Remote sensing of volcanic CO2, HF, HCl, SO2, and BrO in the downwind plume of Mt. Etna

Remote sensing of the gaseous composition of non-eruptive, passively degassing volcanic plumes can be a tool to gain insight into volcano interior processes. Here, we report on a field study in September 2015 that demonstrates the feasibility of remotely measuring the volcanic enhancements of carbon dioxide (CO2), hydrogen fluoride (HF), hydrogen chloride (HCl), sulfur dioxide (SO2), and bromine monoxide (BrO) in the downwind plume of Mt. Etna using portable and rugged spectroscopic instrumentation. To this end, we operated the Fourier transform spectrometer EM27/SUN for the shortwave-infrared (SWIR) spectral range together with a co-mounted UV spectrometer on a mobile platform in direct-sun view at 5 to 10 km distance from the summit craters. The 3 days reported here cover several plume traverses and a sunrise measurement. For all days, intra-plume HF, HCl, SO2, and BrO vertical column densities (VCDs) were reliably measured exceeding 5×1016, 2×1017, 5×1017, and 1×1014 molec cm−2, with an estimated precision of 2.2×1015, 1.3×1016, 3.6×1016, and 1.3×1013 molec cm−2, respectively. Given that CO2, unlike the other measured gases, has a large and wellmixed atmospheric background, derivation of volcanic CO2 VCD enhancements (1CO2) required compensating for changes in altitude of the observing platform and for background concentration variability. The first challenge was met by simultaneously measuring the overhead oxygen (O2) columns and assuming covariation of O2 and CO2 with altitude. The atmospheric CO2 background was found by identifying background soundings via the coemitted volcanic gases. The inferred 1CO2 occasionally exceeded 2× 1019 molec cm−2 with an estimated precision of 3.7× 1018 molec cm−2 given typical atmospheric background VCDs of 7 to 8× 1021 molec cm−2. While the correlations of 1CO2 with the other measured volcanic gases confirm the detection of volcanic CO2 enhancements, correlations were found of variable significance (R2 ranging between 0.88 and 0.00). The intra-plume VCD ratios 1CO2 /SO2, SO2 /HF, SO2 /HCl, and SO2 /BrO were in the range 7.1 to 35.4, 5.02 to 21.2, 1.54 to 3.43, and 2.9× 103 to 12.5× 103, respectively, showing pronounced day-to-day and intra-day variability.

Abstract. Remote sensing of the gaseous composition of non-eruptive, passively degassing volcanic plumes can be a tool to gain insight into volcano interior processes. Here, we report on a field study in September 2015 that demonstrates the feasibility of remotely measuring the volcanic enhancements of carbon dioxide (CO 2 ), hydrogen fluoride (HF), hydrogen chloride (HCl), sulfur dioxide (SO 2 ), and bromine monoxide (BrO) in the downwind plume of Mt. Etna using portable and rugged spectroscopic instrumentation. To this end, we operated the Fourier transform spectrometer EM27/SUN for the shortwave-infrared (SWIR) spectral range together with a co-mounted UV spectrometer on a mobile platform in direct-sun view at 5 to 10 km distance from the summit craters. The 3 days reported here cover several plume traverses and a sunrise measurement. For all days, intra-plume HF, HCl, SO 2 , and BrO vertical column densities (VCDs) were reliably measured exceeding 5 ×10 16 , 2 ×10 17 , 5 ×10 17 , and 1 ×10 14 molec cm −2 , with an estimated precision of 2.2 ×10 15 , 1.3 ×10 16 , 3.6 ×10 16 , and 1.3 ×10 13 molec cm −2 , respectively. Given that CO 2 , unlike the other measured gases, has a large and wellmixed atmospheric background, derivation of volcanic CO 2 VCD enhancements ( CO 2 ) required compensating for changes in altitude of the observing platform and for background concentration variability. The first challenge was met by simultaneously measuring the overhead oxy-composition, the contributing degassing mechanisms, and the dynamics of mass transport in the volcano interior. Thus, monitoring volcanic gas emissions can help constrain subsurface processes and estimate fluxes of the geological carbon cycle (e.g., Allard et al., 1991;Goff et al., 2001;Burton et al., 2013). Furthermore, it holds the promise for improved eruption forecast since enhanced CO 2 / SO 2 emission ratios have been shown to precede eruptive volcanic activity with a lead time of hours to weeks (e.g., Aiuppa et al., 2007Aiuppa et al., , 2010. While remote sensing of volcanic plume composition has been demonstrated to be a valuable tool for various gases including HF, HCl, SO 2 , and BrO (e.g., Mori et al., 1993;Love et al., 1998;Francis et al., 1998;Bobrowski et al., 2003;Grutter et al., 2008;Stremme et al., 2011Stremme et al., , 2012Theys et al., 2013), measuring volcanic CO 2 remotely faces the particular challenge to discriminate the volcanic enhancements from the background concentration of about 400 ppm (parts per million per volume). Therefore, measurement techniques for volcanic CO 2 mostly rely on in situ gas analyzers (e.g., Shinohara et al., 2008), open-path or closed-path spectroscopic techniques (e.g., Burton et al., 2000;Gerlach et al., 2002) deployed in the proximity of the source region, thereby avoiding downwind dilution of the plume. However, deployment close to the source typically comes with great costs and logistics effort, various hazards for instruments and operators, and, depending on the employed sampling strategy, limited representativeness for the volcanic source as a whole. Pioneering remote sensing experiments relied on detecting the absorption of volcanically emitted CO 2 along an atmospheric light path using the infrared emission of hot volcanic material (Naughton et al., 1969;Mori and Notsu, 1997) or exploiting the thermal contrast between the hot volcanic plume and the background sky (Goff et al., 2001). Aiuppa et al. (2015) demonstrated a LIDAR technique for scanning a volcanic CO 2 plume. However, these remote sensing techniques require path-integrated CO 2 enhancements in excess of several ppm or thermal contrast of several degrees between the plume and the background.
