OClO and BrO observations in the volcanic plume of Mt . Etna – implications on the chemistry of chlorine and bromine species in volcanic plumes

Spatial and temporal profiles of chlorine dioxide (OClO), bromine monoxide (BrO) and sulfur dioxide (SO2) of the volcanic plume at Mt. Etna, Italy, were investigated in September 2012 using Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS). OClO was detected in 119 individual measurements covering plume ages up to 6 min. BrO could be detected in 452 spectra up to 23 min downwind. The retrieved slant column densities (SCDs) reached maximum values of 2.0 × 10 moleculescm (OClO) and 1.1 × 10 moleculescm (BrO). Mean mixing ratios of BrO and OClO were estimated assuming a circular plume cross section. Furthermore, ClO mixing ratios were derived directly from the BrO and OClOSCDs. Average abundances of BrO= 1.35 ppb, OClO= 300 ppt and ClO= 139 ppt were found in the young plume (plume age τ < 4 min) with peak values of 2.7 ppb (BrO), 600 ppt (OClO) and 235 ppt (ClO) respectively. The chemical evolution of BrO and OClO in the plume was investigated in great detail by analysing the OClO/SO2 and BrO/SO2 ratios as a function of plume age τ . A marked increase of both ratios was observed in the young plume (τ < 142 s) and a levelling off at larger plume ages showing mean SO2 ratios of 3.17 × 10 −5 (OClO/SO2) and 1.65 × 10 (BrO/SO2). OClO was less abundant in the plume compared to BrO with a mean OClO/BrO ratio of 0.16 at plume ages exceeding 3 min. A measurement performed in the early morning at low solar radiances revealed BrO/SO2 and OClO/SO2 ratios increasing with time. This observation substantiates the importance of photochemistry regarding the formation of BrO and OClO in volcanic plumes. These findings support the current understanding of the underlying chemistry, namely, that BrO is formed in an autocatalytic, heterogeneous reaction mechanism (in literature often referred to as “bromine explosion”) and that OClO is formed in the reaction of ClO with BrO. These new findings, especially the very detailed observation of the BrO and OClO formation in the young plume, were used to infer the prevailing Cl-atom concentrations in the plume. Relatively small values ranging from [Cl] = 2.5 × 10 cm (assuming 80 ppb background O3) to [Cl] = 2.0 × 10 cm (at 1 ppb O3) were calculated at plume ages of about 2 min. Based on these Cl abundances, a potential – chlorine-induced – depletion of tropospheric methane (CH4) in the plume was investigated. CH4 lifetimes between 14 h (at 1 ppb O3) and 47 days (at 80 ppb O3) were derived. While these lifetimes are considerably shorter than the atmospheric lifetime of CH4, the impact of gaseous chlorine on the CH4 budget in the plume environment should nevertheless be relatively small due to plume dispersion (decreasing Cl concentrations) and ongoing mixing of the plume with the surrounding atmosphere (replenishing O3 and CH4). Published by Copernicus Publications on behalf of the European Geosciences Union. 5660 J. Gliß et al.: OClO and BrO observations in the volcanic plume of Mt. Etna In addition, all spectra were analysed for signatures of IO, OIO and OBrO. None of these species could be detected. Upper limits for IO/SO2, OIO/SO2 and OBrO/SO2 are 1.8 × 10, 2.0 × 10 and 1.1 × 10 respectively.


40
In the past years, improved measurement techniques, especially remote sensing methods, gained importance for the study of the chemical composition of volcanic plumes. In this study we present MAX-DOAS measurements (e.g. Hönninger et al., 2004) performed in the volcanic plume of Mt. Etna on Sicily, Italy. MAX-DOAS is an established method to simultaneously study a variety of chemical species in volcanic plumes by analysing scattered 45 sunlight spectra. Furthermore, it is easily possible to monitor the volcanic emissions over a wide range of different plume ages, which is of particular importance for studies related to the chemical evolution of the emitted species. It is well known, that volcanic gases can have significant impacts on atmosphere and climate both on local and global scales (e.g. acid rain, stratospheric sulfur aerosols, see e.g. Robock, 2000). Furthermore, a detailed knowl-50 edge of the chemical composition of volcanic plumes -especially in the young plume -is helpful to give insights into the degassing behaviour of the magma and may even be related to the state of activity of the volcano (e.g. Donovan et al., 2014)). In addition, the environment of volcanic plumes provides an unique possibility to study details of complex chemical 3 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | reaction mechanisms related to atmospheric ozone/oxidant chemistry in the presence of 55 reactive halogen species (RHS).
The focus of this article is on the chemical evolution of volcanic halogens, especially on the formation of RHS (e.g. BrO, ClO, OClO) from the primarily emitted species (e.g. HCl, HBr) and their evolution in the ageing plume. The scientific interest in volcanic RHS increased dramatically, when large amounts of bromine monoxide (BrO) were detected in 60 the plume of Soufriére Hills volcano, Montserrat (Bobrowski et al., 2003). Today, we have only gained a rough understanding of the chemical processes involved in the RHS-formation in volcanic plumes and possible dependencies due to the presence of other species (e.g. ozone or nitrogen oxides). Especially the conversion of the emitted HCl into reactive chlorine is still poorly understood. This is both due to a lack of measurement data and the complexity 65 of the chemical processes involved.

Initial plume composition
The main constituents of volcanic plumes are H 2 O, CO 2 and sulfur gases (dominated by SO 2 , H 2 S). Apart from these species, volcanoes also emit a certain amount of halogen 70 species which is largely dominated by chlorine emissions (e.g. Textor et al., 2004). Volcanic halogens are mainly released in the rather un-reactive form of hydrogen halides such as HCl, HF, HBr, HI (e.g. Carroll and Holloway, 1994;Francis et al., 1995;Gerlach, 2004). Pyle and Mather (2009) reviewed past measurements (∼ 1980-2008) of arc-related volcanic halogen emissions around the globe and found that HCl emissions contribute most 75 with an estimated flux of 4.3 (±1) Tg a −1 . HBr and HI emissions are orders of magnitude smaller with fluxes of 5-15 Gg a −1 and 0.5-2 Gg a −1 respectively.
In case of Mt. Etna, SO 2 / HCl-ratios between 2-7 were found in past measurements (e.g. Francis et al., 1995, La Spina et al., 2010, Voigt et al., 2014. SO 2 appears to dominate 80 the total sulfur emissions of Etna with SO 2 / H 2 S-ratios of the order of 10 1 − 10 3 (Jaeschke 4 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | et al., 1982, Aiuppa et al., 2005. A certain amount of the emitted hydrogen halides is converted into RHS whereas the conversion from HBr into BrO appears to be much more efficient than the analogous 85 reactions for volcanic chlorine. A key question related to these conversion mechanisms is the production of the halogen radicals (i.e. Br, Cl, I) in the plume. Once these are provided, oxidised halogens such as BrO and ClO are formed in reaction with ozone (O 3 ). 90 A certain amount of RHS (e.g. Cl, Br) can be produced in the hot initial plume via high temperature oxidative dissociation processes as suggested by model studies (e.g. Gerlach, 2004;Martin et al., 2006). Furthermore, Br can be formed via the reaction of HBr with OH in the very young plume (Roberts et al., 2009). However, the corresponding amounts are by far too small to explain the BrO amounts observed. In fact, the largest part of BrO is formed 95 in atmospheric reactions including the photolysis of Br 2 and BrCl (e.g. Oppenheimer et al., 2006;Kern et al., 2009). This is further supported by direct observations showing a strong increase of the BrO levels in the young plume (e.g. Bobrowski and Giuffrida, 2012) and the virtual absence during night-time . 100 Nowadays, the underlying chemical reaction processes are mostly understood and likely driven by a heterogeneous and partly auto-catalytic reaction mechanism often referred to as "bromine explosion" (e.g. Lehrer et al., 1997;Wennberg, 1999), which includes the following reactions (note that the subscript "aq" denotes species in the aqueous phase on particles):

Formation of RHS in the plume -the bromine explosion
The "bromine explosion" encompasses the uptake of hypobromous acid (HOBr) from the gas into the aqueous phase. After reaction of HOBr with bromide, Br 2 is released into the gas phase where it is rapidly photolysed forming BrO in reaction with O 3 . Once formed, the self reaction of BrO induces a catalytic destruction of O 3 . Noteworthy in this context are the 115 similarities to observations of bromine emissions in polar regions (e.g. Barrie et al., 1988;Simpson et al., 2007). Measurements performed at Mt. Etna and Stromboli volcano (Aeolian islands, Italy) indicate that up to 11% of the total emitted bromine is converted into BrO already within the first five minutes downwind (Wittmer et al., 2014). 120

