Volcano electrical tomography unveils edifice collapse hazard linked to hydrothermal system structure and dynamics

Catastrophic collapses of the flanks of stratovolcanoes constitute a major hazard threatening numerous lives in many countries. Although many such collapses occurred following the ascent of magma to the surface, many are not associated with magmatic reawakening but are triggered by a combination of forcing agents such as pore-fluid pressurization and/or mechanical weakening of the volcanic edifice often located above a low-strength detachment plane. The volume of altered rock available for collapse, the dynamics of the hydrothermal fluid reservoir and the geometry of incipient collapse failure planes are key parameters for edifice stability analysis and modelling that remain essentially hidden to current volcano monitoring techniques. Here we derive a high-resolution, three-dimensional electrical conductivity model of the La Soufrière de Guadeloupe volcano from extensive electrical tomography data. We identify several highly conductive regions in the lava dome that are associated to fluid saturated host-rock and preferential flow of highly acid hot fluids within the dome. We interpret this model together with the existing wealth of geological and geochemical data on the volcano to demonstrate the influence of the hydrothermal system dynamics on the hazards associated to collapse-prone altered volcanic edifices.

: View of the La Soufrière volcanic massif from the south (1998). The La Soufrière lava dome was emplaced in 1530 AD within superimposed horseshoe-shaped structures from edifice collapses of the last 3,200 years directed south-southwestward (see Fig. 1b for details). Stars: orange main thermal springs 30-49 • C with Galion (GA), Tarade (TA) springs discussed in text; yellow: main thermal springs 24-35 • C with Pas du Roy (PR), Bains Jaunes (BJ) springs discussed in text. See Fig. 4b for details and a cross sectional model of the volcano with the position of the conductor regions discussed in the text. In the 1998 photograph, the main vertical gas plume emanates from Cratère Sud (CS). The low flux intermitent degassing that characterized Tarissan crater (TAR) in 1998 has increased recentt years to become permament and the second largest source of gas flux from La Soufrière. The brown area on the dome is the area of vegetation killed subsequently to the onset of high-flux degassing of chlorine-rich hydrothermal and magmatic fluids from Cratère Sud in early 1998. Photo JC Komorowski (IPGP).
The spatio-temporal evolution of surficial manifestations of the hydrothermal system may reflect two main scenarios. Firstly, a reorganization of the pattern of fluid circulation inside the dome might reflect a response to progressive sealing of the formerly active flow paths [36] by hydrothermal alteration. Both the intense hydrothermal activity and the heavy rains (∼6-7 m.yr −1 ) supply the hydrothermal reservoirs that favour fluid mineralization by magmatic gas and the formation of clayey material that progressively fills and blocks open fractures in the edifice decreasing its macropermeability [37,36,15]. The resulting sealing causes fluid confinement and overpressurization, which eventually lead to the opening of new flow paths inside the edifice. A second possible scenario has been proposed in which the increased flux of summit chlorine-rich degassing and the episodic chlorine spikes recorded in the Carbet and Galion hot springs reflect the sporadic injection of acid chlorine-rich fluids and heat from the magma reservoir or magma intrusions at depth into the hydrothermal reservoirs [38,39,15,16].

