The origin and evolution of breakouts in a cooling-limited rhyolite lava flow

Understanding lava flow processes is important for interpreting existing lavas and for hazard assessments. Although substantial progress has been made for basaltic lavas our understanding of silicic lava flows has seen limited recent advance. In particular, the formation of lava flow breakouts, which represent a characteristic process in cooling-limited basaltic lavas, but has not been de-scribed in established models of rhyolite emplacement. Using data from the 2011–2012 rhyolite eruption of Puyehue-Cordón Caulle, Chile, we develop the first conceptual frame-work to classify breakout types in silicic lavas, and to describe the processes involved in their progressive growth, inflation, and morphological change. By integrating multiscale satellite, field, and textural data from Cordón Caulle, we interpret breakout formation to be driven by a combination of pressure increase (from local vesiculation in the lava flow core, as well as from continued supply via extended thermally preferential pathways) and a weakening of the surface crust through lateral spreading and fracturing. Small breakouts, potentially resulting more from local vesiculation than from continued magma supply, show a domed morphology, developing into petaloid as inflation increas-ingly fractures the surface crust. Continued growth and fracturing results in a rubbly morphology, with the most inflated breakouts developing into a cleft-split morphology, reminiscent of tumulus inflation structures seen in basalts. These distinct morphological classes result from the evolving relative contributions of continued breakout advance and inflation. The extended nature of some breakouts highlights the role of lava supply under a stationary crust, a process ubiqui-tous in inflating basalt lava flows that reflects the presence of thermally preferential pathways. Textural analyses of the Cordón Caulle breakouts also emphasize the importance of late-stage volatile exsolution and vesiculation within the lava flow. Although breakouts occur across the compositional spectrum of lava flows, the greater magma viscosity is likely to make late-stage vesiculation much more important for breakout development in silicic lavas than in basalts. Such late-stage vesiculation has direct implications for hazards previously recognized from silicic lava flows, enhancing the likelihood of flow front collapse, and explosive decompression of the lava core. uneven overgrowths. The globulites are almost entirely crystalline and the space between the crystals is occupied by cristobalite.


INTRODUCTION
Studies of hazards from silicic volcanic eruptions usually focus on explosive activity partly because effusive silicic eruptions are rare events. However, the emplacement of rhyolite lava flows has been previously associated with hazards such as explosions from their surface (Jensen, 1993;Castro et al., 2002), lava flow front collapse Baum et al., 1989), and the potential generation of pyroclastic density currents (Fink and Kieffer, 1993). Of six such effusive eruptions in the 20th and 21st century (Katsui and Katz, 1967;Reynolds et al., 1980;Fierstein and Hildreth, 1992;Singer et al., 2008;Bernstein et al., 2013;Tuffen et al., 2013), three occurred at Puyehue-Cordón Caulle, southern Chile (in 1921-1922, 1960, and 2011Katsui and Katz, 1967;Lara et al., 2004;Singer et al., 2008;Tuffen et al., 2013). Thus, for contemporary hazard as-sessment, as well as for interpreting prehistoric deposits, improving our understanding of silicic lava emplacement processes may be more pressing than previously considered.
Lava flows initially advance in the direction of steepest gradient and their preliminary paths are thus relatively easy to forecast. However, after the lava flow front advance ceases, sec ondary lava flows, or breakouts, can form if effusion continues. Throughout this study we consider a breakout to be a new, morphologically distinct region of lava flow advance, formed from the core of an otherwise stopped or slowed portion of the lava flow. Such breakouts extend or widen the inundated area and uncertainties in where they will occur pose a challenge for forecasting lava inundation hazard. In basaltic lava flows, continued supply of mobile lava to a stalled flow front leads to inflation, and breakouts occur through rupturing of the surface crust (e.g., Walker, 1971;Kilburn and Lopes, 1988;Blake and Bruno, 2000;Applegarth et al., 2010b). Flow lengthening can be substantial when such breakouts are long-lived and fed by thermally preferential pathways that mature into lava tubes (e.g., Calvari and Pinkerton, 1998). However, breakouts in silicic lava flows have only been identified from a collapsed portion of a dacite lava flow margin on Santiaguito volcano, Guatemala (Harris and Flynn, 2002;Harris et al., 2004). Consequently, little is known about breakouts in silicic lavas. Here, we present the first conceptual model for their formation and consider the processes involved.
Previous studies of silicic lavas have interpreted their emplacement in terms of the slow spreading of crystal-poor domes, e.g., for Holocene rhyolite lava flows in the western USA (Fink, 1980a(Fink, , 1980b(Fink, , 1993DeGroat-Nelson et al., 2001) and older rhyolites in Australia (Dadd, 1992;Smith and Houston, 1995) and New Zealand (Dadd, 1992;Stevenson et al., 1994aStevenson et al., , 1994b. As a lava flow advances, cooled surface fragments cascade down the lava flow front and are overridden by the core in a tanktread-style motion typical of block and 'a'ā lava flows (Fink, 1983;Harris et al., 2004). Rhyolite flows form a cooled surface crust that buckles into ogives as the lava flow front slows (Fink, 1980a;Castro and Cashman, 1999). The crust insulates the lava flow core, potentially maintaining core mobility long after effusion ceases (Manley, 1992;Farquharson et al., 2015). Just as for basalts, the crust can therefore exert a controlling influence on lava flow advance and emplacement (Magnall et al., 2017), making the formation of breakouts a possibility. Although a lobate lava flow front (Bonnichsen and Kauffman, 1987;Manley, 1996) and late-stage features such as pumice diapirs (Fink, 1980b;Fink and Manley, 1987) have been previously observed in rhyolite lava flows, breakout processes have not been discussed for rhyolite lava flows. Nevertheless, aspects of these processes could be linked; pumice diapirs are inferred to be caused by vesiculation of the flow core and buoyant rise to the surface, and vesiculation may also be important in breakout formation.
Advances in our understanding of the emplacement of silicic lava flows have been hampered by their infrequency. The 2011-2012 rhyolite eruption of Puyehue-Cordón Caulle in Chile (69.8-70.1 wt% SiO 2 ; Castro et al., 2013) enabled some of the first detailed scientific observations of emplacement processes. This lava flow produced numerous breakouts and provides a new resource to assess breakout processes in a cooling-limited rhyolite lava flow. Here, we use satellite and field observations to understand the processes occurring prior to, and during, breakout formation. Microtextural characteristics from thin sections and synchrotron computed tomography help constrain crystallization and vesiculation histories, allowing us to create a conceptual model for breakout formation in a silicic lava flow.

