptive activity following 4 May, where large amounts of SO2 were emitted into the atmosphere, and the volcanic gas plume could be traced by GOME-2 for several days. After the intensity of the eruption picked up in the beginning of May, an SO2 rich eruption plume was detected on 5 May that was distributed to the southeast and covered North-Ireland and parts of the UK (Figure 2). The following day the plume could be detected directly to the south of Iceland. On 7 and 8 May the GOME-2 observations show that part of the volcanic SO2 cloud was transported toward Spain. SO2 and ash were well collocated during this period of the eruption, therefore the SO2 plume could be used as tracer for the eruption cloud [Thomas and Prata, 2011]. The airspace over parts of southern and central Europe was closed the following day (9 May), due to this volcanic cloud. The next days (9–12 May) the SO2 plume can be observed south of Iceland over the Atlantic leaving European airspace mainly unaffected, with only small parts of the plume touching Ireland on 11 May and France on 12 May. From 14 to 17 May GOME-2 SO2 shows that, except for 15 May, the volcanic plume is again distributed toward northern Europe by the prevailing winds (Figure 2). During the airspace closure over Ireland and the UK on 16 and 17 May, GOME-2 observations show the volcanic SO2 plume over the northern UK. SO2 emissions from the Eyjafjallajökull eruption could be detected until the eruption ceased on 23 May.
 During the whole eruptive activity the total SO2 loadings were moderate with maximum total columns being ∼20 DU, while on most days the maximum did not exceed 10 DU. Figure 3 shows the total daily SO2 amount calculated from the GOME-2 observations for the area depicted in Figure 2. The highest SO2 amount can be observed on 9 and 10 May with the total SO2 mass close to 0.15 Tg. Including all daily emissions, the total emissions can be estimated to ∼1.2 ± 0.5 Tg, which is comparable to the total SO2 emissions estimated from GOME-2 observations during the Kasatochi eruption in August 2008 [Rix et al., 2009]. This result corresponds reasonably well with the Falcon estimates of 3 (0.6–23) Tg for the total SO2 emissions [Schumann et al., 2011].
 Apart from SO2, also enhanced BrO amounts could be observed in the eruption plume of Eyjafjallajökull in May 2010 (Figure 4). Unlike SO2, BrO is not directly emitted by volcanoes, it forms in the plume mainly due to heterogeneous reactions. The main reactions involved in the production of BrO [Bobrowski et al., 2003; Oppenheimer et al., 2006] are:
The reaction inside the plume can only take place under daylight conditions (equation (5)) and if the ozone concentration is sufficiently high (equation (6)). Ozone is mainly obtained from mixing of ambient air into the plume and the depletion of ozone has been observed in volcanic plumes [Vance et al., 2010]. Other studies report that due to these constraints, the BrO concentration increases with increasing distance from the crater and it is higher toward the plume edges [Bobrowski et al., 2007].
 In order to separate the volcanic BrO from background atmospheric BrO, the total BrO vertical column densities (VCDs) were normalized through subtraction of an average BrO column calculated in the region from 0°–35°W and 40°N–70°N using all observations outside the volcanic plume. For identification of the region outside the volcanic plume, the SO2 columns have been used, excluding all measurements with SO2 values exceeding twice the standard deviation. The average background BrO VCDs applied in this study is about 3.5 × 1013 molec/cm2.
 A signal from volcanic BrO is detected from 5 to 8 May and from 14 to 17 May (Figure 4). It can be seen that the shape of the BrO plume is similar to that of the SO2, though it is less clear and drawn out, since most of the BrO total columns were close to the detection limit of the GOME-2 instrument. The maximum BrO VCD observed was ∼7.7 × 1013 molec/cm2 (after subtraction of the background BrO column).
