Causes of unrest at silicic calderas in the East African Rift: New constraints from InSAR and soil‐gas chemistry at Aluto volcano, Ethiopia

Restless silicic calderas present major geological hazards, and yet many also host significant untapped geothermal resources. In East Africa, this poses a major challenge, although the calderas are largely unmonitored their geothermal resources could provide substantial economic benefits to the region. Understanding what causes unrest at these volcanoes is vital for weighing up the opportunities against the potential risks. Here we bring together new field and remote sensing observations to evaluate causes of ground deformation at Aluto, a restless silicic volcano located in the Main Ethiopian Rift (MER). Interferometric Synthetic Aperture Radar (InSAR) data reveal the temporal and spatial characteristics of a ground deformation episode that took place between 2008 and 2010. Deformation time series reveal pulses of accelerating uplift that transition to gradual long‐term subsidence, and analytical models support inflation source depths of ∼5 km. Gases escaping along the major fault zone of Aluto show high CO2 flux, and a clear magmatic carbon signature (CO2‐δ13C of −4.2‰ to −4.5‰). This provides compelling evidence that the magmatic and hydrothermal reservoirs of the complex are physically connected. We suggest that a coupled magmatic‐hydrothermal system can explain the uplift‐subsidence signals. We hypothesize that magmatic fluid injection and/or intrusion in the cap of the magmatic reservoir drives edifice‐wide inflation while subsequent deflation is related to magmatic degassing and depressurization of the hydrothermal system. These new constraints on the plumbing of Aluto yield important insights into the behavior of rift volcanic systems and will be crucial for interpreting future patterns of unrest.


Introduction
Caldera complexes are characterized by regular unrest events (i.e., elevated seismicity, deformation and gas emissions) and infrequent large eruptions [Dvorak and Dzurisin, 1997;Biggs et al., 2014], making them some of the most dangerous and unpredictable volcanic systems on Earth [Acocella et al., 2015]. Understanding what causes unrest at these systems is vital for deciphering the behavior of the volcano during intereruptive periods and forecasting eruptions [Parks et al., 2012[Parks et al., , 2015. The fundamental challenge for volcanologists is to link the unrest observed at the surface to an unseen physical process taking place at depth. This requires knowledge of the location and geometry of the magmatic and hydrothermal reservoirs as well as their connections and interactions. Often a high-temporal resolution monitoring effort, combining geophysical and geochemical observations, is required to elucidate this [e.g., Chiodini et al., 2010Chiodini et al., , 2012Chiodini et al., , 2015a.
The East African Rift System (EARS) hosts a number of large (>10 km diameter) silicic caldera systems [Mohr et al., 1980;Acocella et al., 2003;Rampey et al., 2010Rampey et al., , 2014Robertson et al., 2015]. Unlike similar-sized volcanoes in developed nations, none of the EARS calderas are permanently monitored and in almost all cases knowledge of the frequency and magnitude of past eruptions is extremely limited . Without this information, volcanologists are reliant on regional satellite remote sensing surveys to detect caldera unrest. In the EARS, Biggs et al. [2009aBiggs et al. [ , 2011Biggs et al. [ , 2016 used Interferometric Synthetic Aperture Radar (InSAR) to  Siebert and Simkin, 2002]. Aluto volcano, the focus of this study, is shown in red, other deforming volcanoes [after Biggs et al., 2011] are shown in yellow, and those that have undergone historical eruptions are shown in pink [see Rampey et al., 2010 and references therein]. Note that ground deformation has not been detected at volcanoes that have undergone historical eruptions [Biggs et al., 2011]. Fault structures [modified after Agostini et al., 2011] are shown as black lines. The red shaded areas mark the extent of volcanic segments of the MER [cf. Ebinger and Casey, 2001]. Black arrows show the extension direction after Saria et al. [2014]. Rectangles indicate the coverage of the four SAR data sets used in this study. WSM: Envisat Wide Swath Mode and IM: Envisat Image Mode.
Geochemistry, Geophysics, Geosystems 10.1002/2016GC006395 of a restless caldera system, and reveal important coupling between magmatic and hydrothermal processes that could be widely applicable throughout the EARS.

Geological Setting
The 500 km long Main Ethiopian Rift (MER, Figure 1) accommodates extension between the Nubian and Somalian Plates and constitutes the northernmost segment of the EARS [Corti, 2009]. The MER is often considered to be the type example of a continental rift [Ebinger, 2005] and is traditionally separated into three Northern, Central, and Southern sectors [Corti, 2009]. These rift sectors developed asynchronously and display clear along-axis variations in fault architecture, magmatic processes, and modification of the crust [see Keir et al., 2015, and references therein]. Rift maturity increases northward along the MER toward Afar, where the overall physiology changes from continental rifting to incipient oceanic spreading [Beutel et al., 2010;Ebinger et al., 2010;Ferguson et al., 2013]. The MER is an ideal setting to develop models of continental rift evolution and it has been demonstrated that early stages of extension are accommodated by deformation on large boundary faults and that over time magma-assisted rifting becomes increasingly dominant with extension narrowing toward the rift axis [Ebinger, 2005;Corti, 2009]. The major transition in rifting dynamics, from boundary fault-dominated extension to localized axial volcanic segments (Figure 1), appears to take place in the Northern and Central MER (NMER and CMER) at around 3-1.6 Ma [Boccaletti et al., 1998;Le Turdu et al., 1999;Ebinger and Casey, 2001], while in the Southern MER active deformation remains largely fault-controlled [Hayward and Ebinger, 1996;Corti et al., 2013;Philippon et al., 2014].
Aluto is a silicic peralkaline volcano and is located in the CMER 100 km south of Addis Ababa. It is bounded and dissected by NNE-SSW-trending faults (Figure 2a), commonly referred to as Wonji Fault Belt [e.g., Acocella et al., 2003;Agostini et al., 2011;Hutchison et al., 2015]. The complex has been targeted for geothermal development and eight exploration wells were drilled during the 1980s (LA-1-LA-8; Figure 2a) with the deepest reaching 2500 m below the surface [Gianelli and Teklemariam, 1993;Gizaw, 1993;Teklemariam et al., 1996]. Of the eight wells drilled, only two (LA-3 and LA-6) are productive to date. Hutchison et al. [2015] identified two major structural features at the surface of the Aluto, a 500 m long fault, referred to as the Artu Jawe fault zone (AJFZ), which is aligned with local NNE-SSW-trending Wonji structures, and a 2500 m long remnant of a caldera rim on the east of the complex. Deep well correlations suggest that faulting preceded volcanic activity at Aluto [Hutchison et al., 2015] and that the silicic complex was initially built up as a low-relief shield before undergoing a period of caldera collapse. Significant postcaldera activity then took place at Aluto progressively infilling the caldera and the most recent eruptions, which include pumice fall, pyroclastic density current and lava flow units, have exploited structural weaknesses created by the pre-existing volcanic and tectonic structures (e.g., the AJFZ). Diffuse volcanic degassing also takes place at a number of sites across the volcano and it is evident that the pre-existing structures also dictate gas and hydrothermal fluid ascent to the surface. Aluto has undergone multiple uplift and subsidence events since 2003 [Biggs et al., 2011]. The cause of these unrest events is the focus of this study (section 7) and was also the topic of a recent magnetotelluric survey by Samrock et al. [2015] at Aluto.