THE 2011-2012 ERUPTION OF PUYEHUE-CORDÓN CAULLE
Puyehue-Cordón Caulle is a basaltic to rhyolite volcanic complex located in the southern volcanic zone of the Chilean Andes (Fig. 1) umn , followed by coupled effusive-explosive activity with simultaneous emission of lava and pyroclasts from a common vent Schipper et al., 2013;Tuffen et al., 2013;Bertin et al., 2015). Initial discharge rates were up to 50-70 m 3 s -1 but reduced to ~10-20 m 3 s -1 after ~30 days (Bertin et al., 2015;Coppola et al., 2017). Lava effusion was identified first in synthetic aperture radar satellite images on 19 June 2011 (Bignami et al., 2014) and formed two branches that advanced to the north and south, before stalling and forming numerous (>80) breakouts from the lava flow front and margins Magnall et al., 2017). Although the eruption ended between 15 March 2012 and 24 April 2012, as indicated by the cessation of seismicity at the vent (SERNAGEOMIN/OVDAS, 2012; Tuffen et al., 2013) and the end of ash generation (Global Volcanism Program, 2012), respectively, some local areas of advance were still observable in January 2013 .

METHODOLOGY
Our multiscale data analysis of breakouts and the main lava flow brings together satellite imagery, digital elevation models (DEMs), 3-D photogrammetry models, as well as scanning electron microscopy (SEM) and synchrotron X-ray computed tomography (CT) imaging of samples collected in the field. Fieldwork at Cordón Caulle was carried out during and after the eruption, in 2012, 2013, and 2015 (always in January), and allowed for sample collection and a detailed examination of the structure of the lava flow and breakout morphology. Approximately 50 in situ samples were collected from the lava flow front and top surface, and from fallen fragments at the lava flow front. Fieldwork in 2015 was complemented by aerial photographs acquired from a helicopter.

Satellite Imagery
The lava flow evolution was tracked using satellite images from the Earth Observing 1 (EO-1) Advanced Land Imager (ALI; e.g., Digenis et al., 1998), Landsat 5 and 7 Enhanced Thematic Mapper Plus (ETM+), and the Terra Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER; Yamaguchi et al., 1998). EO-1 and Landsat data are available from the U.S. Geological Survey, and ASTER data are distributed by the Land Processes Distributed Active Archive Center (LP DAAC), located at the U.S. Geological Survey Earth Resources Observation and Science Center (USGS/EROS), Sioux Falls, South Dakota, USA, (http:// lpdaac .usgs .gov). Google Earth imagery was also used. All images, except those from Google Earth, were processed in ENVI to create true color images, or false color images from the visible, near infrared (VNIR) and the shortwave infrared (SWIR) bands, with spatial resolutions between 15 and 30 m. Where possible, the resolution was improved by combining the images with their corresponding panchromatic image (EO-1 10 m resolution, Landsat 7 ETM+ 15 m resolution) using the NNDiffuse-PanSharpening function in ENVI, allowing the detection of lava flow features larger than a few tens of meters across. The resulting imagery was imported into ArcGIS and the coregistration refined using fixed points, such as roads, lakes, and older lavas that were assumed to be stationary throughout the duration of the eruption.
The qualitative thermal evolution of the lava flow surface was tracked using thermal infrared (TIR) images from Landsat 5, 7 (Band 6), and ASTER (L1T) sensors, visualized in ArcGIS. These TIR images have coarser resolution (Landsat 30 m, ASTER 90 m) than their VNIR and SWIR counterparts, and dense time-series analyses were restricted by the eruptive plume and cloud cover, which often obscured parts of the lava flow.

3-D Models from Field Photos and DEMs
To determine the detailed structure of five selected breakouts, 3-D surface models were generated from ground-based field photos using structure-from-motion photogrammetry (e.g., James and Robson, 2012). Approximately 100 photos were taken on foot at regular intervals in an arc around a feature of interest using a Canon 450D digital singlelens reflex camera, with a fixed focus 28 mm lens. The photos were processed in Agisoft PhotoScan Professional (v1.3.2) to generate dense 3-D point clouds, which were manually cleaned to remove outlying points. The models were scaled by recognizing common features in georeferenced satellite images of the lava, giving an overall scale accuracy of 5-10%, suitable for qualitative assessments of lava morphology. Cross sections were obtained by first gridding the 3-D point clouds into 10-cmresolution DEMs, then extracting the sections.
The morphology of the largest breakouts could not be observed fully from the ground so such areas were analyzed using a post-eruption Pléiades DEM (Elevation 1 data set) with a 1 m horizontal resolution and 1.5 m horizontal and vertical accuracy. The DEM was generated from stereo satellite images, so the bases of deep and narrow features, such as fractures, may not have been reconstructed due to being occluded in either one or both images. 2-D profiles were extracted by sampling the DEM at a 2 m resolution, to help reduce the noise from the highly uneven and blocky nature of the lava surface.

