The copyright line for this article was changed on 6 May 2015 after original online publication.
 A combined study of mapping, observational, age constraint, and geochemical data at the summit of Axial Seamount, Juan de Fuca Ridge, has revealed its recent petrological history. Multiple basalt types erupted at the summit in a time sequence. At least three different magma batches have been present beneath the Axial Summit caldera during the last millennium, each with a range in differentiation. The first, prior to 1100 CE, was compositionally diverse, dominantly aphyric T-MORB. The second, from ∼1220 to 1300 CE, was dominantly plagioclase-phyric, more mafic N-MORB erupted mostly in the central portion of the caldera. Since ∼1400 CE, lavas have been more differentiated, and nearly aphyric T-MORB mostly erupted in the caldera's rift zones. Parental magmas vary subtly due to small coupled differences in the degree of melting and sources, but all share a uniform differentiation trend indicating pooling at similar depths. Thus, melts percolate through melt-rich lenses that remain partially isolated in space and/or time. Centennial magmatic timescales at Axial Seamount are similar to those for fast spreading ridge segments. The fluctuation between aphyric and plagioclase-phyric lava likely reflects different pathways or velocities of melt migration.
 Oceanic spreading ridges account for more than 70% of the annual magmatic output of Earth, are essential in defining global energy, material, and nutritional fluxes, and have important implications for marine ecosystems [Kelley et al., 2002]. A great deal has been learned about ridge tectonic and magmatic processes since the advent of seafloor observation. Increasingly detailed geological, seismological, geochemical, and chronological studies on various aspects of ridge system evolution have revealed insights about eruption recurrence and the scales of magmatic episodicity, melt lens segmentation, ridge construction, and feedbacks between thermal and morphological structure of the crust. An increasing number of studies indicate that melt production, transport, and evolution vary over time scales similar to that of eruption recurrence intervals [e.g., Bergmanis et al., 2007; Goss et al., 2010; Rubin et al., 2005; Waters et al., 2013]. A fundamental understanding of fine-scale crustal accretion [e.g., Perfit and Chadwick Jr, 1998] therefore requires a high temporally and spatially resolved analysis of variance within the mosaic of lava sequences at a ridge axis [Rubin et al., 2009]. Intermediate spreading-rate ridges such as the Juan de Fuca Ridge are well suited to high fidelity study of the variability in crustal accretion because they are the most temporally and spatially sensitive to variations in magmatic and tectonic conditions [Canales et al., 2005]. Moreover, resolving the fine-scale patterns in magmatic throughput and volcanism at oceanic ridges is important for interdisciplinary studies of complex and interlinked biological, hydrothermal, and geological processes, for the multidecadal cabled observatory at the summit of Axial Seamount, and for comparative studies of the scales of magmatic variability globally at volcanic arcs and ocean-island basalts.
 In this study, we reconstruct the recent petrological history of Axial Seamount, a large submarine volcano on the intermediate-rate Juan de Fuca mid-ocean ridge (MOR) (Figure 1). We aim to better understand the behavioral patterns of magma reservoirs by investigating the record of short-term variability in the basaltic output of a single ridge-centered submarine volcano. To do so, we integrate new detailed geochemical analyses with 1 m resolution seafloor mapping, ROV observations, and age constraint data [Clague et al., 2011; Caress et al., 2012; Clague et al., 2013] from the summit region of Axial Seamount where the magmatic flux and recent volcanic activity are greatest. We present evidence for petrological variability in lavas over hundreds of meters and hundreds of years that reflects minor variability in source composition, magma parentage, and episodicity in melt production and focusing.
 The ∼500 km long intermediate spreading-rate (5–6 cm/a full spreading rate) Juan de Fuca Ridge is composed of six discrete tectonomagmatic segments, from north to south: Endeavour, Northern Symmetric/Cobb, CoAxial, Axial, Vance, and Cleft. Spreading at Axial is accommodated beneath its summit caldera and its north and south rift zones [Zonenshain et al., 1989; Embley et al., 1990, 2000]. Northwesterly absolute plate migration led to intersection of the ridge with the Cobb hotspot at <0.5 Ma and the formation of Axial Seamount [Karsten and Delaney, 1989]. Axial Seamount is thus the locus of magma oversupply and the broadest crustal inflation on the Juan de Fuca Ridge [Carbotte et al., 2008]. The magma chamber beneath Axial Seamount is the largest and shallowest along the Juan de Fuca Ridge [West et al., 2001; Carbotte et al., 2008 and references therein]. Axial Seamount is the focus of NOAA's long-term NeMO Project and the site of a primary node on an OOI regional cabled observatory.
 Summit volcanism at Axial Seamount has been widespread and frequent during the last several millennia and has produced a range in lava morphology and abundant pyroclasts [Embley et al., 1990; Clague et al., 2009; Helo et al., 2011; Clague et al., 2013], similar to historical eruptions elsewhere along the ridge [e.g., Chadwick et al., 2001; Chadwick and Embley, 1994; Chadwick et al., 1995; Embley et al., 1995; Embley et al., 2000]. The diversity in effusion style over short timescales and distances attests to varying physical (e.g., temperature, viscosity, crystallinity) and chemical properties in the shallow crustal magmatic system. In the following sections, we place compositionally different lava groups within the detailed temporal and spatial geological framework presented by Clague et al. [2011, 2013] and discuss their petrogenesis.
3. Samples and Methods
3.1. Sample Locations and Petrography
 Well-located kilogram-sized lava samples were collected with ROVs from the summit caldera, upper south rift, and flanks over a depth range of ∼1400–1800 m below sea level (bsl) (Figure 1). Detailed maps and identification and descriptions of lava flow units at the summit caldera and flanks of Axial Seamount are presented in Clague et al. ; lava flow unit identifications for samples in this study are reproduced in Table 1. Summit lavas range from aphyric to ∼35% plagioclase-phyric ± up to a few % olivine. For convenience, we classify lavas as aphyric or plagioclase-phyric (≥10 vol % plagioclase phenocrysts of at least several millimeters). Approximately 20% of the lavas that we analyzed for major element abundances are plagioclase-phyric. They are examples of the plagioclase ultraphyric basalts of Cullen et al.  and Nielsen et al. .
Table 1. Geochemistry of Summit Lavas, Axial Seamounta
Major (in wt %) and trace element (in ppm) concentrations and isotope ratios of glasses from the summit of Axial Seamount. Group 1 lavas are defined as having glass MgO <7.9 wt %, typically with K2O/TiO2 > 0.1. Group 2 lavas have glass MgO >7.9 wt %. Plagioclase-phyric lavas have ≥10 vol % plagioclase phenocrysts that are at least several millimeters (see also Figure 1). Lavas with <10 vol % plagioclase are aphyric, including those with sparse (<3%) and rare (<1%) plagioclase. Details of the electron probe major element oxide analyses are reported in Clague et al.  and reproduced here for convenience. Solution- and laser ablation-ICP-MS trace element analyses were performed at UCSC. Measurements of Pb isotope ratios and U and Th isotope ratios and isotope-dilution concentrations were by MC-ICP-MS at UCSC. Precision and accuracy estimates are based on multiple analyses of international basaltic standards treated as unknowns. See text for additional details.
