Sulfur dioxide in the atmosphere can be measured from ground-based instruments and by instrumentation onboard the spacecraft and can be estimated with the help of models. The Brewer is an automated, diffraction-grating spectrophotometer that provides observations of the sun's intensity in the near-UV range. The spectrophotometer measures the intensity of radiation in the ultraviolet absorption spectrum of ozone at five wavelengths (306.3, 310.1, 313.5, 316.8 and 320.1 nm) with a resolution of 0.6 nm. These data are used to derive the total ozone column (Kerr et al., 1980). Because sulfur dioxide has strong and variable absorption in this spectral region, the Brewer spectrophotometer has additionally been proposed to derive SO2 columns (Kerr et al., 1980). About 200 Brewer spectrophotometers around the world contribute high-precision ozone data to the global ozone monitoring network (Kumharn et al., 2012). The existing Brewer network also delivers frequent SO2 columnar measurements as well, which can be used for analyses, but with caution. This is because the signal-to-noise ratio for the SO2 absorption is usually quite low and therefore well-calibrated instruments are required to monitor nominal SO2 columnar amounts (Koukouli et al., 2014). Details on the method with which SO2 is measured by the Brewer spectrophotometer can be found in Kerr et al. (1980, 1985, 1988) and De Backer and De Muer (1991). According to Fioletov et al. (2016), the uncertainty of the Brewer direct sun (DS) SO2 measurements is about 1 DU and is typically insufficient for air quality applications. A more accurate method (with an uncertainty as low as 0.13 DU) based on Brewer “group-scan” spectral direct sun radiation measurements at 45 wavelengths from 306 to 324 nm was developed (Kerr, 2002) but not implemented for routine operations due to its complexity (Fioletov et al., 2016). Although the Brewer instrument has difficulties in detecting low columnar SO2 concentrations, in extreme cases, such as volcanic eruptions, the SO2 levels typically rise well above the instrumental noise and can be identified with the Brewer instrument as shown in this paper and in Fioletov et al. (1998).

Before proceeding to the analysis of Brewer measurements, the methodology to derive columnar SO2 is first presented. To determine ozone and SO2 column amounts, the measured raw photon counts at the five operational channels in the Brewer instrument are converted to radiation intensity. The Beer–Lambert absorption law is applied at each wavelength λ, and the measured intensity of direct sunlight is given by the following formula: log⁡Iλ=log⁡I0λ-βλμR-δλμp-αλO3μ-σλSO2μ, where Iλ is the measured radiation intensity at wavelength λ, I0λ is the measured extraterrestrial spectrally resolved intensity at λ, βλ is the Rayleigh scattering coefficient at λ, δλ is the particulate scattering coefficient at λ, αλ is the ozone absorption coefficient (square centimetres per molecule) at λ, O3 is the total ozone column (molecules per square centimetre), σλ is the SO2 absorption coefficient at λ, SO2 is the column amount of sulfur dioxide, μR, μp and μ are the optical path lengths (air masses) corresponding to the effective heights of molecules, particles and ozone respectively.

According to the Brewer retrieval algorithm, the following ratios are formed: F=F0-ΔβμR-ΔαO3μ and F′=F0′-Δβ′μR-Δα′O3μ-Δσ′SO2μ, where F is the weighted ratio of direct sun measurements at four (or six for double Brewer) spectral channels, F=log⁡I2-0.5log⁡I3-2.2log⁡I4+1.7log⁡I5, and F0, Δβ, and Δα are the same linear combinations for log⁡I0λ, βλ, and αλ. The F′ is the SO2 ratio and F′=log⁡I1-4.2log⁡I4+3.2log⁡I5 and F0′, Δβ′, Δα′ and Δσ′ the corresponding linear combinations for log⁡I0λ, βλ, αλ and σλ. Both of these functions have weights which eliminate the effects of particulate scattering, while the function F is weighted to remove SO2 absorption effects as well. The extraterrestrial constants F0 and F0′ are determined from a long series of intercomparison measurements as well as zero air mass (μ) extrapolations.

The total ozone column is determined by the formula O3=F0-F-ΔβμRΔαμ and the SO2 by the formula SO2=1AF0′-F′-Δβ′μRΔα′μ-O3, where A is the ratio of the SO2 absorption coefficient to the O3 absorption coefficient; A=2.44.

