**1. Introduction**

Active arc volcanoes that erupt with evolved magmas commonly display long periods (decades to centuries) of dormancy between their magmatic eruptions. However, these quiescent periods are often characterized by hydrothermal manifestations (fumaroles, boiling pools, thermal springs, etc.) that attest to a persistent heat and gas supply from a magmatic source at depth. While interacting with shallow groundwater and host rocks, this continued magmatic supply sustains a hydrothermal system (e.g., [1])) and generates acid hydrothermal fluids that promote intense alteration of the host rocks. This could also lead to mechanical weakening of the volcanic edifices and their potential collapse. Moreover, an increasing gas supply from depth or/and gradual self-sealing of a constantly fed hydrothermal system are processes that can lead to shallow overpressurization followed by violent eruption of a closed-conduit volcano (e.g., [1–3]). As recorded by the cases of Ontake in Japan in 2014 [4] and Tongariro in New Zealand in 2012 [5], even purely phreatic eruptions at volcanoes displaying prolonged unrest can involve major hazards and risks owing to their sudden and often unpredictable onset [6]. Therefore, monitoring the spatial distribution and temporal evolution of hydrothermal manifestations at such volcanoes, in combination with geophysical surveys, is crucial to detect and interpret precursors of either non-magmatic explosive activity or, instead, a magmatic eruption.

Compositional changes in fumarolic exhalations have in fact been recognized as signals of unrest or even precursors of several eruptions at dormant volcanoes (e.g., [2,7–10]. However, unequivocal interpretation of these chemical changes is often challenging, owing to the complexity of chemical reactions and bu ffering e ffects involved in water-gas-rock interactions (e.g., [1]). Additional insight into the significance of chemical changes can be obtained by quantifying the emission rate of fumarolic gases (e.g., [11,12]). Nevertheless, flux measurements of fumarolic manifestations are not straightforward, for two main reasons: (i) hydrothermal gas emissions are often too weak to generate a sizable volcanic plume and, hence, to allow gas flux quantification with remote sensing or airborne measuring tools; and (ii) low-temperature (<100–300 ◦C) fumarolic gases generally contain little SO2, which impedes the use of UV sensing tools commonly applied to quantify SO2-rich gas emissions from hotter vents or erupting volcanoes (e.g., [13]). The prevalent sulfur species in hydrothermal gas emissions is usually H2S (e.g., [8]) whose remote detection still remains challenging (e.g., [14]). Alternative approaches targeting the fumarolic fluxes of H2O and CO2 were tested with success on a few volcanic sites: these include ground-based eddy gas profiling [15], CO2 plume imaging with tunable diode laser spectroscopy [16] or di fferential absorption lidar [17]. However, these methods are relatively di fficult to carry out in the field, because they require gentle volcano topography, easy access to fumarolic fields and favorable weather conditions. Moreover, the high abundances of H2O and CO2 in the atmospheric background require substantial volcanic enrichments of these two components to allow reliable quantification.

Recently, Allard et al. [12] demonstrated that the flux of H2S-bearing fumarolic gases at dormant volcanoes in hydrothermal activity can reliably be determined from in situ gas plume concentration profiles measured with a portable Multi-component Gas Analysing System (MultiGAS). MultiGAS is a light and compact device composed of an infrared spectrometer and electrochemical sensors (plus air temperature, atmospheric pressure, and relative humidity sensors) that allows simultaneous and analysis of H2O, CO2, SO2, H2S and H2 mixing ratios in air-diluted volcanic plumes (e.g., [18–20]). The MultiGAS can be used for discrete measurements but also for permanent gas surveys such are currently operated on several volcanoes worldwide (e.g., [2,21]).