Historical hydrothermal explosive eruptions at La Soufrière of Guadeloupe volcano
At la Soufrière of Guadeloupe, exegesis and re-analysis of historical chronicles and geological studies have shown [2,4] that of six historical non-magmatic explosive eruptions that have been documented, the major eruptions of 1797-1798, 1836-1837 and 1976-1977 all have produced laterally-directed explosions and emplacement of associated small volume highly mobile high-energy dilute turbulent pyroclastic density currents that reached probably up to ∼1.5-2 km from the volcano [2]. This information is summarized in Fig. S2 and references therein. Following explosive depressurisation of specific areas of the hydrothermal system through eruptive vents and fractures cross-cutting the dome (Fig. S2), exurgence of pressurised warm to hot acid hydrothermal fluid occurred repeatedly in all of those historical eruptions as well as during the minor 1956 eruption [2]. These observations, coupled with other surface manifestations of the hydrothermal system [2,15,16]: fumarollic degassing, thermo-mineral springs, areas of passive degassing, surficial exposures of hydrothermally altered rock ( Fig. S5) provide unequivocal evidence of the presence, within the La Soufrière of Guadeloupe volcanic edifice, of numerous possibly interconnected reservoirs of agressive acid hydrothermal fluid contained within otherwise non porous host-rock with low permeabilities. Data from detailed studies of acid hydrothermal systems worldwide (e.g. [39,40,41,42]) show that these fluids promote extensive rock dissolution and alteration. Magmatic or non-magmatic unrest associated with significant increases in heat and gas flux, shallow-depth seismicity, and deformation could trigger physico-chemical instabilities in these reservoirs or their interaction with meteoric recharge water that could promote transient overpressurization of the hydrothermally-altered core of the volcano (Fig. S5) leading to violent phreatic eruptions with associated laterally-directed explosions [43,8,9,2,11] and potentially threatening the mechanical stability of the edifice to trigger partial edifice collapse that can reach distances of a few kilometers from the dome and be associated with potential devastating mudflows. [44].
These observations from historical eruptions confirm the presence of large volumes of hydrothermal fluids at shallow depth within the edifice and provide ground truth validation of the different conductor regions inferred from three-dimensional imaging of the resistivity structure of the La Soufrière dome and upper hydrothermal system (Fig. S4).  [24,25,26,27,28,14,1,29,2,3,30,4,31,32,33,34,35]. This map was generated using the Esri ArcMapI 10.1 software (http:www.esri.com).    (Fig. 1b) that was triggered on 19 November 2009 by one of the most intense rainfall events recorded in 24 hours on La Soufrière in 60 years [20]. The scar is located just above anomaly D (Fig. 1b, 2b). Photo 21-12-2009, JB de Chabalier, IPGP/OVSG. b) Typical grey plastic clay-rich (see imprint of fingers, white arrow) mechanically weak texture of material found inside the dome about 100 m below the summit that constitutes the top of the hydrothermally altered core of the volcano and corresponds to regions of the dome with conductivity values between 0.01 and 0.1 S.m −1 (Fig. 2a, 3). This material resulted from prolonged and extensive acid dissolution and alteration of dome rock by rising hydrothermal and magmatic fluids typical of those found in the Cratère Sud acid pond in 1998 (Fig. S4) and at the bottom of the Tarissan pit (Fig. 3). c) The hydrothermally altered core of the edifice is seen at the base of the collapse detachment surface as a grey layer surmounted by moderately altered but fractured coherent dense dome rock that forms the carapace of the dome and that corresponds to the blue region in Fig. 2a, and 4b with conductivity values < 0.01 S.m −1 . d) Detail of the transition form the altered core to the moderately altered but fractured coherent dense dome rock that forms the carapace of the dome. This recent rockslide provides a valuable control point to derive mechanical rock properties (e.g. density, strength, permeability, porosity) for regions of the dome for which we have obtained and inverted resistivity data. Photos b-d: 15-12-2009, JC Komorowski, IPGP/OVSG. 7 3 Inverse and forward mesh Figure S6: Central part of the tetrahedral mesh used for the forward and inverse problems. The dark regions correspond to the mesh refinement around electrodes. To avoid boundary problems, the mesh extends 100,000 m to the sides and in depth (not shown). 8   5 Cross-validation of inverted data Figure S9: a) and b) Convergence curves and validating RMS of the 2 cross-validation tests done to estimate an appropriate data fit. Data were randomly divided in 2 sets and separately inverted. As the number of iteration increases, the validating data set RMS does not decrease accordingly to the inverted data. This indicates that the model is starting to overfit the data c) Relative RMS difference between the inverted data set and the one used for the cross-validation, for the 2 data sets inverted. The dashed red line indicates a relative difference of 5%. The last iteration to have a difference smaller than this value in both inversions is number 11 (Fig. S9c). The corresponding combined RMSs (considering both the inverted and validated data sets) are 8.0 and 7.6. We thus set a threshold of 7.8 in the inversion of the complete data set. d) Convergence curve corresponding to the inversion of the complete data set. Iteration number 16 (Fig. S9d) provides the first model with an RMS smaller than the threshold (red dashed line), and we therefore considered this model as the final result of the inversion.

Measurements and errors
6 Data sensitivity to electrical conductivity distribution . d-f) Mesh cells whose center is located below more than 400 m (d), 500 m (e) and 600 m (f) depth from the volcano summit were assigned a conductivity value of 0.001 S/m (the value used for the homogeneous starting model). The data misfits indicate that conductivity values larger than 1 S/m may not be well constrained. Similarly, changing the conductivity values at depths below 600 m from the summit does not significantly affect the RMS. Thus no interpretation should be done about the conductivity structures below 600 m from the summit since the model is not constrained by the data below these depths. Fig. 2 in the manuscript shows the conductivity model until 550 m below the summit. 12 Figure S11: Iso-conductivity surfaces of 0.1 S/m with computing mesh for spatial reference. a,c) View is from west. The depth limit corresponds to the sensitivity limit of the data (Fig. S10)