From the above-described operational Brewer algorithm it is evident that the estimation of columnar SO2 is the result of the difference between two columnar terms (O3+ SO2) and O3. Both terms have uncertainties (weighting functions, calibrations, random errors, systematic errors). Systematic negative values could be the result of a systematic offset in the measurements that can be related to the calibration of the instrument (usually optimized only for the ozone measurements). Randomly varying positive and negative values around zero, suggest that the signal of SO2 is small (and thus the difference of two terms should be close to 0), but since both terms have uncertainties, negative values are possible, indicating that the amount of SO2 in the atmosphere is below the detection limit of the instrument and could be considered as noise. In this work we have repeated our analysis excluding the negative values, and the results remained the same; i.e. a positive increase after a major volcanic eruption was confirmed as described in the following sections. Finally, we need to point out that perturbations by ash present in the volcanic plumes have been shown not to affect the Brewer SO2 measurements. This is based on the result of the Pappalardo et al. (2013) paper based on EARLINET (European Aerosol Research Lidar Network) observations following the Eyjafjallajökull eruption, in which they found that the Ångström exponent of the volcanic ash optical depth is close to 0. This indicates that the effect of ash in the UV and visible region on the aerosol extinction is almost independent of wavelength. The Brewer SO2 measurements taken in a narrow wavelength band in the UV are therefore not expected to be influenced by the presence of volcanic ash considering the weights already applied in the operational Brewer algorithm.

In this study we analysed 23 stations located in Europe, 6 Brewer stations in Canada, 2 in the USA and 1 in Taiwan; their geographical positions are shown in Fig. 1. SO2 measurements were averaged over a large number of instruments and datasets during periods following volcanic eruptions. Random errors in the measurements of individual Brewer stations are reduced significantly by the averaging processes to calculate regional means.

Daily SO2 columns at Churchill, Goose, Edmonton, Regina, Saturna Island and Toronto in Canada, T'aipei in Taiwan, and Boulder and Mauna Loa in the US were obtained from the World Ozone and Ultraviolet Radiation Data Centre (WOUDC; http://www.woudc.org/) and the NOAA-EPA Brewer Spectrophotometer UV and Ozone Network (NEUBrew; http://www.esrl.noaa.gov/gmd/grad/neubrew/). The data have been checked for quality assurance/quality control by the individual data providers. It is important to note the participation of most of the European Brewer data providers in a recent EU COST Action (EUBREWNET, http://www.eubrewnet.org/cost1207/) programme. Its focus is to establish a coherent network of European Brewer Spectrophotometer monitoring stations in order to harmonize operations and develop approaches, practices and protocols to achieve consistency in quality control, quality assurance and coordinated operations.

In our analysis only DS measurements satisfying the following criteria have been used: a Brewer DS measurement was included if and only if for every measurement cycle of five sets of measurements (from which also total columnar ozone is derived) the standard deviation of O3 and SO2 was less than 2.5 DU, the total columnar ozone was between 250 and 450 DU, and the solar zenith angle was less than 73.5∘. To exclude erratic data of SO2 from our analysis, values exceeding ±6σ of the mean of all SO2 individual Brewer measurements were considered erroneous and were not included in the calculations. Therefore, the range of analysed values was limited to a maximum of ±35 DU for an individual measurement (i.e. 6σ, with σ being equal to 5.8 as estimated from all available sub-daily SO2 values). Then we calculated daily SO2 columns at each station only if at least three individual measurements passed these criteria for each day. Brewers are useful because they provide more than one observation per day. For plumes which change rapidly, more than one observation per day would be useful, especially to complement satellites which typically have just one local overpass.

Daily sulfur dioxide (SO2) columns were analysed in four bimonthly periods, namely August–September 2008, April–May 2010, June–July 2011 and September–October 2014; these include the volcanic eruptions of Kasatochi (2008), Eyjafjallajökull (2010), Nabro (2011) and Bárðarbunga (2014) respectively. For the case of Kasatochi, Eyjafjallajökull and Bárðarbunga, we analysed daily SO2 columns at 30 sites (listed in Table 2), while for the case of Nabro, whose impact was mostly seen over low latitudes in the Northern Hemisphere (e.g. Bourassa et al., 2012), we analysed SO2 columns at three low-latitude sites in the Northern Hemisphere, namely Izaña, Mauna Loa and T'aipei.