Here, we demonstrate that our field-deployable spectroscopic instrumentation enables remote sensing of volcanic CO 2 enhancements simultaneously with co-emitted HF, HCl, SO 2 , and BrO. Our setup builds on direct-sunlight spectroscopy using portable spectrometers that sample the plume several kilometers downwind of Mt. Etna's craters. The key enabling element is the Fourier transform spectrometer (FTS) EM27/SUN that allows for measuring path-integrated CO 2 enhancements well below 1 ppm exploiting shortwaveinfrared (SWIR) absorption bands (Gisi et al., 2012;Hase et al., 2015;Frey et al., 2015;Klappenbach et al., 2015). The FTS also allows for measuring path-integrated HF and HCl concentrations. We operated two of the FTS, one stationary and another one on a small truck together with a co-mounted grating spectrometer targeting SO 2 and BrO in the ultraviolet (UV) spectral range. The mobile FTS and the UV spectrometer were fed by the same beam of direct sunlight from a Figure 1. Photograph of the remote sensing observatory on the mobile platform, a small truck. The observatory consists of the Fourier transform spectrometer (FTS) EM27/SUN and the UV spectrometer ULS2048x64-ENV5. A common solar tracker feeds sunlight into the two spectrometers. The observatory sits on a rubber and foam suspension. A 12 V battery supplies power to the spectrometers, the solar tracker, two control laptops for the spectrometers, and a low-power consumption PC for the solar tracker.
custom-built solar tracker (Klappenbach et al., 2015). During a 3-week field campaign in September 2015, the truck carried the mobile remote sensing observatory to various sampling locations in the vicinity of Mt. Etna while the stationary FTS observed background variability for most of the time.

Campaign deployment at Mt. Etna
Mt. Etna is a quiescently degassing volcano on Sicily, Italy, exhibiting four active summit craters in a confined source region above 3 100 m altitude. Typical passive, continuous noneruptive degassing amounts to more than 5 kt of CO 2 per day (e.g., Aiuppa et al., 2008;Burton et al., 2013). Eruptive activity occurs occasionally but not during the campaign operations reported here, starting on 5 and ending on 25 September 2015. Whenever weather conditions were favorable for direct-sun view, the custom-built solar tracker, the mobile FTS, and the UV spectrometer were mounted on the truck. Figure 1 illustrates the setup of the mobile observatory. In the morning, the truck was taken to a location where visual inspection of the sky indicated Etna's plume roughly overhead. During sunrise, at large solar zenith angles, the observatory remained stationary collecting absorption spectra while the sun was rising. Later, at smaller solar zenith angles, we operated the observatory in stop-and-go patterns to collect scans through the volcanic plume. Operations were suspended whenever the sun was obscured by cloud cover, which persisted for several days in early September due to rainy conditions. Generally, cloud cover appeared around local noon. The early phase of the campaign was dedicated to testing the setup and identifying suitable measurement pro- The white circular grid illustrates horizontal distances from the crater region in steps of 2.5 km as indicated by the ruler in the lower right. The four active craters on the summit at about 3300 m altitude are indicated by black triangles and the base station at Milo is illustrated by the black circle on the eastern slope. The blueish trajectory in the middle of the map shows the motion of the mobile observatory on 22 September, between 08:00 and 08:50 h UTC (typical solar zenith angle SZA = 55 • ) in the vicinity of Rifugio G. Sapienza (black cross). The color code is a rough measure for the observed SO 2 VCDs as indicated by the color bar. Note that SO 2 VCDs are plotted at the position of the observer at the time of measurement.