 Lava flows in the northern and southern portions of the caldera are dusted by less than a few centimeters of sediment whereas caldera lavas in the central portion of the caldera have more. Sediment cover on flows on the rims and flanks is greatest overall, ranging from a few tens of centimeters to ∼2 m. Sediment pushcores collected in several locations are the basis of radiocarbon age constraints that are summarized below and presented in detail in Clague et al. . Core samples also contain pyroclastic glassy shards up to ∼1 cm. Two cores analyzed in detail are from the center of the caldera. These cores were divided into thirteen 2 cm thick depth-horizons and provide stratigraphic and absolute age-constrained glass chemistry. One additional pyroclastic sample is from the southwestern flank, and the pyroclastic grains are from a single 2 cm horizon 14.1–15.3 cm below the core top.
3.2. Analytical Methods
 We report solution ICP-MS trace element and MC-ICP-MS Pb and U-Th isotope analyses of 16 (nine for Pb isotopes) lava glasses and the composite pyroclastic sample from the SW flank (Table 1). We also report laser-ablation (LA-) ICP-MS analyses for 76 glasses (Table 1), of which 14 also have solution ICP-MS data, and XRF and ICP-MS analyses on seven whole rocks from the caldera walls (supporting information Table S1).1 Major element glass analyses of 235 pre-2011 lava samples are reported in full in Clague et al. . For convenience, we present major element data (wt %) for the subset of samples with trace element (ppm) and isotope ratio data in Table 1. Major element analyses of 256 stratigraphically controlled fluidal pyroclastic glass fragments, each an average of four spots on the analyzed shard, are presented from two cores from the central portion of the caldera floor of Axial Seamount (supporting information Table S2).
 Approximately 2–3 g of insonicated and dried glass chips up to ∼1 cm size were hand picked under magnification for microbeam analysis or for solution ICP-MS. ICP-MS solution trace element analyses were performed at University of California Santa Cruz (UCSC) following Dreyer et al. . Calibrations used Standard reference materials of basaltic composition: DNC, W-2, JB-2, BHVO-2, and BCR-2. Instrumental drift was monitored and corrected, as necessary, with multiple analyses of an in-house reference basalt, BW-13. Precision and accuracy are estimated by repeat analyses of BIR-1 treated as an unknown over several days (Table 1). Samples analyzed by LA-ICP-MS at UCSC were ablated with an 86 μm circular spot at 10 Hz with 8–12 J/s for ∼2 min. Calibrations used four USGS synthetic glasses spiked with trace elements (GSA, GSC, GSD, GSE), with 43Ca as the internal standard. Drift was monitored with repeated analyses of BHVO-2G, and a correction was applied as necessary. Typically, three spots were analyzed for each sample. Accuracy and precision were estimated with multiple analyses over several days of BIR-1G and KL2 treated as unknowns (Table 1). Analyses of BIR-1 by laser and solution ICP-MS methods at UCSC agree within 2 standard deviations of their mean values for all trace elements except Cr, Co, and Pb, which agree to within 3–4 sigma. The differences we find in the Pb concentration of BIR-1 with each method are consistent with similar differences documented between analyses of powdered and fused glass forms of BIR-1 [Jochum et al., 2005, 2006]. The concentration of Pb is ∼5–10x higher in BIR-1 powder and fused glass than is typical for samples of Axial seamount.
 Isotopic analyses were performed at UCSC with a Neptune MC-ICP-MS. Leaching of up to 2.5 g of glass in acetone, H2O, 0.1M oxalic acid + 2% H2O2, 0.1M HCl + 2% H2O2, and H2O prior to dissolution follows Sims et al. . Pb was purified from digested sample solutions using mixtures of HBr-HNO3 and ∼150 µL AG1-X8 anion exchange resin in Teflon microcolumns, modified after Lugmair and Galer . Lead isotopes were measured by solution MC-ICPMS in static mode using internal spiking with NBS 997 Tl for mass fractionation correction, assuming 203Tl/205Tl = 0.418911. Solutions contained ∼50 ppb Pb and had 208Pb/205Tl between 2 and 10. Long-term results for NBS 981 are 206Pb/204Pb = 16.931 ± 0.002 (2σ external reproducibility), 207Pb/204Pb = 15.481 ± 0.004, and 208Pb/204Pb = 36.671 ± 0. 007. External uncertainty for basaltic unknowns is 0.001–2 based on triplicate dissolutions analyzed on different days. Uranium-series methods are discussed in Section 3.4.
3.3. High Resolution Bathymetric Mapping to Identify Individual Lava Flows
Clague et al. [2011, 2013] and Caress et al.  present 1–2 m horizontal and 0.1 m vertical resolution bathymetric maps of the summit region of Axial Seamount (∼75 km2). They combined mapping data with dive observations, relative and absolute age data, and glass compositions to define lava flow boundaries, with high fidelity for flows on the caldera floor and upper south rift zone. Contacts on the rim are more subdued, and less certain, because of the 1–2 m thick volcaniclastic deposits that blanket much of the caldera rim. Several flows on the caldera rims are on top of most of the volcaniclastic section and can be mapped from backscatter and side scan data since they have higher backscatter than the surrounding sediment-covered flows. For example, the boundaries shown for a large flow on the NE flank (flow Eg of Clague et al. ), which erupted from a south rift zone fissure, are based on side scan and bathymetric data collected during the autonomous underwater vehicle surveys. Most of the flow was initially mapped by Embley et al.  based on SeaMarc II side scan. Two similar higher-backscatter flows on the west rim, flows Wa and Wb, were identified using amplitude data from EM300 data collected in 1998 and mapped using amplitude data from EM302 data collected in 2011.
3.4. Age Dating Methods
Clague et al.  presented marine-calibrated radiocarbon ages of foraminifera from basal sediment in pushcores collected above lava flows. We use those ages to assign minimum ages to lava flows on the caldera floor and rim. Briefly, sediment pushcores were collected from submeter scale depressions atop lava flows. The basal 1 cm was extracted, sieved, and hand picked under magnification to recover foraminifera for 14C dating. These ages were marine calibrated using methods described in Clague et al.  and are reported in calendar years for samples more recent than 400 CE or, for older samples, in calibrated (cal) years before present (yr BP), where present is 1950 CE.
 Supplemental age constraints were derived from parent-daughter uranium-series radioactive disequilibria measured in lavas. The decay toward equilibrium depends on the half-life of the daughter product (e.g., 230Th t1/2 = 75 kyr). Uranium and Th were separated with anion exchange and their isotope-dilution concentration measurements used 233U and 229Th spikes [Lundstrom et al., 1998]. The external uncertainty in the Th/U ratio was ∼2% (2σ) based on multiple digestions of Table Mountain Latite (TML). Analytical methods for U and Th isotopes followed Sims et al.  and Ball et al. . The activity ratio of 234U to 238U was within 1% of unity for all samples, and so the effect of seawater alteration of glass was negligible. 232Th/230Th measurements of standard reference materials and unknowns were bracketed using UCSC “ThA” 232Th/230Th = 170765. Accuracy and precision of 232Th/230Th for TML were better than 1%, and its (238U/234U) and (238U/230Th) were within <1% of radioactive equilibrium.