Only for the case of the Bárðarbunga eruption in 2014 were the columnar SO2 measurements over Europe also compared with surface SO2 measurements from ground-based European stations. The surface SO2 data were obtained from the European Environment Agency database (AirBase; http://www.eea.europa.eu/data-and-maps/data/aqereporting-1#tab-european-data) covering the bimonthly period September–October 2014. Only rural background stations, i.e stations in classes 1–2 according to the Joly–Peuch classification methodology for surface sulfur dioxide (Joly and Peuch, 2012), located at a distance of less than 150 km from the nearest Brewer station, were used in the analysis. A total of seven stations in Europe (see Table 3) fulfilled the above-mentioned criteria and were included in the current analysis. Observed data from the AirBase network were available in hourly resolution, from which we calculated daily surface SO2 values. We note here that SO2 in the troposphere over western Europe is very low (e.g. Zerefos et al., 2009; Wild, 2012), and therefore plumes from volcanic eruptions are easy to detect against a lower background level.

The columnar SO2 records from remote-sensing spectrophotometers over Europe, Canada, USA and Taiwan were compared with space-borne measurements from (a) the OMI aboard the EOS (Earth Observing System)-Aura (e.g. Ialongo et al., 2015) satellite and (b) the GOME-2 aboard the MetOp-A (e.g. Rix et al., 2009) satellite. We use MetOp-A instead of MetOp-B because it covers a longer time period. Both OMI and GOME-2 satellite SO2 data products were downloaded from the Aura Validation Data Center (AVDC) (available from http://avdc.gsfc.nasa.gov/index.php?site=_245276100). GOME-2 level-2 overpass data have been processed with the GOME Data Processor (GDP) version 4.7. We analysed station overpass data for the various midlatitude stations listed in Table 2 and for the low-latitude stations at Mauna Loa, Izaña and T'aipei. The available OMI version 1.2.0 overpass (collection 3) data analysed in this study include pixels within 50 km radius from the nearest Brewer site and is not affected by OMI row anomalies. The available GOME-2 level-2 overpass data include pixels within 100 km radius from the Brewer sites.

For the case of OMI, the SO2 data are provided from October 2004 to the present. There are four SO2 products: (1) the planetary boundary layer SO2 column (PBL), corresponding to a centre of mass altitude (CMA) at 0.9 km; (2) the lower tropospheric SO2 column (TRL) corresponding to CMA of 2.5 km; (3) the middle-tropospheric SO2 column (TRM), usually produced by volcanic degassing, corresponding to CMA of 7.5 km; and (4) the upper-tropospheric and stratospheric SO2 column (STL), usually produced by explosive volcanic eruptions, corresponding to CMA of 17 km. Details on OMI SO2 columns can be found in various studies (Levelt et al., 2006; Yang et al., 2007; Fioletov et al., 2011, 2013; McLinden et al., 2012; Li et al., 2013; Ialongo et al., 2015). In this study, we made use of the product for the middle-tropospheric SO2 column (TRM) following the recommendation that the TRM retrievals should be used for volcanic degassing at all altitudes because the PBL retrievals are restricted to optimal viewing conditions and TRL data are overestimated for high-altitude emissions (> 3 km) (Ialongo et al., 2015). Also, we analysed the STL data which are intended for use with explosive volcanic eruptions where the volcanic cloud is placed in the upper troposphere/stratosphere. The standard deviation of TRM retrievals in background areas is reported to be about 0.3 DU in low and midlatitudes and about 0.2 DU for the STL retrievals. This is similar to the standard deviations (indicative of typical uncertainties of the measurements) that we find for the TRM and STL retrievals in the four bimonthly periods under this study. For the best data quality, we used data from the scenes near the centre of the OMI swath (rows 4–54) as recommended by Ialongo et al. (2015), who found that data from the edges of the swath tend to have greater noise.

For GOME-2, we analysed the total SO2 columns from April 2007 to the present. The standard deviation found in our study for the GOME-2 retrievals is on the order of 0.4 DU. We analysed satellite SO2 measurements whenever the O3 column was between 250 and 450 DU and the solar zenith angle was less than 73.5∘. We used SO2 data defined as having a cloud radiance fraction (across each pixel) of less than 50 %, as they were found to have a smaller standard deviation than all sky data. Moreover, a range of SO2 values between -35 and 35 DU was used to screen for outliers. In cases of multiple daily data matched to the station overpass, all available measurements within a radius of 50 (100) km from the Brewer site in the case of OMI (GOME-2) are averaged.