Samples and Thin Sections
Microstructural characterization of key representative lava facies was carried out by SEM analysis using a LEO 435VP variable pressure digital scanning electron microscope. Mineral identification was aided by qualitative observation of energy-dispersive X-ray spectra recorded simultaneously during SEM observation, using an Oxford Instruments INCA energy-dispersive X-ray microanalysis (EDXA) system. Images were captured from thin sections in back scattered electron mode for further image analysis, which included determining microlite contents using CSDCorrections software (Higgins, 2000). Up to ~400 microlites in SEM images of ~15 representative samples (from four breakouts) were first isolated by tracing the crystal edges. The resulting image was analyzed in ImageJ and CSDCorrections to determine total microlite volume fractions. The crystal volume fractions presented here (as percentages) are the stereologically corrected intersection data from CSDCorrections (Higgins, 2002) and have been compared with similar data from Schipper et al. (2015). Vesicularity (Table 1) was estimated from the 2-D vesicle fraction in SEM images using ImageJ. This straightforward method only gives an approximate value of vesicularity, but enables comparisons among different lava facies.
Textures in breakout samples were examined in 3-D using synchrotron X-ray source computed tomography (CT), performed at the Australian Synchrotron (Clayton, Victoria) in hutch 3B of the Imaging and Medical Beamline. Images were captured using the in-house "Ruby" detector, with X-ray energy of 45 keV. Scans included the collection of 1800 radiographs with 0.3 s exposure time over 180° rotation and a pixel size of 13.7 µm. The length of the samples required three partially overlapping scans, which were reconstructed separately using the in-house MASSIVE supercomputing cluster, and then stitched together using ImageJ (Schneider et al., 2012). All renderings were performed with the Drishti visualization software package (Limaye, 2012).

Lava Flow Emplacement
Satellite observations from EO-1 ALI have been previously used to show that the ~40-mthick lava flow was emplaced into a topographic depression and, within ~600 m of the vent, split into two main branches 600-1000 m wide, of which one advanced north and the other south ( Fig. 1; Magnall et al., 2017). The southern branch moved down a relatively constant slope of ~7° before interacting with a topographic barrier toward the end of its emplacement. The northern branch was similarly impeded by a ridge within the first 30-60 days of the eruption (from the eruption start on 4 June 2011; Magnall et al., 2017). This ridge halted the northern advance of the lava and led to a widening of the lava flow front (~200 m, yellow and turquoise outlines in Fig. 1A).
The first breakouts were observed at the western margin of the northern branch, 153 days after the eruption started (90-120 days after stopping against a ridge; Fig. 1C), and following the development of flow front lobes 50-250 m wide and 50-200 m long separated by large fractures that extended 25-200 m backward from the flow front (Fig. 1C, Data Repository file 1 1 ). Here, a lobe is distinguished from a breakout through forming via continued but spatially uneven advance of a lava flow front, rather than the effusion of core material from a slowing or stalled lava flow margin. In optical satellite images (Figs. 1C, 1D), the breakouts appear darker than the main lava flow but, in ASTER TIR data, breakouts appear as bright features at the lava flow front (Figs. 2C, 2D; Coppola et al., 2017). Some of the earliest breakouts were subsequently enveloped by the forward advance of the main branches ( Fig. 1), and became subsumed into the main body of the lava flow. Most breakouts range in size from ~15 to 40 m thick, 30 to 150 m wide, and 30 to 450 m long; the longest occur in areas with little topographic constraint. Breakouts from the southern branch developed predominantly from its sides, following lobe generation as seen in the northern branch; some small breakouts (up to 60 m long and 50 m wide) formed from the front toward the end of the eruption (Fig. 1A).
After 200 days (first observed 23 December 2011, 110-170 days after stopping against a ridge), a breakout formed from the eastern margin of the northern branch, and developed a substantial length (~2 km; Magnall et al., 2017); hereafter this breakout is referred to as the northeast breakout ( Fig. 1A). This new breakout entered a relatively steep (~15°) and narrow (200-350-m-wide) valley, and halted only when the eruption ended ( Fig. 1). In TIR imagery, the northeast breakout is brighter than the bulk of the main lava flow throughout its lifetime and for several months after the eruption ceased (Figs. 2B, 2C). The breakout developed distinct levees and appears to have sections of partially drained channel, which likely relates to this breakout being volume limited and stopping once supply to it ended. After the establishment of this breakout, there was a noticeable reduction in the rate of initiation of new breakouts from the northern branch ( Fig. 1). Between December 2011 and January 2012, tens of breakouts formed from the northern branch but, after 26 January 2012, when the northeast breakout had become established ( Fig. 1), only five new breakouts were formed in the following two months before effusion ended on 15 March 2012.
Breakouts continued to form and evolve for some time after the eruption ended. In the western margin of the northern branch, some of the breakouts continued to grow (Figs. 3B, 3C), but some showed only limited post-eruption advance (<50 m). Instead, fractures (up to ~100 m long) developed within the breakouts and parallel to their advance direction (Figs. 3B, 3C). From the southern branch, ~5 new breakouts formed at its eastern margin (Figs. 1A and 3) and developed similar, long fractures (up to 300 m long; Fig. 3E). One of these breakouts was observed advancing in January 2013, ten months after the eruption ended Farquharson et al., 2015).
In total, ~90 breakouts formed ( Fig. 1), accounting for ~12% (±2%) of the total lava flow volume (as estimated from the areal coverage in satellite imagery and assuming an average breakout thickness of ~80% of the main branch thickness, i.e., ~32 m). The volume of the breakouts is thus equivalent to a 4 m increase in thickness of the main branches.