InSAR Data and Processing
InSAR is a geodetic technique that measures phase difference between two Synthetic Aperture Radar acquisitions [e.g., Simons and Rosen, 2007, and references therein]. The technique has been successfully applied in many volcanic settings, including rift zones [e.g., Iceland: Sigmundsson et al., 2010and East Africa: Wright et al., 2006Biggs et al., 2009aBiggs et al., , 2011Biggs et al., , 2016Wauthier et al., 2012Wauthier et al., , 2013. Here we use four sets of SAR data acquired by Envisat and ALOS satellites from 2002 to 2012 (summarized in Table 1, and in Figure 1 where swath coverage is shown) and have generated 100 interferograms in total.
Interferograms for Envisat Image Mode (IM) and ALOS data were generated using the Repeat Orbit Processing software (ROI_PAC) [Rosen et al., 2004]. For the Envisat Wide Swath Mode (WSM) data we processed the interferograms using Gamma software [Wegm€ uller and Werner, 1997]. For all data sets, we removed the topographic contribution using an identical 90 m SRTM DEM. The Envisat WSM data has the lowest spatial resolution of SAR data sets included in this study (150 m) and so a 90 m DEM is sufficient to ensure that this data is not downsampled. It is also important to note that although a 2 m lidar DEM is available for Aluto Geochemistry, Geophysics, Geosystems 10.1002/2016GC006395 [Hutchison et al., 2015] it does not cover the non-deforming areas away from the main edifice and therefore cannot be used to generate reliable displacement time series (section 3.2). Linear ramps were found in the phase data for several interferograms and were likely caused by inaccuracies in the satellite orbital positions. In these cases, orbital ramps were sampled, modeled (fitting a linear or quadratic surface), and then subtracted from the interferogram [e.g., Ebmeier et al., 2010]. Topographically correlated atmospheric delay errors (i.e., water vapor effects) were very minor because the relief over the Aluto complex is low (<700 m).

Displacement Time Series and Components of Ground Motion
InSAR measures displacement along a single line of sight (LOS). For each track, we constructed a LOS deformation time series by using a linear least squares inversion of the displacements for each interferogram to Geochemistry, Geophysics, Geosystems 10.1002/2016GC006395 find incremental displacements between the acquisition dates [e.g., Lundgren et al., 2001;Berardino et al., 2002;Biggs et al., 2010]. Each displacement time series (Figures 2-5) is referenced to a pixel (asterisk, Figure 2a) located 5 km north-west of the main edifice of Aluto. This reference locality was coherent throughout the survey period for all data sets and is assumed not to be deforming (there is no evidence of recent fault scarps within the vicinity of this site). We found that referencing the time series to coherent regions east of the main edifice produced very similar time series results to those shown in Figures 2-5. Uncertainties in the time series were estimated using a Monte Carlo approach, where randomly generated noise of mean amplitude 1 cm was added to every pixel in each interferogram before inversion, neglecting the effects of spatial correlation [Ebmeier et al., 2013a]. Error bars in LOS displacement time series are variable and reflect how well each acquisition date is linked to the network of interferograms [e.g., Ebmeier et al., 2013b].
The interferograms processed for each orbital track were acquired from different satellite viewing geometries (Table 1, note differences in heading and look angles). Using LOS displacement measurements made from different satellite look directions, it is possible to combine the images and resolve the horizontal (Ux) and vertical (Uz) components of motion [e.g., Wright et al., 2004;Biggs et al., 2009b]. To achieve this, we identified interferogram pairs with different look directions (descending Envisat IM and ascending ALOS, Table 2) that cover a similar time period (overlapping to within 1 month) and hence identical deformation signal. We selected separate interferogram pairs for the uplift and subsidence periods and use these to evaluate whether analytical source models fit the deformation profiles (section 4.2).
It is important to recognize that as we only have two interferograms (from ascending and descending orbits) we cannot resolve the volcano's true 3-D deformation field. We assume that north-south motion was negligible and effectively resolve the deformation into a 2-D plane with axes running vertically and eastwest [Biggs et al., 2009b]. This assumption is valid because ascending and descending satellites look from close to due east and west, respectively, and will only capture a very minor component of north-south motion [Ebmeier et al., 2010]. Further, when comparing analytical source models (outlined below) to the components of motion (section 4.2) we only compare east-west profiles through the deformation center, thus minimizing any north-south motion.
At present there are no additional constraints on the radius or shape of the deep reservoirs (e.g., from seismology or petrology) and so we only use simple analytical models (point source and pennyshaped crack geometries) [Mogi, 1958;Fialko et al., 2001] to generate first-order constraints on the depth and geometry of the deformation sources. These models assume that crust is linearly elastic (i.e., they do not take into account crustal heterogeneities or the role of pore fluids), and in the case of the Mogi point source it is assumed that source depth is significantly larger than radius. Although these models are an oversimplification of the real earth they are an important first step toward understanding sources of volcanic deformation [Segall, 2010] and are appropriate when distinguishing between shallow-and deep-seated deformation mechanisms [e.g., Mann and Freymueller, 2003]. Note: The majority of our data was acquired by the European Space Agency's (ESA) Envisat satellite. In this study, we use Envisat data collected in Image Mode (IM) and Wide Swath Mode (WSM). IM is Envisat's conventional operation mode. The Scanning Synthetic Aperture Radar (ScanSAR) technique of the Envisat ASAR instrument, hereafter referred to as the WSM mode, permits radar observations with a much larger swath (400 km) but with a reduced spatial resolution compared to IM. WSM covers the extended area by using five different overlapping antenna beams (subswaths) [Moore et al., 1981]. We also processed Phased Array type L-band Synthetic Aperture Radar (PALSAR) data from the Japanese Aerospace Exploration Agency (JAXA) ALOS satellite. a Data covering the Ethiopian Rift extends until December 2008.

