Magma and Volatile Supply to Post-collapse Volcanism and Block Resurgence in Siwi Caldera (Tanna Island, Vanuatu Arc)

Siwi caldera, in the Vanuatu arc (Tanna island), is a rare volcanic complex where both persistent eruptive activity (Yasur volcano) and rapid block resurgence (Yenkahe horst) can be investigated simultaneously during a post-caldera stage. Here we provide new constraints on the feeding system of this volcanic complex, based on a detailed study of the petrology, geochemistry and volatile content of Yasur^Siwi bulk-rocks and melt inclusions, combined with measurements of the chemical composition and mass fluxes of Yasur volcanic gases. Major and trace element analyses of Yasur^ Siwi

Siwi caldera, in the Vanuatu arc (Tanna island), is a rare volcanic complex where both persistent eruptive activity (Yasur volcano) and rapid block resurgence (Yenkahe horst) can be investigated simultaneously during a post-caldera stage. Here we provide new constraints on the feeding system of this volcanic complex, based on a detailed study of the petrology, geochemistry and volatile content of Yasur^Siwi bulk-rocks and melt inclusions, combined with measurements of the chemical composition and mass fluxes of Yasur volcanic gases. Major and trace element analyses of YasurŜ iwi volcanic rocks, together with literature data for other volcanic centers, point to a single magmatic series and possibly long-lived feeding of Tanna volcanism by a homogeneous arc basalt. Olivine-hosted melt inclusions show that the parental basaltic magma, which produces basaltic-trachyandesites to trachyandesites by $50^70% crystal fractionation, is moderately enriched in volatiles ($1wt % H 2 O, 0·1wt % S and 0·055 wt % Cl). The basaltic-trachyandesite magma, emplaced at between 4^5 km depth and the surface, preserves a high temperature (1107 AE158C) and constant H 2 O content ($1wt %) until very shallow depths, where it degasses extensively and crystallizes. These conditions, maintained over the past 1400 years of Yasur activity, require early water loss during basalt differentiation, prevalent open-system degassing, and a relatively high heat flow ($10 9 W). Yasur volcano releases on average ! 13·4 Â10 3 tons d À1 of H 2 O and 680 tons d À1 of SO 2 , but moderate amounts of CO 2 (840 tons d À1 ), HCl (165 tons d À1 ), and HF (23 tons d À1 ). Combined with melt inclusion data, these gas outputs constrain a bulk magma degassing rate of $5 Â 10 7 m 3 a À1 , about a half of which is due to degassing of the basaltic-trachyandesite. We compute that 25 km 3 of this magma have degassed without erupting and have accumulated beneath Siwi caldera over the past 1000 years, which is one order of magnitude larger than the accumulated volume uplift of the Yenkahe resurgent block. Hence, basalt supply and gradual storage of unerupted degassed basaltictrachyandesite could easily account for (or contribute to) the Yenkahe block resurgence.

I N T RO D UC T I O N
Caldera collapse, produced by massive magma withdrawal during explosive eruptions, corresponds to the culminating stages in the evolution or eruptive cycles of volcanoes beneath which crustal magma reservoirs are able to develop. Renewed magma recharge after such a stage leads to volcanic unrest, often associated with ground deformation of the caldera floor in the form of 'doming'or uplift of faulted blocks (resurgence). Such phenomena have long been described at numerous caldera-bearing volcanoes, including Long Valley in California, Valles caldera in New Mexico, Phlegraean Fields, Pantelleria and Ischia in Italy, Iwo-jima in Japan, and Tanna in Vanuatu (e.g. Dvorak & Dzurizin, 1997;Molin et al., 2003). Caldera floor resurgence and block uplift can be progressive or discontinuous (e.g. Dvorak & Dzurizin, 1997), can have variable geometry depending of the aspect ratio of the magma intrusions or reservoirs (e.g. Acocella et al., 2000), and can occur without contemporaneous magma extrusion, implying complex relationships between magma replenishment, pressurization of aquifers and hydrothermal systems, as well as local structures and tectonic movements (e.g. Marsh, 1984;Hill et al., 1985). Therefore, improved understanding of these phenomena and their relationships is of fundamental interest for anticipating the eruptive pattern of such caldera-bearing volcanoes and the associated risks.
One suitable site where such processes and relationships can be investigated is Siwi caldera, in Tanna island, where post-caldera eruptive activity (Yasur volcano) and block resurgence have long developed together. Tanna island belongs to the southern segment of the 1200 km long Vanuatu volcanic arc, in the SW Pacific (Fig. 1a). This oceanic island arc has been generated by eastward subduction of the Australian plate beneath the Pacific plate, with associated back-arc extension of the North Fiji basin on the Pacific side (Pelletier et al., 1998;Calmant et al., 2003). The plate convergence rate reaches 11·8 cm a À1 at the latitude of Tanna. The island stands 150 km east of the Vanuatu trench and $130 km above the Benioff zone, and has been built through successive phases of volcanic activity and reef limestone growth. Pliocene^Pleistocene volcanic activity gave rise to the Green Hill and Tukosmeru centers, mainly formed by basalts and andesites (Carney & MacFarlane, 1979;Fig. 1b). A complex pyroclastic series, the Old Tanna ignimbrite (OTI), tracks a major caldera-forming event during this period (Robin et al., 1994). Since the late Pleistocene, volcanic activity has been concentrated in the SE part of Tanna (Fig. 1b). The most important event was a caldera-forming eruption of moderate volume ($1^2 km 3 ) that produced the Siwi ignimbrite (Carney & MacFarlane, 1979;Nairn et al., 1988;Robin et al., 1994;Allen, 2005). This eruption is still undated but is probably less than 20 kyr old (Nairn et al., 1988). The Siwi pyroclastic sequence, well exposed as a near-continuous 11km long outcrop along the eastern coast, consists of subaerial pumice-rich pyroclastic flow deposits, separated by a variably welded lithic-rich spatter agglomerate (Allen, 2005). Caldera collapse along the Siwi ring fractures has been followed by a new unrest cycle, with impressive uplift of the central caldera floor, generating the so-called Yenkahe resurgent block (YRB), and by renewed volcanic activity on that horst and along the ring fractures (Carney & MacFarlane, 1979;Nairn et al., 1988;Robin et al., 1994;Fig. 1c).
The Yenkahe resurgent block is an $6 km long by 3 km wide ENE^WSW elongated horst, with an axial graben that is affected by extensive normal block faulting and widespread hydrothermal manifestations (Fig. 1c). Emerged coral reef terraces present on that structure record its rapid uplift, at a time-averaged rate of 156 mm a À1 over the last 1000 years (Chen et al., 1995). Such rapid deformation, involving sharp episodes of co-seismic uplift (e.g. 10^15 m in 1878; Chen et al., 1995), is two orders of magnitude larger than the ground uplift ($1·6 mm a À1 ) along the NW coast of Tanna island since the late Quaternary (Neef et al., 2003). Therefore, it has a volcanic origin, attributed to replenishment of a magma reservoir, and raises concern about hazards from both further major eruptions and possible landslides (Nairn et al., 1988;Chen et al., 1995). The Yenkahe resurgence ranks amongst the fastest ground deformation observed in caldera-bearing volcanic complexes worldwide: its rate greatly exceeds, for example, the resurgence rates recorded during the 1978^1999 unrest at Long Valley caldera (0·04 m a À1 ; Hill, 2006), and in the past 33 kyr at Ischia (0·024 m a À1 ; Gillot et al., 1982). It is comparable with post-caldera uplift at only a few sites such as Iwo-jima volcano in Japan (0·15^0·25 m a À1 ; Ukawa et al., 2006).
Renewed post-collapse volcanic activity in Siwi caldera has included both effusive and explosive phases, with the formation of pyroclastic cones (Ombus, Yasur). Since about 1400 yr BP, the eruptive activity has been focused at Yasur volcano, located on the western edge of the Yenkahe horst, and rising 360 m above sea level (Fig. 1c). The growth of the Yasur cone itself probably began about 800 years ago, and has continued until now (Nairn et al., 1988). Strombolian or Vulcanian explosive activity has long prevailed at Yasur but has occasionally culminated in more powerful (sub-Plinian) eruptions (Nairn et al., 1988). The 'standard' Yasur activity, first described by Captain Cook in 1774 (Aubert de la Ru« e, 1960;Carney & MacFarlane, 1979), consists of recurrent explosions with an average frequency of 1^3 per minute (e.g. Nabyl et al., 1997;Bani & Lardy, 2007), which generate dense ash clouds (Fig. 2a), and expel vesicular scoriae and decimeter-to meter-sized bombs that occasionally reach the volcano flanks (Fig. 2b). Lava flow effusions are rare (Carney & MacFarlane, 1979).
Until now, there have been few studies of the Yasur activity and its broader relationship with the evolution of the Siwi caldera system. Here, we report new data from a systematic study of solid products and magmatic volatiles from both the early and current Yasur eruptive activity, and the Siwi ignimbrite sequence. Our objectives were to (1) determine the magma feeding composition and differentiation processes, (2) assess the abundances of dissolved and exsolved volatiles, (3) quantify the magma volumes sustaining the current Yasur activity and gas emissions, and (4) elucidate the possible relationships between magma supply and storage and the Yenkahe block resurgence. Rock sampling and gas measurements were performed during a field campaign in October 2007 and different approaches were combined to investigate the magma petrology and geochemistry, dissolved volatiles in melt inclusions and glassy matrices, and the present-day crater gas emissions. Altogether, our results allow us to propose a quantitative model for the magma feeding system of the Siwi^Yasur caldera complex, and to discuss its relationship with continuing Yenkahe block resurgence.