Finally, both for the satellite and the Brewer data, we have considered that during a 10-day period prior to any eruption both the surface and the satellite datasets represent a baseline reference from which subsequent departures after the eruption should be tested as to their significance. Therefore, we calculated averages and standard deviations (σ) of departures from the unperturbed pre-volcanic period, for the three studied periods of volcanic importance at each station, only if at least 25 daily values were available. The bimonthly averages for each station in the examined periods are presented in Table 4a. Table 4b shows the mean and standard error (σ/√N) of all bimonthly averages in each studied volcanic period. Averaging the departures from the pre-volcanic baseline for all Brewer stations and for all bimonthly periods gives a mean SO2 columnar departure of 0.10± 0.03 DU. This estimate is on the same order of magnitude as the corresponding statistics for OMI (TRM) SO2 column departures (0.05± 0.02 DU), OMI (STL) (0.04± 0.01 DU) and that measured by GOME-2 (0.09± 0.02 DU). The standard deviation of the bimonthly averages relative to their baselines, which was calculated from a large sample of data, was taken here as an approximation of the typical uncertainties in the columnar SO2 measurements performed by the group of Brewers, OMI and GOME-2 instruments following volcanic eruptions.

Dispersion of volcanic emissions is simulated with the Lagrangian transport model FLEXPART (FLEXible PARTicle dispersion model; Stohl et al., 2005; Brioude et al., 2013). The model is driven by hourly meteorological fields from the Weather Research and Forecasting (WRF) atmospheric model (Skamarock et al., 2008) at a horizontal resolution of 45×45 km. The initial and boundary conditions for the WRF model are taken from the National Center for Environmental Prediction (NCEP) final analysis (FNL) dataset at a 1∘×1∘ resolution. The sea surface temperature (SST) is initialized from the NCEP 1∘×1∘ analysis. A total of 40 000 tracer particles are assumed for each release in FLEXPART simulations. The use of 1-hourly WRF meteorological fields at a 45×45 km spatial resolution allows a more detailed representation of the volcanic plume dispersion but also implies a significant increase in computational time. To overcome this computational time cost, source–receptor relationships between station measurements and volcanic activity are also analysed with the use of HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) model trajectories (Stein et al., 2015) of long-range transport driven by the 3-hourly meteorological dataset Global Data Assimilation System (GDAS) at a resolution of 1∘×1∘.

Bárðarbunga was continuously active during September–October 2014, but it was only during 18–26 September when meteorological conditions favoured transport towards Europe as shown by back trajectory analyses. A detailed description of the transport of Bárðarbunga plumes towards the station of Hohenpeißenberg is provided using the FLEXPART Lagrangian particle dispersion model offline coupled with the WRF_ARW atmospheric model. The simulation period is 18–26 September 2014. We assume a constant SO2 release rate of 119 kilotons per day as reported by Gíslason et al. (2015) from near the source SO2 measurements during the first weeks of the eruption. Similar emission rates are also suggested by Schmidt et al. (2015) through comparisons between NAME simulations (UK Met Office's Numerical Atmospheric-dispersion Modelling Environment) and OMI satellite retrievals. The emission height is set to between 0 and 3500 m above ground level, consistent throughout the simulation period. The establishment of an anticyclonic flow over the British Isles on 21 September 2014 (not shown here) resulted in the separation of the volcanic SO2 field into two distinct plumes (Fig. 2a). On 22 September the primary plume (plume 1) becomes stagnant over the topographic barrier of the Alps (Fig. 2b). The secondary plume is advected southwards by the intense northerly winds over the North Sea. The two plumes overlap at about 09:00–11:00 UTC. Taking a closer look at the surface SO2 values sampled during this event by surface air quality stations in the Netherlands, several days of enhanced SO2 were discovered, which indicate an area of stagnation or blocking of the flow. Trajectory calculations performed at the Royal Netherlands Meteorological Institute (KNMI) correspond well to the calculations shown in Fig. 2 but also show that the air parcels stayed over northern Europe for some time after a very fast flow over the North Sea, which agrees with the spikes found in the surface SO2 records observed over the Netherlands during a period of several days.

The high SO2 concentrations, which were recorded almost simultaneously at stations over Europe at various sites during the period 21–29 September 2014, are thus associated with the activity of Bárðarbunga volcano (Ialongo et. al., 2015; Table A1, see Appendix A). This is also supported by the back trajectory analysis performed with the HYSPLIT dispersion model that is shown in Fig. 3. All back trajectories start at 12:00 UTC on the day of maximum SO2 observations for each of the Brewer stations and indicate that the arrival of air masses originated from Iceland.