Using a MultiGAS instrument, Allard et al. [12] determined the mass output of fumarolic gas emissions from La Soufrière volcano in Guadeloupe, an andesitic volcanic dome of the Lesser Antilles arc that threatens several tens of thousands of people and has raised recent concern due to increasing degassing unrest over the past two decades ([22] and references therein). By coupling the measured gas composition to the horizontal and vertical distribution of H2S in the plume cross-sections and then scaling to the wind speed measured at the vent, Allard et al. [12] found that the total gas flux from La Soufrière had increased by a factor of ~3 in 2012 compared to a first measurement made in 2006 (using one simple H2S electrochemical sensor). Because isotopic tracers demonstrate a persistent supply of magma-derived volatiles and heat to La Soufrière hydrothermal system (e.g., [12,23] and references therein; [24–27]), such an increase in the gas discharge, together with other phenomena recorded by the local volcano Observatory [28], have raised concerns about the evolution of current unrest and, therefore, deserves further investigations.

Here we report on new and more extensive measurements of the gas compositions and fluxes of fumarolic emissions from the La Soufrière volcano in 2016–2017, while degassing unrest at its summit displayed continued expansion of thermal (>50 ◦C) ground areas, new fumaroles and the reactivation of several other vents since July 2014. For these measurements, we used a novel instrumental geometry consisting of an array of three MultiGAS devices operated simultaneously at various heights (1 to 3 m) above the ground. This instrumental array allowed us to accurately determine the fumarolic gases emitted from the different active vents of La Soufrière lava dome, as well as gas emission rates from the three major degassing vents. Moreover, in addition to be compared with previous gas data, our results are interpreted in light of three complementary information: (i) recent electrical conductivity imaging of the underground circulation of hydrothermal fluids inside the lava dome [29], (ii) ground deformation of the lava dome as revealed by extensometric survey of its main fractures since 1995 [22,30], and (iii) seismic data recorded by the Observatoire Volcanologique et Sismologique de Guadeloupe (OSVG-IPGP). With such an approach, our study provides unprecedented insight into the spatio-temporal relationships between the evolution of surface activity (fumarolic degassing and the propagation of ground thermal anomalies) and underground phenomena (hydrothermal circulation, near-field ground deformation and seismicity) at La Soufrière. Thus, we define an improved framework to interpret temporal changes in gas emissions from the volcano during its present unrest phase and in the future, which also bears broader implications for the monitoring of dormant active volcanoes in hydrothermal unrest elsewhere.

#### **2. Volcanological Background and Recent Activity**

La Soufrière of Guadeloupe is one of the most active volcanoes of the Lesser Antilles island arc (Figure 1a). It is the younges<sup>t</sup> eruptive centre of a larger composite volcano, the Grande Découverte massif, located in the southern part of Basse-Terre Island [6,31]. Its main feature is a ~0.05 km<sup>3</sup> andesitic lava dome (1427 m a.s.l.; Figure 1b,d), cut by numerous fractures (Figure 1b,c), that was emplaced during the last major magmatic eruption in 1530 AD [32].

Since then, intense hydrothermal activity has persisted on and around this lava dome (fumaroles, solfataras, hot grounds, thermal springs), culminating in six phreatic eruptions of varying intensity in 1690, 1797–1798, 1812, 1836–1837, 1956, and 1976–1977 [6,33]. The last phreatic events in 1976–1977 were accompanied by an intense seismic crisis (Figure 2) and resulted in a four-month evacuation of 75,000 people from the surroundings [33,34]. Exegesis and re-analysis of historical chronicles have shown that the three most violent phreatic eruptions in 1797–1798, 1836–1837 and 1976–1977 generated small-volume but hazardous manifestations, such as laterally-directed explosions, cold dilute turbulent pyroclastic density currents (PDCs) with runouts ≤1.5–2 km, and rockslides and/or debris avalanches resulting from partial collapses of the dome [6,29,35].

After the 1976–1977 eruption, La Soufrière has become increasingly studied and monitored with multi-parameter networks managed by the Observatoire Volcanologique et Sismologique de Guadeloupe (OVSG-IPGP). Monitoring data are provided by continuous seismic, global navigation satellite system (GNSS), and meteorological networks, as well as periodic extensometric surveys of the evolution of the lava dome fractures [30] and routine sampling/analysis of the fumarolic gases and

thermal springs (OVSG-IPGP, 1999–2017; [10,12,36]. Recorded data are processed and available online on the WebObs internal server [37,38]. Seismic and GNSS data are distributed on the public access Volobsis server of the IPGP Dater Center (http://volobsis.ipgp.fr/). Seismic data and petro-geochemical investigations indicate that the volcano is fed by an andesitic magma reservoir located at about 6–7 km depth beneath the summit [32,33,39–41]. According to C, He and Cl isotopic ratios of the hydrothermal fluids, persistent degassing of this magma reservoir continuously supplies fluids and heat to a shallower hydrothermal system [10,12,24–28,42,43].