Volcanic chlorine
Potential formation processes of reactive chlorine species from the emitted HCl are still little studied. Apparently, the activation of chlorine is much weaker compared to bromine. This is indicated by the comparatively low Cl and ClO x abundances we found (relative to the BrO ratios), indicating that less than 1% of the emitted HCl is converted into reactive chlorine 125 (ClO y ) in the Etna plume. In other words ClO y / HCl is much smaller than BrO/HBr. In our opinion this phenomenon is mainly due to the fact, that Br-oxidation (conversion of bromide to Br, BrO) is a self-amplifying process (the bromine explosion) while Cl-oxidation has no such properties. The reason why Br "explodes" but Cl does not is due to the relatively fast reaction of Cl-atoms with CH 4 (Platt andJanssen, 1995, Platt, 2000). Moreover, the 130 6 dissolved chloride ions are less reactive compared to bromide ions (see R4) (von Glasow et al., 2009). Thus, Cl-release is rather likely to be a by-product of the bromine explosion via formation of BrCl in the reaction of HOBr with chloride. However, the efficiency of this chlorine release channel strongly depends on the Cl − /Br − -ion ratio in the condensed phase (Fickert et al., 1999). A significant release of BrCl is only likely for Cl − /Br − -ratios exceeding 135 10 4 , for instance a Cl − /Br − -ratio of 2 × 10 4 would yield a release of 50 % BrCl and 50 % Br 2 . Direct sampling measurements at Mt. Etna revealed Cl − /Br − -ratios of the order of 10 2 (Martin et al., 2008) up to 10 3 (Wittmer et al., 2014). For these values the HOBr uptake rather yields a release of more than 90 % Br 2 . Note, that this favoured Br 2 release is probably even enhanced in volcanic plumes due to the acid environment (low pH in the aerosol, 140 for details see Fickert et al., 1999, i.e. the pH dependency of the discussed mechanisms). The corresponding reaction rate coefficient is k 8 = 6 × 10 −12 cm 3 s −1 (at 298 K, Sander et al., 2006). Further possible reaction channels for the OClO formation are orders of magnitude slower (e.g. ClO+O 3 , ClO+ClO, Sander et al., 2006) and were not considered within this study. The main daytime sink of OClO is its photolysis:  and General et al. (2014) detected OClO in the plume of Mt. Etna. The corresponding OClO/SO 2 -ratios were between 3-6 × 10 −5 (for spectra related to the plume centre). Simultaneous BrO measurements indicate an OClO/BrO-ratio of approximately 0.25 for Mt. Etna in both studies. Further detections of volcanic OClO are reduced to satellite measurements (Puyehue-Cordón Caulle volcano, Chile) after an erup-160 tion in 2011 (Theys et al., 2014) and most recently, the detection of OClO in the plume of Soufriére Hills volcano (Montserrat) during a hiatus in 2011 (Donovan et al., 2014). In the latter study, comparatively large OClO/SO 2 -ratios (4-6× 10 −4 ) are reported as well as large OClO/BrO-ratios showing values up to five (i.e. about 20 times larger compared to Mt. Etna). 165 A key parameter for the OClO formation in volcanic plumes is the prevailing availability of ClO and BrO molecules. Previous studies reported relatively large amounts of volcanic ClO measured with passive DOAS instruments Lee et al., 2005). The corresponding ClO/SO 2 ratios were of the order of 5 % hence, almost three orders of magnitude more ClO than OClO. However, these measurements have to be treated 170 cautiously due to difficulties and uncertainties in the DOAS evaluation of ClO. Furthermore, to our knowledge it has not yet been possible to reproduce these measurements in model studies (e.g. von Glasow, 2010). Kern et al. (2009)

investigated
ClO and OClO abundances at the vent of Masaya Volcano (Nicaragua) using an active Long-Path DOAS instrument. They did not detect any of both species, most likely due to 175 the proximity of the measurement to the crater (i.e. early stage of the RHS formation). In addition, the halogen-content of Masaya volcano is probably smaller compared to Mt. Etna (Pyle and Mather, 2009).
While the OClO/ClO-ratios should typically of the same order of magnitude in case of Mt.
Etna and Masaya , General et al., 2014, this seems not to be the case 180 for the Puyehue-Cordón Caulle eruption in 2011 which rather indicates a large excess of ClO compared to OClO and even BrO (Theys et al. (2014)).
8 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | The focus of this article is regarding the temporal and spatial evolution of RHS in volcanic plumes (especially BrO, ClO, OClO) and potential impacts on the atmosphere 185 in the vicinity of volcanic plumes. In particular, we use MAX-DOAS data to study the formation of BrO and OClO in the young plume in great detail and to infer typical formation times of these species (for the conditions at Mt. Etna in September 2012). We furthermore estimate mean plume abundances of BrO, ClO and OClO. These results are used to derive Cl-atom concentrations in the plume in order to address the question of a poten-190 tial -chlorine induced -depletion of atmospheric methane (CH 4 ) in the plume environment.

Technical setup
The MAX-DOAS instrument used in this study analyses the solar spectrum in the ultraviolet 195 (UV) and the visible (VIS) range using two spectrographs (UV: Avantes AVA AvaBench-75-ULS2048x64, VIS: Avantes AVA AvaBench-75-ULS2048L) covering a spectral range of 292-578 nm (UV: 292.1-456.1 nm, VIS: 434.7-577.8 nm). Scattered sunlight was collected using a small telescope consisting of a quartz lens (f = 100 mm) which focuses incoming light onto an optical fibre bundle. The latter consists of seven individual fibres with each 200 a diameter of d = 100 µm. Six of these were coupled into the UV spectrograph, the seventh fibre into the VIS, respectively. The measured spectral resolution of both spectrographs was ∆λ UV = 0.51 nm and ∆λ VIS = 0.39 nm. A SCHOTT BG-3 filter was placed behind the entrance slit of the UV spectrograph to reduce stray light. The telescope was focused such, that both spectrographs have approximately the same field of view (UV: 0.15 • , VIS: 0.16 • , 205 full aperture angle). The optical benches of the spectrographs were thermally insulated and temperature stabilised using a Peltier element controlled by a Supercool PR-59 temperature controller. During the whole measurement campaign, both spectrographs were stabilised to a temperature of T meas = 10 • C. The air tight instrument-box was mounted onto 9 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | a tripod. Two motors (azimuth and elevation) allowed to control the viewing direction of the 210 telescope, geo-locations were recorded using a gps-receiver. All hardware elements were remotely controlled using an embedded PC. The software MS-DOAS was used for data acquisition. MS-DOAS was developed by U. Frieß at the Institute of Environmental Physics in Heidelberg and is designed to control standard hardware components used in DOAS instruments (e.g. spectrographs, motors, temperature controller, gps receiver). Furthermore, 215 it includes a scripting feature making it easily possible to automate measurement and scanning routines.

Measurement location and data acquisition
Mt. Etna is the largest and most active volcano in Europe and is situated in the eastern part of Sicily, an island south of the Italian mainland. The activity of Mt. Etna shows a distinct 220 variability including quiescent degassing periods as well as eruptive periods. During the measurement campaign in September 2012, Etna showed a stable quiescently degassing behaviour from the four active craters -North East (NE), Bocca Nuova (BN), Voragine (VOR) and South East crater (SE) -which are located in the summit region at an altitude of approximately 3300 m a.s.l.. The first three days of the campaign (11-13 September 2012) 225 took place at the Etna observatory (Pizzi Deneri) which is located approximately 2.5 km north-east of the active summit at an altitude of 2800 m. Figure 1 shows a photo of the volcano and the emission plumes from the different craters. The photo was taken from the observatory and shows the NE crater (right) and the SE crater (faint in the background) on 13 September at 07:24 UTC. The plume was slightly condensed (see Fig. 1) during most 230 of the measurements performed in September 2012 and showed no visible indications of any ash emissions. In Fig. 2, all measurement locations of the campaign (11-26 September 2012) are indicated. One of the main objectives of this study was to investigate the temporal evolution of oxidised halogens in the volcanic plume. Therefore, the measurements were performed at different locations in order to cover a large variety of different 235 plume ages in the spectra.
10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Plume scanning routines
Three different plume scanning routines ("scans") were performed in order to study the chemical variability of the measured species in the volcanic plume (see sketch in Fig. 3). One "scan" typically consists of a set of plume spectra plus a subsequently recorded solar 240 reference spectrum with the telescope pointing into a volcanic gas free atmosphere (for details see Sect. 2.4).

Plume evolution scans
The purpose of "plume evolution scans" is to study the chemical evolution of the measured species as a function of the plume age. The spectra are therefore recorded at different 245 plume ages along the plume propagation direction (typically in the centre of the plume, see

Plume cross section scans
"Plume cross section scans" are performed perpendicular to the plume propagation axis in order to study chemical variations between the centre and the edges of the plume (Fig. 3b).