As shown in Fig. 4a, the SO2 plume was detected by the Brewer instruments located in the passage of the volcanic SO2 plume and from different ground-based networks. However, no coincident measurements were available from the OMI and GOME-2 overpasses at the time of the high SO2 excursions. Also it should be noted here that no enhanced SO2 columns were detected by the Brewers located outside of the geographical area covered by the volcanic plume (Fig. 4b). In all volcanic cases we have applied a criterion according to which each daily average from either OMI or GOME-2 should be calculated if and only if more than half of the individual overpasses had data on a given day.

The eruption took place at the beginning of September 2014, and several European countries experienced high concentrations of SO2 at ground level during the rest of September. Figure 5 shows the response of ground-level AirBase stations under the plume located within 150 km from the nearest Brewer station plotted together with the coincident Brewer SO2 column measurements. Interestingly, it suggests that the highest amount of the SO2 column measured by the majority of the Brewers on 21 September 2014 due to the volcano reached the surface with a time lag of about 1 day. The high volcanic concentrations were successfully measured by the ground-based AirBase network. Due to strong European efforts over the last decades to reduce SO2 emissions, high concentrations of SO2 are now quite rare in western Europe (e.g. Vestreng et al., 2007) except in the areas affected by industrial or shipping emissions. In situ air quality stations observed high values of SO2 at the ground level on the coast of France, in the United Kingdom, the Netherlands and Germany between 21 and 25 September 2014. This all points towards a volcanic episode with a large spatial extent.

As can be seen from Fig. 4a, the highest SO2 column departures from the pre-volcanic baseline were observed from 21 to 22 September 2014. The mean SO2 column measured by the Brewers under the plume was 2.4± 0.8 DU, which was 5 times greater than the mean column of SO2 measured by the Brewers outside of the plume (-0.1± 0.1 DU) by 2.5 DU on average. The error bars show the standard deviation of the daily SO2 values of all stations during the non-perturbed 10-day period prior to the volcanic eruption. These differences provide rough estimates of the additional SO2 loading induced by the volcanic eruption over Europe which exceeds 3σ. Comparison between satellite data and Brewer are limited for the purposes of interpretation because satellite measurements are sparse, representing an average SO2 column over a relatively large satellite pixel, while the Brewer observations are designed to provide a local point measurement. In spite of the sparsity of OMI observations post Bárðarbunga volcanic eruption, satellite data were used for assimilation in the SO2 analyses and forecasts produced with the MACC (Monitoring Atmospheric Composition and Climate) system (http://atmosphere.copernicus.eu/). This near-real-time forecasting system assimilates satellite observations to constrain modelling forecasts (Inness et al., 2015; Flemming et al., 2015). The OMI instrument aboard the Aura satellite provided information about concentrations of volcanic SO2 emitted by the Icelandic Bárðarbunga volcano on 20 September; these observations were assimilated in 2014 by the MACC system in cases of volcanic eruptions, i.e. when OMI values exceeded 5 DU. As shown in Fig. 6 (the charts of total column SO2 are taken from the website http://atmosphere.copernicus.eu/), the subsequent forecasts capture the transport of the plume of volcanic SO2 southward, while spreading over the continent on 21 and 22 September. The plume stretched all the way from Finland through Poland, Germany and France to southern England. A parallel forecast, for which no OMI data were used (Fig. 6, right), did not show any elevated SO2 values, confirming that “normal” emissions of SO2 (including shipping and industrial activities) could not explain the observed situation.

Finally, it should be mentioned here that the thin aerosol layer that was detected by the PollyXT lidar (Engelmann et al., 2016) over Leipzig at around 2–3 km on 23 and 24 of September 2014 was mostly associated with volcanic ash advection (Fig. 7). A corresponding cluster analysis of all 155-hourly HYSPLIT back trajectories during this period and for the heights of the layer detected by the lidar (∼ 2.5–3.5km) is shown in Fig. 8. The increased wind shear that is evident between these heights does not allow a robust characterization of the air masses. However, the source contribution of about 20 % from Icelandic air masses supports the volcanic origin of the detected plume. During volcanic eruptions, ash and SO2 may be injected to different altitudes and may follow different trajectories for long-range transport. EARLINET lidars can provide alerts on volcanic ash dispersion over Europe, especially when the systems are employed with depolarization capabilities (e.g. Pappalardo et al., 2013). For the Brewer network capabilities and the Hohenpeißenberg station, Figs. 7 and 8 demonstrate that a similar approach can be applied to contribute towards an early warning synergistic tool, as evidenced in the Bárðarbunga case study. The role of the Brewer stations in this system will be the early detection of SO2 plumes transported over continental areas that would trigger the associated forecasting systems (models and networks).

