Volume 4, Issue 9
Free Access

A new model for submarine volcanic collapse formation

Jennifer L. Engels

Jennifer L. Engels

Department of Geology and Geophysics, School of Ocean and Earth Science and Technology, University of Hawaii at Manoa, 1680 East-West Road,, Honolulu, Hawaii, 96822 USA

Search for more papers by this author
Margo H. Edwards

Margo H. Edwards

Hawaii Institute of Geophysics and Planetology, School of Ocean and Earth Science and Technology, University of Hawaii at Manoa, 1680 East-West Road,, Honolulu, Hawaii, 96822 USA

Search for more papers by this author
Daniel J. Fornari

Daniel J. Fornari

Department of Geology and Geophysics, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts, 02543 USA

Search for more papers by this author
Michael R. Perfit

Michael R. Perfit

Department of Geological Sciences, University of Florida, Gainesville, Florida, 32611 USA

Search for more papers by this author
Johnson R. Cann

Johnson R. Cann

School of Earth Sciences, University of Leeds, Leeds, LS2 9JT United Kingdom

Search for more papers by this author
First published: 18 September 2003
Citations: 29

Abstract

[1] Collapse pits and an associated suite of collapse-related features that form in submarine lava flows are ubiquitous on the global mid-ocean ridge crest. Collapse pits, the lava tube systems they expose, and lenses of talus created by the collapse process combine to produce a permeable region in the shallow ocean crust and are thought to contribute significantly to the 100–300 m thick low velocity zone observed at intermediate to fast-spreading mid-ocean ridges. This horizon of low-density, high-porosity material is likely to be an important aquifer for the transfer of hydrothermal fluids in the upper ocean crust. In a recent survey of the East Pacific Rise at 9°37′N, we used photographs, video and observations from the submersible Alvin, and DSL-120A side scan data to determine that 13% of the 720,000 m2 of seafloor imaged had foundered to form collapse pits. In 98% of the images collapse pits occurred in lobate flows, and the rest in sheet flows. On the basis of our observations and analyses of collapse features, and incorporating data from previous models for collapse formation plus laboratory and theoretical models of basalt lava behavior in the deep ocean, we develop a detailed multistage physical model for collapse formation in the deep ocean. In our model, lava extruded on the seafloor traps pockets of seawater beneath the flow that are instantly vaporized to a briny steam. The seawater is transformed to vapor at temperatures above 480°C with a 20 times expansion in volume. Bubbles of vapor rise through the lava and concentrate below the chilled upper crust of the lava flow, creating gas-filled cavities at magmatic temperatures. Fluid lava from the cavity roofs drips into the vapor pockets to create delicate drip and septa structures, a process that may be enhanced by water vapor diffusing into the magma and reducing its melting point. As the vapor pocket cools, the pressure within it drops, causing a pressure gradient to develop across the upper crust. The pressure gradient often causes the roof crust to collapse during cooling, though vapor pocket geometry may be such that the roof remains intact during subsidence of the underlying lava. Alternatively, drainaway of the molten lava may cause collapse in locations where inflated lava roof crusts are not supported from below by bounding walls or lava pillars. Post-eruption seismicity, lava movement, or hydrovolcanic explosions may cause continued collapse of the lava carapace after the eruption.

1. Introduction

[2] The term “submarine collapse features” refers to a suite of volcanic structures that are related to the inflation and subsequent draining of fluid lavas erupted at the seafloor (Figure 1). Collapse features include pits within the volcanic carapace or crust (Figure 1), and associated features such as: cylindrical lava pillars, overhanging lava roofs, piles of talus from foundered roof material, and delicate lava drips or septa located on the undersides of roofs and attached to lava pillars (Figure 2). The dimensions of lava pits range from a few centimeters deep and <1 m in diameter up to ∼25 m deep and hundreds of meters in length or width. Submarine collapse features have been observed in lobate and sheet flows at all mid-ocean ridges (MORs) regardless of spreading rate (though they are most common in sheet flows associated with fast-spreading ridges) and at other sites of submarine volcanism by many authors since systematic photographic surveys of the seafloor began in the 1970s [e.g., Ballard et al., 1979; Wilkins et al., 1991; Embley and Chadwick, 1994; Fornari et al., 1998a; Lagabrielle and Cormier, 1999; White et al., 2000]. On the basis of detailed submersible and high-resolution side scan data for the East Pacific Rise (EPR) in the 9°–10°N region, Fornari et al. [1998a] suggest that the axial summit trough (AST) may in some locations be created entirely by collapse processes and more appropriately be characterized as an axial summit collapse trough (ASCT). From our own research, and catalogs of collapse in other references, collapse features are a dominant morphology at the seafloor in zones of active volcanism, and therefore important to our understanding of submarine eruption dynamics and construction of the upper oceanic crust.