S A M P L E S A N D M E A S U R E M E N T S Solid volcanic products Sample location and description
The early (4 1400 to 800 yr BP) Yasur volcanic products were sampled from a410 m thick sequence on the northern bank of Siwi river, about 1km NNW from Yasur crater (19·51618S, 169·44708E;Fig. 1c). This sequence was exposed to the open air after catastrophic drainage of the former Siwi water lake in 2000, which deeply eroded the Siwi riverbed. It includes two unconformities ascribed to pauses in volcanic activity. An analogous sequence had previously been identified by Nairn et al. (1988) and correlated with a more distal section (4 km NNW from Yasur cone) where the pauses have been dated at 1409 AE 80, and 804 AE 56 years BP, respectively ( 14 C conventional ages). Correlation between our samples and the stratigraphic logs provided by Nairn et al. (1988) is shown in Fig. A in the Electronic Appendix (available for downloading at http://www.petrology.oxfordjournals.org). Deposits older than 1400 years consist of ash beds containing fibrous lapilli lenses (TAN 3), and scattered centimetre-sized scoriae (TAN 8). Deposits between 1400 and 800 years BP are mainly formed of alternating centimetres-thick fall layers made of either fine to coarse ash or fibrous to equant scoria lapilli  Nairn et al. (1988) and Allen (2005). Diamonds indicate the sample sites of the Siwi ignimbrite sequence, and Yasur products of the early (4 1400 to 800 yr BP sequence; S) and present-day activities.
(TAN 4 and 5). The presence of accretionary lapilli and vesiculated tuffs points to repeated involvement of external fluids (groundwater or the hydrothermal system) in driving the explosive activity. The uppermost 60 cm thick layer, made up of equant scoria lapilli (TAN 6), is believed to be equivalent to the sub-Plinian deposit described by Nairn et al. (1988). On the southern side of the Siwi river, we sampled two additional scoria fall layers (TAN 20 and 21) that directly overlie a plagioclase-phyric lava flow forming the riverbed (TAN 19;19·51688S, 169·44578E). These deposits should be related to the early Yasur activity.
As representative samples for the past 800 years of persistent eruptive activity at Yasur cone itself, we collected and analyzed both fibrous lapilli (TAN 14 and 15) deposited on the surface $1·5 km NW from the vent, along the main ash dispersal axis, and proximal explosive products. The latter include a bomb (TAN 10, Fig. 2b), and centimetre-sized highly vesicular scoriae (TAN 12) that fell on the crater rim.
We also collected and analyzed pumices from the ignimbrite deposits related to Siwi caldera collapse (hereafter referred to as the Siwi sequence). Pumices from the upper pyroclastic flow deposit (TAN 23;19·55348S,169·48308E), intermediate spatter-rich agglomerate (TAN 24), and lower pyroclastic deposits (TAN 25;19·54938S,169·48268E) were sampled along the SE coast (Fig. 1c). This succession is equivalent to that described by Allen (2005). Additional pumice samples were taken from meter-scale pyroclastic flow deposits cropping along the eastern side of Port Resolution bay (TAN 9;19·52678S, 169·50408E), and on the Siwi ring fracture (TAN 17 and 18;19·41268S, 169·48428E). These two latter superimposed pyroclastic flows are attributed to late stages of the Siwi ignimbrite eruption on the basis of the texture of their junvenile components. Finally, we sampled another $7 m thick pyroclastic sequence along the Yenkahe horst southern border at Port Resolution bay (TAN 1;19·52928S, 169·49088E), formed by coarse-grained pumice-rich lenses interbedded with ash layers. This sequence, which has not been previously described, overlies laminated black sands with shell fragments. It was most probably emplaced in shallow seawater inside the Siwi caldera and is considered here as related to post-caldera activity, from volcanic vents located on the Yenkahe horst, and preceding the building of the Yasur cone.

Analytical procedures
Major and trace element analyses of Siwi and Yasur bulk-rocks were carried out at the Centre de Recherches Pe¤ trographiques et Ge¤ ochimiques (SARM) in Nancy (France), by inductively coupled plasma optical emission spectroscopy (ICP-OES) and inductively coupled plasma mass spectrometry (ICP-MS), respectively (Carignan et al., 2001). A second set of trace element data was acquired on pairs of Yasur bulk scoria and mechanically separated matrices using ICP-MS at Laboratoire Pierre Su« e (Saclay, France). Matrix fragments were handpicked under the binocular, then aliquots of 50 mg of glass and powdered host bulk-rocks were analyzed during the same run. The former trace element dataset is reported in Table 1, and the second one as Electronic Appendix 1.
Major elements of minerals, melt inclusions, and matrix glasses were analyzed using a SX50 CAMECA electron probe (Camparis, Jussieu, France). In glasses and minerals, they were measured with a 10 nA defocused beam and a 30 nA focused beam, respectively, and a counting time varying from 10 to 25 s. The concentrations of S, Cl and P were determined with a 30 nA defocused beam and a longer counting time (120^200 s) on peak, and their accuracy was checked by analyzing reference glasses (basalts Ba/Th   Analytical uncertainties (1s) are 510% for Th, Ta, Hf, Cs, Co and Cr; 58% for U and Ba; 55% for the others. Bas, basalt; Bas-Ta, basaltic-trachyandesite; Ta, trachyandesite. LOI, loss on ignition; b.d.l. below detection limit; n.d., not determined.
*Yasur distal and prox. products refer, respectively, to distal scoria deposits and proximal bombs and scoriae on the volcano flanks and crater rims.
yLapilli sample (13/1-B) from the sub-Plinian fall deposit described by Nairn et al. (1988), equivalent to our sample TAN 6. AlV981R23 and Vg2, pantellerite KE12). Fluorine was determined by coupling two TAP crystals, in a scanning mode, with 10 kV voltage, 60 nA beam current, 15 mm beam size, and 500 s counting time. Analyses were calibrated against reference glasses containing from 950 to 4250 ppm F (Spilliaert et al., 2006). The relative errors are 10% for S and Cl, and 10% for F. The detection limits are 80 ppm for S and Cl, and 170 ppm for F. Some glassy matrices and minerals were also analyzed and mapped with a Philips XL30 scanning electron microscope equipped with an EDAX DX 4 system (Dpt Scienze della Terra, Pisa, Italy).
Dissolved H 2 O contents were determined with micro-Raman spectroscopy, using a Renishaw inVia instrument (Saclay, France). The spectrometer is fitted to a Leica microscope with a 100Â objective to excite and collect the signal from a very small area (about 1 mm in diameter). Analyses were performed at the sample surface in confocal mode. The excitation source was a green diode laser (532 nm) filtered to deliver a low power (c. 7 mW) on the sample during short counting times (5 100 s). Selected analytical conditions minimize scatter from the embedding medium and prevent sample heating and oxidation. The intensity of the H 2 O T band (3580 cm À1 ) and its scaling with respect to the band for the silicate network (980 cm À1 ) were determined with external analytical procedures (Mercier et al., 2009). Calibration was made against a set of basaltic and andesitic glass standards (Mercier et al., 2010). One to three analyses were performed, prior to any other measurements to avoid H 2 O loss, for each melt inclusion and then averaged. Errors propagated on the external calibration procedures range between 6 and 10%.
Finally, dissolved carbon concentrations were determined with the Saclay nuclear microprobe (France), using the 12 C(d, p) 13 C nuclear reaction (Me¤ trich . Analyses were made in scanning mode, with a 5 mm beam. Areas in melt inclusions previously analyzed for S, Cl, F with a high electron beam current were avoided. The integrated charge varied from 2 to 1·5 mC, implying a corresponding minimum detection limit of 60^70 ppm of CO 2 . Carbon concentrations were calibrated against a scapolite standard (6800 ppm C) and a waterbearing Etna basaltic glass (EtII-1) containing 300 AE 30 ppm C (Spilliaert et al., 2006). The latter glass concentration is reproduced with 10% precision.
The temperatures of melt entrapment were determined from optical thermometry measurements on melt inclusions hosted in olivine and plagioclase in the Yasur products. Experiments were carried out under a Leitz heating stage purged with Ar^1% H 2 gas flow. Temperatures were measured with a Pt^Pt 90 Rh 10 thermocouple (AE108C), and calibrated against the melting points of Ag (961·88C) and Au (1063·88C). It is known that the homogenization temperature (T hom ) of olivine-hosted melt inclusionsç when the shrinkage bubble vanishesçincreases with time because of olivine crystal deformation (Massare et al., 2002). Actually, the temperature (T hom ) measured on melt inclusions hosted in TAN 6 olivines is 10^508C higher when several (1^9) heating^cooling cycles have been completed on a single crystal. A heating rate of 18·58C min À1 was thus chosen to minimize this effect. Experiments were also carried out with a slower heating rate (68C min À1 ) from 10508C to T hom to verify that no kinetic effects may have affected our measurements. A total of 30 heating experiments were conducted on double-faced polished crystals whose inclusions were preserved enclosed.