cedures. In total, we report on 3 days (18, 22, 23 September 2015) in the second half of the campaign, for which we focus on stop-and-go patterns in the vicinity of Rifugio G. Sapienza (37.700 • N, 14.999 • W; 1910 m a.s.l.) at the southern slope of Mt. Etna. The typical distance between the volcanic source region at the summit and the positions of the sun-viewing spectrometers amounted to 5-10 km. Figure 2 shows an illustrative trajectory of the truck on 22 September 2015. Since the truck was restricted to use general purpose roads, deployment options were limited and it was impossible to collect cross sections that were strictly perpendicular to the plume direction. Figure 2 also demonstrates that, due to rough topography in the vicinity of Mt. Etna, lateral displacements of the truck imply changes in observer altitude. Note that the mobile spectrometers were deployed south of the craters in such a way that the observation vectors towards the sun had southward components, sampling the plume at greater distances than the deployment locations. In addition to the mobile observatory, we operated another FTS at our base station in Milo (37.731 • N, 15.113 • W; 806 m a.s.l.) at the eastern slope of Mt. Etna. For the days reported here, the stationary FTS mostly observed background air masses providing a reference for the background variability of CO 2 . Figure 3. Sketch of the telescope used to feed sunlight into the UV spectrometer. The telescope was mounted into the outer part of solar light beam collected by the solar tracker. A prism feeds sunlight into the telescope (0.5 in. tube diameter, ∼ 120 mm tube length), which consists of a UV filter (Hoya U330), a lens (focal length f = 40 mm), and a circular aperture (800 µm diameter) aligned to the focus of the lens. A polytetrafluoroethylene (PTFE) diffuser plate illuminates the glass fiber (400 µm diameter), which then guides the sunlight into the UV spectrometer.

Instrumentation
Both the mobile and the stationary FTS are of the type EM27/SUN developed by Bruker Optics and the Karlsruhe Institute of Technology. The EM27/SUN delivers solar absorption spectra at a spectral resolution of 0.5 cm −1 in the range 5000 to 11 000 cm −1 without the need for detector cooling. Gisi et al. (2012), Hase et al. (2015),  showed that stationary deployment of the EM27/SUN enables measuring the CO 2 column concentrations with high accuracy and precision. Klappenbach et al. (2015) detailed the performance of the mobile instrument used here, for deployment on a research vessel. Previous studies typically coadded 10 double-sided interferograms with a total exposure of roughly 60 s. We adopted the same measurement pattern for the stationary FTS, but for the mobile FTS we processed pairs of double-sided interferograms with a total exposure of 12 s in order to keep temporal averaging small and to increase measurement frequency. The total field of view of the FTS amounts to 0.27 • , resulting in the central part of the solar disk being observed.
The UV grating spectrometer is an AvaSpec-ULS2048x64-ENV5 manufactured by Avantes. It features a 1800 lines mm −1 grating and 2048 pixel linear array CCD detector covering the spectral range 294 to 457 nm. The entrance slit of the spectrometer is 200 µm wide, supporting an optical resolution of 0.8 nm (FWHM). The spectrometer is placed inside a housekeeping box (EnviMeS TSE 1.1) equipped with a Peltier cooling that keeps the spectrometer temperature stable at 15.00 • C with a precision of 0.02 • C. The housekeeping box disposes of a glass fiber coupling that allows for externally connecting a glass fiber for light intake without the need to open the box. During our campaign, the exposure times for individual spectra under clear sky were around 15 ms. Typically, we co-added 30 spectra, leading to effective exposures of several hundred milliseconds.
While the stationary FTS was equipped with the standard solar tracker delivered by Bruker Optics (Gisi et al.,4 A. Butz et al.: Remote sensing of volcanic CO 2 2011), the mobile observatory used a custom-built variant, initially developed for mobile applications by Klappenbach et al. (2015) and further enhanced here, to improve on the response times and to feed the mobile FTS and the UV spectrometer with the same light beam. Two rotatable tracking mirrors direct a light beam of 40 mm diameter into the mobile FTS through a wedged window. Given that the FTS only uses the central part of the beam, a portion of the outer part of the beam is fed into the UV spectrometer through a telescope. Figure 3 shows a sketch of the telescope assembly. A prism (10 mm side length) reflects the incoming sunlight into a 0.5 in. lens tube mounted on a kinematic platform (not shown). The light beam first passes a UV filter (Hoya U330) that shields unused parts of the solar spectrum, then a lens (focal length f = 40 mm) focuses the beam on an 800 µm wide circular aperture. Further downstream, a polytetrafluoroethylene (PTFE) diffuser plate illuminates a 400 µm glass fiber that takes the light into the spectrometer box via the fiber coupling. To support optical alignment, the length of the lens tube is adjustable, the prism can be rotated, and the attitude of the whole telescope can be controlled through the kinematic mount. Compared to previous stationary and shipborne applications, volcanic CO 2 plume detection benefited from enhancing the response time of the solar tracker. At moderate driving speed (up to ∼ 30 km h −1 ) of the platform and benign road surface, the solar tracker was able to reliably stay on target (the sun) while driving and tracking stability was sufficient to enable recording of absorption spectra by the fast UV spectrometer. Real-time analysis of the SO 2 absorption signal on the passenger seat guided the stop-and-go patterns. However, tracking stability while driving was not sufficient to record CO 2 absorption spectra by the mobile FTS, which has substantially longer exposures than the UV spectrometer. Thus, all CO 2 measurements discussed here were recorded while the truck was stopped.

Data analysis
The solar absorption spectra recorded by the FTS and the UV spectrometer provide information on the gas concentrations, integrated along the path from the sun to the ground-based observer. We first retrieved the column gas abundances of CO 2 , HF, and HCl from the SWIR (Sect. 4.1) and the abundances of SO 2 and BrO from the UV spectra (Sect. 4.2). Then, we inferred the volcanic enhancements (Sect. 4.3).