Joint Inversion of Geodetic Observations
Our study has focused on generating a dense coverage of observations between 2007 and 2011 as this covers an episode of uplift at Aluto [Biggs et al., 2011] when both ALOS and Envisat satellites were operational and acquiring data (Table 1). If we assume a fixed source geometry then we can use a joint inversion technique [e.g., Biggs et al., 2010;Parks et al., 2015] to combine the InSAR data acquired from different satellites and tracks, to produce a time series of subsurface volume change. The advantage of this approach is that it allows us to combine independent deformation measurements (made at different times and from distinct viewing geometries) and thus greatly improve the temporal resolution of the data set .
Following Biggs et al. [2010], we make two main assumptions: (1) that all deformation is related to a single fixed source and (2) this source approximates a point pressure variation at depth within an elastic crust (i.e., a Mogi source, Mogi [1958] is used to convert our displacement time series to volume change). Previously, Biggs et al. [2011] used penny-shaped crack geometries [Fialko et al., 2001] to model deformation at Aluto. However, as we shall show in section 4.2, penny models cannot reproduce the large horizontal components of deformation and in addition the shallow depth (<2 km) and large radius (5-10 km) required for these models are well beyond the known limits of the geothermal field (section 6). Mogi models provide a To find the best-fitting source location and depth, we carried out a three-dimensional grid search by minimizing the misfit between the observations and the Mogi model. In the joint inversion, we assume that the source location is fixed through time and found our best-fitting sources converged at a depth interval of 4.8-5.4 km depth (section 4.3). In reality, the plumbing system of Aluto is likely to be significantly more complex than can be represented by a single Mogi source, and although our modeling in section 4.2 suggests that a slightly shallower Mogi source at 3.5 km provides a notionally better fit for the subsidence, for simplicity and because the majority of our data are from the uplift period, we model all volume change with a single 5 km deep source that can reproduce the bulk of the deformation across this unrest period.
We estimate the incremental volume change of the Mogi source, using equations (7) and (8) of Biggs et al. [2010] and then integrate these to provide a time series of volume change. We solved for several different sources of noise or nuisance parameters: those that were temporally and spatially correlated and those that were uncorrelated. More detailed information regarding both the inversion technique and error estimation may be found in Biggs et al. [2010].

Gas Chemistry and CO 2 -d 13 C Analysis
Diffuse degassing through soil and low temperature fumarole vents (<1008C) is the main outlet for volcanic gases on Aluto [Hutchison et al., 2015]. Between January 2012 and February 2014, we made 800 soil-gas CO 2 flux measurements to capture the large scale structural controls on degassing and resolve smaller scale  Hutchison et al. [2015]). In February 2014, we also collected 100 soil-gas samples from nine sites along AJFZ while conducting diffuse CO 2 degassing surveys. We adopted the method of Chiodini et al. [2008] to sample gas for laboratory analysis. At each measurement station, we set up an accumulation chamber [Chiodini et al., 1998] and inserted a T-connector with a pierceable septum in the flowline just after the infrared gas analyser. During the CO 2 flux measurements, we pierced the septum with a syringe and extracted 10 ml of gas. The syringe had a built-in shut off valve and so the sample gas was sealed, and then subsequently injected into an evacuated vial through a pierceable butyl rubber septum. Two gas samples were extracted during each measurement. The first was taken after a few seconds allowing homogenization of gases within the chamber; the second sample was collected 40 s later, at higher CO 2 concentration. The first sample (0 s) allows local anthropogenic air pollution effects to be characterized [e.g., Chiodini et al., 2008] (this was minimal at Aluto), and more importantly, it can be used to distinguish contributions from other sources (e.g., biogenic) before volcanic gases inundate the chamber.
Bulk gas compositions were measured for seven samples (Table 3) at the Department of Earth and Planetary Sciences at University of New Mexico (UNM). A combination of gas chromatography (GC) and quadrupole mass spectrometry (QMS) were used to measure the composition of gases. CH 4 , CO 2 , H 2 , and CO concentrations were measured using GC, while Ar, He, N 2 , and O 2 were determined using QMS. An analytical uncertainty of <0.1% was reported for gas analyses by the QMS system at UNM by de Moor et al. [2013a]. Further, Lee et al. [2016] used the same GC and QMS system for bulk gas analyses of vials collected from an accumulation chamber and estimated the analytical uncertainty of the GC measurements to be 62% based on repeat measurements. It is important to recognize that the pierceable glass vials are not designed for storage of H 2 and He gas species and although these were analyzed at UNM we believe that these species were likely to have been lost either by diffusion through glass or through the rubber septum. Therefore, the values presented for H 2 and He are minima.
Carbon isotope measurements of CO 2 (CO 2 -d 13 C) were made for the majority of samples at the Geochemistry Laboratory of INGV-Osservatorio Vesuviano (Naples). Gases were analyzed within a few days of sampling, thus minimizing CO 2 gas loss and isotopic fractionation through the septum [Tu et al., 2001]. The samples were analyzed using a continuous flow isotope ratio mass spectrometer (Thermo-Finnigan Delta XP) interfaced with a Gasbench II device equipped with autosampler. CO 2 -d 13 C measurements were also made for seven samples at the Center for Stable Isotopes, UNM (the same samples that were also analyzed for bulk gas chemistry, Table 3). At UNM CO 2 -d 13 C were measured by Isotope Ratio Mass Spectrometer (Finnigan Delta XL) with a gas bench and auto-sampler. Results for all CO 2 -d 13 C measurements are shown in delta notation as per mil values (d&) relative to Pee Dee belemnite (PDB) using an internal standard and are characterized by a d 13 C standard error of 60.1& (supporting information Table S1).
It was not possible to make a direct comparison of an identical soil-gas sample measured for CO 2 -d 13 C between the different laboratories at UNM and Naples. However, at site 06A (supporting information Table  S1) where the natural CO 2 -d 13 C variation was well constrained by samples analyzed in Naples (i.e., all nine gas samples were between 24.2& and 24.5&) we found that the corresponding measurements made at UNM were 24.2& and 24.3& (Table 3). The overlapping values suggest that any systematic error in isotopic measurements between the different laboratories were within the natural variability of the soil-gas sampling sites on Aluto (section 5).  (Table 1). In Figure 2a, the main structural features on and around the Aluto volcanic complex are shown, they include NNE-SSW-trending Wonji faults, an elliptical caldera rim and ring fault that is hypothesized to underlie the Aluto complex [Hutchison et al., 2015]. The extent of volcanic deposits from the complex is also highlighted by the irregular dotted outline.

Ground Deformation at
The Envisat IM data provide the longest displacement time series and suggest there have been two major uplift events at Aluto between 2004 and 2011 of magnitude 115 cm (2004) and 110 cm (2008); in both cases, these were followed by a period of slow subsidence (25 cm) which took place over several years ( Figure 2b) [Biggs et al., 2011]. The available ALOS and Envisat WSM data (Table 1)  ) and further evidence that the long-term deformation pattern of Aluto is characterized by short uplift events (<1 year) followed by more gradual subsidence (over several years). Envisat WSM data provide the highest temporal resolution snapshot and suggest that there are two phases to the uplift pulse (marked U1 and U2, in Figures 4a and 5a). The initial phase (U1) has a displacement of 12 2 3 cm over a 3 month period between April and July 2008, the rate then increases between July and October (U2) where 15 cm of deformation occurs over 3 months. We suggest that the WSM interferograms indicate a two-step accelerating deformation trend.
The Envisat IM and WSM interferograms all show an elliptical deformation signal with east-west to ESE-WNW elongation. The center of the deformation (for both uplift-subsidence events, Figure 2) is remarkably consistent throughout the observation period and, based on the Envisat look direction, is located a few kilometers east of the caldera center. The deformation signals for the ALOS data appear more circular ( Figure  3c) and the subsidence pattern shows only slight elongation east-west ( Figure 3d). Comparing the deformation signals on Aluto to the major volcanic and tectonic features, we note several important features.