Point measurements
A certain number of spectra are taken at a fixed point in the plume without changing the viewing direction of the telescope. This measurement type is suited for the analysis of temporal variations in the plume composition (Fig. 3c). 255 The spectra were analysed using the software package DOASIS (v. 3.2.4422, Kraus, 2006). Details on the scanning routines can be found in Sect. 2.3. In order to improve the detection sensitivity, several hundred up to 1500 individual spectra were co-added for the DOAS analysis. A standard DOAS fit (see Platt and Stutz, 2008) was performed for the UV and 11

Data acquisition and DOAS-evaluation
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | VIS spectra in order to retrieve slant column densities (SCDs) of the chemical species in 260 the plume (in this study mainly: OClO, BrO, SO 2 , IO, OBrO, OIO). A Fraunhofer reference spectrum (FRS, I 0 (λ)) was included into the fitting routines to account for solar absorption lines in the spectra (Fraunhofer lines) and atmospheric background absorption. The FRS were recorded with the telescope pointing in the direction of a volcanic gas-free atmosphere and close in time to the corresponding plume spectra (usually subsequently to each scan). 265 The latter is important to keep potential additional stratospheric signals at a minimum (for details see Sect. 2.8). Each potential FRS was pre-evaluated regarding its SO 2 -content using a literature solar background spectrum as FRS (Chance and Kurucz, 2010) which was convolved with the instrumental line-spread-function (LSF). Only FRS candidates showing SO 2 -SCDs (S SO 2 ) smaller than S SO 2 ,FRS < 5 × 10 16 molecules cm −2 were used as FRS. 270 In the following, the implemented steps to retrieve the SCDs from the raw-spectra are described. Further details to individual topics regarding the data evaluation can be found in Appendix A.
Prior to the DOAS-evaluation, all FRS and plume spectra were corrected for electronic offset and dark current. Two Ring spectra (R, R4) were included into the fitting routine to 275 account for inelastic scattering effects (Raman scattering) in the atmosphere (see e.g. Vountas et al., 1998). The first Ring spectrum (R) was calculated in the usual way from the respective FRS using the function of the evaluation software DOASIS (Kraus, 2006). The second ring spectrum (R4) was determined following the suggestions from Wagner et al., 2009 (for details see Appendix A1). Literature cross sections of the individual absorbers 280 (σ i , see Table 2) were convolved with the LSF of the respective spectrograph. During the convolution, the σ i were corrected for the solar I 0 effect and for spectral saturation (Platt and Stutz, 2008) using the corresponding functions in DOASIS. The latter was performed assuming typical SCDs for the respective species (e.g. S SO 2 = 2 × 10 18 molecules cm −2 ). In order to correct for any misalignment of the spectrograph, a slight shift (±0.1 nm) and 285 squeeze (±5 %) was allowed for all fitted species (i.e. FRS, R, R4, σ i ). Shift and squeeze of all σ i were linked to the strongest absorber and the two Ring spectra were linked to the corresponding FRS, respectively in order to minimise the degrees of freedom during Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the fit-process. A 3rd order polynomial was included in the fitting routine to remove broad band extinction. An additional 0th-order polynomial residing in intensity space was included 290 (also referred to as offset polynomial) to account for intensity offsets in the spectra (e.g. due to stray light, for details see Kraus, 2006, Platt andStutz, 2008). The measurement uncertainty (δ meas ) was estimated conservatively by multiplying the retrieved fit errors (δ fit ) with a factor of U = 4 to account for potential abundances of fit residuals structures (Stutz and Platt, 1996, see e.g. Fig. A2). In case of good fit results (which were assessed by the 295 peak-to-peak values of the fit residuals ∆ res ) the correction factor was reduced down to U = 3 (i.e. for ∆ res ≤ 1.2 × 10 −3 , see e.g. Fig. 4). Details regarding the error treatment are discussed in Sect. A2. The detection limits of the SCDs were defined to be twice the measurement uncertainty (2 · δ meas ) thus, representing a detection certainty of 95 %. 300

Evaluation Routines
The focus of this study was on the data collected with the UV spectrograph (i.e. the evaluation of OClO, BrO, SO 2 , IO). In order to find the optimum evaluation range for each species, detailed sensitivity studies were performed including DOAS fit contour plots ("Re-305 trieval wavelength mapping", for details see Vogel et al., 2013). For the VIS data (i.e. the OBrO and OIO evaluation) these sensitivity studies were not performed since this data was of secondary interest. Therefore, we used a fixed correction factor of U = 5 for the estimation of the corresponding measurement uncertainties of OBrO and OIO (for details see Sect. A2). All evaluation routines used in this study are summarised in  Figure 4 including the two ring spectra (R, R4) and all additionally included absorbers as well as the corresponding fit residual. This example rather shows an unstructured residual with a 320 peak-to-peak value of ∆ res = 9.65 × 10 −4 (1.200 co-added scans per spectrum) thus, in this case a fit correction factor of U = 3 was used.
OClO was evaluated in a second "upper wavelength range" (∆λ OClO,uwr =363.6-391.3 nm) in order to verify the retrieved SCDs in the standard range with respect to 325 possible influences due to radiative transfer phenomena and / or cross correlations between different absorbers (for details see Appendix A3). However, this "upper" range was found to be influenced by larger fit uncertainties and was therefore not used for the discussion of our results. Nonetheless, both OClO retrievals show a good correlation (see Fig. A1). 330 Formaldehyde (CH 2 O) was included in the fitting routine in order to account for potential background abundances. In addition, CH 2 O could also be formed in the volcanic plume itself, for instance via CH 4 oxidation (in the plume most likely initiated by the reaction with Cl atoms) or in the presence of nitrogen oxides (Platt and Stutz, 2008).

335
SO 2 was evaluated in two different wavelength ranges. The "lower wavelength range" (lwr) between ∆λ SO 2 ,lwr =314.8-326.8 nm (e.g. Vogel, 2011) was used for SO 2 -SCDs below 3 × 10 18 cm −2 . In this wavelength region, especially below 320 nm, SO 2 shows strong absorption features. In order to avoid the well known evaluation problems related to strong SO 2 absorption in this "lower" wavelength range (see e.g. Kern et al., 2010;Bobrowski et al., 340 2010), SO 2 was evaluated in a second, "upper" wavelength range of ∆λ SO 2 ,uwr = 349.8-372.8 nm for SO 2 -SCDs exceeding 3 × 10 18 cm −2 (Hörmann et al., 2013). These problems 14 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | -originating in the non-linear nature of the Beer-Lambert law -lead to an underestimation of the SO 2 -SCDs in the "lower" wavelength range. This is clearly visible in the scatter plot of the SO 2 retrieval in both evaluation ranges shown in Figure A3 (i.e. flattening of the trend 345 at large SO 2 -SCDs). Furthermore, an exemplary fit result of the upper wavelength range is shown in Figure A2)

IO, OIO and OBrO evaluation
In addition to the evaluation of BrO, OClO and SO 2 , abundances of IO, OBrO, and OIO were investigated. The details of the corresponding evaluation routines for these species 350 can be found in table 1.

Estimation of OClO and BrO concentrations from plume cross section scans
The data from plume cross section scans was used to estimate mean concentrations (c i ) of BrO and OClO in the plume. This was done assuming a circular plume cross section and straight line absorption light paths through the plume. Any potential deviations due to 355 radiative transfer effects (RTE, e.g. multiple scattering, light dilution, for details see e.g. Kern et al., 2010, Mori et al., 2006 or deviations from the assumed circular shape were neglected in this estimation. The plume diameter (Ø pl ) was estimated from the angular extend of the SO 2 -SCD profile and the distance to the plume (see also e.g. Bobrowski et al., 2003;Lee et al., 2005). The corresponding SO 2 -SCDs were used as a proxy for the lengths 360 of the absorption light paths (l eff,i ) in the plume, whereas the largest SO 2 -SCD of the scan was assigned to Ø pl . Based on that, the l eff,i could be estimated for all scan spectra i: l eff,i =Ø pl /S SO 2 ,max × S SO 2 ,i . Using this, the mean concentrations (c j,i ) of the measured species j (e.g. OClO, BrO) were estimated:c ji = S j,i /l eff,i . The corresponding uncertainties were determined from the 365 DOAS-fit errors and the uncertainties in the estimation of the plume diameter using Gaussian error propagation.