Spectral retrieval in the SWIR
The FTS delivered interferograms, from which we generated absorption spectra through Fourier transformation assuming Norton-Beer's medium strong apodization function (Norton and Beer, 1977). The DC part of the interferogram is a good measure for brightness fluctuations of the light source during recording of the interferogram. As described in Klappenbach   Table 1. Spectral retrieval windows and their properties used for processing the FTS absorption spectra in the SWIR. Parentheses following absorber molecules indicate the used spectroscopic databases: TCCON refers to the spectroscopic database used by the Total Carbon Column Observing Network (Toon, 2014), HI-TRAN2008 refers to Rothman et al. (2009), and HITRAN2009 refers to Rothman et al. (2009) with published updates and our own empirical modifications. LM refers to the line-mixing database by Lamouroux et al. (2010) (2015), we used the DC part to discard spectra which suffered from variable illumination due, for example, to unstable tracking of the sun or thin clouds passing by. Small fluctuations of the DC part that passed our quality filter were corrected as described in Keppel-Aleks et al. (2007). In addition to the interferograms with severe brightness fluctuations, we discarded FTS measurements in cases of generally inferior quality of the spectra (as judged on basis of fitting residuals) and unsuccessful convergence of the spectral retrieval.
Overall, the quality filtering removed virtually all the spectra recorded by the mobile FTS while the truck was moving. After generation of the absorption spectra, a line-by-line atmospheric transmittance model retrieved the concentrations of the target gases integrated along the vertical, the vertical column densities (VCDs). We employed the software package PROFFIT developed for ground-based directsun spectroscopy (Hase et al., 2004) and previously used for processing EM27/SUN spectra (Gisi et al., 2012;Frey et al., 2015;Hase et al., 2015;Klappenbach et al., 2015). PROFFIT evaluates Beer-Lambert's law assuming a curved hydrostatic atmosphere with horizontally homogeneous layers. We considered several spectral windows in the SWIR with various absorbing molecules as detailed in Table 1. Molecular absorption was taken into account through a Voigt line-shape model including corrections due to line mixing for CO 2 . Table 1 also lists the databases for the driving spectroscopic parameters. An empirical line list modeled the solar Fraunhofer lines. The parameters to be estimated for each measurement comprised scaling factors for the vertical a priori profiles of the absorbers and auxiliary parameters such as spectral shift parameters per window and parameters to fit the transmittance baseline. For CO 2 , HCl, and HF, we only scaled the lower tropospheric part of the vertical profile (four layers be-tween 3.2 and 4.9 km altitude) and adopted the a priori for the rest. VCDs are a posteriori calculated by summing up the layers of the scaled vertical profiles. The volcanic VCD enhancements discussed below thus assume the case of a horizontally homogeneous, extended plume.
Since the molecular absorption lines in the SWIR are optically thick as well as pressure and temperature dependent, the spectral retrieval in the SWIR (unlike DOAS (differential optical absorption spectroscopy) in the UV) requires a detailed representation of the vertical profiles of meteorological parameters and gas concentrations. The profiles of pressure, temperature, and humidity were derived from global NCEP (National Centers for Environmental Prediction) fields on 1 • × 1 • (latitude × longitude) via interpolation in space and time. Surface pressure, surface elevation, latitude and longitude of the soundings were measured by pressure transducers and GPS recorders deployed next to the FTS and subsequently used to construct the input meteorological profiles and the a priori absorber profiles. For O 2 and H 2 O, the latter were taken from the meteorological pressure and humidity data. For CO 2 , HCl, HF, and CH 4 , we used climatological profiles taken from the recommendations of the Infrared Working Group (IRWG) of the Network for the Detection of Atmospheric Composition Change (NDACC) for the mid-latitudes. Note that all retrievals irrespective of sampling intra-plume or background conditions assumed background absorber profiles; i.e., the retrieval had no a priori information on the volcanic plume except for the fact that we only allowed for scaling of the lower tropospheric part of the vertical profile.

Spectral retrieval in the UV
The spectral retrieval for the soundings of the UV spectrometer was based on the DOAS technique (Platt and Stutz, 2008). It builds on Beer-Lambert's law and exploits narrow-band absorption bands of optically thin absorbers like SO 2 and BrO. Table 2 lists details of the spectral windows and the absorption cross sections used for the target gas retrieval. The tabulated high-spectral-resolution absorption cross sections were convolved by the instrument spectral response function and then used to fit the recorded transmittance spectra after subtraction of dark current and offset spectra. A least-squares fitting routine determined the target absorber concentrations integrated along the slant light path through the atmosphere, the slant column densities (SCDs). Ancillary fitting parameters were a cubic background polynomial, a linear additive offset accounting for spectrometer stray light as well as the SCDs of interfering absorbers ozone (O 3 ), nitrogen dioxide (NO 2 ), the oxygen collisional complex (O 2 -O 2 ), and formaldehyde (CH 2 O). UV spectra were discarded in cases of unstable temperature of the spectrometer and strong intensity fluctuations between adjacent spectra.
