Constraints on the behaviour and content of 1 volatiles in Galápagos magmas from melt 2 inclusions and nominally anhydrous minerals 3

12 Despite their relatively low concentration in most oceanic basalts, volatile species (e.g. H2O, CO2 and 13 S) have a disproportionately large influence on a wide range of mantle and magmatic processes. 14 However, constraining the concentration of H2O (and other volatiles) in basaltic magmas is not 15 straightforward as submarine glass analyses are influenced by assimilation of hydrothermal brines, 16 and the melt inclusion record is often reset by post-entrapment processes. Nevertheless, in this 17 study we show that it is possible to reconstruct a detailed history of the volatile content of basaltic 18 magmas through integration of multiple discreet volatile records and careful consideration of 19 secondary processes. We present new analyses of volatiles in olivine-hosted melt inclusions, melt 20 embayments and nominally anhydrous minerals (NAMS, clinopyroxene and orthopyroxene) found in 21 basalts erupted on Floreana Island in the south-eastern Galápagos Archipelago. Our results indicate 22 that the Floreana magmas, which are characterised by the most radiogenic Pb and Sr isotope 23 signatures in the Galápagos Archipelago, contain H2O concentrations between 0.4 and 0.8 wt% (at a 24 melt Mg# of 0.65, where Mg# = Mg/(Mg + Fe) molar). These are marginally greater than the H2O 25 contents of magmas beneath Fernandina in the western Galápagos Archipelago (cf. 0.2–0.7 wt% H2O 26 at Mg# = 0.65). While the volatile content of magmas from the western archipelago follow trends 27 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. defined by concurrent mixing and crystallisation, NAMs from Floreana reveal the presence of rare, 28 volatile-rich magmas (~2 wt% H2O) that form as a consequence of reactive porous flow in mush29 dominated magmatic systems beneath the south-eastern Galápagos. Furthermore, the Floreana 30 magmas have similar H2O/light Rare Earth Element ratios to basalts from the western Galápagos but 31 contain F/Nd and Cl/K ratios that are ~2 – 3 times greater, indicating that the mantle source of the 32 Floreana lavas might represent an important halogen reservoir in the Galápagos mantle plume. 33

The influence of low-pressure degassing on the H2O and S contents of silicate melts can be mitigated 48 by analysing the glassy exteriors of lava flows that are erupted under 100s to 1000s of metres of 49 water, as the pressure of the overlying water column prevents significant loss of H2O and S to the 50 vapour phase (Dixon, 1997 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. analyses (Gleeson et  study have mean Ba and Nb contents of ~310 ppm and ~25 ppm, respectively, relative to Fernandina 289 glass contents of Ba ~91 ppm and Nb ~23 ppm -data from Peterson et al., 2017). In addition, the 290 Floreana glasses have concave-up rare-earth element (REE) patterns, with high light REE/middle REE 291 ratios (e.g. [La/Sm]n) and relatively low middle REE/heavy REE ratios (e.g. [Sm/Yb]n) compared to 292 basalts from the western Galápagos, in agreement with previous geochemical data from Floreana 293 (Fig. 4). Notably, the deviation between the PEC-corrected trace element (and volatile element) data 294 is typically smaller than the error associated with SIMS analysis, so the measured values are used in 295 all discussions below. 296 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.

Volatile elements 297
Substantial heterogeneity is observed in the volatile element concentrations of the Floreana glasses, 298 with embayments containing lower concentrations of S and H2O (~100-1000 ppm and 0.05 -0.35 299 wt%, respectively) than the melt inclusions (~1250 ppm S and 0.54 -0.77 wt% H2O; Fig. 5). A positive 300 correlation is observed between the H2O and S concentrations of the Floreana glasses ( Fig. 5 & 6). 301 The CO2 concentrations measured in the melt inclusions range from ~700 to ~8800 ppm, whereas 302 the CO2 concentration of the melt embayments are consistently ≲2000 ppm (Fig. 5). However, correlations are observed between Cl and other highly incompatible trace elements, such 309 as Ba, Nb and K (Fig. 6). This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.