**Figure 1.** (**a**) Map of the Lesser Antilles volcanic arc; (**b**) main structures and manifestations of the hydrothermal system at La Soufrière volcano showing the locations of thermal springs (blue circles), sites of exurgence of pressurized hydrothermal fluid (green triangles) during 1797–98, 1836 and 1976–77 phreatic eruptions [6,29], the main faults (blue-green), historical eruptive fractures and craters (black) and collapse structures (purple triangle on trace), the region of highest electrical conductivity (>1 S/m, light purple area) determined by Rosas-Carbajal et al. [29], active fumaroles (small and big yellow circles), 10 m DEM from GeoEye image, Latitude Geosystems; (**c**) 1 m resolution orthophoto (GeoEye) of the lava dome showing the main active fumaroles of the summit area (yellow circles) TAS: Tarissan crater; NAPN: Napoléon Nord; NAP: Napoléon 1 NPE1: Napoleon Est 1; NPE2: Napoléon Est 2; CS: Cratère Sud, which is divided into northern (CSN), central (CSC) and southern (CSS) vents; G56: Gouffre-56; LCS: Lacroix Supérieur, that is divided into LCS-1 and LCS-2; BLK1: Breislack fumarole; TY: Morne-Mitan fumarole along the Ty fault; (**d**) aerial photo of La Soufrière lava dome (October 2016) showing vegetation impacted by prolonged H2S- and HCl-rich acid gas emissions, photo taken by A. Anglade, OVSG-IPGP, with a drone from OBSERA and with permission by the Parc National de Guadeloupe).

After the 1976–1977 events, fumarolic degassing around and on top of the lava dome strongly diminished to ultimately disappear by 1984, synchronously with a gradual decline of seismic activity (Figure 2a). A period of deep rest extended up to 1992, with only minimal fumarolic activity persisting along the volcano-tectonic Ty fault at the SW base of the dome [6,36,44,45]. However, towards the end of 1991, the Tarade (TA) thermal spring that was dry since 1977 reactivated and a new thermal spring (Pas du Roy, PR) appeared at the southern base of the dome (Figure 1b). Then, in May 1992 a new phase of degassing unrest began on top of the lava dome in concomitance with a renewed increase of shallow seismicity (Figure 2). For about five years, increasing fumarolic degassing remained focused at the Cratère Sud (Figure 1b–d), but subsequently extended along the Napoleon fracture (March 1997) then at Gouffre Tarissan pit crater (late 1998). Gas emissions from both Cratère Sud and Tarissan gradually became intense enough to generate a permanent volcanic plume, visible from

several kilometers distance always on clear days. Moreover, in 1998 fumarolic exhalations from both craters started to become extremely acidic (mean pH of 0.95 ± 0.64 at CSN between 1998–2001) due to their marked enrichment in chlorine. A surficial boiling acid lake (pH= −0.8 to 1.6 and T ◦C = 88.8 ± 8.6) formed and persisted from April 1997 to about December 2004 in Cratère Sud (see Figure S4 in Rosas-Carbajal et al. [29]), while another boiling acid lake (pH= −1.3 to 0.8 and T ◦C = 78.3 to 100.3) developed since late 2001 at the bottom of the 30 m wide and 80–100 m deep Gouffre Tarissan (OVSG-IPGP, 1999–2018). Up to now, this acid lake in Tarissan has remained active and been regularly sampled by OVSG [6,10,28]. Acid gas emissions from both vents over the past two decades have considerably impacted the vegetation growing on the downwind summit and W-SW flanks of the dome (Figure 1b).