To compare the UV-retrieved SCDs with the VCDs retrieved in the SWIR, we translated the SO 2 and BrO SCDs into VCDs. To this end, we approximated the geometric assumptions of the SWIR retrieval by ratioing the SCDs by the air mass factor 1 cos(SZA) (solar zenith angle, SZA). If the volcanic plume was horizontally extended and if our lines of sight crossed the plume perpendicularly to its propagation direction, the reported VCDs were truly representative of the vertical column enhancement in the plume. If the volcanic plume has a more complicated geometric shape, that shape would need to be considered for inferring volcanic enhancements and volcanic emissions in absolute units. Ratios of the various gases as discussed below, however, are not affected by the geometric conversion since all gases were measured in the same light path.

Deriving volcanic VCD enhancements
Whenever we detected significant VCDs of SO 2 , HCl, HF, and BrO, we considered those to be volcanic enhancements since atmospheric background concentrations of these species are small. Atmospheric CO 2 , however, is a generally well-mixed gas with background concentrations of roughly 400 ppm (VCD ∼ 8.5 ×10 21 molec cm −2 at 1 bar surface pressure). Therefore, determining volcanic CO 2 enhancements faces the challenges that (A) the measured VCDs co-vary with surface pressure, for our setup mainly variable due to variable observer altitude, and (B) the atmospheric background needs to be removed. We addressed challenge (A) by simultaneously retrieving the VCD of molecular oxygen (O 2 ) and calculating the column-averaged dry-air mixing ratio XCO 2 via where square brackets indicate observed VCDs and 0.20942 is the atmospheric O 2 mole fraction. Referencing by retrieved O 2 is a standard method that aims at canceling instrumental effects and retrieval artefacts (e.g., Wunch et al., 2010;Klappenbach et al., 2015). Assuming that a change in observer altitudes causes the same relative change in the O 2 and CO 2 VCDs, XCO 2 derived from Eq. (1) is independent of observer altitude and, thus, serves to address challenge (B). The atmospheric XCO 2 background varies due to variable meteorological conditions transporting distant source-sink signals to Mt. Etna. Further, imperfect knowledge of spectroscopic parameters and other instrumental or retrieval effects can cause a spurious dependence of XCO 2 on viewing geometry (e.g., Klappenbach et al., 2015). Therefore, we fitted a function P linear in time to the background XCO 2 records. Background soundings are the ones collected at most 30 min before or after a plume scan and with small HF VCDs ([HF] < 1×10 16 molec cm −2 ). Intra-plume measurements are those in-between background soundings exceeding the HF threshold. The volcanic column-averaged mixing  (Meller and Moortgat, 2000) ratio enhancement XCO 2 then reads The volcanic VCD enhancement CO 2 is given by Figures 4 through 6 illustrate the step-by-step derivation of volcanic XCO 2 for the 3 considered days. The variation of the retrieved CO 2 VCDs (upper panels in Figs. 4 through 6) is mostly due to changes in observer altitude. On 18 September, the stop-and-go operations started at around 06:00 h UTC at high altitude close to Rifugio G. Sapienza corresponding to low CO 2 VCDs. The local roads then led the truck downhill toward the southwest causing increasing CO 2 VCDs until we returned to Rifugio G. Sapienza at the end of the morning at about 09:00 h UTC. On 22 September, stop-and-go patterns were restricted to the closer vicinity of Rifugio G. Sapienza (see also Fig. 2) with less altitude-induced changes in CO 2 VCDs than on the other days. On 23 September, observations started around 06:00 h UTC in stationary configuration at Rifugio G. Sapienza waiting for the sun to rise. Around 06:45 h UTC, we took up the stop-and-go patterns, which again led us down and back up Mt. Etna's southern slope until 09:00 h UTC.
Calculating XCO 2 according to Eq. (1) removes the altitude-induced variability in the observed CO 2 VCDs (middle panels in Figs. 4 through 6) and reveals distinct volcanic enhancements between the background measurements and for the sunrise observations on 23 September. Data scatter grows toward late morning, which might be due to some spectra with non-perfect solar tracking escaping our quality filters. Typically, clouds developed in late morning and disrupted the solar tracking. The XCO 2 records observed by the stationary FTS deployed at Milo on the eastern slope of Mt. Etna (blue crosses) confirm that the linear background polynomial (red) is well suited to account for background XCO 2 variability. Except for late morning on 23 September, the stationary FTS at Milo sampled background air masses (according to the HF threshold). On 23 September after about The middle panel shows the column-averaged dry-air mixing ratio XCO 2 (black and grey dots) calculated via Eq. (1). Background measurements (grey) are used to fit the linear background polynomial illustrated by the red line. Blue crosses show the measurements collected by the stationary FTS at the eastern slope of Mt. Etna. The lower panel depicts the volcanic XCO 2 enhancement ( XCO 2 ) calculated according to Eq. (2). Red circles are considered intra-plume enhancements. 08:40 UT, the volcanic plume started drifting toward the lines of sight of the stationary FTS. We refrained from directly using the stationary measurements for background removal since the stationary FTS was placed at about 10 km distance from the truck trajectory at the eastern slope of the mountain. An improved setup might benefit from operating the stationary FTS closer to the trajectory of the mobile observatory.