Details are in the caption following the image
Photographs showing examples of common collapse types observed at the 9°37′N small overlapping spreading center site. (a) 6 m × 4 m. Plan view of small collapse confined within individual lobate lobe. (b) Side view of small collapse through crust measuring ∼2 cm thick. (c) 6 m × 4 m. Plan view of multiple skylight collapse. Depth of collapses ∼1 m. (d) 2 m × 2 m. Side view of skylight penetrating large lobate lobe. (e) 6 m × 4 m. Plan view of multilayered roof collapse. Collapse ∼2 m in depth. (f) 4 m × 2 m. Side view of lava pond collapse through roof crust, uppermost roof crust >10 cm thick. (g) 6 m × 4 m. Plan view of multistory collapse. Collapse surfaces are separated by ∼0.5 m. Total collapse depth ∼3 m. (h) 6 m × 4 m. Plan view of lava pond collapse showing two remnant lava pillars capped by lobate crusts. Depth of collapse ∼3 m.
Details are in the caption following the image
Photographs showing physical features on the interior of lava roof crusts that may be evidence for the interaction of lava with a vapor phase. (a) Interior of a lobate roof crust (sample 2357-1). Sample is ∼2 cm thick with a microvesicular surface texture and has cuspate and drip structures up to 2 cm in length. (b) 8 cm thick collapse roof (2497-6) showing drip structures on the interior surface up to 3 cm in length. Note the massive interior, cooling fractures, and vesicular zone below the drips. (c) 10 cm × 7 cm. Fragment of top of a lava pillar (sample 2354-V) with >3 cm cuspate structures that appear to be the walls of lava bubbles. The interior surfaces have a dull residue indicative of interaction with a high temperature vapor [Perfit et al., 2003]. (d) 20 cm × 15 cm. Jumbled sheet flow (sample 2759-7) containing a 3 cm diameter elongate vesicle of unknown orientation, and a water-filled bubble.

[3] Many recent studies of MOR crests have described an association between collapse features and sites of hydrothermal venting [e.g., Haymon et al., 1993; Fornari et al., 1998a; Von Damm, 2000]. In some locations, heated vent waters issue directly from collapse pits or lava pillars, while in other locations diffusely venting fluids exit piles of collapse talus [Haymon et al., 1993]. Collapse pits occur within narrow axial troughs and in the roofs of lava tubes, therefore we suggest that pits are windows into a subsurface network of conduits that once transported lava along and across the ridge crest, and may subsequently have served as conduits for hydrothermal fluids. Researchers conducting rapid-response surveys to locations of historical eruptions along the MOR have reported seeing bacterial mats spewed out of collapse pits by the force of hydrothermal venting [Haymon et al., 1993, Embley et al., 1995] suggesting that collapse pits may provide critical habitats for chemosynthetic species and aid in distributing them between sites of active venting. Several authors have suggested that collapse features such as water-filled pits and talus lenses contribute to zones of porosity identified within the upper 100–300 m of oceanic crust at intermediate to fast-spreading MOR [e.g., Vera et al., 1990; Wilkins et al., 1991; Cochran et al., 1999; Johnson et al., 2000]. As pointed out by Johnson et al. [2000], the presence of substantial void space within the upper oceanic crust may have major implications for the layer as an aquifer and as a reservoir of hydrothermal fluids in both axial and off-axis zones.