15
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Determination of ClO concentrations
Following Kern et al. (2009), we estimated ClO concentrations from the retrieved BrO and OClO-SCDs assuming steady state between the formation of OClO (R7) and its photolytic 370 destruction (R8): Since BrO and OClO were evaluated in the same wavelength range, differences in the retrieved SCDs (S i ) due to differences in the radiative transfer can be neglected. We therefore assume, that the ratio of the OClO and BrO concentrations is approximately the same as the 375 ratio of the respective SCDs (Eq. 1). The OClO photolysis frequencies J OClO used for the calculation of the ClO concentrations were simulated for our dataset by E. Jäkel (Leipzig Institute for Meteorology). For the simulation, the 1-D radiative transfer model libRadtran (Mayer and Kylling, 2005) was used. The photolysis frequencies were determined for a set of chosen spectra from the field campaign and were between 5.1 × 10 −2 s −1 (SZA ≈ 62 • ) 380 and 7.1 × 10 −2 s −1 (SZA ≈ 34 • ), slightly slower than typical values in the stratosphere (e.g. J str,OClO 7.6 × 10 −2 s −1 , Birks et al. (1977)). Uncertainties in the determination of the ClO-concentrations were estimated using Gaussian error propagation.

Determination of the plume age (τ ) and meteorological data
The plume age (τ ) was estimated using meteorological information (i.e. wind speed and 385 direction) and the measurement geometry (i.e. geo-locations of instrument and craters, telescopes viewing direction). A typical measurement geometry at Mt. Etna is sketched in Figure 5. The intersection of the telescopes viewing direction with the plume determines the distance l. Based on that, the plume age was estimated as follows: Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | The azimuthal alignment of the instrument was performed using a compass. Due to possible disturbances of the planetary magnetic field by the volcano, we estimated the instruments azimuth-uncertainty to ±3 • (gray shaded area in Fig. 5). Wind directions were estimated using own observations/notes and -on clear days -satellite pictures from the MODIS network (Aqua, Terra satellites). Wind velocities were partly retrieved from simultaneously performed 395 SO 2 -camera measurements and from own observations. From the 16. September we additionally monitored meteorological data using a meteorological station, which was installed on the southern side of the craters. Uncertainties in the plume-estimation were determined using Gaussian error propagation, a detailed discussion of these, especially relative and absolute errors can be found in Appendix A5.

Correction for stratospheric BrO
Typical vertical column densities (VCDs) of stratospheric BrO are of the order of several 10 13 molecules cm −2 (e.g. Schofield et al., 2004;Sinnhuber et al., 2005). Therefore, MAX-DOAS measurements of volcanic BrO (using scattered sunlight) can be significantly disturbed by stratospheric BrO signals under certain conditions. Based on the geometrical 405 air mass factor (AMF: X = 1/ cos(Θ)) and by assuming a constant stratospheric BrO-VCD of V str,BrO = 4.0 × 10 13 molecules cm −2 (Sinnhuber et al. (2005), Schofield et al. (2004)) a correction was implemented to account for additional stratospheric BrO signals in our retrieved SCDs. A detailed discussion including simplifications and sensitivity studies can be found in Appendix A6. 410 For our dataset we found, that deviations due to the superimposed stratospheric BrO signals are smaller than 5 % in 85 % of the analysed spectra. Only 8 % of the retrieved BrO-SCDs showed deviations exceeding the corresponding fit uncertainty. All of these spectra were either recorded before 08:15 LT or after 16:45 LT (64.6 • < SZA < 83.2 • ), which shows 415 the importance of this correction, especially for measurements performed in the early morning and late afternoon. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2.9 SO 2 as volcanic plume proxy -Analysis of X m O n /SO 2 ratios In order to study spatial (and temporal) variations of the retrieved halogen species X m O n , molar SO 2 ratios of these species were analysed (i.e. X m O n /SO 2 -ratios) whereas SO 2 was 420 treated as volcanic plume proxy due to its comparatively long tropospheric lifetime (e.g. McGonigle et al., 2004;Lee et al., 2011;Beirle et al., 2013). This is a common method to avoid signal variations due to atmospheric dilution effects (e.g. Bobrowski et al., 2003;Kern et al., 2009). Furthermore, compared to the individual SCDs, the X m O n /SO 2 -ratios are much less affected by radiative transfer 425 effects (RTE) such as light dilution or multiple scattering (Lübcke et al., 2014) and we therefore neglect any potential influences of these effects on our retrieved trace gas ratios. The discussion of our results mostly relates to the measurements performed at the Etna observatory. Especially for these data, potential influences due to RTE on the retrieved ratios should be negligible because of the proximity to the plume, the relatively high altitude (i.e. 430 low plume dilution, see e.g. Mori et al., 2006 and the fairly good visibility during most of the measurements (i.e. low aerosol scattering, for details see Sect. 2.2).
The errors and detection limits of the X m O n /SO 2 -ratios were calculated from the SCD errors using Gaussian error propagation. 435 Most of the significant OClO detections (i.e. 99.2%) are related to the measurements performed at the Etna observatory (11-13 September 2012), where the largest SCDs can be found due to the proximity to the craters (little plume dispersion). Out of 677 significant SO 2 -detections during the whole campaign, OClO could be detected in 119 spectra up to plume ages of six minutes. BrO was detected in 452 spectra (269 of those recorded at 440 the observatory) at plume ages up to 23 min. As expected, OClO was less abundant compared to BrO with a mean OClO / BrO-ratio of 0.16. The retrieved SCDs of OClO, BrO and SO 2 ranged between 0.4-2.0 × 10 14 molecules cm −2 , 0.3-11.4 × 10 14 molecules cm −2 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | and 0.03-8.77 × 10 18 molecules cm −2 respectively. Furthermore, the DOAS evaluation of IO, OIO and OBrO was investigated but none of these species could be detected significantly. 445 Upper limits of IO, OIO and OBrO were determined and are presented in Sect. 3.2.

OClO and BrO results
In Figure 6 we plotted all retrieved OClO and BrO-SCDs as a function of the corresponding SO 2 -SCDs (A, B) and furthermore OClO vs. BrO (C). Both, BrO and OClO showed a good correlation to SO 2 (6A, B), indicating, that these species could only be detected in 450 volcanic plume spectra. Average ratios of 1.65 × 10 −4 for BrO/SO 2 (6A) and 3.17 × 10 −5 for OClO/SO 2 (6B) were found (linear regression). These values are in good agreement with previous findings (e.g. Bobrowski and Giuffrida, 2012;General et al., 2014). For the linear regression, only significant detections at plume ages exceeding three minutes were considered (blue dots in Fig. 6). Measurements at plume ages smaller 455 than three minutes (green dots) were excluded because in this plume age range, the formation of BrO and OClO is yet not fully developed and therefore the X m O n /SO 2 -ratios are smaller (for details see Sect. 3.1.3).
The corresponding average OClO/BrO-ratio (at plume ages exceeding three minutes) is 0.16 ± 0.08 and indicates a very good correlation between both species in this plume age 460 range (R 2 = 0.9447). Young plume measurements (green dots, τ < 3 min), however, rather indicate a stronger fluctuation of the OClO/BrO-ratio (R 2 = 0.4717).

Results from individual scans
In order to study the chemical evolution of BrO and OClO, we analysed the corresponding ratios with SO 2 (BrO/SO 2 , OClO/SO 2 -ratio). In Fig. 7 examples of plume evolution scans 465 of both ratios are plotted for different plume age ranges (i.e. Fig. 7a-e). Furthermore, an exemplary plume cross section scan is shown in Fig. 7f. As discussed in Sect. 2.7 and Appendix A5, plume age errors were separated into a geometrical contribution (x-error bars) and a percentage contribution due to uncertainties in the wind velocity (∆τ , plot header).

19
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | We explicitly point out again, that these are not to be treated as independent random plume 470 age errors between each of the scan spectra but are strongly correlated and thus, rather indicate a stretch/squeeze of the whole dataset towards smaller or larger plume ages.
The plume evolution scans of the BrO/SO 2 -ratio ( Fig. 7a-c) show a strong increase in the young plume (Fig. 7a) stabilising after approximately 150 s downwind. This trend could be observed in six individual measurements performed in the young plume (i.e. τ < 5 min), 475 details are discussed in section 3.1.3. At larger plume ages, the BrO/SO 2 -ratios show a rather constant behaviour with only slight variations.
The discussion of potential trends of the corresponding OClO/SO 2 -ratio in the young plume (at τ 2 min) is more difficult for the data retrieved from the individual scans (see e.g. Fig. 7d-e), since the retrieved values are often below our (conservative choice) of the detection 480 limit. Therefore, we refer to section 3.1.3 were we statistically analyse these apparent trends of both ratios in the young plume. For plume ages exceeding two minutes, we found rather low variations in the retrieved OClO/SO 2 -ratios with increasing plume age (similar to our BrO observations for this plume age range, i.e. compare Fig. 7e,b). 485 Due to the higher S/N -ratio, BrO could also be analysed at larger plume ages (i.e. τ > 5 min). An exemplary BrO scan in the aged plume is shown in Fig. 7c. It covers a plume age range between 8-22 min and shows rather stable BrO/SO 2 -ratios around 1.7 × 10 −4 .
A slight but not significant decrease of approximately 17 % might be observable between eight and ten minutes downwind. 490 The retrieved BrO/SO 2 -ratios in Fig. 7a-c) range from 8 × 10 −5 to 1.8 × 10 −4 in the τ > 3 min-regime (i.e. after reaching steady state). These variations could -for examplebe caused by superimposed diurnal profiles (note: the scans were performed on different days and at different times) or varying volcanic activity.