**Figure 2.** (**a**) Number of earthquakes per year (grey bars) and associated released seismic energy (red line). (**b**) Spatial distribution, magnitude (circles size) and depth (false color scale) of 1799 seismic events recorded in 2007–2017 with longitudinal and latitudinal projections of their hypocenters (OVSG-IPGP, 1999–2018).

Since 1998 volcanic seismicity at La Soufrière has fluctuated in terms of number of events and released energy (Figure 2a). The prevalent seismicity was characterized by numerous swarms of volcanic earthquakes of very low magnitude (dominantly Md ≤ 1), lasting over periods of a few days to a few weeks, most of which originated within 1–4 km depth right below the summit lava dome (Figure 2b). Since 1992, however, 20 felt volcanic earthquakes were recorded, among which 5 in 2013, 1 in 2014, 1 February (1st) 2018, 1 in April (16) 2018, and 2 on 27 April 2018 when the strongest (M = 4.1) seismic event occurred in 42 years [22,28]. Concurrently, the degassing unrest phase has continued to evolve, with gradual reactivation of other vents and the recent opening of new vents. While fumarolic activity remained dominantly concentrated at Cratère Sud and Gouffre Tarissan until 2007, more recently fumarolic degassing resumed or initiated at several other vents, gradually migrating from the lower eastern flanks of the volcano, along the main October 1956–8 July 1976 fracture and the Breislack fault system (Figure 1b,c), up to the summit area. In particular, fumarolic degassing progressively renewed and increased at y Gouffre-56 (G56, the site of 1956 phreatic eruption [46]), then propagated to the nearby Lacroix Supérieure fumarole (LCS- and LCS-2) in December 2011 and, in October 2013, further east in the vicinity of the Breislack crater (Figure 1b, site of the 1797, 1812 and 1836 phreatic eruptions [6]). On top of the lava dome, a new fumarole (NAPN) appeared in July 2014 north of the Napoleon fracture that was reactivated in numerous sites along its 200 m stretch. Between 8 and 10 February 2016, two new vents (NPE1 and NPE2) opened further northeast (Figure 1c). Increasing fumarolic activity in that sector of the lava dome and along the Breislack fault system over the past decade also coincides with enhanced shallow seismicity beneath that part of the volcano (Figure 2b). The Breislack fracture system was involved in all of the phreatic eruptions of La Soufrière since 1797 [6,29,35].

A regular survey of La Soufrière fumarolic gases has long been conducted at the Cratère Sud Central (CSC), the only vent accessible for gas sampling [10,12,43]. The new fumarole opened in 2014 north of Napoleon crater (NAPN; Figure 1c) has been sampled occasionally [28]. Otherwise, chemical data available for the other fumarolic vents unaccessible to gas sampling were obtained recently from in-situ MultiGAS survey (March 2006 and March 2012 campaigns [12]). In this study we report on gas compositions for all the fumarolic vents (see Figure 1c) that were active on top of the lava dome in 2016–2017, as well as gas fluxes from the three major emitting vents (Tarissan, Cratère Sud and Gouffre-56). We also report data on fumarolic degassing at the base of the lava dome, Morne Mitam site, along the Ty fault. Our chemical and flux results are then compared to previously obtained data [10,12,43]. From here in, we use the acronyms listed in the caption of Figure 1 for the fumarolic vents, whereas full names will be maintained for major structural elements such as faults, fractures and craters.