Lava of the Main Northern and Southern Branches
The main northern and southern branches have coarse, red-brown rubble surfaces composed of fragments up to several meters across, and are bounded by steep and blocky levees, comprising a mixture of loose rubble and coherent, dense, core lava (Fig. 4). Parts of the lava surface near the vent appear lighter in color and less blocky and represent areas where collapsed portions of the tephra cone have been rafted on the lava surface. The surface of the lava is systematically folded into ogives with a wavelength of ~40 m and an amplitude of order several meters (Magnall et al., 2017). The lava flow surface is also split by fractures, most of which are oriented perpendicular to the lava flow front (Data Repository file 1; see footnote 1), tens to hundreds of meters long and several meters deep, and many expose a surface crust of coherent and dense lava beneath the loose surface fragments.

Textures of the Main Branch Lava
The fragments that comprise the top surface and lava flow front of the main branches are a mixture of dark vesicular and glassy rhyolite, golden colored lava-flow-top-pumice, and gray to red-brown dense core lava (Fig. 4). The dark glassy rhyolite is the most abundant on the lava surface, whereas the lava-flow-top-pumice occurs as isolated patches (~1-20 m wide) on the lava surface (Figs. 4B, 4E, and 4F). The core of the lava is exposed in areas where portions of the lava flow front and margins have collapsed ( Fig. 4A) and is microcrystalline, with sparse sub-millimeter vesicles (Table 1; Figs. 4C, 4D). Millimeter-spaced, parallel, platy fractures, associated with vapor-phase cristobalite formation (Figs. 5A, 5B; Schipper et al., 2015), are laterally extensive over tens of centimeters (Fig. 4D).

Microstructure
Samples from the northern and southern branches include those from the lava flow core, surface, and lava at the vent. SEM analysis identified two main crystal populations in this lava, phenocrysts and microlites. The core samples are microcrystalline (~70-80 vol%) and display the extensive millimeter-spaced platy fractures. In some cases, the fractures have a number of 1 GSA Data Repository item 2018238, (1) Separated flow outlines showing ogive and fracture networks on the surface of the Cordón Caulle lava flow; (2) 3-D PDF of the 3-D model in Figure 7 (note this only works in AdobeAcrobat); (3) Timeline of feature development, and crystallization history, of the Cordón Caulle rhyolite lava flow, is available at http:// www .geosociety .org /datarepository /2018 or by request to editing@ geosociety .org. s-shaped shear arrays (0.5-1 mm long and 0.05-0.1 mm wide, Fig. 5A) and, in many places, the two sides of the fractures are connected by thin (1-8-µm-wide) glassy filaments.
The surface samples from the main branches comprise vesicular flow-top-pumice (Figs. 5C, 4B, and 4F). Pumice samples have highly contorted, elongate vesicles (Table 1; Fig. 5C) and a microlite volume of ~5 vol%. Samples from the vent represent some of the last effused lava. The samples are vesicular (Table 1) and have poorly developed flow-bands marked by subtle variations in microlite content (~5 vol% variation, Fig. 5D). The microlite content of the vent samples is ~20-25 vol% (Table 1).

Breakouts
The large number of breakouts gives the overall lava flow a lobate outline in plan view, and their darker, more slabby and blocky morphology contrasts with the redder, more rubbly sur-face of the main branches . Analysis of breakouts used a mixture of satellite imagery (>50 breakouts analyzed), aerial photography (10 breakouts), topographic cross sections from a high-resolution DEM (20 breakouts), detailed field observations (20 breakouts), and 3-D photogrammetry models (5 breakouts). Although a comparison of bulk composition between breakouts and main branch samples has not been conducted, analyses of groundmass glass of breakout and main branch lava shows little variation (Schipper et al., 2015). Phenocryst assemblages (constituting plagioclase, pyroxene, magnetite, and ilmenite) were also found to be identical in lava samples and pyroclasts from the eruption .

Sources
Field and satellite observations suggest that breakouts were sourced from either (1) the base or part way up the lava flow margin, or (2) large fractures extending back into the main branches. For some breakouts (particularly those in the southern branch that was obscured by the eruptive plume in satellite images), the source region is difficult to determine. Some source regions were also sites of preferential outgassing, indicated by a white precipitate covering many of the nearby fragments (Fig. 6A) and associated gas plumes observed in 2012. Similar outgassing was observed at stalled portions of the main branches (Fig. 6C).
Small breakouts from the base of the lava flow margin (Fig. 6A) emerged as thin (~10-15-m-thick) lava bodies that then underwent inflation and advance. In some cases, a region of the main branch behind these breakouts subsided to form an amphitheater-like structure up to 10 m deep (Figs. 6A and 7). Such amphitheaters are typically associated with the smallest breakouts, suggesting that the breakout is sourced by drainage from only a limited area of the main branch rather than being connected to an extended, and more continuously supplied, source (Fig. 7).
Other breakouts were sourced from fractures (≤4 m deep, 5-10 m wide) that extended ~50-150 m back into the main branch (Fig. 6B). The satellite data showed that breakouts from fracture zones were preceded by the formation of lobes and extensional fractures at the lava flow margin (Fig. 1, Data Repository material [see footnote 1]). Breakouts sourced from these fractured regions formed the largest breakout complexes (multiple potentially branched breakouts, up to hundreds of meters long).