A major eruption of Mt Nabro, a 2218 m high volcano on the border between Eritrea and Ethiopia (13.37∘ N, 41.7∘ E), occurred on 12–13 June 2011. The volcanic eruption injected ash, water vapour and an estimated 1.3–2.0 Tg of SO2 into the upper troposphere and lower stratosphere (Fairlie et al., 2014, and references therein). In the first phase of the eruption, the main transport pattern of emitted SO2 followed the strong anticyclonic circulation over the Middle East and Asia associated with the Asian summer monsoon at that time of year (Clarisse et al., 2014, and references therein). In the first month after the eruption stratospheric aerosols were mainly observed over Asia and the Middle East and by day 60 covered the whole Northern Hemisphere. Estimated aerosol altitudes from various instruments were between 12 and 21 km (Clarisse et al., 2014). By July 2011 Nabro had cumulatively emitted 5 to 10 % of what was released by Mount Pinatubo in 1991 (∼ 20 Tg), ranking it among the largest SO2 emissions in the tropical stratosphere (up to at least 19 km) since Pinatubo (Krotkov et al., 2011). Sulfur dioxide signals of volcanic origin were detected both by Brewer and satellite measurements over East Asia, where the volcanic SO2 plume was transported, as demonstrated in Figs. 9 and 10a. Measurements were taken by Brewer in T'aipei, Taiwan, Asia. This is also evident from the back trajectory analysis performed with the HYSPLIT dispersion model for T'aipei (Taiwan) as shown in Fig. 10a. The analysis indicates that the upper-tropospheric air masses arriving at T'aipei on 19 June, when the peak in SO2 is observed, originate from Africa.

The Nabro volcanic plume was mainly transported to East Asia and was detected by various satellite instruments which provide better spatial coverage than the Brewers. A special case study focuses on discrepancies found between ground-based and satellite observations of the volcanic SO2 plume. A Brewer located in Tenerife, Spain, detected an increase in the SO2 column, which was not clearly detected by the OMI and GOME-2 satellite overpasses.

More specifically, Fig. 10b shows back trajectories from Izaña (Tenerife) during 19–29 June 2011 at 15, 17.5 and 20 km heights. It appears that the upper-tropospheric–lower-stratospheric air masses arriving at Tenerife during 19–29 June originated from Nabro. In June 2011 the Nabro volcano ash plume was detected by the Micropulse Lidar (MPL) located at Santa Cruz de Tenerife (Canary Islands, Spain). The volcanic plume height ranged from 12 km on 19 June to 21 km on 29 June (Sawamura et al., 2012). Figure 11 shows the columnar SO2 departures from the unperturbed 10-day pre-volcanic baseline measured by the Brewer at Izaña following Nabro. The daily mean SO2 departures (Fig. 11) show a 0.3 DU increase at the beginning of the event (19 June), reaching 0.6 DU on 29 June when the layer is found at a higher altitude. The signal is not strong and is near the error of 0.5 DU estimated for SO2 measurement (Stanek, personal communication, 2016), but the observations are consistent (independent of the ozone and air mass), since at Izaña about 100 O3/SO2 measurements per day are performed resulting in reduced standard errors associated with daily means as compared to individual observations. The Langley calibration is tracked between calibrations by measurements of the internal lamp (Langley and lamp are shown in Supplement Fig. S1). The increase in SO2 due to the passage of the Nabro volcano plume over the Canary Islands is significant using both methods (Fig. 11).

In this case the Brewer at Izaña has been able to detect an SO2 plume at a high altitude from a volcano located 7000 km from the Canary Islands, indicating that the Brewer network is sensitive enough to be incorporated into columnar SO2 monitoring from volcanic eruptions in worldwide networks.

The case of the 2011 Nabro eruption shows an example of the importance of the Brewer spectrophotometers in measuring and detecting changes in SO2 amounts in the atmosphere due to volcanic eruptions, in cases where signal in the satellite overpasses is low. This is true for the case of Izaña (Tenerife), where it appears that OMI and GOME-2 did not clearly detect increases in SO2 column of volcanic origin between 19 June and 1 July as was the case with the Brewer instrument (Fig. 11). During some days between 19 June and 1 July, the Brewer SO2 columns at Izaña rose above the uncertainty of 0.5 DU for the Brewer SO2 measurements at this station, whereas the satellite SO2 columns stayed mostly within the uncertainty of 0.4 DU estimated for OMI and GOME-2 satellite retrievals.