4.1. Distribution of Aphyric and Plagioclase-Phyric Lavas
 The locations of highly plagioclase-phyric lavas are outlined in Figure 1 and presented in greater detail and with flow unit identifications in Clague et al. . Within the caldera, plagioclase-phyric lavas mostly form pillow mounds and inflated sheet flows with a hummocky, domed appearance with greater surface relief and sediment cover than surrounding aphyric lava units Clague et al. . In the caldera center, plagioclase-phyric lavas have lobate flow fronts in close proximity to inflated lineated sheet flows [Embley et al., 1990; Appelgate and Embley, 1992; Paduan et al., 2009]. A small but prominent post-caldera cone in the northwest of the caldera is also plagioclase-phyric (flow Ng2 of Clague et al. ). The largest plagioclase-phyric lava flow lies outside of the caldera on the NE flank with up to ∼15 cm of sediment in localized pockets (flow Eg).
 Sparsely phyric to aphyric lavas dominate elsewhere in the caldera and the rift zones. Volcanism for the last several centuries has occurred along fissures aligned with the SE caldera, further to the south in the rift zone [Caress et al., 2012], and in the northern caldera [Embley et al., 1990; Clague et al., 2013] (Figure 1). These young lavas are aphyric and typically form thin sheet or lobate flows with pillowed margins. Lava channels and collapse structures are common and overall topography is subdued. Aphyric lavas are also dominant on the flanks of the caldera, where they crop out through ∼1 m or more of volcaniclastic sediment [Clague et al., 2009; Helo et al., 2011; Clague et al.2013]. Rarely, pockets of aphyric lava are found within the bounds of highly plagioclase-phyric lava flows; in these instances, the glass compositions of aphyric and plagioclase-phyric lavas are indistinguishable from one another.
4.2. Age of Aphyric and Plagioclase-Phyric Lavas
 The oldest lavas on the caldera floor are highly plagioclase-phyric (i.e., >20% plagioclase). They erupted in the center of the caldera, and their minimum ages range from ∼1264 to 1305 CE (Clague et al. , Table 1) (Figure 1). The age of the highly plagioclase-phyric flow on the NE flank (flow Eg) is at least ∼1264+72/-56 CE. Two sparsely phyric flows that abut the northeastern wall of the caldera date to at least ∼1264 CE and are the oldest “aphyric” lavas on the caldera floor. Given the uncertainty in ages, these plagioclase-phyric and aphyric lavas likely erupted within ∼100 years of each other. Lavas of this period are mostly plagioclase-phyric and have higher MgO glass contents than nearly all other Axial Seamount summit lavas (that is, lavas on the caldera floor and the uppermost flank). A few high-MgO aphyric lavas that flow along the south rift may have erupted during this period although they lack absolute ages.
 Eruption of aphyric lavas has been dominant at the summit of Axial Seamount since ∼1400 CE and within the caldera since ∼1650 CE. An aphyric flow that erupted on the SE rim flowed SE and is dated at ∼1398+71/-55 CE. The youngest flow with a radiocarbon date is a large aphyric flow ∼1650 CE (1664+117/-58 CE and 1631+98/-61 CE). This flow (flow Ng) lies in the center of the caldera where sediment is generally only a few cm thick but can accumulate locally in pockets up to ∼10 cm thick. The remaining aphyric sheet flows in the caldera and uppermost south rift have sediment cover too thin to collect (<4 cm) and are inferred to be younger than the ∼1650 CE flow.
 Flows outside the caldera and uppermost south rift are generally older with the exception of aphyric flows more recent than ∼1650 CE in the upper south rift (i.e., 2011, 1998, and “Bag City” / flow Sa). To the west of the caldera, multiple dates of two aphyric lava flows indicate minimum ages of ∼800 and ∼1000 CE. Along the upper south rift zone, a sparsely phyric pillow cone dates to 411+109/-123 CE, and a kipuka is likely much older than the date of ∼6440 cal yr BP acquired from the base of a pushcore that did not penetrate to the sediment-lava contact. The next older dated group of lavas underlies the thick clastic section that blankets most of the rim of the caldera. These flows yield minimum ages between ∼12.8 and ∼31.2 kyr BP, are dominantly aphyric, and erupted before the formation of the caldera or from fissures in the flank post-caldera Clague et al. . Other lava samples collected from the caldera walls are inferred to be older than 31 ka, although by how much is not known, and most are aphyric.
 All 17 samples (all but one aphyric) measured for U-Th isotope ratios exhibited (230Th)-(238U) disequilibria. 230Th has a half-life of 75 kyr, and so measurable (230Th)-(238U) disequilibria means that these samples are unequivocally <350 kyr and are likely <15 kyr because their (230Th)/(232Th) ratios correlate positively with their (232Th)/(238U) ratios within analytical uncertainties [see Sims et al., 2003, for rationale]. Goldstein et al. , Volpe and Goldstein , and Rubin et al.  reported 230Th-238U, 226Ra-230Th, and 210Pb-226Ra disequilibria in a few Axial lavas, respectively, to constrain the timing of magmatic events. Such young ages (≤10 kyr) of summit lava flows are consistent with location in the caldera, youthful appearance of their glass rinds, thin sediment cover (<25 cm), and radiocarbon ages Clague et al. . Future studies employing short-lived radionuclides should improve the summit eruption chronology.
 In summary, the sequence of summit basalt types discussed herein is, from oldest to youngest: aphyric lavas exposed only in the caldera walls (>31.2 kyr BP); pre-caldera and post-caldera heavily sedimented (≥1 m) aphyric and rare plagioclase-phyric lavas that outcrop on the flanks (∼12.8–31.2 cal kyr BP); moderately sedimented (up to ∼20 cm) aphyric lavas on the east and west flanks (∼410–1000 CE); high-MgO sparsely phyric (older than ∼1264 CE) in the NE section of the caldera and southern rift zone and high-MgO plagioclase-phyric pillow lavas in the central portion of the caldera and east flank (1300–1220 CE); sparsely phyric flow near the upper south rift zone on the eastern flank (∼1440 CE cal yr BP); and numerous thin, exclusively aphyric lavas in the caldera and upper south rift zone (younger than ∼1650 CE) (a summary is presented in Table 3). Thus, the oldest flows on the caldera floor are all high MgO. Most of them are plagioclase-phyric although a few are aphyric, and they erupted during approximately the same brief period (∼1300–1220 CE) but from various locations on the summit. These lavas were preceded and followed by the eruption of aphyric lava with lower-MgO. Compositional groups among the lavas are discussed in the following sections.
4.3. Geochemistry of Axial Summit Lavas
4.3.1. Major and Trace Elements
 Lavas at Axial Seamount and rift zones include normal (N)- and transitional (T)-MORB with slight enrichments in alkalis and incompatible trace elements [Desonie and Duncan, 1990; Rhodes et al., 1990; Chadwick et al., 2005]. Glasses in this study from the more restricted summit caldera floor and near-rim flanks range from N-MORB to slightly enriched T-MORB, which we define here as K2O/TiO2 ≤0.10 and K2O/TiO2 >0.10, respectively. The K2O/TiO2 range is typical for most of the southern Juan de Fuca Ridge and slightly higher than in the adjacent CoAxial segment (Figures 2 and 3). Although there is a relatively large spread in K2O and TiO2 for any given MgO, trends generally follow those typical of fractional crystallization. Variation in La/Yb along the ridge mimic those of K2O/TiO2, with mean (La/Yb)N >1 at Axial Seamount and typically lower values at CoAxial and Northern Cleft (Figure 2 and Table 2). CaO/Al2O3 is negatively correlated with MgO, and this reflects plagioclase (and olivine) fractionation toward the clinopyroxene-in reaction (Figure 4). MgO in lava ranges from 6.8 to 8.7 wt % (Figure 3-5). Aphyric lavas span this entire range in MgO, whereas plagioclase-phyric lavas are more restricted and typically have >8.0 wt % MgO. Pyroclastic glasses span an even wider range, 6.7–9.8 wt % MgO [Clague et al., 2009; Helo et al., 2011].