Morphologies
The breakouts show a range of morphologies, which we describe in four general classes: domed, petaloid, rubbly, and cleft-split ( Table 2; Fig. 8). Domed breakouts are usually sourced from the base or part way up the lava flow margin (Figs. 7 and 8). These breakouts have a domed shape and are wider than they are thick (Fig. 8). The surface comprises slabs (several meters across, Table 1) of dark, glassy, sur-face crust (Figs. 7 and 8), which retain a jigsaw-fit pattern but are disrupted by fractures tens of meters long and several meters deep. Commonly, sections of the breakout have collapsed, to expose the lighter gray-colored core lava.
Petaloid breakouts are steep sided and have an uneven surface due to a large number of upturned surface slabs (Fig. 8), which are up to tens of meters tall (Table 2)   into lighter colored core lava. Regions between the slabs and spines are filled with rubble (tens of centimeters to a meter across), and fragments form a relatively narrow (<10 m) talus apron surrounding breakouts. The surface slabs of petaloid breakouts are more isolated than in domed breakouts, but locally retain jigsaw-fit patterns.
Rubbly breakouts (including the large northeast breakout) have a surface that is almost entirely covered in rubble and disrupted slabs (Table 2; Fig. 8). The rubble comprises brecciated crust and core lava, and includes large slabs that are similar to the surface slabs and spines observed in petaloid breakouts (Fig. 8), giving an uneven surface profile. However, slabs are sufficiently disjointed that any semblance of a jigsaw-fit pattern is lost. Lateral collapses form larger talus aprons than around domed or petaloid breakouts (~20-40 m wide) and, in some cases, expose the lighter colored core lava (Fig. 8).
Cleft-split breakouts have a large central cleft that is orientated parallel to the direction of advance, typically tens of meters long and several meters deep, and runs most, or the entire length, of the breakout (Fig. 8). Cleft surfaces are striated on a decimeter scale, marking repeated transitions from rough to smooth fracture surfaces. Outside of clefts, the breakout surface comprises of either slabs or rubble ( Table 2). Some of the most complex cleft-split breakouts are also branched and, in a few cases, subsidiary breakouts are observed (Fig. 8).
The four breakout morphologies present part of an evolutionary continuum and intermediate morphologies are also observed. For example, some rubbly breakouts show features associated with petaloid breakouts, such as disjointed slabs of surface crust amongst the rubble. Post-effusion satellite images of breakouts (Fig. 3) also demonstrate the continuing evolution of breakouts, whereby rubbly and petaloid breakouts develop into cleft-split morphologies (as indicated by the red lines in Fig. 3). Furthermore, the fracturing and upturning of surface slabs, associated with the transition from domed to petaloid morphologies, was observed during the eruption .

Textures of Breakout Lava
Breakouts comprise dark, glassy lava with a black scoriaceous surface crust that grades into a lighter gray interior (Figs. 9A, 9B). In handspecimen, the lava is vesicular, with a mixture of undeformed, near-spherical and more elongate vesicles (Fig. 9C) top few tens of centimeters of the surface crust, many of the vesicles display white walls, representing regions in which interstitial glass has been corroded during a vapor-phase cristobalite forming process (Schipper et al., 2015(Schipper et al., , 2017. Flow-bands are also common; lighter and darker bands are typically a few millimeters thick but laterally extensive over tens of centimeters (Figs. 9B and 9D). In many cases, these bands are folded and wrap around vesicles (Fig. 10D).
A slightly different texture is observed in one of the longest (400 m) rubbly breakouts at the western margin of northern branch. Here, the lava is glassy and has a large population of white-walled, corroded vesicles, typical of breakout lava, but is also cut by the millimeterspaced platy fractures (coated and partially filled with cristobalite) that are most commonly associated with the core of the main branch (Fig. 9F). The texture represents an intermediate between that of the fractured main branch core and the vesicular breakout lava.