These findings can provide clues to the detection limits of such events from a well-calibrated Brewer network and a space-borne instrument. They need further clarification with more Brewers and a larger number of cases.

The Eyjafjallajökull volcano, Iceland (63.63∘ N, 19.6215∘ W; 1666 m a.s.l.), erupted explosively on 14 April 2010 and continued to emit ash and gas until 24 May (Flentje et al., 2010; Thomas and Prata, 2011; Stohl et al., 2011; Flemming and Inness, 2013). Despite the relatively modest size of the eruption, the prevailing wind conditions advected the volcanic plume toward the southeast leading to unprecedented disruption to air traffic in western Europe. This caused significant financial losses for the airlines and highlights the importance of efficient volcanic cloud monitoring and forecasting. Results demonstrate that the eruption can be divided into an initial ash-rich phase (14–18 April), a lower-intensity middle phase (19 April until early May) and a final phase (4–24 May), where considerably greater quantities of both ash and SO2 were released (Thomas and Prata, 2011).

Figure 12 shows the responses of Brewer stations under the volcanic SO2 plume and the average of Brewer stations outside of the plume together with OMI and GOME-2 satellite observations. We determined 10 stations as being under the plume in 2010 and 10 stations as being outside of the plume based on the analysis of forward and backward trajectories of air masses following the volcanic eruption. The stations determined to be under the plume in 2010 (shown in Fig. 12a) are Sodankylä, Obninsk, Manchester, De Bilt, Uccle, Belsk, Reading, Hohenpeißenberg, Davos and Arosa. The stations determined to be outside of the plume are Vindeln, Oslo, Norrköping, Copenhagen, Hradec Králové, Aosta, Kislovodsk, Rome, Thessaloniki and Athens (Fig. 12b).

We should note here that volcanic clouds can be rather narrow plumes with diameters on the order of a few tens of kilometres (e.g. Stohl et al., 2011; Webley et al., 2012; Thorsteinsson et al., 2012; Kristiansen et al., 2012; Kokkalis et al., 2013), and thus it is possible that a volcanic layer detected at a specific station is not observed by neighbouring stations. The measurements at Uccle and De Bilt that are located at a horizontal distance of 150 km are different during the Eyjafjallajökull episode and provide a very good example. On 2 May 2010 the mean daily SO2 is 5.8 DU at De Bilt and -0.2 DU at Uccle. As seen in the corresponding back trajectories for that day in Fig. 13, air masses originating from Iceland arrive at De Bilt at heights of 6–7 km and have probably transported the volcanic cloud over that station. In contrast similar back trajectories on the same day for the case of Uccle indicate transport of air masses from Iceland but at lower heights (3–4 km) that were probably not affected by the volcanic emissions. In another case of transport on 11 May 2010, Uccle was outside of the plume (see Fig. 13), while the mean daily SO2 for De Bilt was 0.9 DU. On 18 May 2010, both De Bilt and Uccle stations detected a volcanic cloud with SO2 daily means of 1.7 and 1.2 DU respectively. To summarize, in spite of the proximity of Uccle and De Bilt, the transport heights and trajectories can have a different result in transporting volcanic gases.

In Table A1 of Appendix A, we present the dates when the examined Brewer stations were either under or outside of the volcanic SO2 plume. Careful analysis of the trajectories of the volcanic plumes in 2010 and 2014 helped verify these analyses. The distinction between stations outside of the plume and stations under the plume was done as follows: whenever SO2 at each station measuring exceeded 2 DU (2σ), back trajectories were calculated and the origin was compared to the location of the volcanic eruption. All these stations have been considered to be under the SO2 plume. All other stations, for which columnar SO2 amounts were within 2σ and did not originate from the area of the eruption, were considered to be outside of the volcanic SO2 plume.

As we can see from Fig. 12, the columnar SO2 departures at stations located under the passage of the volcanic SO2 plume exceeded 0.3 DU (reaching 1.5 DU in some cases), whereas at stations located outside of the plume, the columnar SO2 departures did not exceed 0.3 DU. Moreover, during the explosive phase 2, there were three main periods in which the volcanic aerosol content was observed by EARLINET over Europe: 15–26 April, 5–13 May and 17–20 May. These periods were determined from measurements of the integrated backscatter at 532 nm in the volcanic layers (Pappalardo et al., 2013). We estimate high SO2 columnar departures measured by the Brewers under the plume during 14 April and 23 May 2010 of up to 6.0 DU (e.g. Arosa, 18 May 2016).