Table 2. Geochemical Variability of Axial Seamount Summit, Segment and Nearby Ridge Segmentsa
Axial Summit pyroclasts
Axial Group 1 lava
Axial Group 2 lava
This Study (n)
This Study (n)
Compositionally distinct magma sources are required to explain observed differences in mean element ratio between each portion of the Juan de Fuca Ridge. The magnitude of the standard deviation is an estimate of the degree of variability within each portion of the ridge. Glass from Axial summit is both compositionally distinct and homogenous (K2O/TiO2 coefficient of variation = <0.15) relative to nearby CoAxial (c.v. = 0.29) and northern Cleft (c.v. = 0.55) segments. Despite the overall homogeneity at Axial Seamount, we distinguish two main compositional groups among summit lavas. Data shown are means, coefficients of variation, and number of samples analyzed. “n.d.” indicates not determined. Data were filtered to include only electron probe (major elements) and unspiked ICP-MS (trace element) methods, with the exception of Th/U of northern Cleft, which are by isotope dilution [Goldstein et al., 1992; Rubin et al., 2005]. Our mean isotope dilution Th/U analyses for Axial summit lavas are indistinguishable from those of Rubin et al. . Like our LA-ICP-MS analyses, these isotope dilution Th/U analyses show that Axial summit is distinct from nearby ridge segments. Additional data from Chadwick et al. , Smith et al. , C. Helo (unpublished data, 2012), and M. Perfit (unpublished data, 2008).
0.074 ± 0.29 (153)
0.11 ± 0.13 (289)
0.10 ± 0.13 (538)
0.10 ± 0.14 (253)
0.110 ± 0.058 (178)
0.089 ± 0.086 (58)
0.095 ± 0.55 (243)
0.88 ± 0.28 (35)
1.50 ± 0.11 (43)
1.42 ± 0.13 (42)
1.51 ± 0.052 (56)
1.21 ± 0.10 (22)
1.17 ± 0.13 (21)
2.61 ± 0.12 (33)
3.16 ± 0.079 (117)
3.04 ± 0.14 (39)
3.06 ± 0.025 (56)
2.83 ± 0.043 (22)
3.01 ± 0.60 (29)
2.13 ± 0.20 (32)
2.87 ± 0.12 (39)
3.03 ± 0.050 (56)
3.01 ± 0.050 (22)
2.41 ± 0.03 (12)
Table 3. Summary of Characteristics of Group 1 and Group 2 Lavas, Axial Seamounta
Lava flow indentifications, age data, and additional maps are given in Clague et al. . See text for additional details.
MgO range wt %
0.11 ± 0.06
0.089 ± 0.86
3.06 ± 0.025
2.83 ± 0.043
Geographical location of eruption
Ubiquitous at rift zones fissures; N and S caldera sections
Central portion of the caldera; NE flank; SRZ
prior to ∼1100 CE; since ∼1400 CE; voluminous since ∼1650 CE
Nearly all are aphyric to sparsely aphyric (<3% vol)
Most plag-phyric (10–35% vol); few aphyric flows that cluster along SRZ
Lava morphology (most recent)
large thin sheet or lobate flows with pillowed margins
2011, 1998, Sa (Bag City), Sb (Ashes), Na (CASM); transitional: Ne
Ng1, Ng2, Eg, Nj, Sg1
Smaller degree melt; extended shallow storage, more degassed (?), and greater differentiation
Larger degree melt; more primitive, undegassed (?), relatively rapid ascent from mush zone, less differentiation
 Axial Seamount summit lavas have similar primitive mantle-normalized incompatible trace element patterns within a narrow concentration range (Figure 6). Their rare-earth element (REE) patterns are ∼5–10x primitive mantle, and their higher LREE concentrations result in ∼20% steeper slopes than in the N-MORB of Sun and McDonough . LaN/YbN ranges from 0.73 to 1.27, and most samples are >1 (i.e., enriched). Primitive mantle-normalized patterns are slightly concave down (LaN/SmN = 0.7–1.0). These REE concentrations overlap, and primitive mantle-normalized patterns are similar to basalts from the central and southern Juan de Fuca Ridge [Chadwick et al., 2005, Figure 7]. Mantle-normalized trace element diagrams have the usual MORB depletions in Pb and Th. The average Th/U∼ 3.0 is higher than in other Juan de Fuca N-MORB from south of the Endeavour segment [Goldstein et al., 1992]. Our data demonstrate that summit lavas have lower variance in K2O/TiO2, La/Yb (Figures 2 and 3), Th/U, Zr/Y, and Pb-isotope ratios than Axial summit pyroclasts, lavas from the Axial segment, and lavas from adjacent ridge segments (Table 2; see also Figures 7 and 8). That is, summit lavas are more chemically homogeneous.
4.3.2. Isotope Ratios
 Our Tl-spiked Pb isotope ratios have a narrower range than previous data (Figure 8). Our data lie just below the Northern Hemisphere Reference Line, have slightly higher 206Pb/204Pb than published Axial data, and fall well within the data field of Juan de Fuca MORB and the TIMS data field of the Cobb-Eickelberg chain [see also Chadwick et al., 2005, Figure 5]. Our one plagioclase-phyric sample is indistinguishable from the field of aphyric lavas. Axial Seamount has the lowest 230Th/232Th ratios on the Juan de Fuca Ridge axis [Goldstein et al., 1992; Rubin et al., 2005]. 230Th-excesses (defined as (230Th)/(238U)> 1) are greater than previously reported [Rubin et al., 2005] and are of similar magnitude to neighboring segments on the ridge (Figure 9). The coefficient of variation of (230Th)/(238U) for all Axial summit lavas is 3.1%. It is even lower, 2.0%, and within analytical uncertainty, among all lightly-sedimented summit lavas (i.e., the youngest). Our data for four samples of the 1998 flow are indistinguishable from one another within analytical uncertainty, although they differ from those of Rubin et al.  by ∼3.5% and <1% in (238U)/(232Th) and (230Th)/(232Th), respectively (conormalized to the same value of UCSC ThA standard). The composite sample of pyroclastic grains from the SW flank has the highest (230Th)/(232Th) ∼1.22, and a lava from the ASHES hydrothermal field has the lowest (230Th)/(232Th) ∼1.16. Summit lavas define a positive linear array on the (230Th/232Th)-(238U/232Th) equiline diagram (Figure 9), in common with most ridges globally [e.g., Lundstrom, 2003], albeit with an atypically shallow slope of ∼0.2–0.3 (see discussion below). There is a robust negative correlation between Th/U and 206Pb/204Pb and a weaker correlation between (230Th)/(238U) and 206Pb/204Pb.