Microstructure
SEM images of breakout samples verify the presence of two vesicle populations and show the three crystal populations of phenocrysts, microlites, and devitrification-related crystalline masses. For the purposes of this study, the phenocrysts (<5 vol% of the lava) will not be assessed further as they are unlikely to have had much, if any, influence on emplacement processes. The microlites primarily comprise plagioclase, some of which are subtly zoned (Fig. 10E), and some pyroxenes as well as accessory apatite and metal oxides (Fig. 10E). Microlites are 2-10 µm wide and 5-50 µm long, and typically aligned parallel to the flow-bands. They comprise 30-40 vol% (measured from four breakouts with petaloid, rubbly, and cleft-split morphologies) in glassy flow-bands where devitrification crystallization is absent. The microlite abundance varies slightly (~5 vol%) between separate millimeter to submillimeter bands.
Darker, spherical, crystalline masses (0.03-0.4 mm diameter, Fig. 10), some of which surround vesicles (Fig. 10C), are found in the core of breakouts. These crystalline masses are surrounded by small vesicles or fracture networks 1-2 µm wide but lack the radiating crystal structure commonly observed in spherulites and are best classified as globulites, which form as a result of devitrification (Lofgren, 1971;Schipper et al., 2015). Globulites comprise ~70 vol% micro lites (with overgrowths of 1-2 µm in width, Fig. 10F) and ~30 vol% cristobalite. Where globulites merge, they form substantial flow-bands up to several centimeters thick. Globulites are observed in all breakout samples except those from the quenched surface crust.  The two vesicle populations comprise (1) sheared, partially collapsed vesicles and (2) spherical and relatively undeformed vesicles, both of which can be observed in hand specimen (Fig. 9C), thin section (Figs. 10B-10D), and CT renderings (Fig. 11). Sheared vesicles are elon-gate (Table 1) and many are pinched at one end, suggesting partial collapse. Some form vesicle trails (Fig. 10B), but many are found in isolation (Fig. 10C). Sheared and collapsed vesicles are more abundant in samples from deeper in breakouts than at the surface. In contrast, spherical (Table 1) and undeformed vesicles (Figs. 9C, 10D, and 11) are found throughout most breakouts but are more common near the surface and in flow-bands (Fig. 11A).
Many of the vesicles located deeper within the breakouts have uneven walls (Fig. 11C) that have undergone partial preferential dissolution of interstitial groundmass glass due to gas flux through the vesicle network (Schipper et al., 2015). Dissolution led to the formation of a diktytaxitic halo (0.05-0.22 mm wide) that produced the white-walled appearance in hand specimen (Fig. 10D). However, haloes are absent from vesicles in the uppermost (upper tens of centimeters) glassy crust.

DISCUSSION
Breakouts represent an important part of the emplacement process and constitute ~12% of the final lava flow volume at Cordón Caulle. Furthermore, evidence from satellite images, field observations , and numerical modeling (Farquharson et al., 2015) indicate that they can remain active, and hence hazardous, many months after effusion ended.

Conditions for Breakout Formation
For breakouts to form, the local conditions must evolve to favor the initiation of a new path for lava advance rather than continuing with the current scenario. Initiation of a breakout arises when flow front advance slows or ceases, despite continued lava supply from the vent, and can result in a change in the direction of lava propagation. More specifically, breakouts form when the internal pressure within the lava flow exceeds the local confining force, due to increased core pressure and/or decreased crust strength. Crust weakening is enhanced by lateral spreading of lava flow front lobes (Applegarth et al., 2010a) and higher effusion rates at the vent can increase core pressure by propagating an increased supply rate to the lava flow front, even if the front is still advancing. Within-flow crystallization (e.g., second boiling) can also pressurize the core by increasing the melt vapor pressure, leading to vesicle growth.

Breakout Initiation
At Cordón Caulle, both internal pressure increase and crustal weakening are likely to have contributed to breakout formation. The breakouts that formed from the completely stalled lava flow margins, in areas without clear crustal weakening (as indicated by crust fractures) such as the static margin of the southern branch (e.g., Figs. 6A and 7), suggest that the core pressure of the lava exceeded the yield strength of the surface crust, which, based on previous work is likely to be between 10 6 and 10 8 Pa (Griffiths and Fink, 1993;Bridges, 1997;Fink and Griffiths, 1998;DeGroat-Nelson et al., 2001;Magnall et al., 2017). Such pressure increases could have resulted from the continued supply of lava through thermally preferential pathways (or "fluid pathways") beneath the static surface crust. Alternatively, small domed breakouts formed from static portions of the lava flow margins, where thermal pathways were likely absent or potentially only locally developed (Fig. 1), suggest that more local processes, such as margin collapse or late stage core vesiculation, could also have been important.
Late-stage vesiculation occurred within most breakouts, indicated by the presence of a relatively undeformed vesicle population (Figs. 9C, 10, and 11); the limited deformation interpreted to reflect a short distance of flow, e.g., from breakout advance only. Consequently, these vesicles probably result from volatile exsolution either due to a pressure decrease during breakout extrusion, or due to crystallization occurring between the vent and lava flow front. Lava crystallinity increases from <25 vol% at the vent (as low as <5 vol% in some lavaflow-top-pumice samples) to ~30-40 vol% in breakout samples. The increase in crystallinity as lava moved toward breakouts will have increased the volatile concentration in the melt and potentially driven second boiling  or enhanced vesicle nucleation and growth, if the melt was already at volatile saturation. Some support for distal volatile exsolution from the lava flow core is provided by the observations of outgassing plumes at  the stalled lava flow front (Figs. 6C and 12) but, strictly, this is proof only of gas migration and escape, and concurrent vesicle growth for contributing to pressure increase can only be inferred. Nevertheless, this gas remobilization likely contributed to the precipitation of cristobalite throughout the lava flow (Schipper et al., 2015) and late-stage volatile exsolution, driven by in-flow crystallization, has also been inferred to form the vesicular layers observed in many prehistoric rhyolite lavas (e.g., Fink, 1980b;Fink and Manley, 1987;Manley and Fink, 1987). The large extensional fractures associated with lobe formation at the lava flow front were often preferential regions of breakout formation; breakouts appear to exploit the weakened crust (Figs. 6B and 12B). A similar process has been observed in laboratory experiments, where tensile surface fractures in a spreading and lobate analogue lava flow front enabled upwelling of the viscous core (Applegarth et al., 2010a).