 For the eight days where the signature of volcanic BrO could clearly be identified, the ratio of the total BrO and SO2 mass has been calculated (Figure 3) to check the validity of the GOME-2 BrO observations, as these total BrO columns were mainly close to the GOME-2 detection limit. The BrO/SO2 ratio ranges from 1.1 × 10−4 to 2.1 × 10−4. These values are well within the BrO/SO2 ratios that are typically observed in volcanic plumes [Bobrowski and Platt, 2007]. The ratio for 16 May has been compared with observations of the CARIBIC aircraft on the same day. The BrO/SO2 ratio observed by GOME-2 is 2.1 × 10−4, while the one measured by the CARIBIC aircraft is 1.3 × 10−4 [Heue et al., 2011]. This disagreement, however, could be explained by the way both ratios were calculated. In case of GOME-2, high BrO values close to the volcano (see Figure 4) were taken into account, whereas in case of CARIBIC only values over Great Britain and Ireland were measured, resulting in higher BrO/SO2 ratios for the GOME-2 observations.
4. SO2 Plume Height Estimation
 The correct determination of the volcanic plume height is important for aviation safety, as well as for air quality and climate studies. Furthermore, plume height information is required for the correct quantitative retrieval of the SO2 total columns, as the DOAS method uses a priori assumptions on the vertical SO2 distribution. These profile assumptions are typically characterized through a central SO2 plume height. If no additional plume height information is available at the time of the retrieval, errors are introduced due to differences between the assumed and the actual plume height.
 Recently the direct retrieval of the plume height from OMI [Yang et al., 2009] and GOME-2 [Nowlan et al., 2011] spectral measurements has been demonstrated using a direct fitting technique. The direct retrieval of the plume height from satellite observations is possible, as the height of a volcanic SO2 plume determines the proportion of backscattered photons that pass the absorbing layer on their way to the satellite instrument. The higher the plume the more photons are Rayleigh scattered below and therefore pass through the SO2 plume resulting in more prominent SO2 absorption structures in the measured radiances. If the layer is in the lower atmosphere more photons are scattered above. Therefore the measured radiances contain information about the SO2 plume altitude besides the total column information.
 The direct fitting technique is computationally very time consuming and for several applications, especially aviation safety, the plume height information is required in near-real time (NRT). Therefore, a fast retrieval scheme has been developed through a combination of the DOAS method to retrieve the total SO2 column and a minimization method for matching simulated GOME-2 spectra to retrieve the plume height.
4.1. SO2 Plume Height Rapid Inversion With GOME-2
Figure 5 gives an overview of the SO2 Plume Height Rapid Inversion (SOPHRI) retrieval algorithm. GOME-2 spectra in the wavelength range between 312 and 325 nm are simulated with LIDORT-RRS [Spurr et al., 2008b], which allows the inclusion of rotational Raman scattering, for different assumed central plume heights z between 1 and 15 km. Further input parameters for the simulation of the Earthshine spectra are ozone profile, satellite viewing geometry, surface albedo, cloud fraction and cloud top albedo. The ozone total column information taken from the operational GOME-2 product [Loyola et al., 2011b] is used to select an ozone profile from a column-classified ozone profile climatology [Wellemeyer et al., 1997]. In the same way cloud information based on the OCRA and ROCINN algorithms (see section 2.2) is obtained from the operational GOME-2 product. To allow a NRT estimation of the plume height a look-up table (LUT) approach is used with pre-calculated GOME-2 spectra. The simulated GOME-2 spectra are stored in the LUT according to total SO2 column, SO2 plume height, total ozone column, albedo, solar zenith angle, viewing angle and relative azimuth angle.
 The first guess SO2 vertical column is taken from the operational GOME-2 product, as described in section 2, with an assumed plume height of z0 = 15 km. The simulated spectra based on this first guess column are derived through interpolation in the LUT and the difference R is calculated for the set of simulated spectra and the measured spectrum as:
The plume height z of the simulated spectrum Imod(z), that shows the smallest difference R to the measured spectrum, is selected as the new plume height. If the retrieved plume height z differs from the plume height z0, which has been assumed for the DOAS retrieval, the vertical column is recalculated with an updated air mass factor, which uses the new plume height as reference. The iteration stops when the assumed and retrieved plume height converge. The best convergence with a small number of iterations (usually 3–4) is found with a starting height z0 of 15 km. This is related to the fact that in the wavelength region used for the plume height retrieval with GOME-2 (312–325 nm), the sensitivity for larger plume heights is lower than for medium and small plume heights [Yang et al., 2009]. The retrieval of the SO2 total column and plume height takes approximately 20 min for one GOME-2 orbit (∼16000 single measurements in ∼100 min) on a 2.4 GHz processor.