495
Previous studies showed increased BrO/SO 2 -ratios at the edges of the plume (e.g. Louban et al., 2009;General et al., 2014). These are likely due to a 20 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | limited transport of tropospheric O 3 and HO 2 radicals towards the plume centre (see also Sect. 1.1). In order to elaborate this issue, plume cross section scans perpendicular to the plume propagation axis (see Fig. 3b) were performed. Unfortunately, we were not able to 500 confirm these observations significantly due to enhanced measurement uncertainties at the edges of the plume (i.e. at low SO 2 -SCDs). However, analysing the retrieved ratios in dependency of the corresponding SO 2 -SCDs (which indicates, whether the spectrum was recorded in the centre/edge of the plume) we found indications of enhanced ratios at the edges of the plume. An exemplary plume cross-section scan of the BrO/SO 2 -ratio is shown 505 in Fig. 7f. This example visualises the problems related to the plume-edge spectra: the BrO/SO 2 -ratios show increased values at low SO 2 -SCDs but considering the larger errors (due to low BrO and SO 2 -SCDs) it is not possible to draw certain conclusions. However, in 76 % of all 25 suited cross section scans we observed this trend (enhanced BrO/SO 2 -ratios at low SO 2 -SCDs). In case of OClO/SO 2 it was even more difficult to draw confident con-510 clusions due to the weaker OClO signal. Nonetheless, in five of -in total -nine suited cross section scans indications of enhanced OClO/SO 2 -ratios at low SO 2 -SCDs could be found.

Statistical analysis of the young plume evolution
All plume evolution scans performed in the young plume clearly showed increasing 515 BrO/SO 2 -ratios at plume ages (τ ) smaller than 150 s (see e.g. Fig. 7a). In case of OClO, we found strong indications of a similar trend in the young plume. However, from our individual scans this could not be validated with certainty due to comparatively large measurement uncertainties in the OClO retrieval (i.e. in most cases the OClO/SO 2 -ratios appeared to be below the detection limit in this plume age range, see e.g. Fig. 7d). 520 Therefore, in order to further elaborate this issue and especially the young plume evolution of OClO, we performed a statistical analysis of the retrieved X m O n /SO 2 -ratios as a function of the plume age. The plume was subdivided into six plume age intervals between zero and 250 s downwind (i.e. ∼ 42 s interval −1 ) and the retrieved X m O n /SO 2 -ratios were 21 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | assigned to the corresponding plume age interval accordingly. Only spectra related to the 525 plume-centre were considered by including only measurements showing SO 2 -SCDs larger than 1.5 × 10 18 molecules cm −2 . This was done to avoid possible falsifications due to potentially enlarged ratios at the edges of the plume (for details see Sect. 3.1.2). Further, we did not distinguish between measurements above or below the respective detection limits of BrO and OClO. Based on this selection, we determined the mean-value of the retrieved 530 X m O n /SO 2 -ratios for each τ -interval. The corresponding uncertainties (∆, i.e. y axis errors) for the averaged ratios were determined from the mean of the individual errors (σ i ) divided by the inverse square-root of the number N of averaged spectra in each interval: The results of this statistical approach are plotted in Fig. 8  The similarities in the trends of OClO and BrO in the young plume (i.e. increase in the young plume and steady state after about 2-3 min) strongly support the assumption that OClO is mainly formed via the "BrO+ClO"-reaction (R7).

Photochemical formation of BrO and OClO at low solar radiances
The "bromine explosion" includes the photolysis of the  Fig. 9 and clearly shows an increase of the BrO/SO 2 -ratio with time between 05:20 and 05:32 UTC (see Fig. 9, top) and a constant ratio afterwards. A similar trend can be observed for the corresponding OClO/SO 2 -ratios (see Fig. 9, bottom) which were averaged (nearest neighbour averaging) due to larger 570 measurement uncertainties. Compared to BrO/SO 2 , the increase of OClO/SO 2 appears to be delayed by approximately 30-40 min reaching a plateau around 06:15 UTC. One possible explanation for such a delayed increase could be that the availability of Cl-atoms is delayed with respect to Br during this time of the day. BrO/SO 2 -ratio of (1.15 ± 0.2) × 10 −4 (20 spectra) which is in good agreement with the values shown in Fig. 9 after reaching the steady state. OClO could not be detected in this scan.

BrO, OClO and ClO mixing ratios
As described in Sect. 2.5, average BrO and OClO concentrations (volume number densi-600 ties) were estimated from plume cross section scans assuming a circular plume shape. The plume diameter could be estimated in 61 (from in total 90) cross section scans performed during the campaign. Furthermore, ClO concentrations were calculated as described in Sect. 2.6. The corresponding number densities were converted into mixing ratios and the results are plotted in Fig. 10 as a function of the plume age τ . Only BrO, ClO and OClO 605 concentrations above the detection limit were considered. Furthermore, only measurements during clear meteorological conditions were included to avoid potential impacts on the radiation light path, for instance caused by clouds or high background aerosol concentrations.

615
The comparatively large errors of the derived mixing ratios (see Fig. 10) are due to our conservative estimation of the SCD errors and the uncertainties in the plume diameter estimation. More detailed radiative transfer effects (RTE, e.g. multiple scattering, light dilution, for details see Kern et al., 2010;Kern et al., 2012) were neglected in the determination of the mixing ratios as well as potential deviations from the circular plume cross section. 620 Hence, the reported numbers are rather an estimate of the order of magnitude of the average abundances of these species in the plume. Nonetheless, for the observatory data (11-13 September 2012, i.e. τ < 4 min range in Fig. 10) potential deviations due to RTE are most likely smaller than a factor of two, relying on the findings of Kern et al., 2012 andMori et al., 2006. This is due to the fairly good measurement conditions (i.e. low plume con-625 densation, see e.g. Fig. 1) and furthermore, because of the proximity to the plume (mean distance to plume:d = 2.03 km, d max < 3.2 km) and the high altitude at the measurement location (i.e. 2.800 m a.s.l., i.e. lower scattering probability). Moreover, a strong eccentricity of the plume (i.e. pronounced elliptical plume shape) is unlikely in this plume age regime (see e.g. Turner, 1970). 630 The measurements performed in the aged plume (i.e. data points at τ > 4 min in Fig. 10) are most likely stronger influenced by deviations from the assumed circular cross section and by the light dilution effect since they were partly performed at sea level and at greater plume distances (up to 17 km). Thus, the observed decrease of the BrO mixing ratios in the ageing plume (see Fig. 10) is most likely not solely due to the decreasing concentra-635 tions (plume dilution) but to a certain degree also influenced by light dilution and elliptical plume shapes. However, in most cases the latter two effects should counteract each other 25 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | to a certain degree since light dilution induces a decrease in the "true" signal whereas the lengths of the effective light paths through the plume are most certainly overestimated. The latter is due to the pronounced elliptical shape of the plume (i.e. σ hor >σ vert , for typical con-640 ditions) and large plume viewing angles (i.e. close to 90 • ). Hence, rather σ hor is estimated from the scan and as a result, the individual lengths of the absorption light paths (l eff,i ) are overestimated. A rough estimation assuming slightly stable conditions (i.e. Pasquill stability class E, see e.g. Turner, 1970) would approximately yield a factor of σ vert / σ hor ≈ 0.16 at 20 km distance from the source, yet underestimated concentrations up to a factor of six.