Removing the background via Eq.
(2) yields the volcanic enhancement XCO 2 (lower panels in Figs. 4 through 6), which is further processed to yield the volcanic VCD enhancement according to Eq. (3).

Volcanic CO 2 , HF, HCl, SO 2 , and BrO
Figures 7 through 9 collect time series of the volcanic enhancements of CO 2 , HF, HCl, SO 2 , and BrO found in the plume of Mt. Etna. Our mobile observatory detected the volcanic plume for a single stop-and-go traverse on 18 September, three traverses on 22 September, and another traverse as well as the sunrise measurements on 23 September. Since the truck carrying the observatory was constrained to use the local roads, the traverses were not strictly perpendicular to the plume but exhibited displacements along plume direction and in altitude (see also Fig. 2). Early on 18 September, for example, Etna's plume was sampled while the observatory was moving to the southwest until 08:00 h UTC, when a dead end forced us to turn back and to move the truck to the east.
We find volcanic CO 2 occasionally exceeding 2 ×10 19 molec cm −2 , which, depending on observer altitude, amounts to column-averaged mixing ratio enhancements of a few tenths of a ppm (see also lower panels in Figs. 4 through 6). Thus, given background VCDs of 7 to 8 × 10 21 molec cm −2 , the detected volcanic signal corresponds to a sensitivity of about 1 : 400. We estimated the overall CO 2 precision as a measure for the random error component by calculating the standard deviation (1σ ) of all the background measurements identified via the HF threshold. It amounts to 3.7 ×10 18 molec cm −2 for CO 2 (0.20 ppm for XCO 2 ), suggesting that individual measurements exceed the precision estimate by a factor 5 to 6. Detection of CO 2 is corroborated by correlated enhancements in HF, HCl, SO 2 , and BrO. Precision estimates for the latter species amount to 2.2 ×10 15 , 1.3 ×10 16 , 3.6 ×10 16 , and 1.3 ×10 13 molec cm −2 , respectively, where we first averaged the UV-measured SO 2 and BrO VCDs on the 12 s integration time of coincident FTS spectra and then calculated the standard deviation of the background measurements identified through the HF criterion. Figure 10 further examines the correlation between CO 2 and co-emitted SO 2 for the intra-plume measurements, while Figs. 11 through 13 show the correlations between SO 2 and HF, HCl, and BrO. As above the UV-measured SO 2 and BrO VCDs were averaged for the integration time of coincident FTS spectra. Table 3 summarizes the CO 2 / SO 2 , SO 2 / HF, SO 2 / HCl, and SO 2 / BrO VCD ratios (as well as R 2 ) obtained by fitting straight lines to the correlations. For 22 and 23 September, we investigated the correlations for all the intra-plume measurements of the respective day as well as for three subsets on each day. On 22 September, these subsets correspond to the three stop-and-go transects (before 08:00 h UTC, between 08:00 and 08:30 h UTC, after 08:30 h UTC). On 23 September, the correlations group into three subsets that can be attributed to the sunrise observations, the early stop-and-go operations (between 07:00 and 07:45 h UTC), and the later stop-and-go operations (after 07:45 h UTC).
The CO 2 / SO 2 ratios show considerable day-to-day and intra-day variability ranging between 7.1 ± 1.5 and 35.4 ± 1.3 when considering the intra-day subsets. The errors correspond to the standard deviations of the fitted slopes. On 18 September, R 2 documents a good correlation between CO 2 and SO 2 . On 22 September, R 2 is fair for the two earlier plume transects but vanishes for the third transect and  Table 3. Intra-plume VCD ratios found by fitting a straight line to the correlations shown in Figs. 10 through 13. The errors are the standard deviations of the fitted slopes. The R 2 of the linear fit is shown in parentheses. On 22 and 23 September, we inferred VCD ratios for the whole day and three subsets separately. The lowermost row provides the campaign-averaged VCD ratios and standard deviations calculated from the daily subsets. The CO 2 / SO 2 ratios for the measurements after 07:45 UTC (bracketed) on 23 September need to be considered with care since CO 2 is negative. For details see text.  For the UV-measured species SO 2 and BrO, we only consider those intra-plume soundings that occur within the integration time of a coincident FTS measurement. Therefore, some SO 2 and BrO measurements that are clearly intra-plume are not red colored. The precision estimated from the standard deviation of all background soundings is shown as an error bar in the upper left corner of each panel.