 The SOPHRI retrieval has been applied to synthetic GOME-2 spectra to estimate the accuracy of the new method for different atmospheric conditions [Rix, 2011]. In most cases an accuracy of about 1 km could be achieved, which is comparable to the accuracy of the direct fitting method used for the plume height retrieval during the Kasatochi eruption [Nowlan et al., 2011]. An important parameter that affects the accuracy of the retrieved plume height is the total SO2 column. The retrieval works best for high total columns, as a clear SO2 signature in the spectrum is needed. The SOPHRI retrieval provides very good results for total SO2 columns >10 DU, an overestimation of 1–2 km can sometimes be observed for total columns in the range of 5–10 DU. Larger errors of 2–4 km arise for low total SO2 columns (<2 DU).
 Additional uncertainties arise due to volcanic ash in the plume or if the GOME-2 pixel is not completely covered by the plume. To minimize the influence of volcanic ash and aerosols, differential GOME-2 spectra are used for the plume height retrieval, which filters out the broad extinction structures of the ash/aerosols. However, large amounts of volcanic ash can lead to a strong increase in the optical thickness, which reduces the retrieval accuracy.
4.2. The Eyjafjallajökull Eruption
 The new method to estimate the plume height has been applied to the eruption of Eyjafjallajökull. For analysis of the plume height, GOME-2 measurements from 5 May 2010 have been selected, as the SO2 emissions were high on that day and the eruption plume was broad compared to other days. Only measurements with a total column >5 DU have been used in the plume height retrieval.
Figure 6 (left) shows that the retrieved SO2 plume heights on 5 May are between 8–13 km, and that for most of the GOME-2 measurements a plume heights of ∼11 km is found. Lower heights in the range of 3–8 km are retrieved in the eastern part of the plume. The maximum plume height reported at the volcano that day was around 9 km and around 5 km the previous day, inferred from visual observation and weather radar [Hjaltadóttir et al., 2010; Marzano et al., 2011]. These heights are somewhat lower than the retrieved plume heights, however they explain the lower SO2 plume heights retrieved in the eastern part of the plume, as it is likely that the SO2 in this part of the plume has already been emitted on the previous day. The estimation of the ash emission height on 5 May using a Lagrangian dispersion model suggest emissions up to 10 km [Stohl et al., 2011], which, as the reported plume heights, indicates emissions somewhat lower than the retrieved SO2 plume heights. However, as both the dispersion model as well as the reported plume heights, refer to ash, it is possible that the SO2 was at higher altitudes during the eruption.
 For the measurement closest to the volcano a plume height of 15 km is retrieved, which seems too high compared to other height estimations. A possible explanation is that the SOPHRI retrieval is obstructed by high amounts of volcanic ash, which are typically found close to the volcano. A further difficulty for the analysis of the eruption plume of Eyjafjallajökull is the narrow character of the SO2 plume with only parts of the GOME-2 pixel area covered by the SO2 plume. However, for other parts of the plume the results of the SOPHRI plume height estimation mostly agree within 1–3 km with the reported plume heights and the results of the ash dispersion modeling.