Yasur crater gas emissions Chemical composition of emitted gases
The chemical composition of Yasur gas emissions was determined from in situ analysis and sampling of the volcanic plume at the crater rim, $150 m downwind of vent A (Fig. 2a). This vent was the most vigorously active (one explosion every minute, on average) and the main source of degassing during our two series of measurements, on 21 and 29^30 October 2007.
The plume concentrations in water vapor, CO 2 and SO 2 were measured continuously with MultiGas instruments (Shinohara, 2005;Aiuppa et al., 2007Aiuppa et al., , 2008Shinohara et al., 2011), which combine different gas sensors in a home-made box (INGV-type used on 21 October and GSJ-type used on 29^30 October). Volcanic gases were pumped at an average flow rate of 0·6 l min À1 and analyzed with these systems. An LI-840 closed-path NDIR spectrometer measured both CO 2 (range 0^3000 ppmv; accuracy AE1·5%), and H 2 O (range 0^80 ppt; accuracy AE1·5%). SO 2 was measured with different electrochemical sensors (3ST/F-Cod.TD2D-1A and KTS-512P, respectively) that have similar performances in concentration range, noise level and response (Shinohara et al., 2011). The signals from all sensors were recorded and stored every 2 s (INGV-type) and 1s (GSJ-type) on a data-logger.
MultiGas sensing on 21 October 2007 provided $4500 measurements of each gas species during 135 min of volcanic activity. The data for CO 2 and H 2 O were post-processed by subtracting the concentrations in ambient air from the bulk recorded signals. Air background concentrations were constrained by fitting a polynomial function to the subset of CO 2 and H 2 O data for which SO 2 concentration is zero (given that SO 2 ça typical component of magmatic gas emissionsçis absent in standard atmosphere). The resulting air concentrations were fairly constant for CO 2 (384^387 ppm) but more variable for H 2 O (19^4402 1^020 ppm). Similar data handling was applied to the 29^30 October datasets; however, these were affected by higher and more randomly changing background H 2 O.
During this second survey, H 2 and H 2 S could additionally be determined with the sensor configuration described by Shinohara et al. (2011).
Filter pack sampling of the volcanic gas was carried out in tandem with MultiGas sensing to measure time-averaged SO 2 , HCl and HFconcentrations. Sampling was performed for 45^60 min, by pumping the volcanic plume at an average flow rate of 5 l min À1 through alkali-impregnated filters (e.g. Aiuppa et al., 2005). Six filter series were collected on 21 October and seven on 29^30 October. In the laboratory, the filters were leached with bi-distilled water and H 2 O 2 , to elute S, Cl and F. The leachates were then analyzed by ion chromatography for SO 2À 4 and Cl À and by ion selective electrode for F À .

SO 2 emission rate
The SO 2 emission rate from Yasur was determined by differential absorption UV spectroscopy (DOAS), during repeated traverses under the volcanic plume with a four-wheel-drive vehicle. The UV spectrometer (Ocean Optics USB2000, Serial No. 2196) spanned a 280^400 nm spectral range, with a resolution of 0·5 nm (full width at half-maximum). It was coupled by a fiber optic bundle to a telescope fixed outside the vehicle and pointed to zenith. Exposure time for single spectra was 200 ms, and we co-added eight spectra to enhance the signal-to-noise ratio. The position of each UV spectrum was obtained from the log of a continuously recording global positioning system (GPS) unit. SO 2 column amounts were retrieved following standard DOAS spectral procedures (e.g. Galle et al., 2003;Bani & Lardy, 2007). The best-fitting window for SO 2 retrieval (307·7^335·1nm) was constrained from near random fit residual structure with minimal standard deviation. The SO 2 reference spectrum was obtained by convolving a high-resolution SO 2 spectrum with our instrument line shape, and then removing the low-frequency component using the Ocean Optics WinDOAS software. The observed SO 2 column amounts were then projected onto the plane perpendicular to the plume transport direction, and the SO 2 emission rate for each traverse was obtained by scaling the integrated column amounts across this plane with the plume speed. Plume speeds were directly obtained from simultaneous wind speed measurements on top of Yasur volcano, using a hand-held anemometer (AE0·1m s À1 ). The overall error on each SO 2 flux series (given in Electronic Appendix 2) includes errors on the wind speed and DOAS spectral analysis (Galle et al., 2003).

M I N E R A L O GY A N D M AG M A G E O C H E M I S T RY Bulk-rocks and matrices
Juvenile products from the early ( 1400 years) and present-day Yasur activity show a limited compositional range, with SiO 2 between 54·5 and 56·8 wt % (Table 1). In the Total Alkali^Silica (TAS) diagram, they plot in the field of basaltic-trachyandesite, except one sample (TAN 8), which has a pure trachyandesitic composition (58·6 wt % SiO 2 ; Table 1). Yasur bulk-rocks also display restricted variations in trace elements (Figs 3 and 4), even though distal and proximal explosive products differ in texture and crystal content.
The Yasur distal products contain a few per cent of plagioclase, clinopyroxene and olivine phenocrysts, embedded in highly vesicular glassy matrices with trachyandesitic composition (SiO 2 from 56·5 to 58·5 wt %; Table 2). Systematic analyses (88) of plagioclase crystals in the TAN 6 sample show that most of them are normally zoned, from An 66 to An 57 . Rare crystals were found to have a more calcic core (An 80 to An 87 ). Clinopyroxene (Wo 41·1^40 Fs 12·3^15·3 ) and olivine (Fo 75·2^72·7 in the cores to Fo 75^72·6 in the rims) display limited compositional variations. Olivine also occurs as inclusions in clinopyroxene and plagioclase An 68^60 , together with Fe^Ti oxides (Usp 16%). We attempted to precisely quantify the crystal content of these scoria lapilli by analyzing the trace element content of five pairs of the bulk clasts and their mechanically separated matrices (TAN 4,5,14,15 and 20), representative of the past 1400 years of eruptive activity (Electronic Appendix 1). These selected samples are all similar in their incompatible element concentrations (i.e. U, Th, Nb, La, Ce), normalized trace element patterns ( Fig. 4), large ion lithophile element (LILE) enrichment, and negative Nb anomaly (Nb/Nb* ¼ 0·26 in bulk clasts and 0·31 in matrices, where Nb* refers to the normal mid-ocean ridge basalt (N-MORB) value of 3·507 ppm; Hofmann, 1988). This homogeneity indicates a nearly constant extent of crystallization over time. Using the equation for equilibrium crystallization [ with C i0 and C i being the average thorium concentrations in the bulk clasts (2·1ppm) and matrices (2·5 ppm), Di the Th bulk partition coefficient (D mineral/melt $0·1) and F the proportion of residual melt, we find that the Yasur distal scoriae contain 10% of crystals at most. This proportion agrees with the difference in K 2 O content between the bulk samples and their matrices (e.g. TAN 6 and 21; Tables 1 and 2), indicating that K 2 O is also a reliable tracer of magma differentiation.
Pumices from the Siwi ignimbrite sequence display a wider range in major and trace element compositions ( Table 1). The least evolved samples (TAN 17, 18, 1), with basaltic trachyandesitic compositions (SiO 2 56 wt %, K 2 O ¼ 2·95 wt %), closely resemble Yasur products in terms of both major and trace elements (Fig. 3). Their matrices are slightly more evolved (3·4 wt % of K 2 O, Table 2). Although a complete database for the mineral chemistry of these pumices is not available, the olivine crystals (Fo 74^69 ) analyzed in samples TAN 17 and TAN 1 are either similar to or slightly more Fe-rich than olivines in the Yasur distal products.
The most evolved Siwi samples (SiO 2 ¼ 60 wt %, K 2 O ¼ 3·6^3·8 wt %), with the highest concentrations in incompatible elements (Fig. 3), are trachyandesitic pumices (TAN 23,24,25,and 9). Their matrix glasses and obsidian shards, extensively analyzed in the crystal-poor sampleTAN 23, have a very homogeneous composition (Table 2) equivalent to that of the host bulk pumice, as previously reported by Robin et al. (1994) and Allen (2005).The equilibrium paragenesis is made of Na-rich plagioclase (An 54 ) associated with Fe-rich olivine (Fo 57 ) and clinopyroxene (Wo 39·8 , Fs 21·5 ). Each of these minerals includes Fe^Ti oxide (usp530%), Cu-rich sulfide and fluorapatite micro-crystals that occasionally themselves contain melt inclusions. However, we also discovered in these samples some homogeneous or slightly normally zoned olivine crystals that are noticeably richer in Mg (from Fo 70^72 to Fo 77^80 ). The glassy envelope coating these crystals, when preserved, is also less evolved than the bulk matrix (Table 2).Therefore, these olivines are antecrysts that were mechanically entrained during the ascent and eruption of the Siwi trachyandesite.
Yasur P-products* *D and P-products refer to distal and proximal scoriae, respectively. yAveraged analysis performed in several fragments, with standard deviation values in parentheses (i.e. 42 for 56·54 AE 0·42). zTraverses between bubbles in different vesiculated pumice clasts. §The composition of the olivine is indicated in parentheses. N, number of analysis per fragment; b.d.l., below detection limit; n.d., not deternined.
Altogether, the major and trace element compositions of the Yasur and Siwi bulk-rocks demonstrate that they belong to a single magmatic series, with different degrees of differentiation. This is especially clear from the positive correlations, passing through zero, between incompatible element pairs (Fig. 3), which define a continuous differentiation trend from the products emplaced during the Yasur explosive activity to the pumices of the Siwi sequence. In contrast, there is an overall decrease of Co (Fig. 3). Long-lived feeding of Tanna volcanism by the same magmatic series is further verified from published data for the Old Tanna ignimbrite (OTI; Robin et al., 1994) and three basalts (Peate et al., 1997) dated at 2·4, 0·67 and 0·23 Ma (Carney & MacFarlane, 1979), which define the least differentiated end-members. In all these samples the correlations between incompatible element pairs are excellent, within the error limits, considering that analyses were performed with different analytical methods and in different laboratories. The regression lines obtained for various element pairs, such as Ba^Th, Rb^Nb and Nb^Th but also ThĤ f, Ta^Th, La^Th, La^Sm and Nb^Y, have high correlation coefficients ( Fig. 3; Table 1).