Breakout Evolution
The growth and inflation of breakouts is inferred to have occurred through a combination of continued magma supply and vesiculation. The smaller, morphologically more simple breakouts form from areas of the lava flow without continued supply by thermally preferential pathways (i.e., the lateral margins rather than the fronts; Figs. 1 and 8). Small domed breakouts, some with a drained amphitheater structure in the adjacent levee (Figs. 6A, 12A, and 13D), indicate a limited, local magma supply. In these breakouts, vesiculation caused by decompression after extrusion from the main branches could have played a significant role in the overall inflation, with the observed vesicularities (15-30 vol%, Table 1) representing a potential contribution of meters to inflation (or a lateral advance of tens of meters). Decompression-driven vesiculation of breakout lava likely contributed to the formation of their scoriaceous and glassy crust (Fig. 9) which grades into the lighter, gray, breakout interior with abundant globulites (Fig. 10A). Given that globulites form due to devitrification below the glass transition temperature (Schipper et al., 2015), their growth is unlikely to have driven vesiculation within breakouts and they probably had negligible influence on breakout development.
Breakout inflation causes increased fracturing and upturning of surface slabs . From observations of inflation , intermediate morphologies, and the development of inflationary clefts (Fig. 3), we infer a breakout evolution from domed to petaloid, to rubbly then cleft-split morphologies. Cleft-split morphologies may also develop directly from petaloid breakouts where continued lava supply leads to inflation, rather than advance. The striated surfaces of the resulting clefts mark repeated transitions from brittle fracturing to ductile tearing during the cleft opening (Forbes et al., 2012;Forbes et al., 2014). How- ever, clefts do not have the centimeter-scale striations and convex sides typical of crease structures commonly observed at other silicic lavas and domes, which suggests that the simultaneous upwelling and deformation of core lava asso ciated with crease structure formation did not occur (c.f. Anderson and Fink, 1990;Anderson and Fink, 1992). We therefore propose that the formation of clefts within cleft-split breakouts is similar to that of "lava-inflation clefts" observed in basaltic pahoehoe tumuli, where a surface crust is buckled upwards by an increased core pressure (Walker, 1991). Nevertheless, in contrast to basaltic lavas, we consider that latestage vesiculation at Cordón Caulle provided an important contribution to this inflation. The sustained advance of breakouts further promoted crust brecciation, and breakouts that  were active for months required the establishment of extended thermally preferential pathways to maintain a continued lava supply. In one of the longer rubbly breakouts, the observation of platy fractures suggests that sustained breakouts may begin to develop microstructure characteristics comparable with those of the main branches. With such characteristics not observed in smaller breakouts, it is likely that the presence of platy fractures reflects processes that occur only toward the end of the emplacement of long breakouts or substantial branches.
The surface morphology of a breakout may also reflect the rate at which it grows, just as the transition from pahoehoe to 'a'ā lava can be driven by increased effusion and strain rates (Pinkerton and Sparks, 1976;Rowland and Walker, 1990;Cashman et al., 1999;Hon, 2003).
In analogue experiments of silicic lava dome formation, low effusion rates (relative to cooling rate) aided surface crust formation and favored spiny or platy morphologies (Fink and Griffiths, 1998), whereas higher effusion rates created smoother and lower profiles (Fink and Griffiths, 1998). Thus, we propose that the larger volume rubbly breakouts (Table 2) formed at higher supply rates than domed and petaloid breakouts, with higher supply rates facilitated by efficient thermally preferential pathways. Future work should attempt to systematically relate breakout properties (i.e., volume, effusion rate, rheology) to breakout morphologies and the transitions between different morphological types. Such work would likely combine analysis of breakout types, and comparisons to features generated in analogue experiments (e.g., Griffiths and Fink, 1993;Griffiths and Fink, 1997;Fink and Griffiths, 1998).
Finally, it is likely that breakouts did not all evolve independently, particularly those linked by thermally preferential pathways. For example, the onset of the formation of the large northeast breakout coincided with a decreased rate of new breakout initiation from the main northern lava flow (Fig. 1, Data Repository file 1). This relationship suggests that the supply of lava toward the northeast breakout reduced lava supply to the northern branch. Nevertheless, the drainage (or partial drainage) of thermally preferential pathways enabled continued formation and evolution (e.g., fracture growth) of some breakouts even after effusion at the vent had ceased (Fig. 3). The rubbly northeast breakout ultimately attained a final length of ~2 km and it represents an example of the challenge for hazard forecasting that late-stage breakouts present.