4.3. Comparison With IFS Plume Simulations
 Simulations of the Eyjafjallajökull eruption plume have been performed using the IFS model, in the framework of the Monitoring Atmospheric Composition and Climate (MACC) project (http://www.gmes-atmosphere.eu/news/volcanic_ash). The IFS calculates the SO2 transport by advection and turbulent diffusion. The SO2 source and sink terms can be provided by a Chemical-Transport-Model (CTM), which is coupled to the IFS [Flemming et al., 2009]. For this study, the IFS was run uncoupled. The SO2 mass was directly injected in the IFS. A fixed lifetime of five days has been assumed (based on SO2 lifetime estimates by Krotkov et al.  and Lee et al. ) to account for the SO2 loss because of reaction with OH and loss by wet-deposition has been simulated. A comparison of the volcanic SO2 plume simulated with the more comprehensive sulphur chemistry of the coupled system IFS-MOZART did not show significant differences for the considered time scale. The injection height is an essential parameter for the simulation of the dispersion of volcanic plumes. For the simulation of the Eyjafjallajökull eruption plume an injection height of 6–13 km was assumed, based on SO2 eruption height calculations using the FLEXTRA trajectory model [Stohl et al., 2001; C. Maerker, personal communication, 2011].
 The IFS was run at a T511 spectral resolution, which corresponds to a horizontal grid-box size of about 40 km. IFS used a vertical discretization of 60 vertical layers up to a height of about 60 km (0.1 hPa). The vertical resolution varied from about 300 m at 1 km to about 1 km at 15 km height. An advantage of using a weather-forecast model for tracer plume simulation is that the wind fields used in the tracer advection are available at high temporal resolution. The IFS model time step was 900 s.
 The SO2 plume heights for 5 May 2010 as calculated with the IFS model are shown in Figure 6 (right). The modeled heights are mainly in the range of 8–10 km in the western part of the plume, with lower heights of 4–7 km in the eastern part. The results match the GOME-2 retrieval reasonably well, especially in the eastern part of the plume where the retrieved altitudes are in the 3–8 km range. In the western part of the plume, the SOPHRI retrieval shows higher plume altitudes which are mainly around 11 km but can reach up to 13 km. Note that the modeled plume shape is very similar to the GOME-2 observations, indicating that the assumed emission heights for the model simulations are correct, and that a SO2 height around 10 km is a good estimation for the Eyjafjallajökull eruption plume on 5 May.
5. Comparison of GOME-2 Data With DLR Falcon Measurements
 During the eruption of the Eyjafjallajökull the DLR research aircraft Falcon performed 17 flights to probe the volcanic clouds over Germany, the UK and Iceland [Schumann et al., 2011]. Measurements include particle concentration and size to infer ash mass concentrations and gas-phase plume tracer including SO2. The Falcon in situ SO2 measurements provide a good opportunity to validate the GOME-2 SO2 data.
 The SO2 instrument deployed on the Falcon was a Thermo Electron SO2 Analyzer (Model 43C Trace Level). Outside air is sampled through an inlet mounted on top of the aircraft fuselage. Teflon (PFA) is used for the inlet tubes. In the instrument detector SO2 is measured using pulsed fluorescence [Luke, 1997]. SO2 is electronically excited in an optical cell by radiation in the wavelength range 190–230 nm generated by a Xenon flash lamp pulsed at 10 Hz. Excited SO2* is partly quenched and photolyzed, and partly fluoresces at 320 nm. This radiation is detected by a photomultiplier tube. The photomultiplier sampling gate is delayed by 30 μs from the trigger of the flash lamp to reduce electronic noise associated with the flash. The fluorescence intensity is linearly proportional to the SO2 mixing ratio.
 The fluorescence technique used for the Falcon measurements includes a weak positive interference from aromatic hydrocarbons, CS2 and NO. Therefore, hydrocarbons are removed from the sample air in the instrument by diffusion through a semi-permeable membrane (hydrocarbon kicker). Interferences from CS2 and NO are negligible considering rejection ratios (ratio of the concentration of interferant to SO2 required to produce an equivalent instrument signal) for CS2 and NO of 20 [Luke, 1997] and 35 [Roiger, 2007], respectively, and observed CS2/SO2 and NO/SO2 concentration ratios of about 0.01 [Cronn and Nutmagul, 1982] and 0.001 [Rose et al., 2006] in volcanic plumes. The precision and accuracy of the individual SO2 measurements are 3% and 5%, respectively. The inferred SO2 column densities from the Falcon profiles have an accuracy of 10%.