Results for IO, OIO and OBrO
We investigated the presence of IO (in the UV spectral range), OIO and OBrO (in the VIS spectral range) but did not detect any of these species. Here, we give both the detection limits for plume ages smaller and larger than three minutes (table 3) since it appears reasonable to assume, that these species -if abundant in the plume -show a similar plume 650 evolution as was observed in case of BrO and OClO (for details see 3.1.3). For plume ages larger than three minutes upper limits of 5.2 × 10 −6 (IO/SO 2 ), 2.8 × 10 −5 (OIO/SO 2 ) and 1.8 × 10 −5 (OBrO/SO 2 ) were found. Note that the UV spectrograph showed a better performance (S/N -ratio) than the VIS spectrograph, which is indicated by the lower detection limits for IO compared to OIO and OBrO. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Furthermore, mean concentrations of the order of several hundred ppt (ClO x ) up to several ppb (BrO) could be estimated. Having in mind, that most of the ClO x and BrO x originates 665 from the initially emitted HCl and HBr, our findings strongly suggest, that the oxidation of chloride is much weaker compared to the bromide oxidation. This is mainly due to the fact, that any potential Cl-release mechanisms are likely less efficient compared to bromine. Moreover, once formed, the Cl-radicals in the plume will rapidly react with CH 4 which may even cause a significant depletion of CH 4 in the plume. This very important question is 670 addressed in the following, where we use our results (i.e. the formation times of ClO x and the estimation of mean concentrations) to derive an estimate of Cl-atom concentrations and from that, the potential of a chlorine-induced depletion of CH 4 .

Cl-atom concentrations and the depletion of atmospheric CH 4 in the plume
Once Cl atoms are produced in a volcanic plume, they will rapidly react either with CH 4 or 675 with O 3 : The corresponding reaction rate coefficients are k 10 = 1.0 × 10 −13 cm 3 s −1 (R9, at 298 K) and k 11 = 1.2 × 10 −11 cm 3 s −1 (R10, 298 K, Sander et al., 2006). Note, that R9 is 16 times 680 faster than the OH + CH 4 reaction (at 298 K) and has a strong positive temperature dependence (Sander et al., 2006). All other Cl sink reactions are much slower and are therefore neglected here.
(3) 690 Actually the true rate of Cl atom production d/dt [Cl] is larger since a fraction of the Cl atoms reacts with CH 4 (R9) and never shows up as ClO y (possible reaction of Cl with other hydrocarbons is likely to be unimportant and therefore neglected here): The corresponding Cl-atom concentration is then given by: whereas the lifetime of Cl (τ Cl ) was estimated 700 Introducing the expression for τ Cl in Eq. (6) and using Eq. (3), an estimate of the Cl atom concentration can be obtained: Based on this the CH 4 -lifetime in the plume (due to reaction with Cl R9) can be derived: corresponds to the methane lifetime in the plume.
For the estimation of ClO y , we determined mean ClO and OClO concentrations from our retrieval considering only plume ages between 120 and 240 s (see also Sect. 3.1.5). We retrieved values of ClO = 2.0 × 10 9 cm −3 and OClO = 3.7 × 10 9 cm −3 , respectively and hence 710 ClO y = 5.7 × 10 9 cm −3 . Based on our findings discussed in Sect. 3.1.3 we estimated the ClO y formation duration to τ 0 = 142 s. The OClO concentrations (used to estimate ClO y ) might carry potential uncertainties due to RTE or non-circular plumes. However, as discussed in Sect. 3.1.5, these deviations should be small (i.e. 2) for the majority of data points recorded in the young plume and should not significantly influence the outcome of 715 this analysis.
The typical tropospheric O 3 -background is 60-80 ppb for this (relatively polluted) region and altitude (Kalabokas et al., 2013). The expected CH 4 -lifetime in the plume is directly proportional to the prevailing O 3 -concentration (Eq. 9). Since O 3 is most likely depleted in the plume (von Glasow, 2010, Kelly et al., 2013 we determined [Cl] and τ CH 4 as a func-720 tion of the O 3 -concentration (assuming O 3 mixing ratios between 1-80 ppb). Our results are summarised in table 4 showing relatively small Cl-atom concentrations (i.e. 10 6 -10 8 cm −3 ) and CH 4 -lifetimes between 14h up to 47 days. These lifetimes are more than two orders of magnitude shorter than the average atmospheric lifetime of CH 4 . However, if O 3 is not strongly depleted, CH 4 destruction by Cl-atoms will probably not lead to a detectable loss 725 of CH 4 in the plume since the Cl-levels derived from our measurements (at plume age τ = 142 s) should decrease rapidly as the plume disperses. Even if these Cl-levels would prevail for a few hours downwind, only a small fraction (less than 1 %) of the CH 4 would be destroyed (assuming that the mean O 3 -levels in the plume exceed 10 ppb). However, in regions of very low O 3 concentrations (i.e. possibly in the plume centre, von Glasow, 2010) 730 a significant loss of CH 4 could be present but the atmospheric impact would probably still be negligible since the effective volume of this potential methane depleting environment would be very small. Nevertheless, we want to remark that our calculations are based on the volcanic conditions at Mt. Etna in September 2012 and we therefore want to stress, 29 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | that it is absolutely possible, that CH 4 depletion may become detectable in plumes of other 735 volcanoes or at different conditions (e.g. due to varying volcanic activity, stronger chlorine emissions, larger Cl − /Br − -ratios, low NMHC (nonmethane hydrocarbons) concentrations or the presence of volcanic particles favouring the chloride oxidation).
Appendix A: Details regarding the data evaluation A1 Determination of the R4 spectrum 740 In addition to the standard Ring spectrum (calculated using the software DOASIS, Kraus, 2006), a second ring spectrum (R4) was determined as follows (for details see Wagner et al., 2009, Appendix B therein): Here, j denotes the pixel on the detector, λ(j) the appropriate wavelength and λ 0 the 745 central wavelength of the evaluation range. R(j) corresponds to the intensity of the Ring spectrum at position j. The R4-spectrum is orthonormalised with respect to λ 0 . It accounts for influences due to multiple scattering and / or scattering on aerosols and cloud particles which are not considered in the determination of R.

750
According to Stutz and Platt, 1996 the error of atmospheric trace gas measurements with the DOAS technique does not purely arise from pure photon (shot) noise and is thus not entirely statistical (Poisson statistics). The fit residuals often show distinct structures which are mainly a result of the limited optical resolution of the instrument or due to uncertainties in the absorption spectra of the fitted species. In this case, the DOAS fit yields underestimated 755 measures of the true fit uncertainty since it is based on the assumption of individual radiance measurements in each pixel on the detector. Thus, in case of structured fit residuals, 30 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | this underestimation has to be accounted for. This can be done by multiplying the retrieved DOAS fit errors with a certain factor (here denoted with U ) which can reach values of up to six according to Stutz and Platt, 1996. The choice of U for a given measurement is mainly 760 dependent on the spectral width of the fitted absorption lines and on the width of potentially abundant residual structures (in channels on the detector, see Fig. 10 in Stutz and Platt, 1996). Since we could observe such residual structures in some of our measurements (see e.g. Fig. A2), we followed Stutz and Platt, 1996 and corrected our retrieved DOAS fit errors with 765 a factor of U = 4. We remark, that this constitutes a conservative estimation of the measurement uncertainty. Therefore, in case of good fit results (i.e. low peak-to-peak value of the fit residual, ∆ res ), this factor was reduced as follows: Fig. A2) 1.2 × 10 −3 ≥ ∆ res ⇒ U = 3 (see e.g. Fig. 4) We found a good correlation between ∆ res and the width of structures in the residual.
However, we want to point out, that this approach constitutes only a rough -but easy 775 to implement and still conservative -implementation of the interpretation of the retrieved DOAS fit errors (e.g. Donovan et al. (2014) used a fixed correction factor of U = 3 for their OClO and BrO evaluation).
In Fig. A2, such an example of a structured residual is shown (one residual structure is marked and has a width of W ≈ 20-30 channels on the detector). Typical absorption lines 780 of the fitted species cover between 15-30 channels on the detector of our spectrograph. Considering these typical widths, we decided to use fit correction factors between three and four, based on the findings of Stutz and Platt, 1996 (see especially Fig. 10 therein).

31
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

A3 Alternative OClO evaluation routine (OClO uwr )
OClO was additionally evaluated in a second range between 363.6 and 391.3 nm cover-785 ing three OClO absorption bands. Besides the two Ring spectra and the FRS, reference spectra of SO 2 , O 3 , O 4 , NO 2 were additionally included. In principle, an advantage of this "upper" wavelength range should be, that it is less influenced by potential cross correlations with BrO, O 3 or CH 2 O in the DOAS fit. However, it was found that the fits often showed distinct residual structures in this wavelength range, resulting in relatively large fit uncertainties 790 compared to the standard evaluation range for OClO (section 2.4.1). These structures are most likely caused by distinct solar Fraunhofer lines in this wavelength region which can cause a strong Ring effect. Nevertheless, the retrieved OClO-SCDs showed a good correlation with slightly larger SCDs (≈ 8 %) in the upper wavelength range (see Fig. A1).