Breakout Formation at Other Lavas
Breakouts are widespread in basaltic lava flow fields where the process of continued flow beneath a static surface crust is a fundamental aspect of the inflation and growth of pahoehoe flow fields (e.g., Self et al., 1996;Anderson et al., 1999;Vye-Brown et al., 2013) and is a recognized process in lengthening 'a'ā flow fields (e.g., Calvari and Pinkerton, 1998). In both basaltic and silicic lava flows, the continued supply of lava to a stalled lava flow front can generate breakout formation through driving a pressure increase and eventual rupturing of the surface crust (Pinkerton and Sparks, 1976;Blake and Bruno, 2000). However, in silicic lava flows, crystallization-driven late-stage vesiculation or vesicle growth is also likely to provide an important contribution to pressurization. This process will not be as important in basaltic lava flows, where lower lava viscosities allow faster bubble migration (e.g., Vergniolle and Jaupart, 1986) and substantial vesicle loss can be measured with increasing distance from the vent (Cashman et al., 1994). However, vesiculation-driven inflation may play a minor role in the observed inflation of some mafic lava flows and lava lakes (e.g., Peck, 1978;Stevens et al., 2001). The prevalence of breakouts in silicic lava flows may be currently underestimated, and other likely candidates are present in an earlier lava at Cordón Caulle (1960). One of the 1960 lavas has a thin lobe (800 m long and 80 m wide, 40.542765°S, 72.235110°W) with an overall morphology similar to the northeast breakout of the 2011-2012 lava (Fig. 1). At the front of this rubbly potential 1960 breakout, a domed and slabby front (~50 m in diameter) resembles a transitional domed-to-petaloid breakout (Fig. 8). In some of the large rhyolite lavas in the western USA, marginal lobes (often interpreted as forming due to the splitting of the lava flow front) have also been identified (Bonnichsen and Kauffman, 1987;Manley, 1992Manley, , 1996 but they have not been discussed in terms of breakout processes. The margin of the dacitic portion of Glass Mountain at Medicine Lake Volcano, California, USA, has a particularly pronounced lobate flow front, and satellite images show that some lobe surfaces partially comprise large fractured slabs, similar to the domed and petaloid breakouts at Cordón Caulle (Figs. 8 and 13)  Continued supply to the breakout, coupled with core vesiculation, leads to inflation and upturning of surface slabs. Globulites formed in the breakout core after emplacement and below the glassy surface crust, these have limited impact on the breakout development. -Nolan et al., 2016), which lacks such a lobate margin (Fig. 13). This difference could suggest that emplacement processes for the Cordón Caulle lava flow had more in common with those of glassy dacitic lava flows than some higher silica rhyolite lava flows. Although breakouts are not readily identifiable at most other rhyolite lavas, other emplacement features have been linked to late-stage vesiculation. In some higher silica (>70% SiO 2 ) rhyolites, diapirs of coarsely vesicular pumice (Fink, 1980b(Fink, , 1983Fink and Manley, 1987) are thought to form due to bubble growth in the core of the lava flow Furukawa and Uno, 2015) and could represent a vertically orientated equivalent of lateral breakouts. Similarly, the formation of large gas cavities capa ble of triggering small explosions (DeGroat-Nelson et al., 2001;Castro et al., 2002), is thought to be caused by the accumula-tion of gas, in association with a buckling surface crust, in a layer directly beneath the surface crust (Jensen, 1993;Castro et al., 2002). The late-stage vesiculation inferred at Cordón Caulle did not lead to any observed pumice diapirs or substantial gas cavities, and further work could explore whether the small compositional differences between Cordón Caulle and more silicic rhyolites are responsible for observed differences in late-stage evolution.

CONCLUSIONS
The 2011-2012 rhyolite lava flow at Cordón Caulle generated ~90 breakouts and provided the first observations of breakout processes during such an eruption. Our multi-scale analyses have enabled a morphological classification of rhyolite breakouts and given insight into their importance as part of the emplacement of silicic lava flows. The breakouts are likely initiated by a combination of pressure build-up in the core of the lava flow (due to continued lava supply and vesiculation, partly driven by in-flow crystal growth), and fracturing of the surface crust. During and after extrusion from the slowed lava flow margins the breakouts developed into a sequence of morphologies from domed, petaloid, rubbly, to cleft-split, which reflected their evolving stages of growth and inflation. Inflation is inferred to be driven by both lava supply and vesiculation within the core of a breakout, and was responsible for the breakup of surface slabs and the formation of inflation clefts.
The continued supply and growth of breakouts was due to the presence of active thermally preferential pathways, supplying fresh lava beneath a stationary lava crust after advance of the main flow had ceased. The potential presence of such pathways should be considered in the interpretation of emplacement processes at other laterally extensive rhyolite lavas. Furthermore, the evidence for late-stage vesiculation in breakouts has substantial hazard implications; volatile accumulation could lead to surface explosions or flow front collapse. The processes of breakout formation and lava supply through thermally preferential pathways that we have showed for a rhyolite lava are well-studied in mafic lava flows, and direct comparisons can be made across the compositions. However, in rhyolites, the driving effects of late-stage vesiculation are likely to be strongly enhanced with respect to those in basalts, due to the high-viscosity inhibiting pressure release and vesicle migration.

ACKNOWLEDGEMENTS
NM is supported by a Natural Environment Research Council-Envision studentship and a British Geological Survey University Funding Initiative grant. HT is supported by a Royal Society University Research Fellowship. CV-B publishes with the express permission of the executive director of the British Geological Survey. CIS acknowledges access to the Australian Synchrotron's Imaging and Medical Beamline, Clayton, Victoria, granted under proposals 2014/1-M7574 and 2017/1-M11688, with travel support from the New Zealand Synchrotron Group Ltd. JMC is supported by the Volcanoes and Atmosphere in Magmatic Open Systems research center, University of Mainz, Germany. We gratefully acknowledge the assistance of Gren Turner (British Geological Survey) with scanning electron microscope imaging of the thin sections. We also wish to thank the BBC Earth team for assistance with aerial photography of the lava flow. Earth Observing (EO-1) data were obtained via the National Aeronautics and Space Administration's (NASA) Volcano Sensor Web at the Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA. EO-1 was managed by the NASA Goddard Space Flight Center, Greenbelt, Maryland, USA. We gratefully acknowledge editorial handling and helpful comments from Brad S. Singer and Jocelyn McPhie and constructive reviews from Jonathan Fink and an anonymous reviewer.