Components of Motion and Deformation Source Models
Previously, Biggs et al. [2011] used Envisat IM data (section 4.1) to evaluate best fitting point source [Mogi, 1958] and penny-shaped crack [Fialko et al., 2001] geometries for Aluto. While their best-fitting model was a shallow (<2.5 km) penny-shaped crack with large radius (3-10 km), it is important to recognize that a point source at greater depths can produce a similar ground deformation to a penny-shaped crack [Fialko et al., 2001].
Here we exploit the different look directions of the interferograms (i.e., Envisat and ALOS) to decompose the LOS motion into horizontal and vertical components (section 3.2). The vertical (Uz) and horizontal (Ux) components of deformation are shown in Figure 6 for both the uplift and subsidence phases of deformation. East-west profile lines taken through the center of the deformation pulse (and hence minimizing the north-south component of motion, section 3.2) are also shown in Figure 6. Profiles reveal that the 2008 uplift Uz max is 12 6 1 cm and Ux max is 6 6 2 cm, while for the subsidence between end-2008 and mid-2010 Uz min is 3.5 6 1.5 cm while Ux max is 2 6 2 cm. Note that for both the uplift and subsidence, the ratio of maximum horizontal to maximum vertical displacement is 0.5, much greater than a penny geometry which is normally less than 0.3, and closer to a spherical (Mogi) source which is about 0.4 [Fialko et al., 2001;Yun et al., 2006].
To further distinguish between the two end members models (i.e., the shallow penny-shaped crack and the deeper Mogi point source), we carry out a nonlinear inversion procedure to find the optimum model parameters that match the components of motion observed along the profile lines ( Figure 6). To evaluate the quality of each fit, we calculate the root mean square error (RMSE) between the model and data along the profile line (shown in Table 2).
Mogi models ( Figure 6) at a depth of 5 km provide a reasonable fit to both the Ux and Uz components for the uplift and subsidence (with misfit of 1.12 and 0.65 cm, respectively, Table 2). When we attempted to minimize the misfit for both the Ux and Uz components for the penny model the inversion procedure did not provide a unique solution. In other words, the large ratio of horizontal to vertical components that we observe in both the uplift and subsidence phases could not be reproduced by a penny source model. Instead we have only inverted the penny model for the Uz component of our data (Figure 6). While the penny model can reproduce the Uz component, it is evident that it greatly underestimates the Ux components (misfits of 2.29 and 0.83 cm, for uplift and subsidence respectively, Table 2). On this basis, we consider it unlikely that a shallow penny source is the main driver for deformation events on Aluto.
Although our analysis of deformation components primarily seeks to discriminate between shallow (penny) and deep (Mogi) model geometries, it is interesting to note that a Mogi model at 3.4 km provides a better fit for the subsidence than a source at 5.1 km as was identified for the uplift phase (misfits are 0.54 and 0.65 cm, respectively, Table 2). Our InSAR dataset has limited temporal coverage of the subsidence phase  Figure 10 and are linked to the Sample ID. CH 4 , CO 2 , H 2 , and CO concentrations were measured using gas chromatography (GC) techniques. Ar, He, N 2 , and O 2 were measured via quadrupole mass spectrometry (QMS). Analytical uncertainty for the GC measurements is at 65%, and QMS analyses at <0.1% (concentration), see section 3.3. CO 2 -d 13 C measurements were also made by Isotope Ratio Mass Spectrometer for seven samples at the Center for Stable Isotopes, UNM. Results for CO 2 -d 13 C are shown in delta notation as per mil values (d&) relative to PDB (Pee Dee belemnite) and are characterized by a d 13 C standard error of 60.1&. Note that H 2 and He concentrations represent minima. b.d.: below detection limit.
Geochemistry, Geophysics, Geosystems 10.1002/2016GC006395 (Figures 2 and 3) and does not allow us to thoroughly investigate whether or not the deformation source remains stable during the deflation period (e.g., evolving to shallower depths with time). A more complete understanding of the temporal evolution of the deformation sources at Aluto will be revealed by recently deployed continuous GPS networks and new radar satellites (e.g., Sentinel-1) that have significantly improved repeat times.

Joint Inversion of Deformation Data
We used the interferograms covering the uplift-subsidence period between 2008 and 2010 to estimate the volume changes that could explain the observed deformation trend (Figure 7). The best fitting spherical source location and optimum depth was found by carrying out a grid search (Figure 7a) and finding the minimum weighted residual for each inversion (Figures 7b and 7c). The best source is shown to be at a depth of 5.1 km broadly consistent with the models in Figure 6. The total volume increase for the uplift period is 113 6 0.5 3 10 6 m 3 , followed by 25 6 1 3 10 6 m 3 of volume decrease between the end of 2009 and 2010 (subsidence period).

10.1002/2016GC006395
Combining the individual interferograms from the different tracks demonstrates that the rate of volume change accelerated through 2008 (shown as the step between U1 and U2 in Figure 7d). This is consistent with the two-step uplift pattern which we identified in the LOS displacement time series from WSM data (Figures 4a and 5a). At the point of maximum source volume change between November 2008 and January 2009, the time series (Figure 7d) shows a distinct ''roll over'' into negative volume change (i.e., deflation). Finally, although there is a considerable gap in measurements during the presumed slow subsidence (deflation) period through most of 2009 and early-2010, the volume time series (Figure 7d) indicates that a significant component of the deflation (50%, equivalent to 22 3 10 6 m 3 ) took place in the 3 months immediately following peak uplift.

Degassing Patterns and Compositional Overview
The large scale pattern of gas emissions on Aluto is shown in Figure 8 where CO 2 flux measurements, made in January to February and November 2012, are overlain on a lidar DEM. As discussed by Hutchison et al. [2015], fumarole vents, hydrothermal alteration, and sites of anomalously high CO 2 gas fluxes (i.e.,   Hutchison et al. [2015] and were mapped using remote sensing imagery or were suggested by previous geological reports and deep well data [Kebede et al., 1985;ELC Electroconsult, 1986;Gizaw, 1993;Gianelli and Teklemariam, 1993]. The dashed ellipses show the size and orientation of a buried ring fault that may explain anomalous degassing zones around the rim of the edifice (see Hutchison et al. [2015] for detailed explanation). The fine dashed ellipses represent 6500 m uncertainty bounds on the central ellipse. The mapped caldera rim is shown in blue and overlaps the proposed ring structure. Fumaroles vents and zones of surface alteration were mapped using a combination of field and remote sensing observations [Kebede et al., 1985;Hutchison et al., 2015]. The black box inset links to Figure 10a, where a dense grid of soil-CO 2 flux measurements was undertaken and all gas samples (section 5) were collected.