347
We collected H2O data from five clinopyroxene and one orthopyroxene crystals from scoria sample 348 17MMSG16 (28 individual analyses in total), as well as clinopyroxene crystals separated from 3 349 wehrlite (7 crystals and 17 individual analyses; 17MMSG02b, 17MMSG02c, 17MMSG03a), 2 dunite 350 (3 crystals and 7 individual analyses; 17MMSG04c, 17MMSG04f), and 2 gabbro xenoliths (6 crystals 351 and 17 individual analyses; 17MMSG03b, 17MMSG04b). SIMS analyses were carried out on the core 352 and rim of each crystal (from both the scoria and xenolith samples) to characterise the variability in 353 H2O concentrations across an individual grain. In addition, a small number of core-to-rim profiles 354 were also collected on the scoria clinopyroxene crystals (  In crystals where multiple core analyses were performed, relatively homogeneous core H2O 368 concentrations were observed (<10% variability) with lower H2O contents, by ~10-40%, at their rims 369 (Fig. 2). While this suggests that low pressure degassing has caused diffusive loss of H2O from crystal 370 rims, the relatively homogeneous nature of the crystal cores indicates that diffusive loss of H2O 371 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.
during low pressure degassing has a very small influence on these core H2O concentrations. 372 Equilibrium melt H2O concentrations calculated from the analyses of the pyroxene cores are typically 373 between 0.30 and 0.80 wt% but extend up to ~1.6 wt% (partition coefficients calculated using the T-  The H2O concentrations measured within clinopyroxene crystals from each of the two gabbroic 388 xenoliths analysed are relatively constant for each sample (209 ±47 ppm (n=7) and 151 ±29 ppm 389 (n=10) -2σ variation). Calculation of equilibrium melt H2O concentrations indicates that these 390 xenoliths last equilibrated with melts containing 1.19 ±0.13 and 0.64 ±0.12 wt% H2O, respectively. This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.

Degassing and diffusive controls on H2O, S and CO2 416
Water is more soluble than CO2 in basaltic melts and OIB magmas are therefore unlikely to degas 417 substantial amounts of H2O until they reach very low pressures, likely within the upper ~1 km of 418 crust (Dixon, 1997 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. The results of our simple diffusion models provide important insights into the volatile content of 455 magmas beneath Floreana. For example, they indicate that the H2O vs S trend in the Floreana 456 embayments can be reproduced when the initial H2O content of the system is between 0.55 and 457 0.75 wt%, corresponding to the range in H2O contents measured in our melt inclusions (Fig. 5). In 458 addition, the diffusion models also recreate the H2O vs CO2 systematics of the Floreana melt 459 embayments (Fig. 5). 460 We do not use our diffusion models to estimate the decompression rate of the Floreana magmas, as 461 we were not able to collect transects along individual embayments (owing to their narrow width and 462 the relatively large spot size of our analyses; Lloyd et al., 2014), and so our measurements typically 463 represent a single analysis from each embayment. In addition, the embayments analysed in this 464 study display a range of morphologies, which will influence the diffusion of volatile species out of the 465 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. basaltic magmas in different parts of the present day magmatic system (Fig. 8). 511 The yellow field on the right-hand axis of Figure 8 shows a kernel density distribution of melt H2O 512 contents in equilibrium with NAM analyses that are uninfluenced by diffusive loss of H2O during low 513 pressure degassing (i.e., excluding rim analyses that return H2O contents >>10% lower than the 514 respective crystal core) and are derived from the present-day Floreana magmatic system (i.e. those 515 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. that show no chemical affinity to the gabbroic xenoliths). The kernel density distribution has a 516 primary peak at ~0.4-0.8 wt% H2O, with a long tail to high equilibrium-melt H2O contents and a 517 secondary peak at ~1.5 wt% H2O. In addition, a kernel density distribution was also constructed for 518 the melt H2O concentrations predicted from the Floreana whole-rock data, using the measured La 519 concentrations (Harpp et al., 2014) and an assumed melt H2O/La ratio of 350 (grey field in Fig. 8). 520 While we acknowledge that there might be small differences in the true H2O/La of the Floreana 521 basalts, the kernel density distribution of melt H2O concentrations predicted from these whole-rock 522 analyses display several similarities with that constructed for the NAMs from the present-day 523 magmatic system, with a primary peak at ~0.4-0.8 wt% H2O and a tail to higher H2O contents (1-2 524 wt%; Fig. 8). 525 The overlap between the primary peaks in the two kernel density distributions validates our 526 calculated equilibrium melt H2O concentrations from the NAMs and likely records the typical range 527 of pre-eruptive melt H2O concentrations in the present-day Floreana magmatic system (0.4-0.8 wt% 528 H2O). This is further supported by the similarity between the location of the kernel density 529 distribution primary peaks and the H2O concentrations measured in our Floreana melt inclusions 530 (0.54-0.77 wt%). However, the subsidiary peak in the NAM equilibrium-melt kernel density 531 distribution and melt H2O concentrations predicted from the whole-rock data record substantially 532 higher melt H2O contents than our Floreana melt inclusions or embayments (Fig. 8). 533 There are two potential origins for the anomalously H2O-rich (and trace element enriched) melts 534 identified from the Floreana NAMs and whole-rock data: (i) they formed from low-fraction mantle 535 melts generated at the base of the melting region and have incompletely mixed with H2O-poor melts 536 produced at shallower depths; or (ii) they derive from magmas that have undergone chemical 537 enrichment via reactive porous flow (that is, disequilibrium melt-mush reaction during melt 538 transport) or in-situ crystallisation in highly-crystalline storage regions beneath Floreana (i.e. where 539