Melt inclusion geochemistry and dissolved volatiles Yasur distal and proximal products
As representative of Yasur scoria from distal deposits, olivines were systematically hand-picked from 5 fallout layers (TAN 4,5,6,20,21), with a focus on TAN 6, a sample from the scoria lapilli layer attributed to a violent (sub-Plinian) event. Crystals are commonly elongated in the [100] direction, and display hourglass (skeletal) morphologies that are delineated by small-sized melt inclusions ( 50 mm) and Fe^Ti oxides (Fig. 5a). From experimental work (Faure & Schiano, 2005) we infer that these skeletal textures track the first stage of olivine crystallization. Larger sealed inclusions (!100 mm), which constitute the most abundant population, and opened glass embayments were trapped during subsequent olivine growth (Fig. 5a). The sealed inclusions are perfectly glassy, with no or only a small shrinkage bubble, which indicates efficient cooling upon magma emplacement. Altogether, melts entrapped in Yasur olivines fall in the compositional domain of the bulk-rocks ( Fig. 5b and c) and, therefore, are representative of the carrier magma. The large sealed inclusions and embayments have a very homogeneous composition (K 2 O ¼ 2·9^3·1wt %; Table 3) and contain uniform amounts of H 2 O (on average 1·07 wt %; 2s/ N ¼ 0·05 with N ¼ 41), S (280^350 ppm), Cl (890^920) and F (330^370 ppm). Their S/Cl and Cl/F (wt) ratios average 0·34 and 0·42, respectively. Their CO 2 content varies from 117 to 232 ppm. However, some smaller melt inclusions depart from this chemically homogeneous population in being richer in MgO and poorer in K 2 O ( 2·7 wt %; Fig. 5c). They also display higher S (9806 80 ppm) and Cl (880^1420 ppm) contents (S/Cl ratio up to 1·0; Fig. 6), whereas their H 2 O concentrations (0·81 ·3 wt %) are comparable (Fig. 7).
To complement our melt inclusion study, we systematically analyzed dissolved volatiles in the glassy matrices of the same samples. These glasses are compositionally analogous to the large melt inclusions and glass embayments (Fig. 5). They also preserved comparable amounts of S, Cl and F (Tables 2 and 3), implying their efficient quenching and minor or negligible loss of these volatiles during fast (explosive) magma extrusion. In particular, repeated measurements along 0·5 mm long profiles (11 traverses, $55 analyses) in vesiculated matrix glasses of sample TAN 4 reveal remarkably constant concentrations of S (130 AE 60 ppm), Cl (870 AE 60 ppm), and F (373 AE 50 ppm), within the error limit (1s). In contrast, H 2 O is largely degassed relative to the melt inclusions and embayments ($0·3 wt % against 1·0^1·2 wt %; Fig. 7). This differing behavior of water with respect to sulfur and halogens is consistent with the much higher diffusivity of water in high-temperature melt (Freda et al., 2005), even during fast magma extrusion.
In proximal scoriae, melt inclusions in olivine (Fo 67 ) and plagioclase (An 56^62 ) represent the most evolved trachyandesite terms of the Yasur series, as do their matrix glasses ( Fig. 5b and c). There is no significant difference between melt inclusions trapped in either olivine or plagioclase, which points to their negligible evolution after entrapment. All of them are enriched in SiO 2 ($59 wt %) and K 2 O (3·7^4·1wt %), strongly impoverished in H 2 O (0·3^0·4 wt %; Fig. 7) and S (590 ppm), and contain no more detectable CO 2 . Their Cl contents (440^590 ppm) and S/Cl ratios (0·1^0·2) are half of those in the distal scoria lapilli (Fig. 6), whereas their F/Cl ratio (0·8^1·0) is twice as high (Table 3). All these features demonstrate extensive degassing of the melt ponding in shallow conduits, during which chlorine is more efficiently released than fluorine. Significant water loss from the inclusions upon magma emplacement is unlikely as the host matrix of these proximal vesicular scoriae is glassy. Instead, these are still more impoverished in H 2 O (0·2 wt %), S (580 ppm, the detection limit), Cl (460 ppm), and partly in F (360 ppm), thereby strongly supporting a pure degassing process.
Systematic optical thermometry measurements on melt inclusions in the Yasur samples reveal a homogeneous temperature of melt entrapment, irrespective of their mineral paragenesis. In distal scoria lapilli (TAN 6), the homogenization temperatures of melt inclusions vary from 1123 to 10938C in olivine (average 1108 AE 88C; 11 measurements), and from 1119 to 11048C (average 1106 AE 68C; six measurements) in plagioclase. In crater rim scoriae (TAN 12), plagioclase-hosted inclusions homogenize at a very similar mean temperature of 1107 AE58C (13 measurements).
Most importantly, we find that the few Mg-rich (Fo 77^80 ) olivine antecrysts present in the Siwi trachyandesite have entrapped melt inclusions with a basaltic composition, comparable with that of the bulk Tanna basalts (Fig. 5b  and c). These inclusions are the richest in CaO (up to 10 wt %; CaO/Al 2 O 3 up to 0·59) and the poorest in P 2 O 5 and K 2 O (P 2 O 5 /K 2 O ¼ 0·18; Fig. 5c), once corrected for the limited effect of post-entrapment crystallization of their host (Table 4). They represent the least evolved melt of the Yasur^Siwi magmatic series, and thus provide unique information on the volatile content of the parental basaltic magma of Tanna volcanism. These scarce basaltic inclusions contain between 1 and 2 wt % of H 2 O (Fig. 7), 1000 ppm of S (without any immiscible sulfide globules), and 550 ppm of Cl (S/Cl ¼1·8 AE 0·2; Fig. 6). Their CO 2 content ranges from 573 to 130 ppm, without co-varying with H 2 O. However, they also contain bubbles into which carbon dioxide could have migrated upon cooling after entrapment. Consequently, their dissolved CO 2 content may represent a lower limit for the amount of that species at the time of melt entrapment. Figure 8 shows the temporal variations of SO 2 and air-corrected CO 2 and H 2 O concentrations in Yasur volcanic gas emissions, as recorded on 21 October 2007. In the first $22 min enhanced plume condensation prevented H 2 O retrieval. The observed variations are due to both the pulsed explosive activity and variable air dilution of the volcanic plume during the interval of measurement. Looking at the relative behavior of the volatile species, however, reveals a very coherent pattern for CO 2 and SO 2 . The best-fit regression line in a scatter plot of CO 2 vs SO 2 concentrations (Fig. 9a) constrains a mean CO 2 / SO 2 molar ratio of 1·6 AE 0·4 (Table 5), with minor and short-lived temporary fluctuations (from 51·5 to 43). A comparable CO 2 /SO 2 mean ratio (1·8^2; Table 5) was measured on 29^30 October, and had previously been reported ($2·0) from Fourier transfor infrared (FTIR) remote sensing in 2005 (Oppenheimer et al., 2006). This parameter thus appears to be rather steady in Yasur volcanic gas emissions. In fact, Oppenheimer et al. (2006) reported no difference in CO 2 /SO 2 ratio during quiescent and explosive degassing in 2005, in sharp contrast to observations   during analogous activity at Stromboli volcano, in Italy (Burton et al., 2007). The relative behavior of H 2 O vs SO 2 is much more variable ( Fig. 9b and c), owing to the inherent complexity in resolving volcanic H 2 O from the high and changing air humidity background on top of Yasur. The volcanic H 2 O/ SO 2 molar ratio was thus assessed by considering only the time windows during which highest SO 2 and air-corrected  2 and H 2 O, we report in-plume concentrations after subtraction of background air contents (384^387 ppm and 19440^21020 ppm for CO 2 and H 2 O, respectively). (b) Time record of CO 2 /SO 2 and H 2 O/SO 2 molar ratios in the Yasur plume. Open squares, high-rate record of CO 2 /SO 2 ratios, calculated by taking the point-to-point ratio of CO 2 and SO 2 concentrations reported in (a) (the ratios were calculated for all the couples of MultiGas measurements for which SO 2 concentration was above the threshold value of 2 ppm); black filled squares, mean CO 2 /SO 2 ratios calculated from subsets of CO 2 and SO 2 concentrations acquired within the same time intervals used above for H 2 O/SO 2 ratios (ratios calculated as described above for H 2 O/SO 2 ); diamonds, mean H 2 O/SO 2 ratios calculated from subsets of H 2 O and SO 2 concentrations acquired within specific time intervals. These time windows (indicated by the vertical grey bars) were selected as those showing the highest H 2 O air-corrected concentrations and peak SO 2 concentrations45 ppm. The mean H 2 O/SO 2 ratios were calculated by drawing, for each time window, a SO 2 vs H 2 O scatter diagram, and calculating the corresponding ratio from the slope of the best-fit regression line.