[4] In this study we develop the first detailed physical model of the formation of the whole range of collapse features observed on the EPR, including pits, pillars, roofs, talus, and drips. Many other authors have presented qualitative models for the formation of collapse features at the large end of the collapse spectrum [e.g., Ballard et al., 1979; Francheteau et al., 1979; Gregg and Chadwick, 1996; Chadwick et al., 1999]. Some authors have numerically addressed the related issues of submarine lava flow inflation [e.g., Gregg and Chadwick, 1996; Chadwick et al., 1999; Gregg et al., 2000; Gregg and Fink, 2000; Fox et al., 2001; Chadwick, 2003], formation of pillars [Gregg and Chadwick, 1996; Gregg et al., 2000], formation of lineated sheet flows in collapse pits [Chadwick et al., 1999], and formation of deep-sea limu o pele (bubble wall fragments) [Clague et al., 2000; Maicher and White, 2001; Clague and Davis, 2002]. However, no one model conceptually incorporates all time-sequenced elements of the growth and foundering of a lava roof, and the resultant pillars, drips, and talus that provide insight into the specific eruption that created them. Using elements of the models described above, analog examples of Hawaiian collapse, laboratory and theoretical modeling and observations of lava flow behavior in the deep ocean, and the results of an Alvin and towed camera investigation of the neovolcanic zone of the EPR at 9°37′N in 2000, we describe in detail each stage in the genesis of collapse features. We have quantified our conceptual model by estimating the cooling behavior of lava, the rise rate of gaseous bubbles, and the density contrasts between lava and gaseous seawater. Central to our hypotheses about collapse formation is the presence of vaporized seawater that becomes trapped between the crust of a lava roof and the molten lava interior of a flow [see also Perfit et al., 2003]. We demonstrate that it would not be possible to form the characteristic suite of observed collapsed features without this vapor phase. The specific purpose of our paper is to lay down the first conceptual framework for the origin of collapse features and their implications for a whole range of submarine lava flow processes.

2. Background

[5] Submarine volcanic collapse has been documented on the slow-spreading Mid-Atlantic Ridge (MAR) [Scheirer et al., 2000], the intermediate-spreading Juan de Fuca Ridge (JdFR) and Galapagos Rift [e.g., Ballard et al., 1979; Embley and Chadwick, 1994], the fast-spreading EPR [Haymon et al., 1991; Fornari et al., 1998a], the superfast-spreading southern EPR (SEPR) [Sinton et al., 2000; White et al., 2000], on seamounts [Scheirer et al., 2000; Fornari et al., 1988], and in shallow water (<30 m) at the location where historic subaerial Hawaiian lava crossed the shoreline [e.g., Moore et al., 1973; Tribble, 1991]. On-land analogs to submarine collapse features have been observed in Hawaii [Greeley, 1987; Hon et al., 1994; Walker, 1991; Byrnes and Crown, 2001], although different mechanisms for collapse formation exist there due to the vast differences in temperature and pressure between subaerial Hawaiian lava flows and the deep ocean. Most collapse features are described as occurring in young lava flows, and most often in sheets and lobates (sheets and lobates as defined by Gregg and Fink [2000, and references therein]). A subset of collapse types known as “pillow implosions” has been observed forming off the coast of Hawaii [Moore, 1975; Tribble, 1991], but we consider these to be genetically unrelated to collapse in lobate and sheet flows, and they will not be addressed further in this study.

[6] The first model to specifically address the formation of collapse pits, roofs, pillars, and pillar selvages (vertical stacks of broken roof remnants that line lava pillars) was developed by Ballard et al. [1979] based on their observations of lava lakes (one of the largest types of collapse features) at the Galapagos Rift. They proposed that topographically-induced inflation of ponded sheet lavas trapped subsurface water, necessitating the growth of large spiracles or water escape pipes (lava pillars), which could in some places support lava roofs when the molten lava at the interior of the inflated flow drained away. They hypothesized that lava selvages resulted from intermittent drainage, or alternatively multiple lava flow occupations of the same lava pond location. Fornari et al. [1998a] extended Ballard et al.'s [1979] model to address the dual volcanic/tectonic nature of seafloor spreading at the crest of the fast-spreading EPR. They presented a 4-stage model for the evolution of the neovolcanic zone based on evidence for repeated inflation and collapse within the AST.