 For the comparison with the GOME-2 observations of SO2 two days have been selected, 17 and 18 May. On these days the Falcon aircraft measured vertical profiles of the SO2 molar mixing ratios inside the volcanic plume as shown in Figure 7. On 17 May, two vertical SO2 profiles were recorded during the plume sampling, one during the aircraft ascent and one during the descent. On 18 May, a vertical profile was recorded during the descent (the plume penetration during the ascent was mainly performed on a constant flight level and is not shown here since only few data points are available for the vertical profile). From the measured Falcon SO2 profiles the SO2 total column in the volcanic plume can be inferred and compared with the GOME-2 SO2 data.
 The Falcon observations show that the SO2 plume was located between 4 and 7 km on 17 May and slightly lower on the following day. For the calculation of the GOME-2 total SO2 columns a plume height of 6 km has been assumed for both comparisons (see section 2.2).
 As described in section 2.1, GOME-2 is located on a sun-synchronous polar orbiting satellite and with a local equator crossing time of 9:30. As the measurements of the Falcon were carried out at different times of the day we calculated backward trajectories for the air masses sampled by the Falcon to the GOME-2 measurement time for the comparison. Figures 8a and 8b show the endpoints of the trajectories for the four Falcon profiles on 17 and 18 May. These endpoints mark the position of the air masses probed by the Falcon at the time of the GOME-2 measurement. On 18 May the times of the Falcon and GOME-2 observations coincides for large parts of the measurements, the maximum time difference is ∼1 h. On 17 May the Falcon flight was in the afternoon, in this case the time difference between the observations was 6–7 h.
 For the calculation of the backward trajectories the Lagranto model [Wernli and Davies, 1997] has been used. The model calculations are based on ECMWF data with a spatial resolution of 0.25° × 0.25°. The trajectories are started for the times, locations and heights of every Falcon measurement that detected enhanced SO2 concentrations in the volcanic cloud and calculated backward for 24 h. The model output has a temporal resolution of 30 min. The trajectory point closest to the GOME-2 measurement time is selected as position of the air mass sampled later by the Falcon and the GOME-2 SO2 value at this position is used for the comparison.
 From the 4 vertical SO2 profiles (P1–P4) measured by the Falcon the total SO2 column have been inferred. For comparison, the GOME-2 measurements have been interpolated to the positions of the air masses sampled by the Falcon at the time of the GOME-2 observations. The conditions were favorable for GOME-2 observations as the measurements were cloud free or with the clouds located below the SO2 plume. In all cases both instruments clearly detect the enhanced SO2 concentrations inside the volcanic plume. Profile P1 corresponds to 4.7 DU of SO2, which is very close to the maximum SO2 amount of 5.0 DU observed by GOME-2 in that region. Averaging the GOME-2 measurements along the Falcon flight path results in a lower total column density of 2.9 DU, as also pixels with low SO2 amounts are located along the flight path. A second profile (P2) was measured by the Falcon on the same day within the same area of the plume. In this case the total SO2 column estimated from the Falcon observations is 3.6 DU, while averaging all GOME-2 measurements along the flight path results in a total column density of 3.1 DU. In this case the two observations show a good agreement. On 17 May 2010 the maximum SO2 total column was high enough to apply the SOPHRI retrieval, providing a unique opportunity to compare the retrieved plume height with a measured SO2 profile (P1). The retrieval shows a plume height of ∼6 km for this measurement, which is only slightly higher than the plume height shown by the Falcon observations (4–6 km).