10.1002/2016GC006395
The chemical composition of the dry (water-free) gas-phase in the soil-gas samples collected from the AJFZ in February 2014 are reported in Table 3, and summarized in the CO 2 -O 2 -N 2 triangular diagram in Figure 9. The gas sampling sites are also shown as colored symbols in Figure 10a where they are overlain on aerial imagery and gridded soil-CO 2 flux (measured at the time) [Hutchison et al., 2015].
The soil gas samples are dominated by N 2 (69-80%) and O 2 (16-21%) and represent air contaminated by a volcanic component rich in CO 2 (1-10%, Table  3). This is corroborated by the triangular diagram in Figure 9 which shows that the Aluto soil-gas samples fall on a mixing line between air and a pure-CO 2 end member. CH 4 was above detection only in the samples with the highest CO 2 concentrations (Table 3) while CO was between 33 and 65 ppm. CH 4 (when above detection limits) was significantly greater than atmospheric Figure 9. CO 2 -O 2 -N 2 triangular plot for soil-gas samples collected in this study (see Table 3) at Aluto volcano (circles and diamonds), as well as gas phase samples collected at Dallol volcano, Afar, Ethiopia (squares)  and throughout the Kenyan Rift (triangles) [Darling et al., 1995]. The Aluto samples fall on a mixing line between air and pure CO 2 , although are evidently dominated by air. Orange circles with blue outlines identify the location of the productive geothermal wells (LA-3 and LA-6). The sampling locations are overlain on an aerial photo and a CO 2 -flux map (blue to red colors, described by Hutchison et al. [2015]). Small gray points represent a discrete CO 2 flux measurement. The black line indicates the Artu Jawe fault scarp (black) and its projected continuation north (dashed). Lesspronounced breaks in topography (10-20 m high) west of the main fault scarp are marked by white dashed lines. pp: the Aluto-Langano geothermal power plant. The range of soil CO 2 flux and CO 2 -d 13 C (for the 40 second samples, section 3.3) for each site are shown in the white boxes around the grid. (b) Carbon isotopic composition (d 13 C) of soil CO 2 efflux plotted against the reciprocal of CO 2 concentration in the gas samples (symbols are linked to map view in Figure 10a). This plot shows all samples that were collected and analyzed for CO 2d 13 C (both the 0 and 40 s samples, section 3.3). The data define a triangular array that can be defined by mixing between between air, an end-member characterized by high CO 2 concentrations and light d 13 C values approaching 225& (i.e., biogenic CO 2 ) and an end-member characterized by high CO 2 concentrations and heavier d 13 C values (i.e., magmatic CO 2 ). (c) Inset of Figure 10b showing CO 2 -d 13 C for the highest concentration samples. CO 2 -d 13 C of gases from geothermal wells on Aluto are shown after Darling [1998]. Note that deep well samples contain 98% CO 2 (dry gas phase). The full CO 2 -d 13 C data set is provided in the supporting information file.

10.1002/2016GC006395
concentrations and it is likely that this species is produced by reducing reactions in the hydrothermal reservoir [e.g., Agusto et al., 2013;Tassi et al., 2013].
In Figure 9, we also show gas-phase analyses made at fumaroles, geothermal wells, springs, and boreholes, at active geothermal areas along the Kenyan Rift [after Darling et al., 1995] and at Dallol volcano located in Afar, Ethiopia [after Darrah et al., 2013]. Although high temperature magmatic gas analyses have been analyzed at the basaltic Erta Ale volcano [de Moor et al., 2013b], there are very few low temperature gas-phase analyses available for other geothermal sites in Ethiopia that we could directly compare to Aluto. The regional comparison shown in Figure 9 does, however, illustrate that geothermal gases sampled from across the EARS can be broadly defined by mixing between air, air saturated water, and a CO 2 -rich end-member that is of magmatic/mantle origin [e.g., Darling et al., 1995;Barry et al., 2013;Darrah et al., 2013;de Moor et al., 2013a].
The samples with highest CO 2 concentrations (sample site 06-A, Figure 10a and Table 3) show a narrow range of CO 2 -d 13 C values from 24.2& to 24.5&, strongly supporting a magmatic origin (Figure 10c). The key inference that the CO 2 -d 13 C data allows us to make is that the magmatic and geothermal reservoirs of Aluto are physically connected. Volatiles released by the magma reservoir degas into a deep geothermal reservoir at >2 km (section 6) [Gizaw, 1993;Teklemariam et al., 1996]. While the exact interactions of the magmatic CO 2 gas with the fluids in the geothermal reservoir are likely to vary (i.e., it may remain as a free gas, dissolve in the fluid and/or precipitate out through mineralization processes) [Chiodini et al., 2015b], ultimately the CO 2 -d 13 C of the gas and boiling fluid efflux that ascends along the AJFZ fingerprints this deep magmatic origin. There are several subtle features to note regarding the CO 2 -d 13 C data set and these are explored in the following sections.

Spatial Variations in CO 2 -d 13 C
There are clear spatial variations in CO 2 -d 13 C at different sample sites along the fault zone (Figure 10c). For example, fumarole vents at sites 06-A and 07-A at the northern end of the grid show distinct isotopic differences from fumaroles at sites 01-A, B, and C at the southern end; the latter show values that are up to 2& heavier (Figures 10a and 10c). While variations in gas flux along the AJFZ have previously been linked to the presence of deep structures and changes in near-surface permeability [Hutchison et al., 2015], the distinct CO 2 -d 13 C at different sites also suggests that there is limited mixing and homogenization of the gases along the fault zone.

Temporal Variations in Gas Flux and CO 2 -d 13 C
Short-timescale variations are also seen in soil-gas flux and CO 2 -d 13 C at the measurement sites. For example, sites 01-A, B, and C were measured over the course of 1 h and were only separated by a few tens of meters. These sites show clear differences in both the absolute magnitude of soil-gas flux (45, 7, and 5 3 10 3 g m 22 d 21 , respectively) and also the flux variation between measurements (30%, 50%, and 5%, respectively, Figure 8a). There is no correlation between soil-gas flux and CO 2 -d 13 C at these three measurements sites: site 01-C, for example, shows a near constant soil-gas flux but 2& range in CO 2 -d 13 C (Figure 10a). These observations clearly reveal that gas emissions at the surface of the fault zone are not stable through time, and that both gas flux and composition can vary over the course of a few minutes.