H2O and La act as incompatible trace components). 540
This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.
To distinguish between these two different possibilities, we can consider the trace element and 541 isotopic composition of the Floreana basalts (including the enriched samples with estimated H2O 542 concentrations >0.8 wt%). Notably, there is greater heterogeneity in the trace element and isotopic 543 composition of the Floreana basalts than at any other location in the Galápagos Archipelago, 544 indicating the presence of a heterogeneous mantle source beneath the island (Harpp et al., 2014). 545 Importantly, any differences in the lithological properties and/or volatile content of the mantle 546 components involved in the genesis of the Floreana lavas will cause offsets in their solidus 547 temperatures and melt productivities. Therefore, variations in the mean melt fraction of the 548 Floreana mantle source is expected to drive changes in the isotopic composition of the resulting 549 basalts (by influencing the relative contribution of melts from the different, isotopically distinct, 550 mantle components) as well as incompatible trace element ratios such as [La/Sm]n (where n 551 indicates normalisation to the primitive mantle composition of Sun and McDonough, 1989). In fact, 552 when we consider the available isotope and trace element data from the Floreana basalts with 553 estimated H2O concentrations <0.8 wt%, we find that a statistically significant correlation exists 554 between [La/Sm]n and 206 Pb/ 204 Pb (Fig. 9). This correlation, however, does not extend to the highly 555 enriched Floreana basalts (i.e., those that have estimated H2O contents >0.8 wt%), which are 556 isotopically indistinguishable from the rest of the Floreana lavas. We therefore suggest that their 557 anomalous trace element signature is a consequence of crustal processing in magmatic mush zones, 558 rather than variations in the extent of mantle melting ( In addition, there is substantial geochemical and textural evidence preserved in the Floreana 561 xenoliths to support the interpretation that reactive porous flow is an important geochemical 562 process in mush zones beneath the island. For example, trace element enrichment in the wehrlite 563 xenoliths cannot be explained through simple fractional crystallisation, but can be explained by 564 models that account for melt-mush reaction during reactive porous flow (Gleeson et al., 2020a). 565 Additionally, the enrichment in the trace element composition of the cumulate clinopyroxenes are 566 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.
commonly more extreme at the crystal rims compared to their cores, consistent with trace element 567 enrichment originating through magma processing in a mush rather than initial crystallisation from 568 anomalously enriched mantle melts (Gleeson et al. 2020a). Finally, all clinopyroxene crystals that 569 have anomalously high equilibrium-melt H2O contents also have incompatible trace element 570 signatures that are too enriched to be in equilibrium with the majority of the erupted Floreana 571 basalts (Gleeson et al., 2020a), and, as a result, we favour the reactive porous flow hypothesis 572 presented above.  (Fig. 1). These basalts have high 3 He/ 4 He ratios that are characteristic of Fernandina 591 This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.
magmas (Harpp and White, 2001;Peterson et al. 2017). In addition, olivine-hosted melt inclusions 592 from a nearby submarine lava flow on the western margin of Fernandina have also been analysed 593 for their volatile contents (Fig. 1) This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.