 On the following day two additional SO2 profiles were recorded by the Falcon aircraft. Profile P3 was measured away from the plume center outside the area with strongly enhanced SO2 values (see Figure 8b). A total column of 0.6 DU was estimated from the Falcon measurements. As in this case the measurements time of the GOME-2 instrument corresponds with the flight time, no trajectory matching had to be applied, to find the corresponding measurement location. The average of the GOME-2 measurements along the Falcon flight path is 0.8 DU, and close to the value obtained from the Falcon observations. It should be noted that the satellite value is close to the detection limit (see section 2.2). The total column derived from the second Falcon plume penetration on 18 May (P4) is 2.1 DU of SO2, while averaging the GOME-2 observations along the flight path results in a total column of 1.2 DU. Here the Falcon measurement is ∼60% higher than the GOME-2 measurement.
 Part of the underestimation by GOME-2 seen for P1 can be explained by the fact that the calculated line of air masses sampled by the Falcon is located at the edge of the volcanic plume. As the GOME-2 pixel size is fairly large, it is likely that some of the GOME-2 pixels were only partly covered by the SO2 plume and the observed differences are due to gradients in the SO2 abundance which cannot be resolved by the GOME-2 instrument. For P4 the main difficulty is that the measurement flight was mainly performed at constant flight level and the total column has been estimated from only a few points, which results in larger uncertainties in the SO2 column calculated from the Falcon data.
 Further, compared to the Falcon measurements, the uncertainties in the GOME-2 measurements are relatively large for low SO2 total columns (see section 2.3) and the degradation of the GOME-2 instrument since the launch in 2007, resulted not only in an increased noise level, but also in a systematic decrease in the fitted SO2 slant column densities. Further differences can arise due to the time difference between the Falcon and GOME-2 measurements, as despite the trajectory modeling small scale dynamical movements in the plume, which cannot be resolved by the model, can influence the observations.
6. Comparison of GOME-2 Data With Brewer Observations
 The overpass of the SO2 plume from the Eyjafjallajökull eruption has also been observed by several Brewer instruments throughout Europe, allowing a comparison with the GOME-2 observations. The Brewer spectrophotometer was developed in the seventies by the University of Toronto and the Atmospheric Environment Service of Canada. The purpose of this development was to supplement a network, which mainly consisted of Dobson spectrophotometers at that time [Brewer, 1973; Kerr et al., 1981]. The modern spectrophotometer type (Brewer Mk. II) has the capability to measure the total column of ozone and SO2 separately by using additionally a specific wavelength, absorbed mainly by SO2 [Evans et al., 1980] and has been used to study SO2 pollution [De Muer and De Backer, 1992].
 For the comparison the two instruments at Valentia (51.93°N, 10.25°W), Ireland and at Hohenpeissenberg (47.8°N, 11.02°E), Germany were selected, as they both clearly show the overpass of the volcanic cloud. The Brewer SO2 data for Valentia have been provided by the World Ozone and Ultraviolet Radiation Data Centre (WOUDC, www.woudc.org) and are only available as daily averages. The individual Brewer SO2 measurements for Hohenpeissenberg have been obtained directly from the station.
Figure 9 shows the comparison of the average SO2 total columns measured by the Brewer instruments at Valentia and Hohenpeissenberg with the GOME-2 observations for May 2010. For the measurements at Valentia, Ireland, the Brewer shows clearly enhanced SO2 values of 8.1 DU on 11 May, on the other days the values are well below the detection limit. GOME-2 also detects enhanced SO2 values on that day. The GOME-2 measurements are interpolated to the measurement site showing a total SO2 column of 4.1 DU which is ∼50% lower compared to the Brewer observations. Due to the narrow structure of the SO2 plume after the Eyjafjallajökull eruption, the comparison is complicated by the fact that the Brewer observations are daily averages while the GOME-2 measurements represent a snapshot of the situation at the time of the overpass. Simulations made with the IFS model (not shown here) indicate that the plume with high SO2 values was located over Valentia for several hours after the GOME-2 overpass. Therefore it is likely that part of the difference between the GOME-2 and Brewer SO2 column is due to the fact that these higher SO2 values are included in the Brewer daily average. Further, differences can be caused by uncertainties in the GOME-2 observations (see section 2.3) and large uncertainties in the Brewer SO2 measurements of up to 50% [Fioletov et al., 1998].