4.3.3. Differences between Lava Types
 In detail, the lavas are chemically bimodal (Table 3). A differentiated mode (referred to as “Group 1” throughout) is slightly enriched T-MORB and has glass MgO ≤7.9 wt %, CaO/Al2O3 >0.80, K2O/TiO2 = 0.10–13, and typically LaN/YbN >1 (Figures 3 and 4). A second, more mafic group (“Group 2”) is N-MORB with glass MgO >7.9 wt %, CaO/Al2O3 <0.80, K2O/TiO2 <0.11, and LaN/YbN <1. The locations of high-MgO Group 2 lava flows are outlined in Figure 1. Among our samples, ∼88% of aphyric lavas are classified as Group 1 (i.e., T-MORB) and ∼90% of the plagioclase-phyric lavas are Group 2 (i.e., N-MORB). All highly plagioclase-phyric flows with minimum eruption ages between 1220 and 1300 CE are Group 2. Pyroclastic glasses also exhibit compositional modality (Figures 3 and 5), but we usually cannot tell whether their magmas were phyric or not. There is also evidence for a third magma type (“Primitive Group”). A few pyroclast glasses collected from the caldera flanks have MgO =9.4–9.8 wt %, systematically lower K2O and TiO2, and generally lower K2O/TiO2 and CaO/Al2O3 than Groups 1 and 2 (Figures 3, 4, and 6) [see also Helo et al., 2011]. Despite extensive sampling at Axial Seamount, lavas with this primitive composition have not been recovered.
 A systematic analysis of pyroclasts from two pushcores above Group 2 lava flows, Si1 and Sg1Clague et al. , from the central portion of the caldera shows that the two main compositional modes are separate stratigraphically (Figure 5). In both cores, pyroclasts nearer the contact with the lava are Group 2, whereas pyroclasts nearer the top of the cores are Group 1. Radiocarbon dating in each pushcore indicates that the transition from Group 2 to Group 1 pyroclast composition occurred between 1223+76/-67 and 1423+77/-45 CE. This range is similar to the age range we determine based on multiple dates of Group 1 and 2 lavas flows within the caldera (see above).
 Despite similar and restricted trace element concentration ranges among the vast majority of summit basalts, there are consistent differences between Groups 1 and 2 (Table 3). The magnitude of the trace element concentration difference scales crudely and positively with an element's incompatibility during mantle melting and crystal fractionation. That is, the more differentiated T-MORBs (Group 1) have similar concentrations of heavy HREE but higher and more varied concentrations of more incompatible trace elements than the mafic N-MORBs (Group 2) (Figures 6 and 7). The few elements that deviate from this general trend, Sr, Rb, and possibly Ba and Pb, have similar concentrations regardless of compositional group. These large ion lithophiles are compatible to moderately incompatible in plagioclase [Aigner-Torres et al., 2007] over a range of conditions and An-content appropriate for MORB formation, suggesting plagioclase control on the concentration of these elements in both Group 1- and 2-type melts.
 Differences in incompatible element ratios between Group 1 and 2 lavas amplify the concentration differences. Mafic N-MORBs (Group 2) have lower mean K2O/TiO2, La/Yb, La/Sm, Nb/Zr, Nb/Y, and Zr/Y than differentiated T-MORB (Group 1) (Figure 7 and Table 2). Group 2 is broadly similar in these ratios to the primitive group of lavas dredged at Axial [Rhodes et al., 1990]. These differences cannot be explained by plagioclase fractionation. Variation within Group 1 and Group 2 at Axial Seamount are detailed in the next section.
4.3.4. Variation Within the Compositional Groups
 Plagioclase-phyric samples (∼1220–1300 CE) are invariably higher-MgO N-MORB (e.g., Group 2). The few plagioclase-phyric samples with Group 1 compositions are either only sparsely phyric or much older (e.g., within the caldera wall). Similarly, a majority of all aphyric lavas, including all flows younger than 1400 CE, are T-MORB (Group 1). Thus, plagioclase abundance in recent lavas is a crude proxy for compositional bimodality, particularly for lavas erupted during the last few centuries.
 Group 2 includes all of the highly plagioclase-phyric lavas and also multiple minor subgroups of aphyric lavas. All of the aphyric Group 2 lavas are more mafic and demonstrably older than the phase of volcanism that is currently producing more evolved Group 1 aphyric lavas, predating it by at least one and likely several centuries (see above and Clague et al. ). Chief among aphyric Group 2 lavas are subgroups of a slightly less depleted N-MORB (MgO 8.3–8.7 wt % and K2O/TiO2 ∼0.09–0.10) and a more depleted N-MORB (MgO 8.0–8.6 wt % and K2O/TiO2 ∼0.07–0.09). The less depleted Group 2 aphyric lavas are found only in an area of limited extent on the NE central portion of the caldera floor. Flows of the more depleted Group 2 aphyric lavas also cluster spatially and do so in areas outside of the caldera at eruptive vents or near fissures that are either aligned with the southern rift zone or circumferential to the caldera. Differences within each of these Group 2 aphyric subgroups are evident in trace element concentration and ratios. For example, there are distinct differences in Cr, Ni, V, Nb, Ba, and Th concentrations and in trace element ratios among the less depleted of the Group 2 aphyric lavas (despite sharing similar MgO content, K2O/TiO2, and location). Chemical differences of this type that occur within the defined subgroups suggest that not all lavas in a subgroup are comagmatic and that distinct magmatic batches with varied histories are required.
 Among the compositional sameness of Group 1 lavas, one aphyric flow warrants special comment because of its stratigraphic and compositional significance. This flow (flow Ne of Clague et al. ) lies on the NE central portion of the caldera amidst several high-MgO Group 2 lava flows. The minimum eruption age of flow Ne is ∼1650 CE, and it is characterized by elevated K2O/TiO2 (∼0.12 and amongst the highest of all lavas) and the highest MgO (∼7.9 wt %) of the Group 1 lavas. Therefore, within the caldera this transitional flow is chronologically intermediate and compositionally distinct from the older Group 2 (higher-MgO and lower K2O/TiO2) lavas and the more differentiated Group 1 lavas that have dominated since ∼1650 CE.
 All of the older aphyric flank lavas sampled are Group 1, but subtle differences from their younger counterpart are found among samples using higher-precision solution ICP-MS and isotope data. Lavas from the heavily-sedimented SW flank generally have lower 206Pb/204Pb ratios and higher Zr, Y, Th/U, and (230Th)-excess compared to the majority of the younger aphyric lavas (Figures 7, 9, and 10). On the NE flank, an aphyric Group 1 lava and a plagioclase-phyric Group 2 lava have intermediate to low 206Pb/204Pb values but slightly lower Th/U and (230Th)-excess relative to the patterns observed in all other lavas (Figure 10). In other words, these NE flank lavas occupy the low-Th/U, intermediate-206Pb/204Pb portion of a negative array created by all the lavas.
 In summary, the youngest differentiated, intermediate mafic, and older diverse lava groups that we recognized stratigraphically also differ geochemically, in modal plagioclase content, and spatially (Table 3). The temporal and spatial patterns in magma types could not have been recognized without high-precision chemical analyses, high-resolution mapping, and dating. The correlation between compositional bimodality and plagioclase content is strongest between the most recent phases of aphyric and plagioclase-phyric volcanism, representing the last several centuries of summit volcanism. The exceptions described above are chiefly among older lavas within each group, some of which predate the caldera formation and reflect additional complexity in the past. We next focus on the petrological origins of the summit lavas. We show that some geochemical systematics can be reproduced by closed-system differentiation and that others require open-system behavior. We then present models of magma genesis and differentiation to explain the spatial, temporal, and compositional patterns at Axial summit.