High CO 2 -d 13 C Values
High CO 2 concentration samples (Figure 10c) collected from sites 01-B and C show a cluster of CO 2 -d 13 C values between 22.5& and 23.0&, above the range normally associated with magmatic gases (23& to 28&). There are a number of explanations as to why these values are slightly heavier than typical magmatic values: Geochemistry, Geophysics, Geosystems 10.1002/2016GC006395 I. Calcite precipitation: Loss of CO 2 due to calcite precipitation has the potential to fractionate CO 2 -d 13 C.
This process is temperature-dependent, such that at temperatures <1928C, calcite is enriched in 13 C relative to residual dissolved CO 2 in geothermal fluids, whereas at higher temperatures >1928C calcite becomes depleted in 13 C relative to residual CO 2 [Bottinga, 1969]. To explain the observations on Aluto would require high-temperature (>1928C) precipitation of calcite. II. Phase partitioning and hydrothermal degassing: Vapor partitioning can be caused by either boiling (>1008C) and/or hydrothermal degassing due to supersaturation of a particular gas species. The loss of CO 2 from geothermal fluids via these processes will cause isotopic fractionation of CO 2 -d 13 C [e.g., Vogel et al., 1970;Mook et al., 1974]. These processes are temperature dependent and at temperatures <1108C dissolved carbon species will be enriched in 13 C, leaving 13 C relatively depleted in the residual gas phases [Szaran, 1997]. At higher temperatures (>1108C), the isotopic fractionation is in the opposite sense and residual gases will consequently be enriched in 13 C. This latter case would again have to be true to explain the observed values of 22.5& at Aluto (Figure 10c). III. Crustal contamination: CO 2 gas could be released by decarbonation of limestone in the subvolcanic system, induced either by heating, or reaction with magma. These gases would yield a CO 2 -d 13 C of 12-6& [Lollar et al., 1997;Tassi et al., 2010;Darrah et al., 2013;Parks et al., 2013], and when mixed with magmatic gases could explain the values of 22.5& observed.
To thoroughly investigate these processes further would require d 13 C analyses of the geothermal fluids, and also additional tracers, such as He [cf. Barry et al., 2013Barry et al., , 2014. However, until this data are available we can make several arguments. First, deep well data from LA-3 and LA-6 ( Figure 10a) confirm that temperatures exceed 3008C beneath the surface of the AJFZ and that boiling takes place from at least 2100 m up to 700 m depth [Gizaw, 1993]. Calcite is also one of the main hydrothermal mineral precipitates at Aluto and has been identified in deep wells cores collected along this fault zone ]. These constraints are consistent with both scenarios (i) and (ii), and favor carbon isotope fractionation by hightemperature calcite precipitation, boiling, and degassing within the geothermal system.
We can investigate the plausibility of crustal contamination (scenario iii) using a simple mass balance calculation. Assuming that the heavy CO 2 -d 13 C values are entirely the result of crustal contamination, taking the magmatic end-member as 24.25& and assuming a carbonate contaminant of 14& (an average of the typical range noted above), it would require 30% carbonate to 70% magmatic input to explain the heaviest gases (22.5&, Figure 10c). There is, however, no physical evidence from the petrography or geochemistry [Di Paola, 1972;Weaver et al., 1972] of the Aluto rock samples to suggest that carbonate materials are incorporated into the melt. In addition, lacustrine sediment layers in the subsurface are thin (50-100 m, Figure 11) and do not contain significant carbonate horizons [Gianelli and Teklemariam, 1993;Teklemariam et al., 1996]. Decarbonation might link to Jurassic-Palaeogene sedimentary rocks that make up basement lithologies in the MER [e.g., Cornwell et al., 2010], however, their thickness and composition beneath Aluto are completely unknown.
Our preferred explanation is that geothermal processes (scenarios i and ii) cause some CO 2 -d 13 C samples to be slightly heavier than the magmatic range; these scenarios can explain the elevated values without invoking contamination by unknown carbonate material. The AJFZ provides an important pathway for fluid upflow and migration from the geothermal reservoir to the surface (section 6). It is likely that upflow rates, as well as boiling, degassing, and mineral precipitation vary along the fault zone, and that these produce the complex spatial (section 5.2) and short-timescale (section 5.3) variations in both gas flux and CO 2 -d 13 C.

Constraints on the Magmatic-Hydrothermal System
We now use constraints on the subsurface structure and geothermal field of Aluto to develop a conceptual model of the magmatic and hydrothermal systems ( Figure 11). The original interpretations of Aluto's magmatic-hydrothermal system were largely based on deep well logs, cores and cuttings [e.g., Gizaw, 1993;Gianelli and Teklemariam, 1993;Teklemariam et al., 1996]. In 2012, magnetotelluric surveys were carried out by Samrock et al. [2015] and their new resistivity model for Aluto has helped clarify many of the previous interpretations (outlined in detail below). Samrock et al. [2015] found no evidence for any deep (>5 km) low resistivity bodies that would be conventionally interpreted as an active magmatic system (i.e., a zone of partial melt) beneath Aluto. This finding appears to run in contrast to geological and geochemical observations Geochemistry, Geophysics, Geosystems 10.1002/2016GC006395 from Aluto, e.g., the CO 2 -d 13 C results, high CO 2 flux measurements (section 5), and young volcanism (<10 ka) [Gianelli and Teklemariam, 1993] which are all consistent with an active magmatic source that contributes significant volumes of volcanic gases and supplies melt for recent eruptions. An explanation that satisfies these apparently conflicting observations is that Aluto's magmatic system is represented by a locked crystal-rich mush rather than extensive zone of partial melt (as was imaged in the Dabbahu magmatic segment) [Desissa et al., 2013]. It is important to recognize that there are have been few geophysical experiments to characterize peralkaline magmatic systems like Aluto and as a result our understanding of the relative volumes of partial melt, magmatic fluid and crystals in these systems, their interconnectivity and hence their detection thresholds is currently limited. We envisage a crystal-melt mush zone at 5-10 km to be the most plausible configuration of the magmatic system at Aluto; as has been suggested at other peralkaline systems in the EARS [e.g., Macdonald et al., 2008]. Volatiles are transferred from the magmatic reservoir through fractures to the geothermal fluids, which then migrate along fault conduits to the surface where they boil and degas.
d 18 O measurements of waters extracted from the deep wells on Aluto reveal that the fluid contained within the geothermal reservoir is largely meteoric in origin (>90%); derived from rainfall on the rift margin [Darling et al., 1996;Rango et al., 2010]. Despite their proximity, the shallow lakes of Ziway, Langano, and Abijata supply minimal water (<10%) to the reservoir [Darling et al., 1996]. Groundwater beneath Aluto flows toward the south, consistent with the hydraulic gradient, and hot spring discharges are focussed along the shore of Lake Langano [Kebede et al., 1985;Darling et al., 1996;Rango et al., 2010;Hutchison et al., 2015]. Figure 11. Schematic west-east cross section of the Aluto geothermal system. Deep stratigraphy and hypothesized subsurface structures on Aluto are summarized after Hutchison et al. [2015]. The deep stratigraphy summarizes results from several publications [Gizaw, 1993;Gianelli and Teklemariam, 1993;Teklemariam et al., 1996] and drilling reports provided by the Geological Survey of Ethiopia [Yimer, 1984;Mamo, 1985;ELC Electroconsult, 1986;Teklemariam, 1996]. Red ''F'' labels indicate major fumarole and degassing regions mapped at the surface of the volcano [after Hutchison et al., 2015]. The unit colors from light to dark gray provide a simple qualitative interpretation of permeability of the geological unit [after Gizaw, 1993]; the least permeable units are shown in darkest gray. Regions of high and low resistivity beneath Aluto were identified by magnetotelluric surveys and are identified in italics (for full details on absolute values the reader should refer to Samrock et al. [2015]). Our main interpretation is that the magmatic and hydrothermal reservoirs of Aluto are connected, and that magmatic volatiles are channelled to the surface by faults and fracture pathways.