A4 Details regarding the SO 2 evaluation
795 SO 2 was evaluated in two wavelength ranges in order to account for radiative transfer effects due to strong absorption at large SO 2 column amounts in the wavelength regime below 320 nm. An exemplary fit result of the evaluation in the alternative "upper" SO 2range (SO 2,uwr ) is shown in Fig. A2 showing a comparatively large SO 2 -SCD of S SO 2 ,uwr = 5.57 × 10 18 molecules cm −2 . The corresponding SO 2 -SCD in the lower evaluation range was 800 found to be smaller (as expected) and amounts to S SO 2 ,lwr = 4.89 × 10 18 molecules cm −2 .
The corresponding plume spectrum was recorded during the early morning point measurement discussed in Sect. 3.1.4 at 06:20 UTC on 13 September 2012. The FRS was recorded subsequently at 06:25 UTC, explaining the low O 3 -SCD. This example clearly shows the significance of the λ −4 dependency of the Ring effect, accounted for by fitting 805 the R4-spectrum as discussed previously (Sect. A1). It furthermore clearly shows the necessity of our conservative approach for the fit error estimation: In contrast to the BrO and OClO fit example (given in Sect. 2.4, Fig. 4) this fit example shows a rather structured fit residual (most likely due to the strong Ring effect in this spectrum). This even causes a "false" detection of CH 2 O in this spectrum, showing a negative (but significant) SCD of 810 32 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | S CH 2 O = −1.89 ± 0.67 × 10 16 molecules cm −2 (i.e. CH 2 O could not be detected in the standard evaluation range in this spectrum). Hence, using only the uncorrected DOAS fit error would yield a significant detection of CH 2 O in this case (even within 3σ confidence). This demonstrates the tremendous importance of applying appropriate fit correction factors to account for these effects (as discussed in Sect. A2).
815 Figure A3 shows a scatter-plot of the retrieved SO 2 -SCDs in both evaluation ranges.
The retrieved values show good coincidence within their errors up to SO 2 -SCDs around 4 × 10 18 molecules cm −2 . For larger SCDs the trend starts to flatten due to the underestimated SCDs in the lower evaluation range.

820
The main uncertainties related to the plume age determination using Eq.
(2) are due to uncertainties in the wind velocity and the determination of l (i.e. mainly due to uncertainties in δ and α, see Fig. 5). We thus subdivided our error-representation of the plume age τ into two contributions: 1. The first (in the following denoted as ∆τ l (α, δ)) is determined from the uncertainties 825 in α and δ (mainly geometrical uncertainties). ∆τ l (α, δ) can vary strongly between different spectra from plume evolution scans due to the nature of the trigonometric functions involved in the calculation of l. ∆τ l (α, δ) is therefore plotted for each spectrum separately in form of x axis error bars (see e.g. Fig. 7).
2. The second contribution (in the following denoted as ∆τ ) to the plume age error is 830 caused by uncertainties in the wind velocity ∆v wind which have a linear effect on the plume age uncertainty (∂τ /∂v wind ∝ ∆v wind ). Since ∆τ is independent of the measurement and plume angles, its relative impact on each spectrum (∆τ /τ ) is constant.
The corresponding contribution is therefore given as a percentage value in the plot header (see e.g. Fig. 7). 835 For simplicity, we reduced the determination of τ to a 2-D problem in the horizontal plane, because for typical scanning geometries (and a wind driven, horizontal plume propagation), 33 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | differences in the altitude between plume and DOAS instrument have only a small influence on the determination of τ .
Furthermore, by reducing the volcanic plume and the telescopes viewing direction to 840 a line (dotted lines in Fig. 5) we did not consider any effects caused by plume dispersion or the telescopes field of view for our estimation of τ . These effects are usually negligible for typical scan geometries (ϕ ≈ 90 • , see Fig. 5) and may be considered when the measurements are performed at small angles ϕ. In this case the analysed light has penetrated a multiple of different plume ages which essentially causes a smoothing of the signal with 845 respect to τ . The corresponding impact with regard to the data interpretation depends on the desired temporal resolution of the respective scan (i.e. ∆τ between individual scan spectra) and on the chemical variability of the analysed species in the analysed plume age range.
A further simplification in our algorithm is the reduction of the four main craters (BN, VOR, 850 NE and SE) to a single emission source point P 0 (i.e. τ = 0 point) located at 37 • 45 6.7 N, 14 • 59 49.6 E between the central craters (BN, VOR) and NE.
Furthermore, chemical processes which may have taken place already within the craters are also not considered in our routine. The latter effect can cause a plume age offset with respect to P 0 . Both effects are strongest for viewing directions close to the vent were the 855 plumes are still separated. However, in most cases the corresponding error was assessed to be relatively small considering the uncertainties in the meteorological data.

A6 Correction algorithm for stratospheric BrO
The BrO-SCDs (S meas ) derived from the DOAS evaluation (see Sect. 2.4) are composed of a volcanic (S plume ) and a stratospheric contribution (dS str ). The latter is due to changes in 860 the zenith angle (Θ) between plume spectrum and FRS.
In order to examine dS str , we used a simple geometrical approach assuming, that the stratospheric air-mass-factor (AMF) X is given by X = cos(Θ) −1 (see also Hönninger et al.,34 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2004). For our purposes, this assumption was assessed to be sufficient, since plume spec-865 tra and FRS were recorded close in time and the SZAs were in most cases (99.1 %) smaller than 80 • . For a more accurate estimation of the AMF, radiative transfer calculations are necessary.
The AMF relates the stratospheric slant column S str,i of a given spectrum i to the corresponding vertical column V str : S str,i = V str / cos(Θ i ). Based on this, the corresponding strato-870 spheric contribution dS str,ij between two spectra i, j can be determined from the vertical column and the difference in the AMF: Of course, the relative impact of the stratospheric contribution increases for smaller (measured) BrO-SCDs (compare e.g. dark blue with green colours in Fig. A4). For BrO-SCDs of the order of 6 × 10 14 molecules cm −2 (green colours) we found that the impact of dS str on 885 the measured signal amounts to a = 6.6 %/γ ij (linear regression in Fig. A4).
In order to estimate the influence of potential variations in the total stratospheric BrO load (V str,BrO ), we additionally determined this slope for two different stratospheric BrO-VCDs of V str,BrO = 2 × 10 13 and V str,BrO = 7 × 10 13 molecules cm −2 .
The corresponding impacts were a = 3.4 %/γ ij and a = 11.8 %/γ ij respectively. This 890 shows, that influences due to stratospheric BrO are in most cases relatively small (i.e. for small γ ij -values), even for considerably large stratospheric VCDs. Nonetheless, one has to keep in mind, that these slopes were determined at considerably large BrO-levels of the order of S meas = 6 × 10 14 molecules cm −2 . For lower measured BrO-SCDs (blue colours in Fig. A4) the impact of stratospheric BrO increases and can significantly influence the vol-895 canic signal, especially at large differences in the SZA between plume spectrum and FRS (i.e. at large γ ij -values).  (1) and (2), see also          Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Figure 6. Retrieved slant column densities (SCDs) of BrO (A) and OClO (B) as a function of the retrieved SO 2 -SCDs and (C), OClO vs. BrO-SCDs. The measurements were subdivided by their plume age τ (i.e. τ < 3 min: green stars, τ > 3 min: blue dots) due to smaller BrO/SO 2 and OClO/SO 2 -ratios in the young plume (see also Sect. 3.1.3). Measurements below the detection limit of BrO (A) and OClO (B) are indicated by red dots. We determined mean ratios in the τ > 3 min range (blue dots) by applying a linear fit and found values of BrO/SO 2 = 1.65 × 10 −4 and OClO/SO 2 = 3.17 × 10 −5 respectively. The OClO/BrO-ratio (C) was found to be 0.16 for τ > 3 min and approximately 0.22 in the young plume. 55 Cross section scan BrO Figure 8. Plume evolution scans (see fig. 2) of OClO/SO2 and BrO/SO2 (a-e) and a sample cross section scan of BrO/SO2 (f). The BrO/SO2 ratio (green circles) and the OClO/SO2-ratio (green triangles) are plotted with their corresponding detection limits (green dotted line). Red error bars indicate measurements below the respective detection limit. The SO2-SCDs are plotted as grey shaded areas (right axis). We observed increasing BrO/SO2 and OClO/SO2-ratios in the young plume (τ 3 min, a, d) levelling off at larger plume ages (b, e). Note that in d the OClO/SO2-ratios are technically below the detection limit, however relative changes in OClO can still be detected from the data (for details see sect. 3.1.1). For plume ages between 8 and 22 minutes (c) we found a rather stable BrO/SO2 ratio with indications of a slight decreasing trend between eight and ten minutes downwind, which is probably due to a superimposed vertical profile (for details see text). In f, a cross section scan of BrO is plotted which was performed right before c and at a plume age of τ = 6 min. The BrO/SO2-ratio increases at the edge of the plume by 30% from 2.0 · 10 −4 (plume centre) to 2.6 · 10 −4 (see also sect. 3.1.2).
the scan shown in figure 8 c (high BrO/SO 2 -ratios) may have been recorded off the plume centre. This is further supported by the plume-cross-section scan shown in 8 f which was performed directly before 8 c at τ = 6 min. The