 We examine major element trends in Axial summit lavas to investigate pathways and conditions of magmatic differentiation. We modeled fractional crystallization trends at variable H2O contents and pressures using Petrolog algorithms [Danyushevsky et al., 2001]. Compositions of the mafic T-MORB and N-MORB glasses used as input melt compositions are presented in supporting information Table S3. Forward model simulations were carried out with oxygen buffered at QFM-1 at static pressures ranging from 1 to 3 kbar.
 Major element trends observed in glasses from nearby CoAxial and Northern Cleft segments, which represent noninflated Juan de Fuca Ridge oceanic crust, are generally reproduced by fractional crystallization of olivine + plagioclase ± clinopyroxene from anhydrous parental magmas at pressures ∼1.5–3.5 kbars [e.g., Smith et al., 1994]. Trends observed in glasses from Axial's summit are reproduced with models at lower pressure (∼1–2 kbar, equivalent to crustal depth of ∼3–6 km) fractional crystallization of parental magmas with ∼0.1 wt % H2O. These water contents are slightly lower than published data for Axial glass rims [Chadwick et al., 2005] and plagioclase-hosted melt inclusions and their host glasses [Helo et al., 2011]. The distribution of samples along fractionation trends illustrates that mafic Group 2 lavas did not reach the clinopyroxene-in reaction (CaO/Al2O3 ∼0.84 and ∼7.4 wt % MgO in Figure 4), whereas the more differentiated Group 1 lavas are more likely to have been saturated with clinopyroxene. Some scatter around the clinopyroxene-in inflection may be due to slightly different depths of fractional crystallization or varied differentiation histories [e.g., O'Neill and Jenner, 2012], but the general dearth of scatter means that any magma mixing must have occurred prior to or near clinopyroxene saturation (e.g., along an olivine-plagioclase cotectic). Crystallization depths of 3–6 km imply that much of the magmatic evolution is occurring at depths greater than the seismically detected melt lens beneath Axial Seamount.
 Notwithstanding the subtle differences described in sections 4.3.3 and 4.3.4, the overall greater trace element and isotope ratio homogeneity at Axial Seamount compared to adjacent ridge segments (Figures 2 and 3 and Table 2), and the uniform differentiation trend in Figure 4 [see also Chadwick et al., 2005, Figure 3], attest to a high degree of petrological coherence at Axial. Thus, Group 1 and 2 lavas repeatedly sampled magmas that are closely related spatially, temporally, and compositionally, and therefore that are linked largely by reproducible closed-system differentiation of compositionally similar parents during at least a millennium. The similar liquid lines of descent also means that the major element composition of parental melts, their volatile content, and the depth of differentiation varied within narrow limits during this time. Both the shallower differentiation and reduced source-inherited heterogeneity observed in summit lavas are consistent with higher mixing efficiency associated with greater magmatic flux [e.g., Rubin and Sinton, 2007] beneath Axial Seamount compared to CoAxial and northern Cleft segments. That is, the higher overall melt supply at Axial favors greater homogenization of parental magma diversity within a comparatively robust magmatic reservoir [Chadwick et al., 2005].
5.2. Origin of Parental Magmas
 Despite a high degree of compositional homogeneity, Groups 1 and 2 differ enough in minor element ratios (e.g., K2O/TiO2) and trace element ratios (e.g., La/Yb, Nb/Y, Zr/Y, Th/U) that they cannot be related to one another solely by crystal fractionation (Table 3). Rather, the origin of these differences may be attributed to differences in the degree of partial melting of the MORB source(s). Simple partial melting models demonstrate that slight differences in the degree of melting from compositionally similar source(s) can reproduce the average Zr/Y of Groups 1 and 2 (Figure 7). Parental source(s) may be slightly different derivative mixtures of more depleted (i.e., the primitive group) and more enriched mantle source (Figure 7a) [see also Chadwick et al., 2005]. Subsequent fractional crystallization leads to higher incompatible trace element but has a much-reduced effect on the ratio. The bimodality in major and trace element trends indicates that Group 1 and 2 lavas largely do not mix during fractional crystallization (Figures 3, 4, and 7). Thus, the basalt groups inherit and preserve a range of parental melt compositions that cross the Moho and remain partially isolated in time and/or space despite similar conditions of melt evolution. Understanding the origins of compositional variability and their expression through episodic eruption will ultimately clarify magmatic processes, timescales, and the geometry of the shallow melt lens network.
 Here we leverage systematic U-Th-Pb relationships in Axial lavas to illuminate the sources responsible for observed diversity. The spread in Th/U values among local MORB is commonly attributed to mixing between melts from heterogeneous sources (i.e., with distinct Th/U) and/or from different depths [Lundstrom, 2003]. At Axial, the correlations between subtle variation in Th/U (and to a lesser degree 230Th-excess) and differences in Pb isotope ratios are evidence of both source heterogeneity and mixing (Figure 10). That these correlations are preserved within Group 1 lava (only one Group 2 lava has been analyzed) is noteworthy because approximately constant incompatible trace element ratios (Figure 7) and homogenous Pb-isotopes (Figure 8) suggest that Group 1 lavas share a similar parental melt or source composition. Therefore, the Th-U-Pb correlations may reflect mixing of melts during partial melting and melt extraction. Recent studies highlight the geochemical effects of mixing between enriched and depleted melts during partial melting of a lithologically heterogeneous mantle [e.g., Stracke and Bourdon, 2009; Brandl et al., 2012].
 While a linear array on the equiline diagram (Figure 9) reflects mixing of melts from sources with contrasting Th/U, the slope of the trend is independent of the Th/U of the sources and can be used to assess petrogenesis [Lundstrom et al., 1998]. Because the trend is anchored at the equiline and the enriched end member has high (230Th/232Th) at a given (238U/232Th), the slope of the Axial array on the equiline diagram is distinctly shallower than the trends observed for lavas from the CoAxial and Northern Cleft segments (Figure 9). The causes of MORB U-series disequilibria trends, and contrasts between trends in different segments, have been linked to a host of parameters including differences in element residence times [e.g., Spiegelman and Elliot, 1993], spreading rate and axial ridge depth (proxies for ascent rate, temperature, melt initiation depth), and source lithology or fertility [e.g., Bourdon et al., 1996; Lundstrom et al., 1998]. Conclusive isolation of these potentially causative factors is a challenge since many are interdependent. The greater 230Th-excesses at Axial compared to CoAxial and Northern Cleft are unlikely to be due to slower ascending mantle (invoked by ingrowth models) chiefly because Axial is the locus of the highest melt supply along the ridge. Instead, differences in melt column length provide a suitable framework to jointly address differences between ridge segments and coupled correlations between (230Th)-excesses, Pb-isotope ratios, and Th/U at Axial.
 The depth of melt initiation is a function of mantle potential temperature, composition (e.g., peridotite versus pyroxenite), and volatile content [Klein and Langmuir, 1987; Langmuir et al, 1992; Niu and O'Hara, 2008]. Mantle lithology is not likely to be significantly different between local segments of the Juan de Fuca because their basalt REE patterns are similar, and there is no evidence of significant differences in modal garnet and pyroxene in their sources. Instead, we attribute the shallower slope at Axial to deeper melt initiation resulting from either a deepening of the solidus due to higher mantle potential temperature and/or a lowering of the solidus temperature due to higher volatile content. The preceding applies specifically to the high-Th/U enriched mantle source. Higher temperatures beneath ridges that are affected by hotspots leads to melt initiation at greater depth [Shen and Forsyth, 1995], and indeed Hooft and Detrick  argued for a mantle thermal anomaly of 30–40°C beneath Axial Seamount based on gravity data. Elevated CO2-contents in plagioclase-hosted melt inclusion [Helo et al., 2011] is evidence for volatile enrichments beneath Axial Seamount.