YA S U R VO L C A N I C G A S E M I S S I O N S Gas composition
H 2 O concentrations were recorded on 21 October (grey vertical bars in Fig. 8). Although it varied widely even during these intervals (from 40 to 120; Fig. 9b), the H 2 O/ SO 2 ratio was more restricted (from 50 to 90) when the volcanic plume was more concentrated (45 ppm SO 2 ; Fig. 9c). Although the lower values may be more representative (given the plume dilution factor of order 10 2^1 0 3 ), we considered here the mean volcanic H 2 O/SO 2 molar ratio of 70 AE 20. Measurements on 29^30 October gave higher and even more variable ratios (120 AE 60 and 160 AE 60, respectively; Table 5), which, as mentioned above, were affected by a higher and more erratic moisture air background. It should be noted that H 2 O could not be determined with FTIR remote sensing at Yasur in 2005 (Oppenheimer et al., 2006) and so our results are the first for this volcano.
Chemical analysis of filter pack samples demonstrates reasonable stability of the volcanic HCl/SO 2 (0·33^0·53) and HF/SO 2 (0·095^0·12) molar ratios during our 21 and 29^30 October surveys (Table 5 and Fig. 10). It should be noted that these ratios characterize the time-integrated emission during both passive and explosive magma degassing. This prevented us from detecting the sharp SO 2 /HCl peak increases (up to 30) during single explosions that were measured with remote FTIR sensing at Yasur in 2005 (Oppenheimer et al., 2006), and in 2008 (Allard et al., in preparation). However, the average SO 2 /HCl molar ratio of 2·4 AE 0·4 derived from our measurements does not differ much from the mean SO 2 /HCl molar ratio ($2·0) for purely passive degassing reported by Oppenheimer et al. (2006), which means that passive gas emissions supply a prevalent fraction of the overall Yasur gas budget.
Combining our results from both MultiGas and filter-pack measurements, and normalizing to 100%, we derive the mean composition for Yasur volcanic gases reported in Table 5. As verified from MultiGas sensing on 29^30 October, the contribution of other gas species such as H 2 S and H 2 is minor (H 2 S/SO 2 50·01 and H 2 / SO 2 50·1). Yasur gas emissions thus appear to be highly hydrous ($98 mol % H 2 O) and oxidized (SO 2 /H 2 S ratio 4100); these are typical features for magmatic gas emissions from arc volcanoes worldwide (e.g. Symonds et al., 1994). Their CO 2 content is relatively modest and their HCl/HF molar ratio ($4) ranks among the lowest ratios measured at arc volcanoes (1·9^160; Aiuppa, 2009).

Gas fluxes
The SO 2 column amounts measured during our DOAS traverses beneath the Yasur volcanic plume occasionally displayed up to three peaks, tracking gas emission from the three active vents in the crater. The mean SO 2 fluxes determined on 20^21 and 29^30 October 2007 are 940 and 620 tons per day (tons d À1 ), respectively (Electronic Appendix 2). Their average (690 AE 330 tons d À1 ) is identical to the time-averaged SO 2 flux of 680 AE 380 tons d À1 constrained by repeated measurements at Yasur in 20042 009 (Bani & Lardy, 2007;Bani et al., in preparation). Therefore, our gas results in October 2007 are representative for the standard eruptive activity of the volcano.
Combining the chemical composition of the volcanic gas with the measured SO 2 fluxes, we directly derive the emission rate of the other volatile species (Table 5). The daily mean fluxes (tons d À1 ) are in the range of 13·4 Â10 3 for H 2 O, 840 for CO 2 , 165 for HCl and 23 for HF. The total gas flux, quantified here for the first time, averages $15 Â10 3 tons d À1 . According to these results, Yasur volcano ranks among typical arc emitters: it contributes c. 1·7^1·9% of global volcanic emissions of H 2 O ($8 Â 10 5 tons d À1 ; Wallace, 2005), and SO 2 ($3·6 Â10 4 tons d À1 ; Andres & Kasgnoc, 1998), $0·3^2·4% of global CO 2 Fig. 9. (a) Scatter diagram of MultiGas-sensed SO 2 and CO 2 concentrations (in ppm), including the entire dataset of Fig. 8a. The mean plume CO 2 /SO 2 ratio of 1·6 is derived from the slope of the best-fit regression line (see Shinohara et al., 2008;Aiuppa et al., 2007, for more details). (b) Scatter diagram of MultiGas-sensed SO 2 and H 2 O concentrations (in ppm), drawn from data acquired within the selected time intervals indicated as grey bars in Fig. 8. (c) Scatter diagram of SO 2 concentrations (in ppm) against the plume H 2 O/SO 2 ratios. The diagram shows that as plume density (and therefore SO 2 concentration) increases the spread of observed H 2 O/SO 2 ratios becomes narrower. The H 2 O/SO 2 ratio range of 50^90 (on average 70) is therefore accepted as the most representative of Yasur gas emissions. emissions (0·4^3·5 Â10 5 tons d À1 ; Arthur, 2000), $1·4% of HCl ($1·2 Â10 4 tons d À1 ; Pyle & Mather, 2009), and 0·5% of HF ($5 Â10 3 tons d À1 ; Pyle & Mather, 2009). However, its volatile source strength is an order of magnitude lower than that of nearby Ambrym volcano, in the central Vanuatu arc (Allard et al., 2009;Bani et al., 2009).

D I S C U S S I O N Long-term feeding of Tanna volcanism by the same arc basalt
We show here that the Yasur^Siwi magmas belong to the same magmatic evolution series as both the Old Tanna Ignimbrite (OTI; Robin et al., 1994) and Tanna basalts erupted as long as 2·4 Myr ago (Carney & MacFarlane, 1979;Peate et al., 1997). This conclusion, deduced from both major and trace elements (Fig. 3) in the bulk-rocks, is also verified by the composition of melt inclusions: these cover a complete range from basaltçcomparable with bulk Tanna basaltsçto the most evolved trachyandesite erupted during the Siwi ignimbrite eruption (Fig. 5b and  c). Hence, we propose that a geochemically homogeneous basaltic magma could have supplied Tanna volcanism for $2 Myr.
Furthermore, our study demonstrates that basaltictrachyandesites erupted during the last 1400 years of volcanic activity in the Siwi caldera have maintained a constant bulk composition, in agreement with previous observations by Nairn et al. (1988). Such a constancy points to steady conditions of magma supply and differentiation  during this time interval. We did not find any textural or mineralogical evidence of basalt input and mixing with more evolved trachyandesitic magmas in the Yasur^Siwi products (Figs 3 and 5). This conclusion is also supported by the S and Cl behavior, as discussed below. Robin et al. (1994) made the same observation for the OTI and Siwi deposits. Therefore, crystal fractionation is the dominant process controlling basalt differentiation towards the compositions of the Yasur and Siwi magmas, even though mixing between cogenetic melts that have differentiated along the same liquid line of descent cannot be totally excluded. Based on their average bulk Th (1·9 ppm) and Nb (1·1ppm) contents, we compute that Yasur basaltic-trachyandesites derive from a basaltic parent magma (Th ¼ 0·9 ppm, Nb ¼ 0·6 ppm) by $50% crystal fractionation (removal). The generation of the most evolved trachyandesitic magma (Th ¼ 3·3 ppm) extruded during the Siwi ignimbrite eruption (TAN 23 and 24), requires a higher extent (!70%) of differentiation. The Yasur^Siwi magma series displays high U/Nb (1 AE0·1) and Th/Ta (23 AE1) ratios typical for a subduction context, and La/Yb ratios (5·7 AE 0·3) that are similar to those for island arc basalts in the southern Vanuatu arc (Monzier et al., 1997). Following Peate et al. (1997), their overall enrichment in Ba and K, together with their low Zr/Yb (43 AE 2) and Nb/Yb (0·53 AE 0·02) ratios, points to the contribution of slab-derived fluids to a depleted MORB-mantle source. The high Th/Nb ratio (1·7) could be compatible with the involvement of a sediment-derived component (Johnson & Plank, 1999;Elliott, 2003), as also inferred for other Vanuatu lavas (Peate et al., 1997). Alternatively, these trace element ratios could be controlled by the stability of accessory phases in the subducting slab (Skora & Blundy, 2010). Sediments subducted beneath Tanna, which derive from the North Loyalty Basin, are prevalently composed of andesitic^dacitic volcano-clastic debris associated with silt^sandstone and conglomerate horizons, ash units rich in nanofossils, radiolaria and clays (e.g. Peate et al., 1997). Such sediments can be expected to contribute a small amount of volatiles to the mantle wedge upon subduction, in agreement with our finding of modest volatile enrichment of the Tanna parental basalts.
According to our data for the most primitive melt inclusions in Mg-rich olivines, the parental basalt of the SiwiŶ asur magma series is moderately hydrous (1^2 wt % H 2 O), weakly enriched in sulfur ($0·1wt %), and rather poor in Cl ($0·055 wt %). In fact, it falls within the domain for Cl-poor arc magmas whose genesis requires the contribution of 5 wt % slab-derived equivalent NaCl to a MORB-mantle source (Wallace, 2005). The low Cl content of this basalt is also consistent with the low HCl/ HF molar ratio (2·8^5·6) of Yasur gas emissions compared with other arc volcanic gases (1·9^160; Aiuppa, 2009).
Moreover, Yasur melts, with an average Cl/H 2 O wt ratio 0·05, are far from being saturated with a hydrous saline liquid (Webster et al., 1999). Finally, the relatively low CO 2 /SO 2 ratio of Yasur arc volcanic gases, the modest CO 2 emission rate ($840 tons d À1 on average) and the rather low dissolved CO 2 content of local thermal waters (Gauthier et al., in preparation), all concur to suggest a moderate enrichment of the parental basalt in carbon dioxide.

Depths and conditions of magma differentiation
The depths and conditions of magma differentiation beneath Siwi caldera were assessed from both our melt inclusion data and MELTS calculations (Ghiorso & Sack, 1995). Assuming volatile saturation, the dissolved CO 2 and H 2 O contents of olivine-hosted melt inclusions in the Yasur^Siwi lavas allow us to estimate the total fluid pressures (P CO2 þ P H2O ) and, hence, the lithostatic depth at the time of entrapment of the various liquids (Electronic Appendix 3). For these calculations we used the solubility model of Papale et al. (2006), taking account of the fact that CO 2 solubility data are still poorly documented for basaltic-trachyandesite compositions. Melt temperatures were derived from MELTS computations and from our thermometry measurements. Finally, redox conditions close to the nickel^nickel oxide (NNO) buffer were found to best fit with the composition and mineralogy of the Yasur^Siwi magmas (see below) and, hence, were used in our calculations.