for the case of a single correlation assumed valid for the whole day. The time series in Fig. 8 confirms that low outliers contaminate the CO 2 record after 08:30 h UTC on 22 September, which might be due to emerging cloud cover disturbing solar tracking stability. On 23 September, we find good correlation for the sunrise observations and a fair R 2 for the early phase of the plume transect (between 07:00 and 07:45 h UTC). For the later phase of the transect (after 07:45 h UTC) CO 2 is negative, which is unreasonable and possibly points to deficiencies of the CO 2 background Atmos. Meas. Tech., 10, 1-14  removal as further discussed below. Assuming a single correlation for 23 September would clearly mask the geophysical variability contained in the records. The SO 2 / HF, SO 2 / HCl, and SO 2 / BrO VCD ratios reveal high R 2 for all intra-day subsets. On 23 September there is considerable variability in the VCD ratios, which is supported by Figs. 11 through 13. The decreasing SO 2 / HF, SO 2 / HCl, and SO 2 / BrO ratios indicate that the compo-sition of the sampled parts of the plume changed from relatively SO 2 -rich to SO 2 -poor in the course of the morning hours. The CO 2 / SO 2 ratios also yield a substantial decrease between sunrise and the early stop-and-go operations (between 07:00 and 07:45 h UTC) but an increase later (after 07:45 h UTC). After 07:45 h UTC, however, CO 2 is negative, which is unreasonable. Thus, we argue that, on 23 September, composition of the observed parts of the plume Figure 10. Correlation of CO 2 and SO 2 intra-plume VCDs observed on 18 September (black circles), 22 September (red squares), and 23 September (blue diamonds) 2015. On 22 and 23 September, we identify three separate groups. On 22 September, these groups are data recorded before 08:00 h UTC (cyan filling), between 08:00 and 08:30 h UTC (green filling), and after 08:30 h UTC (yellow filling). On 23 September, these groups are data recorded during sunrise (cyan filling), after sunrise before 07:45 h UTC (green filling), and after 07:45 h UTC (yellow filling). The solid lines are fits to the daily records, fits for the intra-day subsets are omitted for clarity. The error bars in the upper right corner illustrate the precision estimated from the standard deviation of background soundings. For clarity, data on 22 and 23 September are offset in ordinate by 20 × 10 18 and 40 × 10 18 molec cm −2 , respectively, as indicated by the colored dotted lines. changed from relatively SO 2 -rich to SO 2 -poor and, likewise, from relatively CO 2 -rich to CO 2 -poor while HF, HCl, and BrO levels remained relatively elevated. The argument is supported by the observations of the stationary FTS, which started detecting elevated HF levels at the eastern slope from 08:40 h UTC onward. This finding suggests that, in the later phase of the stop-and-go operations on the southern slope, the plume migrated away from the lines of sight of the mobile observatory toward the lines of sight of the stationary FTS deployed on the eastern slope. Most likely, this change went along with variable contributions of the various craters to the plume composite observed by the mobile FTS.
Given the detected composition variability for 23 September, the negative CO 2 found after 07:45 h UTC could be an artefact of the CO 2 background removal, which is based on the HF threshold for identifying background soundings. If our spectrometers sampled a CO 2 -rich (and HF-rich) plume in the early morning and a CO 2 -poor (but HF-rich) plume later, HF would be a poor CO 2 -plume indicator for the later period and the negative CO 2 could just correspond to undetectably low enhancements. To try the sensitivity of the inferred CO 2 to background removal on 23 September, we deliberately categorized all CO 2 measurements after 07:45 h UTC as background and recalculated CO 2 as described in Sect. 4.3. The inferred CO 2 / SO 2 ratios for sunrise and for the early stop-and-go operations (between 07:00 and 07:45 h UTC) remain largely unchanged, amounting to 34.8 ± 0.9 (R 2 = 0.88) and 12.7 ± 2.7 (R 2 = 0.21), respectively. Thus, our conclusions for the early measurements on 23 September are largely insensitive to background removal. Overall, our assessment emphasizes that short-term composition variability of the observed plume requires careful consideration.

Discussion and conclusion
We demonstrate simultaneous remote sensing of volcanic CO 2 ( CO 2 ), HF, HCl, SO 2 , and BrO VCD enhancements in the plume of quiescently degassing Mt. Etna several kilometers downwind of the source. Our remote sensing observatory combined the portable and rugged EM27/SUN FTS for the SWIR spectral range (observing CO 2 , HF, HCl) with a DOAS spectrometer for the UV (observing SO 2 , BrO), both instruments measuring direct-sun absorption spectra. The spectrometers were supplied with sunlight by a common, fast solar tracker, all together deployed on a mobile platform and supplied by a 12 V battery. The mobile setup enabled sequentially measuring intra-plume and background air masses in stop-and-go patterns. Campaign operations were supported by another, stationary, EM27/SUN FTS, sampling atmospheric background conditions for most of the time. Gener- ally, we focused on the retrieval of CO 2 , since CO 2 , unlike the other measured volcanic gases, has a high atmospheric background concentration that is well mixed.