 At Axial, partial melts of this enriched source have elevated Th/U and (230Th)-excesses that are correlated with slightly lower 206Pb/204Pb ratios. Lavas with these traits crop out through thick sediment on the SW flanks and represent an older, perhaps pre-caldera history of the volcano. Based on current sampling, we do not recognize temporal or spatial patterns in the diversity of chemical composition or crystal content of these older lavas. This period was followed by the eruption of lavas with more uniform Th/U and 206Pb/204Pb ratios; smaller (230Th)-excess; and bimodality in K/Ti, MgO, and plagioclase abundance to a lesser extent. These lavas erupted in the caldera or along fissures aligned with rift zones. Of these, aphyric and more commonly plagioclase-phyric N-MORB (Group 2) erupted between ∼1220 and 1300 CE. Data from melt inclusions and the 1–2 m thick deposits of pyroclasts that they occur within [Helo et al., 2011; Clague et al., 2009] suggest that Group 2-type melts are likely to have also erupted prior to this period, but we do not yet find evidence for them after ∼1300 CE. More evolved aphyric T-MORB (Group 1) lavas have been erupting along summit rift fissures since ∼1400 CE (SE flank) and frequently within the caldera since ∼1650 CE. Despite consistent closed-system differentiation that resulted in lavas with a high degree of petrological regularity and chemical homogeneity compared to nearby segments, the spatial and temporal patterns of source-inherited heterogeneities require subtle open-system magmatic behavior. The following section presents an initial model of the shallow magmatic plumbing system linking inferences from geological, geochemical, and petrological systematics.
5.3. The Subridge Magmatic System
5.3.1. Structure and Processes
 Geophysical studies indicate that melt is present beneath Axial Seamount in a diffuse zone from ∼1.4 to 2.0 km depth bsf to at least 6 km bsf near the Moho [West et al., 2001; Carbotte et al., 2008], and that there is a magma storage reservoir at ∼3.5 km depth beneath the caldera [Nooner and Chadwick, 2009]. We presented evidence that melts have experienced conditions of magma storage and differentiation that have not varied greatly for at least a millennium (Figure 4). Despite this magmatic regularity, lavas and pyroclasts exhibit systematic compositional variations that reveal differences in magmatic throughput in space and time (Figures 3, 7, and 10 and Table 3).
 Results of previous studies and our inference of magma batches can be reconciled by the presence of a network of melt-rich magma reservoirs, composed of multiple sills of restricted composition at various depths within the crystal-rich “mush zone” that approach the Moho [Helo et al., 2011; Wanless and Shaw, 2012]. Group 1 and 2 magmas represent discrete melting events within the mantle that tapped slight differences in mantle compositions. The Group 1 and 2 melts subsequently rise independently through the crust where they partially homogenize and cool to varying degrees under similar conditions to produce uniform differentiation trends, but they do not intermix. The lack of mixing during shallow crystallization and the apparent relationship between lava composition and dominant eruption location (rift versus caldera) are likely due to structural asymmetries in the magmatic system beneath Axial Seamount that result from the intersection of the Cobb hotspot and the Juan de Fuca Ridge [Desonie and Duncan, 1990; Rhodes et al., 1990; Chadwick et al., 2005].
 We present a conceptualized illustration for the magmatic system between Axial Seamount and the generation of Groups 1 and 2 magmas (Figure 11). The magmatic system beneath Axial Seamount caldera and the overlapping northern and southern rift zones produced the voluminous Group 1 aphyric lavas through the majority of the last tens of thousands of years Clague et al.  (Figure 11a). In comparison, the more primitive Group 2 lavas are the result of larger degrees of partial melting of a compositionally similar parent. These magmas commonly scavenge high-An plagioclase during rapid ascent through the crystal-rich mush zone [Zellmer et al., 2012]. Further growth of these plagioclases, and ultimately their preservation in Group 2 lavas, may be related to the nature of their ascent. Group 2 magmas remain isolated in space or time from Group 1 magmas during shallow storage and differentiation (see Figures 3 and 7). We propose that the primary magmatic pathways are temporarily bypassed as the Group 2 magmas are mainly erupted in the center of the caldera rather than the well-established northern and southern rift zones (Figure 11b).
 That Group 2 magmas erupted separately from Group 1 magmas does not provide a priori information as to whether the magmatic network hosts these magma batches simultaneously or sequentially. While the relatively short interval between eruption of the two main basalt groups may seem to favor the former scenario, it is plausible that a pulsed (i.e., sequential) supply of magma through the mush zone can produce a series of slightly different magma batches; this is more consistent with the inferred lack of shallow mixing. Changes in the chemistry of lavas erupted from the same fissure in 1991–1992 and 2005–2006 events on the faster-spreading EPR [Goss et al., 2010] may be evidence that distinct magma batches may be hosted both simultaneously and sequentially.
 Evidence from Axial indicates that crystalline components are complexly superposed on compositional variation (see summary Table 3). Group 2 lavas erupted exclusively for a few centuries; most of these lavas are highly plagioclase phyric, but some are nearly aphyric. Prior to and following (but apparently not during) this brief period, only Group 1 lavas erupted and they are largely aphyric. Taken together, the quasi-independent spatial and temporal variations in compositional and crystalline components suggest that magmatic layering or separation can develop on multiple scales depending on inputs to and the character of the partially molten region.
 Magmatic layering into more and less phyric components may be aided by contrasts in the physical properties (e.g., density or viscosity) or processes (e.g., filter-pressing, flow differentiation, etc.) over a range in depths within the magma reservoir. For example, Helo et al.,  reported volatile-enriched melt inclusions hosted in plagioclase-bearing pyroclasts at Axial and suggested that CO2-supersaturated magmas stagnate and partially exsolve CO2 at depth beneath Axial. Variations in the course of magmatic devolatilization can strongly control the dynamics of magma ascent, crystallization, and eruption [reviewed by Metrich and Wallace, 2008], and this effect may be compounded when paired with variations in the locus and rate of magma recharge.
 The sequential appearance of plagioclase-phyric and aphyric with more- and less-depleted basalts is a dominant feature of magma input and networks at Axial since at least the last millennium. However, this mode of magmatic evolution is unlikely to be inevitable or universal at MOR because aphyric lavas are not always more differentiated and enriched (e.g., higher La/Yb, K2O/TiO2; see Figure 3) than plagioclase-phyric lavas within ridge segments [Hellevang and Pedersen, 2008; Adams et al., 2011 and references therein].
5.3.2. Timescales of Magmatism
 Whether and how different magma types overlap in space or time or both has implications for the dynamic behavior of the magmatic-tectonic system. The same is true for the periodicity of their eruption. Existing age constraints on summit lavas show that the eruption of Group 2-type (N-MORB) magmas dominated for no more than ∼200 years. This was followed first by an apparent repose period of ∼100 years and then by the eruption of Group 1-type (T-MORB) magmas in and close to the caldera. This phase has dominated for the last ∼550 years (Table 3).