Parental basalt
For the basalt, accurate pressure estimates are hampered by the scarcity of our data for the basaltic melt inclusions and their variability in both CO 2 (from 574 to 103 ppm) and H 2 O (from 1 to 2 wt %). Moreover, the measured dissolved CO 2 contents may represent lower limits, as mentioned above. For the most CO 2 -rich inclusion, with 1wt % H 2 O, we infer a minimal entrapment pressure of 155 MPa (Fig. 11, and Electronic Appendix 3), corresponding to $6 km lithostatic depth for a mean host-rock density of 2700 kg m À3 . Alternatively, one can place an upper limit on the initial CO 2 content and saturation pressure of the basalt from its pre-eruptive S content of 0·1wt % and the CO 2 /SO 2 ratio of Yasur gas emissions (1·6^2·0; Table 5). From these parameters we compute a total (dissolved and exsolved) amount of $0·22^0·27 wt % CO 2 in the basaltic system, which, if originally dissolved together with 1^2 wt % H 2 O, would imply volatile saturation at $290^330 MPa or $11^12 km lithostatic depth under NNO redox conditions. It should be noted that such an amount is rather modest when compared with that of basaltic arc melts stored at crustal depths, which typically contain 0·2 to 50·5 wt % of dissolved CO 2 but also coexist with an exsolved CO 2 -rich gas phase (e.g. Wallace, 2005). This is again consistent with the idea of minor carbon recycling beneath the southern Vanuatu arc.
MELTS calculations must be considered with caution as they do not include the effect of CO 2 on the stability field of the mineral phases (Ghiorso & Sack, 1995). However, water exerts the main influence on the crystallization path of magmas that are weakly or moderately enriched in carbon dioxide, such as inferred here for the Tanna basalt. Our MELTS computations, performed at between 1150 and 10908C and fO 2 buffered at NNO, show that this basalt, chemically represented by melt inclusions in olivine Fo $80 (TAN 23-18 and 26; Table 4), cannot differentiate into Yasur-type basaltic-trachyandesite if it initially contains more than $1wt % of water. For 2 wt % H 2 O, as measured in the TAN 23-18 inclusion, both isobaric crystal fractionation at 300 MPa and polybaric melt differentiation from 300 to 200 MPa fail to reproduce the Yasur magma composition and mineral paragenesis. At 11001 0908C, the basalt would crystallize for only 20%, with clinopyroxene being prevalent, whereas plagioclase starts crystallizing at low temperature ( 10908C). As a result, Al 2 O 3 drastically increases (up to 19 wt %) whereas CaO decreases in the melt, which is opposite to what is observed in the natural samples (Tables 1 and 2). For a higher initial H 2 O amount, the basalt crystallization rate is lower or even negligible. These conclusions remain valid under redox conditions more oxidizing than NNO. Instead, with 0·8^1wt % of dissolved H 2 O the basalt is able to crystallize extensively (45^30%), forming plagioclase (An 71^68 ), olivine (Fo 75^71 ), clinopyroxene, and a little spinel in the 1110^10808C range, which does correspond to the paragenesis of the poorly crystallized Yasur basaltictrachyandesite (sample TAN 6). Therefore, the basalt feeding the Yasur^Siwi system must be weakly hydrous to differentiate into basaltic-trachyandesite. Depending on its actual dissolved CO 2 content, we infer that this basalt could be emplaced between about 11km and $6 km depth beneath the Siwi caldera, before differentiating into basaltic-trachyandesite.

Yasur^Siwi basaltic-trachyandesites
Our dataset for dissolved CO 2 and H 2 O in olivine-hosted melt inclusions and glass embayments in the Yasur basaltic-trachyandesite is much larger. It indicates polybaric melt entrapment in olivine at total (P CO2 þ P H2O ) pressures ranging from about 130 MPa to 550 MPa ( Fig. 11; Electronic Appendix 3). A similar pressure range, from $110 to 550 MPa, is inferred for the few analogous inclusions from the Siwi sequence (TAN 1 and 17; Electronic Appendix 3). We thus infer that basaltic-trachyandesite, having on average 10% crystals and $1·0^1·3 wt % of dissolved H 2 O, is emplaced and degasses at between $4^5 km and 5$1·8 km depth beneath Siwi caldera. This magma is directly erupted only during the vigorous explosive activity, which produces the distal fallout deposits. Otherwise, during 'standard' Yasur explosive activity the magma experiences an additional late stage of degassing and crystallization, probably at shallow depth, as demonstrated by the high crystal (32%) and low bulk H 2 O contents (0·2^0·3 wt %) of the proximal scoriae. MELTS calculations do reproduce the crystallization extent and mineral paragenesis of these water-poor ejecta, within a temperature range of 1090^11008C, under NNO redox conditions and a total pressure of 550 to 10 MPa (5 1·8 to 0·37 km depth). Plagioclase (An 56^53 ) is the prevalent phenocryst phase ($27^30%), coexisting with $1^1·5% olivine (Fo 70 ), 1·6^2%, spinel, and a negligible fraction of clinopyroxene, as actually observed in these scoriae. Under more oxidizing conditions (NNO 4 þ0·5), we find that the olivine either disappears or is significantly richer in MgO (Fo $74) than measured in our representative TAN 12 sample.
In summary, distal and proximal scoriae from Yasur share comparable bulk-rock compositions and crystallization temperatures, but differ in their crystal and volatile contents, and in their glassy matrix compositions. These samples provide a clear illustration of the prevalent role of water exsolution, at nearly constant temperature, in promoting shallow crystallization. Crystal-rich degassed magma, expelled as bombs and vesicular scoriae during typical Strombolian (or Vulcanian) activity, is actually stored at shallow depths in the upper Yasur volcanic conduits.
We also outline that the relatively high crystallization temperature of the basaltic-trachyandesites, irrespective of their mineral paragenesis and crystal amount, requires a steady and relatively high heat flow. Basalt crystallization supplies a heat flow ÈH ¼ (C p ÁT þ jL), where C p is the melt heat capacity (1300^1484 J kg À1 K À1 ; Bohrson & Spera, 2001), ÁT the temperature decrease, j the fraction of crystals, and L the latent heat of crystallization [(3·54 ·2) Â 10 5 J kg À1 ; Bohrson & Spera, 2001]. A ÁT of $558C is assessed from the difference between the basalt liquidus temperature, calculated from MELTS (1158^11648C for 1wt % dissolved H 2 O), and the measured temperature of basaltic-trachyandesite crystallization. We obtain a crystallization-related heat release of $2·3 Â10 5 W kg À1 basalt, which could contribute to maintain the relatively elevated temperature of the overlying H 2 O-poor basaltictrachyandesites.