10.1002/2016GC006395
The subsurface structure of Aluto and lithological relationships ( Figure 11) is constrained by data from the deep exploration wells [Gizaw, 1993;Gianelli and Teklemariam, 1993;Teklemariam et al., 1996]. From well temperature and pressure measurements, these authors infer that the geothermal reservoir is >2000 m beneath the surface, and this is supported by magnetotelluric results [Samrock et al., 2015] that confirm the presence of a high resistivity body at these depths. The main aquifer is likely to be a sequence of ignimbrites previously referred to as the ''Tertiary'' ignimbrites (now called Neogene ignimbrites, Figure 11). The reservoir fluids sampled from the deep wells are of alkali-chloride-bicarbonate type, with near-neutral pH, and display geochemical evidence for interaction with rhyolitic volcanic products [Gianelli and Teklemariam, 1993], consistent with the Neogene ignimbrite deposit (Figure 11) being the reservoir for the geothermal field.
Above the Neogene ignimbrites, a thick sequence of basalt lavas (Bofa Basalts) 500-1000 m thick are identified . The basalts are pervasively altered and sealed by deposition of hydrothermal alteration minerals [e.g., Gianelli and Teklemariam, 1993;Teklemariam et al., 1996]; as a result, they exhibit poor permeability [Gizaw, 1993]. Overlying the basalts, a layer of interbedded sediments and volcanic tuffs are encountered, these units are also highly altered and sealed by deposition of clay minerals . This clay rich zone appears to correlate with a low-resistivity zone identified by Samrock et al.
[2015] at depths of 500-1500 m ( Figure 11). It is unclear whether the cap rock for the geothermal reservoir is made up by Bofa Basalt or the altered tuffs and sedimentary layers above. We consider that together the Bofa basalts and altered sediments-tuffs provide a capping lithology for the system (Figure 11, between 500 and 1500 m), although it is possible that the basalt unit serves as a reservoir along the contacts and where it is fractured and brecciated. Above the cap layers, peralkaline rhyolite volcanics erupted from the Aluto volcanic complex are encountered and comprise alternating layers of rhyolite lavas and volcaniclastic deposits. The ignimbrite at the base of these units (Figure 11) is permeable and appears to provide a cool shallow aquifer that has been identified in a number of the deep wells at <700 m [Gizaw, 1993].
The main feature of the Aluto geothermal field is that it is characterized by an upflow zone that coincides with the major NNE-trending AJFZ (Figures 2a and 10). Along this fault zone, the in-hole temperatures (in excess of 3008C), water chemistry (elevated Na/K ratios) [Gizaw, 1993], presence of high-temperature calcsilicate minerals (e.g., epidote, garnet, prehnite, tremolite-actinolite), and resistivity structure [Samrock et al., 2015] all support upflow from the reservoir toward the surface. Temperature profiles from the bottom of these wells (2100 m) up to 700 m show this interval is characterized by boiling [Gizaw, 1993]. Either side of this zone of boiling lateral outflow of the reservoir fluid takes place. This is confirmed firstly, by temperature profiles in wells LA-4, LA-5, and LA-7 that show lower temperature ( 2508C) waters infiltrating at 1500 m (Figure 11), and second by water chemistry which shows the fluids are more concentrated and distinct from the reservoir fluid (i.e., waters have already boiled before entering the wells). Steam also condenses within the upflow zone and this likely supplies water to the cooler shallow (<700 m) aquifer ( Figure  11) [Gizaw, 1993].

Causes of Unrest
Understanding the causes of unrest at calderas remains a key challenge in volcanology [Acocella et al., 2015], and much of the debate focuses on discriminating between magmatic versus hydrothermal processes [e.g., Lowenstern et al., 2006;Gottsmann and Battaglia, 2008;Chiodini et al., 2012Chiodini et al., , 2015a. Given our broad understanding of the geometry and connections between the magmatic and hydrothermal reservoirs (section 6, Figure 11) we now summarize our geodetic data and build a testable hypothesis for the causes of unrest at Aluto.
Ground surface displacement at Aluto is characterized by episodic accelerating uplift that causes edificewide inflation, followed initially by rapid subsidence and then slower deflation (section 4). The uplift location is roughly centered within the caldera and is constant through time (between 2004 and 2011, Figure  2). For the uplift, our analysis of the vertical and horizontal components of ground motion ( Figure 6) and joint inversion of all InSAR observations (Figure 7) were consistent with a best-fitting spherical point source at a depth of 5.1 km beneath the surface. Since deep well observations place the main geothermal reservoir at a depth of >2 km (section 6), we infer that a 5 km inflation source is most likely located between the upper boundary of the magmatic reservoir and base of the geothermal system.