Fernandina submarine matrix glasses of Peterson et al. (2017) show a linear correlation between 617
H2O/Cl and K/Cl, which indicates that the assimilated brine component has a H2O/Cl ratio of ~17.1 618 (Fig. 10). Therefore, using the calculated amount of Cl assimilated by each sample, and the H2O/Cl 619 ratio of the assimilated component, it is possible to calculate the amount of H2O that has been 620 assimilated. 621 The kernel density distribution of the uncorrected Fernandina matrix glass H2O/La ratio is centred at 622 ~500, with a long tail extending to higher values (>800; Fig. 10). However, using the method above, This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta. This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.
but more data is required to confirm this hypothesis. As a result, the high H2O concentrations 668 measured in gabbro 17MMSG03b could indicate that reactive porous flow operates beneath 669 present-day volcanoes in the western archipelago, but is yet to be identified in erupted products 670 (signatures of reactive porous flow in erupted magmas may include an over-enrichment in 671 incompatible trace elements with increasing differentiation; Lissenberg and MacLeod, 2016); the 672 ancient Floreana magmatic system was distinct from the magmatic systems currently underlying the 673 plume-proximal western volcanoes (Harpp and Geist, 2018); or that the H2O content of the gabbroic 674 xenoliths is reset by interaction with more recent Floreana magmas. 675

676
Our new data from the Floreana basalts and xenoliths, and re-evaluation of published data from 677 Fernandina, indicate that these volcanoes have distinct volatile histories: the H2O contents of the 678 Fernandina basalts are primarily controlled by fractional crystallisation and magma mixing, whereas 679 some Floreana basalts are influenced by H2O-rich magmas generated by reactive porous flow within 680 crystal-rich sub-volcanic mush zones. To directly compare the volatile contents of magmas from 681 these locations, we reconstruct initial melt inclusion and whole-rock H2O concentrations (i.e. prior to 682 alteration by secondary processes) using their measured La concentrations and the characteristic 683 H2O/La ratio of each magmatic system as determined above. 684 Comparing reconstructed initial melt inclusion and whole-rock H2O concentrations indicates that the 685 Floreana magmas typically have slightly higher H2O contents than the Fernandina magmas at an 686 equivalent melt Mg# (Fig. 8)  This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.
do not match those expected from melting of a primitive mantle component (Farley et   This manuscript represents a pre-print of a manuscript that has undergone peer-review and has been accepted for publication with Geochimica et Cosmochimica Acta.
Nevertheless, in the absence of accurate source trace element estimates, we can consider the 742 H2O/REE systematics of the Galápagos basalts to determine the relative 'hydration' of the different 743 Galápagos source components with respect to their trace element composition. Specifically, we can 744 use our new data, alongside published volatile analyses from basalts across the archipelago and Overall, these data indicate that the H2O/La ratio of the enriched mantle components in the 757 Galápagos plume are lower than that of the DGM. This 'dehydration' signature indicates that the 758 H2O enr /H2O DGM ratio of the Galápagos mantle plume (i.e. the concentration of H2O in the enriched 759 source components relative to the DGM) is smaller than La enr /La DGM . Notably, the 'dehydration' 760 signature is also observed when we consider the H2O/Ce ratio of the enriched Galápagos basalts 761 instead of H2O/La. In particular, the H2O/Ce ratio of basalts from Fernandina and Floreana are <200 762 (based on our analysis above), whereas depleted plume-influenced GSC basalts have H2O/Ce ratios 763 between 200 and 250 ( Fig. 11; . 764 The origin of these differences in the H2O/REE systematics of the Galápagos plume basalts is 765 uncertain, but it is notable that this is not the only region globally where such variations have been