We note here that the ash cloud caused further disruptions to air transportation on 4–5 May and 16–17 May 2010, particularly over Ireland and the UK. The average SO2 columnar departures measured by the Brewers under the plume in the UK (Manchester and Reading) during these two periods were estimated to 1.1± 0.3 and 1.5± 0.4 DU respectively. These amounts were higher than the amounts measured outside of the plume (-0.1± 0.2 and -0.1± 0.1 DU, accordingly) by almost 1.4 DU on average.

The eruption of Kasatochi volcano on 7–8 August 2008 injected large amounts of material and SO2 into the troposphere and lower stratosphere of the northern middle latitudes during a period of low stratospheric aerosol background concentrations. The Kasatochi volcano in the central Aleutian Islands of Alaska (52.17∘ N, 175.51∘ W) erupted three times between 22:01 UTC on 7 August and 04:35 UTC on 8 August 2008 (Bitar et al., 2010). Aerosols from the volcanic eruption were detected by lidar in Halifax shortly after the eruption (Bitar et al., 2010). The total mass of SO2 injected into the atmosphere by the eruption is estimated at 1.7 Tg, with about 1 Tg reaching the stratosphere (above 10 km a.s.l.) (Kristiansen et al., 2010).

We have studied the columnar SO2 amounts following the Kasatochi eruption in August 2008 from ground-based and satellite data. Figure 14 shows the columnar SO2 departures from the unperturbed 10-day pre-volcanic period over Canada and the USA and Europe during the bimonthly period August–September 2008 as measured by the Brewers in comparison with the satellite observations by OMI and GOME-2.

The SO2 plume was clearly seen by the Brewers in Canada and the USA (Fig. 14), and it was also detected by the majority of the Brewers in Europe with a delay of about 3 days. The total SO2 columnar departures averaged over Canada during the period 12–20 August 2008 are estimated to 0.9± 0.2 DU. Accordingly over Europe, we estimate a mean SO2 columnar departure of 1.0± 0.1 DU during the period 15–22 August 2008. This number gives a rough estimate of the average volcanic SO2 column measured by the Brewers over Europe. We note here that the e-folding time of the Kasatochi SO2, i.e. the time where the volcanic SO2 amount decayed, was estimated to be about 8–9 days (Krotkov et al., 2010).

The high amounts of SO2 and the variability of SO2 measured in Europe by the Brewers after the eruption of Kasatochi in August 2008 are in line with OMI and GOME-2 satellite observations. More specifically, OMI (TRM) shows an average SO2 columnar departure of 0.5± 0.1 DU during the period 15–22 August 2008, OMI (STL) an average SO2 columnar departure of 0.4± 0.1 DU and GOME-2 an average SO2 columnar departure of 0.8± 0.1 DU.

The Brewer data have been correlated with those from OMI and GOME-2. The Pearson's correlation coefficients between the three datasets were all highly statistically significant (> 99 %). The correlation between SO2 from the Brewers and SO2 from GOME-2 at 19 stations averaged over Europe is +0.86 (t value = 12.54; p value< 0.0001; N=59). Accordingly, the correlation between Brewer and OMI (TRM) SO2 data is +0.86 (t value = 11.77; p<0.0001; N=50), and between GOME-2 and OMI (TRM) data, it is +0.92 (t value = 16.32, p<0.0001, N=48). The same correlations are found for the Brewer–OMI (STL) and GOME–OMI (STL) data pairs. These correlations were calculated from 60 daily averages during the Kasatochi volcanic eruption in August–September 2008. The statistical tests gave significant results and verified the capability of the Brewers in detecting natural SO2 emitted by volcanoes when the volcanic plume of SO2 passes over the ground sites. We note here that there is a general consistency between all three datasets (Brewers, OMI and GOME-2) on the changes in SO2 column following the Kasatochi volcanic eruption.

Table 5 summarizes the correlation coefficients between the mean columnar SO2 measured by all Brewers over Europe and provided by the satellite products of OMI and GOME-2 during the globally extended Kasatochi event. The correlation coefficients have high statistical significance explaining more than 70 % of the total variance between the columnar SO2 measurements from the ground and space in the case of Kasatochi. However, the discrepancies found between satellite and Brewer observations during the other volcanic eruptions could be impacted by the sparsity of coincident measurements and thus cannot confirm or deny Kasatochi case findings at high significance levels.