Magma degassing processes and budget Magma degassing processes
Our melt inclusion dataset allows us to characterize the degassing behavior of the main volatile species (CO 2 , H 2 O, S and Cl) during magma differentiation and decompression in the above estimated pressure ranges beneath Siwi caldera.
As shown by Fig. 11, during decompression of the basaltic-trachyandesite magma from about 130^110 MPa to 550 MPa carbon dioxide sharply degasses, whereas H 2 O does not vary much from its initial value of $11 ·3 wt % until very shallow depths. Such a pattern strongly suggests open-system degassing conditions (e.g. Newman et al., 2000). It would imply a CO 2 -dominated (60^80 mol %) gas phase, whereas gas emissions are water-dominated. Moreover, if the basaltic-trachyandesite actually derives from $50% crystal fractionation of a basalt containing $0·8^1wt % H 2 O, it should be as twice as rich in water.
The trend depicted in Fig. 11 cannot be an artefact of mixing, in constant proportions, between the basaltictrachyandesite and an H 2 O-poor component (such as recycled degassed magma or, possibly, a residual melt of Siwi ignimbrite). Indeed, a water-poor degassed component should also be depleted in sulfur and partly in chlorine and, while mixing, would affect the distribution of these two species. This possibility is discounted by the variation of S and Cl in the olivine-hosted melt inclusions, which clearly delineates true degassing trends (Fig. 12). As also mentioned above, we have found no textural or mineralogical evidence of mixing between the low-H 2 O feeding basalt and an H 2 O-enriched trachyandesite.
Furthermore, several observations exclude significant H 2 O re-equilibration of the analyzed melt inclusions with the carrier magma (e.g. Danyushevsky et al., 2002). This process, driven by proton diffusion, can occur on short time scales but is limited during rapid ascent of magma from shallow depths (Johnson et al., 2010). This is typically the case for the Yasur scoriaceous products, which experienced high cooling rates: their matrices are perfectly glassy, the morphology of both their olivine crystals and melt inclusions points to crystal nucleation induced by fast decompression events, and, most importantly, their opened glassy embayments in olivine have the same H 2 O content as the sealed inclusions. All these features demonstrate limited water diffusion during magma ascent and emplacement. Accordingly, the H 2 O concentrations of 1^1·3 wt % measured in the olivine-hosted melt inclusions in the Yasur distal scoria lapilli are considered representative for the basaltic-trachyandesite prior to its crystallization and eruption.
Therefore, the lower than expected H 2 O content in the basaltic-trachyandesite would imply water loss either through the conduit walls or via its fluxing by deep CO 2 -rich gas bubbles derived from the underlying and differentiating parental basalt. This process was recently advocated to explain magma dehydration in both basaltic (e.g. Spilliaert et al., 2006;Johnson et al., 2008;Collins et al., 2009) and silicic  systems. Moreover, recent experiments involving re-equilibration of H 2 O-bearing rhyolitic melts with a CO 2 -rich gas phase show that fluxing by even a small quantity of CO 2 may provoke temporal volatile under-saturation of the melt and the release of large proportions of H 2 O (Yoshimura & Nakamura, 2010). Although further data are needed to confirm this interpretation, it remains likely that H 2 O exsolution and loss (possibly $1wt %) exerts the dominant control upon the genesis and crystallization path of the basaltic-trachyandesitic magma beneath Siwi caldera. In that context, we infer that the gas phase flowing through this magma will be H 2 O-dominated (up to $90 mol %), instead of CO 2 -dominated, as actually observed in the surface emissions.
Focusing on the variations in H 2 O, S and Cl during magma differentiation and decompression, which are the best documented, we identify a multi-stage degassing process during which these three species behave differently (Table 6). Magma differentiation is tracked by K 2 O, the fraction of residual melt (f) being given by the K 2 O 0 / K 2 O ratio (where K 2 O 0 is the concentration of the melt inclusions representative of the parental basalt). During the first degassing stage, while the basalt evolves by crystal fractionation towards the basaltic-trachyandesite (f ¼ 0·46), $40% of the total H 2 O and 63% of the total sulfur become exsolved and lost (Fig. 12a, Stage I in Table 6). In contrast, Cl loss is negligible as the Cl/K 2 O (0·046) remains unchanged (Fig. 12b). Chlorine exolution is recorded by the large sealed melt inclusions in the Yasur distal scoriae (Stage II in Table 6), and by the basaltic-trachyandesitic melts (TAN 1 and 17) from the Siwi ignimbrite sequence (Fig. 12). During this stage of concomitant Cl and S degassing, crystallization is almost negligible. Finally, the ultimate degassing stage (Stage III in Table 6) is represented by the evolved (f ¼ 0·35) trachyandesitic melt inclusions and matrix glasses of the proximal ejecta, which are strongly depleted in both H 2 O ( 0·4 wt %) and S (580 ppm) and extensively impoverished in Cl ($560^460 ppm). Taking account of their extent of crystallization (32%), these ejecta are twice as degassed in chlorine as the magma that is erupted and quenched during periods of more vigorous explosive activity. This difference cannot be ascribed to the limited variation of the [(Al þ Na þ Ca þ Mg)/Si] mole ratio of the melt inclusions (0·35 and 0·45 in proximal and distal samples, respectively). Rather, it indicates a strong control of diffusion kinetics on late-stage Cl degassing, while gas bubbles and/or slugs are flowing across the ponding and crystallizing upper magma column, with chlorine being efficiently partitioned into the H 2 O-rich vapor (Webster et al., 1999).
Simple mass-balance calculations show that, during this multi-stage degassing process, the S/Cl molar ratio of the equilibrium gas phase will vary noticeably, from an average of 5·9 when magma degasses in the $100^50 MPa Fig. 12. Log^log diagrams of S (a) and Cl (b) evolution vs K 2 O in Yasur and Siwi melt inclusions and glassy matrices. f, extent of melt evolution derived from the K 2 O/K 2 O 0 ratio. Sulfur is exsolved during a first stage (Stage I) of basalt differentiation, whereas Cl purely tracks the degassing of the basaltic trachyandesite (Stages II and III). The star represents the bulk volatile content of shallow crystal-rich basaltic-trachyandesite, corrected for the amount of crystals (32%), which demonstrates their marked depletion in Cl. Excluding these latter samples, the variation of the melt inclusion S and Cl contents in samples TAN 1 and TAN 17 (Siwi sequence) overlaps that measured in Yasur samples. D-and P-scoria refer to scoriae from distal and proximal deposits, respectively. pressure interval to $1·0 during the final (shallow) degassing stage. Therefore, deeply derived gas slugs driving the recurrent Yasur explosions should display high S/Cl ratios, as actually observed (Oppenheimer et al., 2006;Allard et al., in preparation), whereas purely residual degassing of the upper magma column should be characterized by a low ( 1) S/Cl ratio. The time-averaged S/Cl ratio of 2·4 we measured for both passive and explosive crater gas emissions in October 2007 thus requires a mixed contribution of shallow ( 1) and deeper-derived (!5·9) magmatic gas. It should be noted that the steady S/Cl ratio of $2·0 for purely passive emissions, as measured in January 2005 (Oppenheimer et al., 2006), also requires a two-component gas mixture, but with enhanced contribution of the shallow gas phase. These observations further support a differential gas transfer (i.e. open-system degassing) across the magma plumbing system, as already suggested by the CO 2^H2 O degassing pattern in Fig. 11.

Magma degassing budget
As a whole, we compute a bulk loss of !94% for H 2 O and S, and of 70% for Cl during magma degassing beneath Siwi caldera (Table 6). Combined with the Yasur gas fluxes, the degassed mass fraction of each volatile species allows us to quantify the amounts of degassing magma that sustain the current eruptive activity (Table 7). From the pre-eruptive S content of the basalt and the time-averaged SO 2 flux of 680 tons d À1 representative for the last 5 years of activity (Bani & Lardy, 2007), we infer a bulk magma degassing rate of 4·1 Â10 3 kg s À1 , or $5 Â10 7 m 3 a À1 , for a mean density of 2650 kg m À3 . Chlorine, which does not exsolve during basalt differentiation, purely tracks degassing of the basaltictrachyandesite between about 5 and 4 km depth and the surface. From its bulk loss (Table 6) and the mean HCl emission rate (165 tons d À1 ), we derive a degassing rate of 2·1 Â10 3 kg s À1 , or 2·6 Â10 7 m 3 a À1 , of basaltictrachyandesite. Hence, the degassing of nearly equivalent amounts of basalt and basaltic-trachyandesite sustains the overall volcanic gas output. Considering the uncertainties involved in these calculations, we find a good agreement between the proportion inferred from volatile species and the rate ($50%) of basalt fractionation needed to generate the basaltic-trachyandesite constrained from the major and trace element data. This implies that, even though the analyzed basalt is not a true primitive magma, one should  not significantly underestimate its pre-eruptive sulfur content and, hence, the computed bulk magma degassing rate. Because about 60% of the total sulfur and thus of the SO 2 flux from Yasur derive from basalt degassing, a differential flow of S-bearing gas bubbles must occur through the overlying basaltic-trachyandesite. This same conclusion applies to deeper exsolved CO 2 but also to H 2 O, 1wt % of which may be exsolved during basalt differentiation. A differential transfer of gas bubbles nucleated in the basalt and already enriched in CO 2 , H 2 O and SO 2 is indeed consistent with the open-type degassing process suggested by the P-related CO 2^H2 O concentrations in the Yasur products (Fig. 11). We note here that the H 2 O/SO 2 molar ratio inferred for bulk degassing of the basalt ($17; Table 6) is much lower than the mean H 2 O/SO 2 ratio (70; range 50^90) measured in Yasur gas emissions. However, we have seen that the latter are affected by a large uncertainty, owing to correction for ambient air background. Moreover, we do not know whether the emitted water vapor is wholly magmatic in origin or may partly derive from entrainment of external steam of meteoric or/and hydrothermal origin into the shallow plumbing system of Yasur. This latter possibility is compatible with the observation of magma^water interactions in the Yasur volcanic deposits, with the presence of a widespread hydrothermal system beneath the Yenkahe horst (Gauthier et al., 2001, in preparation), and with the heavy local rainfall recharge and highly permeable nature of the Yasur volcanic cone.