#### **3. Methodologies for Gas Measurements and Extensometric Survey**

In-situ field determination of fumarolic gas compositions at La Soufrière was performed on several occasions in 2016 (May 10–12, June 18, September 6, and December 9–10) and 2017 (March 26 and October 31) by using MultiGAS. Two types of devices were operated. The first one, built in Palermo University, consists of a Gascard IR spectrometer for CO2 determination (calibration range: 0–3000 ppmv; accuracy: ± 2%; resolution: 0.8 ppmv) and of City Technology electrochemical sensors for SO2 (sensor type 3ST/F; calibration range: 0–200 ppm, accuracy: ± 2%, resolution: 0.1 ppmv), H2S (sensor type 2E; range: 0–100 ppm, accuracy: ±5%, resolution: 0.7 ppmv) and H2 (sensor type EZT3HYT; range: 0–200 ppm, accuracy: ± 2%, resolution: 0.5 ppmv), all connected at a Campbell Scientific CR6 datalogger. The second one, built at Simon Fraser University (SFU, Canada), consists of two Alphasense non-dispersive IR (NDIR) solid-state detectors for CO2 (range: 0–5000 ppmv and 0–5%; accuracy: ± 1 and ± 1.5%; resolution: 0.1 and 1 ppm, respectively) and Alphasense electrochemical sensors for SO2 (sensor type EZT3ST/F; calibration range: 0–2000 ppm; accuracy: 1%; resolution: 0.5 ppm) and H2S (sensor type EZT3H; range: 0–2000 ppm; accuracy: 1%; resolution: 0.25 ppm). Each instrument also includes a relative humidity sensor (Galltec, range: 0–100% Rh, accuracy: ± 2%), coupled with a temperature sensor (range: −30–70 ◦C, resolution: 0.01 ◦C) and atmospheric pressure (Patm) sensor, all fixed externally, that permit to determine the concentration of water vapor (ppmv). The latter was obtained by combining the sensor readings of Rh% and Patm following the procedure described in Moussallam et al. [47]. H2O determination with these external sensors allowed us to circumvent the potential influence of steam condensation in the MultiGAS inlet tubing and, therefore, to avoid underestimating the measured water/gas ratios. Unfortunately, this setting has been used only during the May 2016 campaign when the SFU-type MultiGAS was available.

Prior to field measurements, all sensors were calibrated in laboratory using target gases of known concentration. The di fferent MultiGAS instruments have been tested in the field by measuring the same gas (the inlets were close together), and the results show a good comparison between the measured concentrations. Time-averaged gas compositions (H2O, CO2, SO2, H2S and H2) were determined at all fumarolic vents during stationary measurements lasting a few minutes in the downwind air-diluted plumes. Post-processing of data was performed using the RatioCalc software [48]. Di fferences in the response time of the sensors were taken into account from lag times in correlation analysis of the various time series, and potential interference between SO2 and H2S sensors was calibrated and corrected. CO2 and H2O contents were corrected for the ambient air composition, measured in the clean atmosphere outside the volcanic plumes. Figure 3 shows an illustration of MultiGAS recordings at the three main craters.

**Figure 3.** Example of MultiGAS recording of CO2, SO2, H2S and H2 co-variations in gas emissions from Gou ffre-56 (G56), Cratère Sud (CS) and Gou ffre Tarissan (TAS) fumaroles during walking profiles across the plumes in May 2016.

Fumarolic gas fluxes were determined on six occasions in 2016 (March, May and December) and 2017 (March and October) at the three main degassing craters: CS, TAS and G56 (Figure 1c). As detailed below, gas fluxes were derived by scaling the integrated amount of each gas species in the air-diluted plume cross-sections to the wind speed measured during the gas survey with a hand-held anemometer. During most of our measurements, the volcanic gas plumes were flattened to the ground by relatively strong trade winds (4–16 <sup>m</sup>·s<sup>−</sup>1) and, according to both field observations and video-camera footage, had a maximum height of ca. 3 m above the ground at each of our measuring sites. The horizontal and vertical distributions of gas species in the plume cross-sections were measured during walking traverses orthogonal to the plume transport direction, a few meters downwind to the vents, with hand-held GPS in one-second track mode. One key improvement in our measurements, with respect to previous studies, has been the simultaneous use of multiple MultiGAS devices allowing to record gas concentrations in real time at di fferent heights within the plume cross-sections. On 12 May 2016 we simultaneously operated three MultiGAS (two UniPA-type and one SFU-type) instruments whose

gas inlets were fixed at 1, 2 and 3 m height above the ground along one single (4-m long) vertical pole held vertically (Figure 4a).

**Figure 4.** (**a**) MultiGAS setting for measuring gas concentrations between 1 to 3 m above the ground along a traverse perpendicular to the plume's main axis and resulting (**b**) CO2 concentration profiles crossing G56 volcanic plume acquired on 10 May 2016. (**c**) Calculated fitting surface interpolating CO2 concentrations.