10.1002/2016GC006395
The uplift of active calderas is typically linked to fluid injection from depth into an inflating source region [Pritchard and Simons, 2002;Wicks et al., 2006;Parks et al., 2012Parks et al., , 2015. At Aluto there is no evidence from any of the individual interferograms to suggest contracting sources elsewhere around the complex, so we infer that inflation was fed from a depth greater than 5 km. The uplift (inflation pulse) shows two steps, suggestive of an accelerating deformation trend (section 4.3, Figure 7d). This could either be explained by two pulses of fluid input to this source region [e.g., Parks et al., 2015], or alternatively damage and fracturing, increasing permeability and allowing a greater rate of fluid supply. Deformation data alone do not allow us to investigate these possibilities, nor do they unambiguously differentiate between whether the fluid is gas, aqueous fluid, magma, or a combination of these. However, most peralkaline volcanoes are considered to have a volatile-rich cap at around 5-6 km depth [e.g., Leat et al., 1984;Mattia et al., 2007;Biggs et al., 2009a;Neave et al., 2012;Macdonald et al., 2014]. This zone is consistent with our modeled source depth ( Figure  7c) and our interpretation is that magmatic fluid injection or intrusions into this cap provide the source mechanism for uplift at Aluto (Figure 11).
At the point of peak uplift (maximum source inflation) subsidence (deflation) begins at Aluto (Figure 7d). A key feature of Aluto's deformation is the roll-over from uplift to subsidence which takes place over a timescale of a few months (section 4.3, Figure 7d). This short timescale is strongly suggestive of the migration of magmatic or hydrothermal fluids and degassing [e.g., Todesco et al., 2004;Wauthier et al., 2013;Caricchi et al., 2014]. Ground subsidence is commonly observed during extraction of geothermal fluids [e.g., Vasco et al., 2002Vasco et al., , 2013Allis et al., 2009;Keiding et al., 2010]. However, deformation related to commercial groundwater extraction is often spatially irregular, localized across a few subsidence peaks and most-importantly fault controlled [e.g., Fialko and Simons, 2000;Vasco et al., 2002;Samsonov et al., 2011;Sarychikhina et al., 2011]. At Aluto, deflation patterns are caldera-wide (e.g., Figure 2f), analytical source models suggest deformation occurs at 3.5 km ( Figure 6 and Table 2), and together this strongly suggests that fluids are being removed from deep within the geothermal reservoir.
Since our data suggest that the magmatic and hydrothermal reservoirs of Aluto are physically connected, our favored mechanism for the ground deformation at Aluto is that it represents a coupled magmatichydrothermal process. Our proposed mechanism involves uplift being caused by a fresh magmatic fluid pulse or intrusion into a shallow crustal reservoir at 5 km ( Figure 11). As the inflating source region appears not to be well sealed, fluids and gas may then leak into the geothermal reservoir and ascend to the surface along fault pathways, leading to sharp deflation (Figure 7d). Slow long-term subsidence over the following years may reflect continued fluid loss and depressurization of the hydrothermal system consistent with the timescales predicted by numerical simulations of CO 2 -rich magmatic fluid pulses [cf. Chiodini et al., 2012].
Our hypothesis has many parallels with the suggestions made by Samrock et al. [2015] for causes of unrest at Aluto. Their interpretation is that unrest events are related to pulses of hot fluids entering the geothermal system and pose two physical mechanisms (clay swelling and thermoelastic expansion) that might be responsible. In their opinion, the apparent lack of a hot extended magma reservoir rules out a pure magmatic intrusion as the main cause of unrest. Together our new observations and the magnetotelluric results of Samrock et al. [2015] favor pulses of hot magmatic fluids entering a reservoir at 5 km as the root cause of uplift and subsequent fluid expulsion and degassing as the cause of subsidence.
To rigorously discriminate between whether a pulse of melt (i.e., a magmatic intrusion) or volatile fluids drive unrest would require a long-term monitoring protocol to be put in place at Aluto. For example, repeat geodetic and gravity surveys could be used to infer the density of the intrusive fluids and their relation to the deformation source [e.g., Battaglia et al., 2003Battaglia et al., , 2006Gottsmann and Battaglia, 2008;Tizzani et al., 2009]. With these additional constraints on the subsurface and reservoir geometries increasingly complex and hence realistic source models could be considered for deformation analysis (e.g., finite spherical and ellipsoidal model [McTigue, 1987;Yang et al., 1988] and also finite element models [Masterlark et al., 2010;Hickey et al., 2015]). Further, gas geochemical monitoring (e.g., He/CH 4 ratios [Chiodini et al., 2015a] or repeat CO 2 surveys [Parks et al., 2013]) could help confirm arrival of new magmatic gases at the surface, and when combined with geodetic observations, could reveal whether peak degassing follows maximum edifice inflation. Geochemical measurements and geophysical imaging need to be carried out across a period of unrest and should provide a much clearer understanding of how deep magmatic processes at Aluto transfer energy and fluids to the hydrothermal system, and how the major fault zones facilitate their release to the surface.

10.1002/2016GC006395
It is important to re-emphasize that comparable uplift-subsidence events are taking place at a number of caldera complexes in the EARS [Biggs et al., 2009a[Biggs et al., , 2011[Biggs et al., , 2016, many of which are also being targeted for geothermal development. This provides an invaluable opportunity for government and geothermal industry stakeholders to put in place monitoring schemes that will ensure long-term safe, sustainable exploitation of the resources that these restless caldera systems host.

Conclusions
The Aluto magmatic and hydrothermal systems are physically connected by faults and fractures, such that deep (>2 km), hot (>2508C) geothermal fluids currently receive a continuous input of magmatic volatiles. Fluids migrate along areas of structural weakness, e.g., a proposed volcanic ring fault and major tectonic faults that dissect the complex, and ascend to the surface, releasing high-concentrations of magmatic CO 2 (d 13 C of 24.2& to 24.5&). Episodic uplift-subsidence events typify the Aluto volcanic system and a joint inversion of new and existing InSAR data collected from 2008 to 2010 suggest that the uplift source is located at 5 km, and requires an inflationary volume change of 113 3 10 6 m 3 . Uplift-subsidence events play out in a style that is characteristic of a coupled magmatic-hydrothermal system, and we propose that uplift is caused by a deep magmatic fluid injection or intrusion, and that faults provide key pathways for gas and fluid leakage into the geothermal reservoir. This hypothesis can be tested by monitoring changes in gas flux and composition along Aluto's main fault zone, which provides an important conduit for fluid upflow and sampling of these deeper reservoirs. present, and future). W.H., J.B., T.A.M., and D.M.P. are supported by and contribute to the NERC Centre for the Observation and Modelling of Earthquakes, Volcanoes, and Tectonics (COMET). Envisat data were provided by ESA. ALOS data were provided through ESA third party mission. W.H. funded by NERC studentship, NE/ J5000045/1. Additional funding for fieldwork was provided by University College (University of Oxford), the Geological Remote Sensing Group, the Edinburgh Geological Society, and the Leverhulme Trust. Analytical work at the University of New Mexico was supported by the Volcanic and Geothermal Volatiles Lab at the Center for Stable Isotopes and an NSF grant EAR-1113066 to T.P.F. The full CO 2d 13 C data set is provided in the supporting information file. All other data presented in this paper are archived at the Department of Earth Sciences, University of Oxford, and can be accessed by contacting Tamsin A. Mather (Tamsin.Mather@earth.ox.ac.uk). The airborne lidar data set (collected by NERC ARSF flight ET12-17-321) is freely available and can be accessed from http://dx.doi.org/10. 6084/m9.figshare.1261646. We also thank Giacomo Corti and an anonymous reviewer for their constructive comments that helped to improve the original paper.