Relationship between magma supply and the Yenkahe block resurgence
Coral reef terraces on the Yenkahe resurgent block record a sequence of uplifts with two sharp events dated at AD 1002 and 1878 (Chen et al., 1995). The 10^15 m uplift that occurred in 1878 was consecutive with a local earthquake, not felt 20 km away, and was not registered by coral reefs standing a few hundred meters from the uplifted area, as noted by Chen et al. (1995). Those workers thus concluded that the Yenkahe resurgence was strictly confined within the Siwi caldera and was probably due to magma emplacement or movement at shallow depth. From the geometry of the Yenkahe resurgent block and its average uplift of 15·6 cm a À1 during the last millenium (Chen et al., 1995), we compute a cumulative volume uplift of 2·8 Â10 6 m 3 a À1 , or 2·8 km 3 in 1000 years.
It is of interest to compare this cumulative deformation with the magma degassing rates during the same interval. To maintain the steady eruptive activity of Yasur and the great homogeneity of its products for 1000 years at least, the magma degassing rates must correspond to supply rates, and present-day average values can reasonably be extrapolated to the past. A bulk influx of $5 Â10 7 m 3 a À1 of basaltic magma, as deduced from the time-averaged SO 2 flux and the S content of basaltic melt inclusions, would correspond to a cumulative basaltic supply of $50 km 3 in 1000 years. Our data for chlorine degassing show that about a half ($26 km 3 ) of this supply should have differentiated into the basaltic-trachyandesite that has sustained the Yasur eruptive activity. The proportion of this latter magma extruded in 1000 years is not easy to assess, but can be roughly estimated from the volumes of the Yasur volcanic cone and Yasur-related tephra deposits. The Yasur cone, made of ash, coarse lapilli, bombs and lava, has a base diameter of 1·5 km and a height of 210 m, which, given its morphology, implies a volume of 50·2 km 3 . The Yasur deposits derive from the standard eruptive activity plus occasional more powerful events, about one every 20^30 years, whose explosivity index (VEI) is at most three (Carney & MacFarlane, 1979). For these stronger explosive events, we estimate a maximum production of 0·3^0·4 km 3 of dense magma. For the standard activity, we use the emission rate of volcanic ash produced by Yasur explosions and transported downwind of the crater, estimated as 5 Â10 5 kg d À1 from filter-pack sampling of the particulate matter in October 2007 and the SO 2 plume flux (Allard et al., in preparation). For an ash particle density of between 1500 and 2000 kg m À3 , the corresponding volume of dense erupted magma approximates $0·4^0·6 km 3 in 1000 years. The lava bomb production, roughly estimated as 200 m 3 d À1 (Gauthier et al., 2001), would correspond to a minor dense output of 0·05 km 3 in 1000 years, taking account of the high vesicularity of most of these ejecta. As a whole, therefore, only about 1km 3 of magma may have been erupted from Yasur in that time interval, which means that 95% ($25 km 3 ) of the supplied (degassed) basaltic-trachyandesite should have accumulated beneath the Siwi caldera complex. Such a high proportion of unerupted to erupted magma is typical for open-conduit basaltic volcanoes, such as Stromboli (Allard et al., 2008).
We thus find that, in 1000 years, the volume of degassed basaltic-trachyandesite accumulated beneath the Siwi caldera is about 10 times larger than the cumulative deformation of the Yenkahe block and, therefore, could easily account for its resurgence. The storage of $25 km 3 of degassed and hence dense molten magma beneath the 5·5 km long by 3 km wide Yenkahe horst could have formed an accreting sill-like body of 1·5 km height, crystallizing on its margins and conductively heating the host rocks and the overlying hydrothermal system. The low magmatic CO 2 content of local thermal waters (Gauthier et al., 2001, in preparation) and soil gas emanations (Chaput et al., 2009;P. Allard et al., unpublished data, 2007), attributed to the extensive hydrothermal alteration and sealing of volcanic rocks forming the Yenkahe block (Chaput et al., 2009), would also be consistent with relatively shallow emplacement of the roof of this degassed magma body. The intrusive body, with almost zero eruptive potential, must be connected to the deeper feeding system by some dyke-like pathways that allow gas bubbles and batches of the undegassed basaltic-trachyandesite to rise and sustain the Yasur eruptive activity. The Yasur vent can thus be viewed as a pipe linked to the accreting magma body, through which most of the gas supplied to the system escapes separately (Fig. 13). GPS measurements made in 2002^2005 have revealed short-term episodes of horizontal deformation and rapid inflation then deflation of the Yenkahe horst, reaching an amplitude of 17 cm over a few months and associated with a temporary increase then a decrease in Yasur activity (Erre, 2005). Such rapid and reversed deformations strongly suggest transient pressuring of the magma body owing to an enhanced influx of magmatic gases, rather than a magma intrusion. Temporal variations in magmatic gas influx from either the underplating basalt or the basaltic-trachyandesite, or both, could explain the observed oscillations in Yasur's eruptive regime (Erre, 2005) and SO 2 emission rate (Bani & Lardy, 2007). The associated episodes of rapid inflation and deflation of the Yenkahe horst detected with GPS, possibly amplified by the shallow hydrothermal system, indicate an almost elastic reponse of the medium to these short-term variations in gas supply.
In the long term, however, the accumulated uplift and extension (graben) of the Yenkahe block, with episodes of sharp co-seismic uplift, require magma accumulation with occasional brittle fracturing of the caldera floor. Rock fracturing and tensile failure in volcanic systems are known to depend on several parameters, such as the strength of the volcanic rocks, the deformation mode (elastic or viscoelastic), and the mechanical response of the magma reservoir to its pressurization (e.g. Simakin & Ghassemi, 2010). It has to be stressed that the sudden block faulting and large co-seismic uplift that affected the eastern part of the Yenkahe block in 1878 (Chen et al., 1995) was not accompanied or followed by a major change in Yasur activity. Such a decoupling between violent co-seismic uplift and volcanic activity negates a sudden magma recharge as the trigger of that event. Rather, it is compatible with a sudden mechanical release of accumulated stress, facilitated by previous faulting of a weakened structure. Hence, we propose that Yenkahe block Fig. 13. Schematic representation, not to scale, of the magmatic plumbing system beneath Siwi caldera. Both gas and melt inclusion data lead us to propose that $25 km 3 of degassed basaltic-trachyandesite may have accumulated beneath the Yasur^Siwi volcanic complex during the last 1000 years, which is one order of magnitude larger than the accumulated deformation of the Yenkahe resurgent block over the same time interval. (Note that these relative proportions are not respected in our illustration.) resurgence in the past millennium could result from both continuous and discontinuous (co-seismic) accommodation of the mechanical stress generated by the gradual storage of degassed basaltic-trachyandesite magma, at a mean rate of $0·025 km 3 a À1 , beneath the Siwi caldera. We note that a comparable magma and gas supply rate (0·02 km 3 a À1 ) was inferred to explain the resurgent dome inflation (0·04 m a À1 ) in Long Valley caldera between 1978 and 1999 (Hill, 2006), despite the much larger dimensions of that caldera. The respective influence of magma intrusion, gas supply and hydrothermal systems on ground deformation in calderas, however, remains strongly debated. We emphasize that the Yasur^Siwi volcanic complex, with its long-term and fast deformation of the caldera floor, short-term deflation^inflation phases most probably caused by magmatic gas motion, and persistent eruptive activity, is a most suitable site to address these questions quantitatively.

C O N C L U S I O N S
We provide here new data on the petrology of the Yasur^Siwi volcanic products, the first results on their pre-eruptive volatile contents, and the first complete budget for Yasur gas emissions. Together, the results allow us to quantify the rates of magma and volatile supply to this volcanic complex during a post-caldera stage and to propose initially a structure of its plumbing system. We outline the following main observations and conclusions.
(1) The geochemistry of Yasur^Siwi bulk products and of their melt inclusions, together with literature data for associated basalts, strongly suggests the long-term feeding of Tanna volcanism by a homogeneous arc basaltic magma. This conclusion could be reinforced by further study of the Tanna basalts themselves.
(2) Yasur basaltic-trachyandesites derive from $50% crystal fractionation of a parental basalt moderately enriched in volatiles, and according to their dissolved H 2 O and CO 2 contents, should be emplaced between 5^4 km depth and the surface. Their constant bulk composition and mineralogy during the last 1400 years, as well as their high mean temperature (1107 AE158C), imply a steady regime of crystallization. Their early dehydration and prevalent open-system degassing generate a gas phase gradually richer in H 2 O, S and then Cl upon decompression.
(3) Yasur magmas erupted during normal and more powerful explosive activity share a comparable bulk composition and crystallization temperature, but noticeably differ in their crystal contents, volatile abundances and glassy matrix compositions. These features indicate a prevalent role for water exsolution, at nearly constant temperature, in promoting shallow crystallization of the magma filling the upper Yasur conduits, which is represented by light scoriae and vesicular to dense bombs, rich in crystals, thrown out on the upper volcanic cone. (4) Yasur volcano emits H 2 O-rich (97 mol %) magmatic gas at an average rate of 13·4 Â10 3 tons d À1 , which makes it a typical arc emitter. The amount of feeding basaltic magma needed to sustain this gas output, computed from its initial S content and the time-averaged SO 2 flux (680 tons d À1 ), averages 4·1 Â10 3 kg s À1 , nearly half of which is differentiated into basaltic-trachyandesite. This implies the development of a large deep magma reservoir able to accommodate the accumulated volume of basalt over the last 1000 years. Basalt crystallization would supply a heat flow of $1MW to the overlying basaltictrachyandesite, thereby contributing to maintain the relatively elevated temperature of this otherwise H 2 O-poor magma. (5) The time-averaged HCl flux of 165 tons d À1 is essentially supplied by the degassing of basaltictrachyandesite at a rate of 0·026 km 3 a À1 , most (495%) of which remains unerupted. The respective amounts of degassing basalt and basaltictrachyandesite are in good agreement with independent estimates from petrological data. (6) Over the last 1000 years, $25 km 3 of basaltictrachyandesite would have degassed without erupting and, therefore, been stored in an accreting intrusive body beneath Siwi caldera. This amount is one order of magnitude larger than the accumulated deformation of the Yenkahe resurgent block. We thus suggest that ground deformation of the caldera floor could result from both continuous and discontinuous (co-seismic) mechanical relaxation of the stress induced by this accumulation of unerupted degassed magma.

AC K N O W L E D G E M E N T S
Our fieldwork greatly benefited from local support from B. Pelletier (IRD), the French Embassy in Port-Vila and Air Vanuatu. We thank D. Edouard and A. Beaumais for their involvement in data acquisition, P. Pantani for graphical assistance, D. P. Hill and A. Hirn for useful discussions, C. Erre and M. Lardy for providing us with GPS data on Yenkahe horst, and B. Scott for providing us with the New Zealand Geological Survey 1988 report. Reviews by D. Pyle, P. Ruprecht, M. Humphreys and J. Stix greatly helped in improving the paper.

F U N D I N G
This work was supported by ANR contracts ANR-06-CATT-02 Arc-Vanuatu and ANR-06-CATT-012 Volgaspec. Field missions of A.B. and A.A. were funded by the INGV-Pisa (Italy), and the Universita' di Palermo (Italy), respectively. This paper is IPGP Contribution 3133.

S U P P L E M E N TA RY DATA
Supplementary data for this paper are available at Journal of Petrology online.