 The Brewer spectrophotometer No. 10 at the Meteorological Observatory Hohenpeissenberg has been in use since 1983. Only in the first years significant SO2 values could be sporadically observed with the Brewer, when a coal-fired power plant in the near-by small town Peissenberg was in operation during cold periods in winter season. SO2 values up to 14 DU were the maximum values. In the past two decades, however, the normal SO2 values were always around zero or even slightly negative, that means below the limit of detection, which is estimated to be around 1 DU.
Figure 9 shows that both the GOME-2 observations and the Brewer instrument clearly detect the overpass of the SO2 cloud at Hohenpeissenberg on 18 May. A very good agreement between the two observations is found as the Brewer observations show a total SO2 column of 1.2 DU, while the interpolated GOME-2 observations show a total column of 1.3 DU. (Here it should be noted that an off-set correction has been applied to the Brewer observations after 10 May to account for instrument calibration issues). At the station Hohenpeissenberg, also a comparison has been made between the GOME-2 SO2 observation and the individual Brewer measurements during the overpass of the SO2 cloud on 18 May (Figure 10). The SO2 total column of 1.3 DU derived from the GOME-2 measurements is lower than the maximum column of ∼3 DU detected by the Brewer instrument. However, the GOME-2 SO2 column agrees quite well with the Brewer observation shortly before the GOME-2 overpass. The changes in the SO2 columns measured by the Brewer instrument within a relatively short time period are an indication for strong variations in the SO2 concentration within the volcanic cloud. This shows that, apart from the relatively large measurement uncertainties, the observed difference between the GOME-2 and Brewer measurements is partly due to the inability of GOME-2 to resolve gradients in the SO2 amount within a narrow plume.
7. Summary and Conclusions
 We have presented observations of SO2 and BrO from the GOME-2 instrument during the eruption of the Eyjafjallajökull volcano in April and May 2010 and the retrieval of the plume height using the SOPHRI algorithm. The GOME-2 SO2 data have been compared with model simulations and in situ data from the DLR Falcon research aircraft and with Brewer measurements. During the eruptive period in May 2010, SO2 total columns of up to ∼20 DU (5.4 × 1017 molec/cm2) and BrO columns of ∼7.7 × 1013 molec/cm2 were detected. The BrO/SO2 ratio estimated from the GOME-2 observations of the Eyjafjallajökull eruption varied from 1.1 × 10−4 to 2.1 × 10−4. The application of the SOPHRI retrieval on 5 May 2010 yields SO2 plume heights from 8 to 13 km with an estimated uncertainty of 1–2 km. The GOME-2 observations mostly agree within 1–3 km with estimates of the ash plume height from visual observations, radar data and modeling results [Stohl et al., 2011]. The simulations with the IFS model show SO2 plume heights mainly around 8–10 km in reasonably good agreement with the GOME-2 heights. In the eastern part of the plume, the IFS model results and the GOME-2 SO2 plume heights agree well, with values between 4 and 7 km.
 Volcanic plume height from GOME-2 provided by the SOPHRI retrieval are especially useful for aviation safety [Van Geffen et al., 2007; Ferrucci and Tate, 2011], as an estimate of the plume height is needed in near real time, and an accuracy of 1–2 km is in most cases sufficient to decide whether an interference with air traffic routes and certain flight levels can be expected.
 The eruption plume of the Eyjafjallajökull eruption was also observed by a variety of other instruments. The GOME-2 measurements of SO2 were compared with total columns estimated from SO2 profiles measured by the DLR Falcon research aircraft on 17 and 18 May. The SO2 columns derived from the Falcon SO2 profiles on these days range from 0.6–4.7 DU. The comparison with the GOME-2 measurements show a good agreement: the SO2 total columns mostly agree within 1 DU, larger differences are likely related to GOME-2's inability to resolve fine plume structures. Furthermore, the plume height shown by the Falcon observations on 17 May (4–6 km) agrees very well with the retrieved GOME-2 plume height on this day (∼6 km).