1025
corresponding SO 2 -profile indicates plume-centre-SCDs of the order of several 10 18 molecules/cm 2 for this plume age regime (6-8 min). The retrieved BrO/SO 2 -ratios in the τ > 3 min-regime (i.e. after reaching steady state, see fig. 8 a, Figure 7. Plume evolution scans of the BrO/SO 2 -ratio (a-c) and the OClO/SO 2 -ratio (d-e). A sample plume cross section scan of BrO/SO 2 is shown in (f). The BrO/SO 2 ratio (green circles) and the OClO/SO 2 -ratio (green triangles) are plotted with their corresponding detection limits (green dotted line). Red error bars indicate measurements below the detection limit. The SO 2 -SCDs are plotted as grey shaded areas (right axis). The BrO/SO 2 increases in the young plume and a levelling off at larger plume ages (τ 3 min, a,b) Figure 9. The young plume evolution of the BrO/SO2-ratio (top) and the OClO/SO2-ratio (bottom): The colour code indicates the number of averaged individual measurements. The errors of the ratios were determined from the errors of the individual measurements using gaussian error propagation (for details see text). The horizontal errors denote the respective plume age interval, which was used for averaging. The position of the averaged ratios for each plume age interval represents the mean plume age of the individual spectra included in this range. For both species, we observed an increase in the young plume levelling off at τ = 142 s. For larger plume ages, the BrO/SO2-ratio stays rather constant (at~1.3·10 −4 ) whereas the OClO/SO2-ratio slightly decreases.
The results of this statistical approach clearly show the increase of BrO/SO 2 and OClO/SO 2 in the young plume, confirming our observations from the individual plumeevolution scans (see previous section). Both BrO/SO 2 and 1065 OClO/SO 2 level off after approximately 142 s. However, ter plots of the whole dataset (i.e. BrO/SO 2 1075 OClO/SO 2 = 3.17·10 −5 , see fig. 6). One expla deviation could be that spectra related to the (which showed elevated BrO/SO 2 and OC see sect. 3.1.2) were excluded in the statist Further possible reasons could be that the lo 1080 of both species (i.e. τ > 250 s) is still incr a superimposed diurnal signal might have statistics. Moreover, both long and short-term the volcanic activity might have affected the re Nonetheless, the main objective of this stud 1085 young plume increase of both species could In order to further elaborate the long term tre OClO (and/or diurnal profiles, variations du activity) more measurements are necessary, plume ages larger than τ < 250 s.
1090 Table 3 gives an overview of the retrieved fo (τ 0 ) of BrO and OClO, meaning the plume BrO/SO 2 and OClO/SO 2 start levelling off. T (∆) indicates the difference in the formatio 1095 OClO and BrO (i.e. ∆ = τ 0,OClO − τ 0,BrO individual scans (i.e. rows 1-6) it appears formation reaches its steady state slightly fa to OClO. While τ 0,BrO ranged from 24 − 15 OClO formation lasted about 70% longer w 1100 τ 0,OClO : 85 − 226 seconds (rows 2-6). Only scan showed similar formation times for both 1, τ 0 ≈ 123 s) as did our statistical analysis (ro s, see also fig. 9). Since BrO and ClO are m chemical precursors of OClO, it is in principl 1105 the formation duration of OClO is longer com The variations in the formation times itself co by uncertainties in the plume age determinati see sect. 2.7) but could also indicate a var effectiveness of the reaction mechanisms invo  the colour code indicates the number of averaged individual measurements. The errors of the ratios were determined from the uncertainties of the individual measurements using Gaussian error propagation (for details see text). The horizontal errors denote the respective plume age interval, which was used for averaging. The position of the averaged ratios for each plume age interval represents the mean plume age of the individual spectra included in this range. For both species, we observed an increase in the young plume levelling off at τ = 142 s. For larger plume ages, the BrO/SO 2 -ratio stays rather constant (at ∼ 1.3 × 10 −4 ) whereas the OClO/SO 2 -ratio slightly decreases, which might be due to plume dilution. Figure 9. Early morning point measurement of BrO and OClO: the BrO/SO 2 -ratio increases between 05:17 and 05:32 UTC (top) reaching a plateau afterwards. The corresponding OClO/SO 2ratios also shows an increase which is delayed by approximately 30-40 min with respect to BrO (note that the OClO/SO 2 -ratios of this scan were averaged to increase the detection sensitivity, see also  Figure 10. Mixing ratios of ClO, BrO and OClO as a function of the plume age τ . BrO and OClO mixing ratios were determined directly from the retrieved SCDs assuming a circular plume cross section. ClO mixing ratios were determined from the BrO and OClO-SCDs assuming chemical equilibrium between the formation and destruction of OClO. Y-axis error-bars were derived using gaussian error propagation of the DOAS fit-errors and the uncertainties in the estimation of the plume diameter (BrO and OClO). Values between 70-235 (±44-121) ppt (ClO), 11-2700 (±7-1200) ppt (BrO) and 37-597 (±24-440) ppt (OClO) were found covering plume ages between zero and 17 min. Mean abundances in the young plume (i.e. τ < 4 min) were ClO = 139 ppt, BrO = 1.35 ppb and OClO = 300 ppt. Due to plume dispersion, the concentrations decrease with increasing plume age. In addition, the fit results of the two ring spectra (R, R4) and the additionally included absorbers (SO 2 ,O 3 , O 4 , NO 2 , H 2 CO) are shown as well as the corresponding residual (peak-to-peak value: ∆ res = 1.60 × 10 −3 ). In this example, the residual is rather structured. One typical structure was marked showing a width W between 20-30 channels on the detector. Please note, that CH 2 O was falsely detected in this fit (considering the fit error) due to the relatively structured residual. This shows the importance of the used fit correction factors U . Please also note the improvements due to the fitted R4-spectrum.

62
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |  Figure A2. Retrieved SO 2 -SCDs from the two SO 2 evaluation ranges. The evaluation scheme centred around 360 nm (SO 2,uwr ) is plotted on the x-axis, the scheme centred around 320 nm (SO 2,lwr ) on the y-axis. The red line indicates perfect correlation between both ranges. In case of large SO 2 -SCDs (i.e. S SO 2 > 3 · 10 18 molecules/cm 2 ), the retrieved SCDs in the lwr-range are more and more underestimated. For smaller SCDs, a good correlation is observable with an increased scattering in the uwr-range Figure A3. Retrieved SO 2 -SCDs from the two SO 2 evaluation ranges. The evaluation scheme centred around 360 nm (SO 2,uwr ) is plotted on the x axis, the scheme centred around 320 nm (SO 2,lwr ) on the y axis. The red line indicates perfect correlation between both ranges. In case of large SO 2 -SCDs (i.e. S SO2 > 3 × 10 18 molecules cm −2 ), the retrieved SCDs in the lwr-range are more and more underestimated.

63
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | . Relative deviation (dS str /S meas ) of volcanic BrO from the measured SCDs (S meas , colour coded) due to stratospheric BrO dSCDs. We assumed a vertical stratospheric BrO column of V str,BrO = 4.0 · 10 13 molecules/cm 2 . The results are plotted as a function of ∆SZA represented in the parametrisation γ ij (Θ i , Θ j ) (see eq. 8). We included all spectra from our dataset with significant BrO-SCDs corresponding to the respective detection limit. For 8% of the data, we observed a significant deviation from the measured SCDs (marked with red circles). Significant deviations were only Figure A4. Relative deviation (dS str /S meas ) of volcanic BrO from the measured SCDs (S meas , colour coded) due to stratospheric BrO dSCDs. We assumed a vertical stratospheric BrO column of V str,BrO = 4.0 × 10 13 molecules cm −2 . The results are plotted as a function of γ ij (Θ i , Θ j ) (i.e. ∆SZA, see Eq. A4). We included all spectra from our dataset with significant BrO-SCDs corresponding to the respective detection limit. For 8 % of the data, the stratospheric contribution (dS str ) exceeded the corresponding fit error (marked with red circles). All of these cases were observed at γ ij values exceeding 0.86 the corresponding measurements were performed before 08:15 LT or after 16:45 LT respectively.