For 3 days reported here, the sunrise and stop-and-go observations yielded CO 2 up to about 2 ×10 19 molec cm −2 with an estimated precision of 3.7 ×10 18 molec cm −2 . The other volcanic gases were measured with an estimated precision of 2.2 ×10 15 , 1.3 ×10 16 , 3.6 ×10 16 , and 1.3 ×10 13 molec cm −2 , for HF, HCl, SO 2 , and BrO, respectively. The key to detecting the small volcanic CO 2 enhancements on top of the high atmospheric CO 2 background column (7 to 8 ×10 21 molec cm −2 ) was the simultaneous observation of the overhead O 2 column and of volcanic HF coemitted with CO 2 . The O 2 column was used to compensate CO 2 variations due to changes in observer altitude. The HF columns provided an indication for intra-plume and background measurements. The latter were used to remove the atmospheric CO 2 background. The tightest correlation between CO 2 and SO 2 (R 2 = 0.88) was found for the sunrise observations on 23 September, when the slant absorption path through the atmosphere was longest among our campaign records. Good to fair correlations (R 2 in the range 0.45 to 0.17) were found for four stop-and-go plume transects. One plume transect yielded a vanishing correlation between CO 2 and SO 2 and another plume transect yielded negative CO 2 , most likely related to a change of plume composition from CO 2 -rich to CO 2 -poor during the transect. The correlations between SO 2 and HF, SO 2 and HCl, and SO 2 and BrO were all high (R 2 greater than 0.9 for all except one transect).
The intra-plume VCD ratios of CO 2 / SO 2 , SO 2 / HF, SO 2 / HCl, and SO 2 / BrO match the range of emission ra- tios and molar ratios previously reported for Mt. Etna (e.g., Aiuppa et al., 2007Aiuppa et al., , 2008La Spina et al., 2010;Voigt et al., 2014). During our campaign operations, in situ gas analyzers deployed in the vicinity of the summit craters found molar CO 2 / SO 2 ratios of 17.7 ± 10.3 close to Voragine crater on 18 September, 22.2 ± 5.8 close to Bocca Nuova crater on 22 September, and 13 ± 5 close to Bocca Nuova crater on 23 September, which confirms that Mt. Etna exhibits substantial variability of the emissions and that the various craters show different emission ratios (e.g., Aiuppa et al., 2008;La Spina et al., 2010). Since our remote sensing approach samples the plume several kilometers downwind of the craters, variability of the observed plume composition can be either due to temporally variable source gas emission, varying contributions of the different craters (e.g., induced by variable winds) (Voigt et al., 2014), and possibly flank emissions contributions (Allard et al., 1991) to the plume. Interpretation of our remote sensing data records becomes difficult if such variability occurs on the timescales of a plume transect. For our measurements on 23 September, we find a change in plume composition from CO 2 -rich and SO 2 -rich to CO 2 -poor and SO 2 -poor in the course of the morning hours. Most likely, the detected trend occurred since plume direction changed from southward to eastward, which caused variable contributions of the various craters to the plume composite observed on the southern slope of Mt. Etna.
The employed methodology bares potential for substantial extension and refinement. In view of plume variability, data interpretation would benefit from reducing the time span needed to conduct plume transects by the mobile observatory. During our campaign, the fastest transects took roughly 30 min. Reducing the exposure time of the FTS or the number of spectra collected during a transect would speed up the operations but the signal-to-noise ratio would become worse. A caveat also applies to variability of the volcanic CO 2 , HF, or HCl signal on the timescale of the FTS exposure (12 s for the mobile FTS, 60 s for the stationary FTS). Since the FTS collects interferograms, a change of the volcanic gas signal during interferogram recording has a largely unpredictable effect on the absorption spectra and the inferred gas columns. A grating spectrometer would in good approximation average the volcanic signal over the exposure time. In the view of extending our methodology toward emission estimates, we will aim at linking the VCD ratios observed in the plume to actual emission ratios by combining our measurements with the monitoring infrastructures at Mt. Etna such as the FLAME SO 2 network (Salerno et al., 2009), by operating a dedicated SO 2 camera system together with the direct-sun spectrometers (e.g., Kern et al., 2015;D'Aleo et al., 2016), or by meteorological modeling of the local wind fields. In the view of volcano monitoring, a few of our spectrometers could be operated in stationary observatories, possibly together with a mobile observatory, to set up a network that allows for monitoring Mt. Etna's or any other suitable volcano's plume. If the FTS can be deployed somewhat closer to the source than the local infrastructures allowed for during our campaign, the plume would be less diluted and CO 2 would show greater enhancements, allowing for better relative precision than reported here.
Overall, our approach allows for detecting volcanic CO 2 enhancements with good confidence and for measuring volcanic HF, HCl, SO 2 , and BrO with high precision using a setup that can be easily deployed in the field in safe distance from the craters. Therefore, further refinements such as discussed above could make our approach a valuable tool in volcano monitoring. Inevitable drawbacks of direct-sun spectroscopy, however, are the required daytime and clearsky conditions.

Data availability
The data are available from the author upon request.
Author contributions. Nicole Bobrowski and André Butz developed the research question. André Butz, Anna Solvejg Dinger, Nicole Bobrowski, Julian Kostinek, Lukas Fieber, Constanze Fischerkeller, Giovanni Bruno Giuffrida, Jonas Kuhn, Peter Lübcke, and Lukas Tirpitz took an active part in the field campaign by operating instrumentation, collecting, analyzing, and sharing data. Frank Hase and Friedrich Klappenbach supported preparations for the field campaign through instrument developments. André Butz, Anna Solvejg Dinger, Nicole Bobrowski, Frank Hase, Friedrich Klappenbach, and Qiansi Tu contributed to the spectral retrievals. André Butz wrote the paper. Anna Solvejg Dinger produced the figures for an earlier version of the manuscript.