 Further, the GOME-2 SO2 data have been compared with Brewer observations of the plume overpass at Valentia and Hohenpeissenberg. Both GOME-2 and the Brewer instruments detected the overpass of the SO2 cloud at the two stations. A good agreement is found for the observations at Hohenpeissenberg, where both instruments measured a daily averaged SO2 column of ∼1.3 DU. The Brewer instrument at Valentia measured SO2 columns of ∼8 DU on 11 May, which are ∼50% higher than the SO2 columns detected by GOME-2. A possible explanation for the relatively large difference at Valentia is that GOME-2 did not capture the part of the plume with high SO2 columns, that was located over Valentia for several hours after the GOME-2 overpass.
 Although the eruption of Eyjafjallajökull volcano can be considered fairly small (VEI = 2–3) it caused unprecedented air traffic problems not only in Europe. It left millions of travelers stranded and also affected the economy. The estimated economic loss due to the eruption is approximately 2.5 billion € in total [Gudmundsson et al., 2010]. Iceland has seen 9 eruptions of similar magnitude in the past 40 years (Hekla, 1971, 1980, 1991 and 2000, Gjalp 1996, Grímsvötn 1998, 2004 and 2011). In six of these cases the eruption plumes were dispersed toward the north and northeast leaving Europe unaffected. However, it is to be expected that a similar eruption will cause air traffic disruption again in large parts of Europe in the future. The eruption underlined the importance of establishing a close monitoring system for volcanic eruptions and the necessity to combine different monitoring tools, i.e., dispersion modeling, ground-based and satellite observations, and aircraft measurements to obtain the best possible picture of the distribution of the volcanic emissions.
 The eruption of Eyjafjallajökull has shown that atmospheric satellite instruments, like GOME-2, provided valuable near-real time information about the volcanic plume during the eruption giving an overall picture of the location of the eruption plume and a good estimate of the plume height, with an accuracy of 1–2 km. They can assist and complement model studies in the prediction of the plume development, therefore assuring aviation safety, while avoiding unnecessary disruptions in air traffic. As polar orbiting satellites like GOME-2 can only provide snapshots, modeling is still an essential part of aviation safety after volcanic eruptions. A combination of satellite data and other observations has proven to be useful for close monitoring of the eruption cloud. The eruption demonstrated that the synergetic use of aircraft and satellite observations is of great advantage. Satellite observations were used to direct the aircraft to the volcanic plume and aircraft observations provide the valuable opportunity to validate the satellite data. Additionally, aircraft observations are especially valuable in terms of providing detailed information on the vertical profile of the volcanic plume, which is often difficult to obtain. Further, the comparison with the aircraft and Brewer observations shows that the strength of these observations is that they can resolve the finer structures in the plume, which GOME-2 misses due to the coarse spatial resolution. However, major improvements can be expected with the launch of the upcoming European atmospheric composition missions, Sentinel-5 precursor, Sentinel-4 and Sentinel-5, which will provide a much better resolution in time and space.
 The development of the GOME-2 SO2 and BrO column products and their validation has been funded by the O3M-SAF project of EUMETSAT, the Exupery project of the German Geotechnologien program, and by national contributions. The research on the GOME-2 SO2 plume height algorithm (SOPHRI) was funded by DLR. Thanks to the BIRA colleagues J. van Geffen, J. van Gent, N. Theys and M. Van Roozendael for the fruitful collaboration on the SO2 and BrO retrieval algorithms. We would like to thank W. Thomas (DWD) and M. Koukouli (University of Thessaloniki) for providing the Brewer observations for Hohenpeissenberg and Valentia. We thank DLR colleagues W. Zimmer and L. Butenko for the development work on the UPAS system and R. Spur for providing the LIDORT-RRS program. We would like to thank S. Kiemle for the development work on the DIMS system; J. Jaeger, K. H. Seitz, B. Huber and T. Ruppert are kindly acknowledged for the daily operations of the O3M-SAF facility at DLR. We thank EUMETSAT for the provision of GOME-2 level 1 products.