 Observational, geochronological, and geodetic data suggest that the eruptive recurrence interval at Axial is decadal to multidecadal [Caress et al., 2012; Chadwick et al., 2012; Clague et al., 2013]. Because only a small fraction of the total volume of shallowly stored magma may erupt during any eruption event [e.g., West et al., 2001; Caress et al., 2012; Chadwick et al., 2012], each magma type can be tapped multiple times for centuries to millennia in a closed system. The short interlude during which only Group 2-type magma were erupted demonstrates punctuated open system magmatic recharge. Their limited spatial and temporal extent suggests that Group 2 magma is currently subordinate to Group 1 magma within the shallow melt reservoir, but little is known about their source volumes. Similar magmatic episodicity is a characteristic of other volcanoes associated with robust magmatism, most notably at well-documented Kilauea [e.g., Holcomb, 1987; Heliker et al., 1998; Garcia et al., 2003; Marske et al., 2008].
 Multi-centennial petrological periods within the Axial magmatic system are similar to the residence time of melts within the crust proposed in a study by West et al.  that combined seismological observations of low-velocity zones with assumptions about local magma flux. Similar centennial to multimillennial magmatic timescales have been estimated for the southern Cleft segment [Cordier et al., 2012]. Compositionally distinct magma batches that have erupted over short (decadal) time periods have been described for MORB elsewhere, including the fast-spreading East Pacific Rise [e.g., Sims, et al., 2002; Bergmanis et al., 2007; Goss et al., 2010] and intermediate-rate Galapagos Spreading Center [Colman et al., 2012]. Magma batches at Axial summit are erupted over similar spatial scales as the Galapagos Spreading Center, which are smaller than at the East Pacific Rise.
5.4. Dynamics of the Magmatic System at Axial Seamount
 If variability in spatial, temporal, and compositional elements of magmatic batches is related to melt focusing and shallow magma dynamics, then it may be possible to further relate our observations and interpretations of shallow melt reservoirs to deeper or larger scale processes operating at spreading segments. Our observations at Axial show that the highly plagioclase-phyric Group 2 lavas that erupted in the center of the caldera did so largely from virtual point sources. This contrasts with the eruption of most other young lavas that erupted through the extensive network of fissures (primarily extra-caldera) associated with the northern and southern rift zones (Figure 1 and Table 3) Clague et al. . Thus, the eruption of plagioclase-phyric Group 2 lava reflects a temporary divergence of the magmatic pathway into the center of the caldera and away from rift zones. Deflection of the magmatic pathway may have resulted from intrusion of a new batch of parental magma.
 Additional insight into magma dynamics can be gained from analyses of high-An plagioclase that are commonly abundant in Group 2 lavas. Based on disequilibria profiles in plagioclase [Zellmer et al., 2012] and high volatile contents in hosted melt inclusions [Helo et al., 2011], it is probable that the plagioclases formed in mush zone and were entrained by melts that ascend rapidly. Detailed models of plagioclase-ultraphyric basalts at MORs elsewhere have also emphasized links between the formation of An-rich plagioclase in the mid-to-deep crustal mush zone and the dynamic cycles within the upper crustal magma reservoir [e.g., Hansen and Gronvold, 2000; Hellevang and Pedersen, 2008].
 Conversely, we posit that rift-fissure eruptions of more differentiated melts largely without phenocrysts, dominant at Axial and more typical of Pacific MORB generally, are associated with periodic partial draining of more evolved melt from shallower parts of the magma network. The differentiated character of the aphyric basalts may be due to longer residence or more effective (hydrothermal) cooling in the crust. The reestablishment of Group 1 volcanism following the brief period of Group 2 volcanism suggests that displacement and return of the primary magmatic conduit system and locus of volcanism can occur within a few centuries. Adjustments in primary magmatic pathways or significant changes in their throughput may be in response to magmatic or tectonic reorganization, such as dike injection event(s) [e.g., Hellevang and Pedersen, 2008] and/or rifting [e.g., Hansen and Gronvold, 2000]. At Axial Seamount, we speculate that temporary modifications in the behavior of the magmatic system may be associated with changes in strain partitioning on the overlapping rift zones enveloping the caldera.
 Future work to constrain the ages, compositions, and structural relationships of older suites of volcanic products at Axial may address if or how stress state can affect magmatism and crustal morphological evolution, perhaps including caldera formation. Detailed in situ microanalyses of plagioclases and melt inclusions may reveal further insight into crystallization depths and magmatic conditions during genesis and pre-eruptive processing of plagioclase-phyric lavas [e.g., Nielsen et al., 1995; Lange et al., 2013; Zellmer et al., 2012; Wanless and Shaw, 2012] and whether conditions differed between plagioclase-phyric lavas sourced from the caldera center and rift zones.
6. Summary and Conclusions
 A combined study of mapping [Caress et al., 2012], observational, age constraint Clague et al. , and petrological and geochemical data of Axial Seamount demonstrates that magmatic recharge is variable in time, space, and composition. We describe a sequence of compositional modes in summit lavas. From oldest to youngest, they are: (1) compositionally diverse, dominantly aphyric T-MORB from several kyr BP to ∼1100 CE, (2) ∼1220–1300 CE highly plagioclase-phyric mafic N-MORB in the central portion of the caldera and E flank and chemically similar but aphyric magma erupted from rift zone fissures, and (3) rift zone eruptions of differentiated nearly aphyric T-MORB since ∼1400 CE becoming particularly voluminous since ∼1650 CE. Magma beneath the summit of Axial oscillates between more and less differentiated and more or less depleted compositions on the scale of a few centuries (Table 3). Parental magmas varied subtly, and all lavas share a uniform differentiation trend. The fluctuation between aphyric and plagioclase-phyric lava likely reflects different pathways or velocities of crustal melt migration.
 We propose the following scenario for the petrogenesis of recent magmas beneath Axial Seamount. Melting initiates deeper beneath Axial due to a warmer or more enriched mantle compared to elsewhere on the Juan de Fuca Ridge. Parental melts ascend to and may stall near the base of the crust [Helo et al., 2011] where they may scavenge high-An plagioclase from the mush zone prior to or during rapid ascent [Zellmer et al., 2012]. Ascending melt batches remain partially isolated in time and perhaps in space and differentiate to varying extents in the crust under conditions that do not vary substantially over millennia; their eruptive pathways occasionally differ (Figure 11). The brief eruption of highly plagioclase-phyric lava may represent the periodic disgorgement of more primitive magmas that ostensibly elude extensive shallow differentiation and bypass the well-established rift zone fissure network. We posit that their eruption may reflect sympathetic response to variations in magma recharge and/or strain partitioning across the spreading-rift zone. In this context, highly plagioclase-phyric lavas are an integral though less common and transient expression of magmatism at intermediate-rate spreading ridges. More generally, constraining the temporal and spatial distribution and diversity of lava can illuminate magmatic and structural evolution of oceanic spreading centers at increasingly finer scale.
 We thank Ken Rubin, Michael Perfit, Bob Embley, Bill Chadwick, Christoph Helo, and Ryan Portner for fruitful discussion. We thank Roger Nielsen, Brian Cousens, Mike Perfit, Georg Zellmer, associate editor David Peate, and an anonymous reviewer for their thoughtful efforts. We thank Chris Russo, Doug Pyle, Caroline Harris, Dan Sampson, Rob Franks, Jenny Paduan, and Julie Fero Martin for technical assistance. We are grateful to the pilots of the Tiburon and Doc Ricketts ROVs and pilots and crew of the R/V Western Flyer. We acknowledge support from the David and Lucile Packard Foundation and the National Science Foundation under OCE grant 1061176.