On 9 December 2016, we used the same configuration but with only two UniPA-type MultiGAS instruments whose inlets were positioned at ~1.7 and ~3 m heights. On 26 March 2017, multiple (2–3) plume transects at the three main vents were made using one single UniPA-type MultiGAS but sequentially switching the gas inlet position from ~0.8 to ~2.7 m height above the ground. The maximum height of ~3 m for our measurements corresponds to the observed maximum elevation of the volcanic plumes at each measuring site, recurrently grounded by strong winds, and allowed us to accurately quantify gas distributions within the entire, or almost entire, plume cross-sections.

While H2S, SO2 and most of the measured H2 are purely volcanic-hydrothermal in origin, H2O and CO2 concentrations need to be corrected for the atmospheric background in which these two species are abundant. Accordingly, each of our walking profiles (Figure 4b) was initiated in pure atmosphere, upwind of the volcanic plumes, in order to characterize and then subtract the ambient air composition from our recorded data. The concentrations of purely volcanic CO2 and H2S retrieved in each plume cross-section were interpolated with a smoothing interpolant function with an R<sup>2</sup> >0.95. The interpolating surfaces (Figure 4c) were integrated to obtain integrated concentration amounts (ICAs, in ppm·<sup>m</sup>2) in each plume cross-section (Table 2). Our series of traverses for each vent reveal a similar CO2 and H2S distribution pattern at the different heights, with maximum concentrations at half plume height (see Figure 4b), indicating steady plume structures with maximum gas density centered at between ~1.5 and 2 m above the ground (Figure 4b). In agreemen<sup>t</sup> with both field visual observations and video-camera footage, volcanic gas concentrations were considered below the detection limit at above the maximum plume elevation of 3–3.5 m. Both our MultiGAS procedure and the field conditions thus provided us with a good coverage of the plume structures and gas emissions from the three main vents on top of La Soufrière lava dome.

Wind speed is one key parameter and a main source of error in quantifying volcanic gas fluxes. On La Soufrière, we repeatedly measured the wind speed, as well as atmospheric pressure, temperature and relative humidity, with a hand-held weather sensor device. The measurements were performed inside the gas plume in order to achieve a fair match between plume and wind speeds. In fact, these two velocities can significantly differ from one another if measured at different sites, contributing to high and usually unquantified errors [49]. Hence, the measured wind/plume speed was used (error 1σ) in our calculations of the volcanic gas fluxes (Table 2). Weather conditions varied rapidly during our measurements: sunny intervals alternated with episodes of fog and occasional rain, and trade winds blowing from the northeast varied in speed from moderate (3–5 <sup>m</sup>·s<sup>−</sup>1) to strong (12–18 <sup>m</sup>·s<sup>−</sup>1). Strong winds were less variable (relative standard deviation of ~30%) than moderate winds (RSD of ~60%). Despite the fact that the derived gas fluxes are a ffected by these changing conditions in terms of variability, our measured chemical compositions show, instead, constant values at single vents on time scales of a few days, months and even one year.

The flux of H2S and CO2 are thus obtained from the integrated column amounts (ICAs) of H2S and CO2 directly measured in each plume cross-section and then by multiplying with the average wind speed (Table 2). The fluxes of other gas species were derived by multiplying the H2S or CO2 fluxes by the average Xi/H2S or Xi/CO2 weight ratio of each gas emission (calculated from the molar ratios shown in Table 1).

We integrate here our volcanic gas observations with data from extensometric survey of the fractures of La Soufrière lava dome (Figure 1b–c) monitored since 1995 by OSVG manual recording of the width of 15 fractures performed every three months [22,30]. Each of the monitored sites is equipped with two stainless steel hooks anchored to rock on both side of the fracture. A tape-type extensometer allows operators to measure the distance in between the two hooks. The instrument consists of a steel tape, a tape tensioning apparatus and an embedded caliper. An indexing mark is used to apply tension on the tape at constant values for each measurement. The standard error ranges between 0.1 and 0.5 mm, mainly depending on contemporaneous wind conditions.