[7] With the specific goal of studying collapse features, in 2000 we carried out a survey over the crest of the EPR at 9°37′N where the full spreading rate is 110 mm/yr [Carbotte and Macdonald, 1992] (Figure 3). This location has recently been the subject of a detailed geochemical and geomorphological study by Smith et al. [2001] and has been characterized in the larger context of the 9°–10°N EPR by other authors [e.g., Haymon et al., 1991; Macdonald et al., 1992; Von Damm, 2000; Dunn and Toomey, 2001]. Of relevance to the issue of collapse is the presence of a small overlapping spreading center (SOSC) [Smith et al., 2001] offsetting two limbs of the AST by ∼450 m in a right-lateral sense at ∼2500 m depth (Figure 4). The SOSC separates inflated ridge crest morphologies to the north and subdued ridge crest morphologies to the south; Smith et al. [2001] hypothesize that the two limbs of the SOSC represent dying and propagating phases of ridge volcanism. On the basis of evidence of black, glassy lavas, abundant collapse, and subcritical vapor phase chemistries for a hydrothermal vent in the area [Von Damm, 2001], Smith et al. [2001] conclude that the 9°37′N SOSC was the southernmost point on the EPR crest to be affected by the 1991 eruption discovered in the 9°50′N area [Haymon et al., 1993].

Details are in the caption following the image
Map showing location of the study area relative to major plate boundaries and continental landmasses. The Clipperton and Siqueiros Transforms, which offset the East Pacific Rise, are indicated.
Details are in the caption following the image
Left half of figure shows DSL-120A side scan data collected during the AHA-Nemo2 cruise. Side scan swaths measure ∼1 km in width. The SOSC is centered at 9°37′N. Note the large collapsed channel trending east off axis from the SOSC. Right half of figure shows morphologic data from Alvin and the WHOI-TCS collected during the ALGRAV2000 cruise. N-S data are WHOI-TCS digital still images, E-W data are Alvin video. Lobate morphologies are red, sheets are green, pillows are blue, and the AST is shown in black and was digitized from Argo-II 1989 visual and acoustic data [Haymon et al., 1991]. Sheet flows are concentrated within the AST, and around the base of the eastern (propagating) limb of the SOSC. Note the prominent pillow ridges to the east of the AST.

3. Data Collection and Analyses

[8] On the basis of descriptions of collapse occurrence relative to the AST and ridge flanks [Fornari et al., 1998a; Gregg et al., 2000], we designed a nested survey that densely covered the two overlapping ASTs at 9°37′N, and ∼300 m to each side of each trough extending ∼6 km north and south of the SOSC. We collected four types of complimentary data: DSL-120A side scan sonar imagery with a resolution of 2 m, digital photographs collected from a towed sled, Alvin video footage, and collapse pit roof samples (Figures 2, 4, and 5). By collecting data sets with different resolutions, we delimited the size and shape of collapse features, and at the same time provided a broader geologic context of seafloor textures to highlight patterns of collapse zonation. On the basis of our knowledge of the differing chemistries, eruption histories, and tectonic styles of the two overlapping ASTs at 9°37′N, our survey allowed us to test hypotheses about the association of collapse with young versus old volcanic structures and distance from the locus of volcanism at the ridge axis within the AST.

Details are in the caption following the image
Left panel shows DSL-120 side scan map of the 9°37′N SOSC site overlain with data showing the locations of collapse features <2 m in diameter in blue. Location of the Argo-II 1989 digitized AST is in white. Right panel shows the same site with locations of collapse features >2 m in diameter in red, AST in white. Schematic in right corner shows that all collapse abundances peak in the same location relative to the AST, but collapse features <2 m in diameter are relatively evenly distributed throughout the ridge crest area, while collapse features >2 m in diameter cluster near the AST.

3.1. Data Collection and Navigation

[9] The Woods Hole Oceanographic Institution's (WHOI) towed camera sled (TCS), equipped with battery-powered strobes and a black and white digital still video camera, collected photographic images of the ridge crest at 15 second intervals along north-south survey tracks (Figure 4). The sled was manually flown at the end of a steel cable behind the ship at 5–7 m above the seafloor as determined by a 12 kHz pinger trace. The depth of collapse pits was estimated based on Alvin imagery of lava pillar heights, and a 100 kHz altimeter when traversing over the pits in Alvin. Survey speed for TCS surveys was ∼0.25 kt. The sled navigated using a long-baseline (LBL) transponder network that located the images to within 4–5 m. Eight TCS tows were conducted, one in each AST, and the others at ∼25 m spacing (Figures 4 and 5). Alvin video footage was collected during six dives that crossed the SOSC and ASTs from east to west (Figures 1, 4, and 5). Representative samples of lava were also recovered during these traverses [Smith et al., 2001]. Alvin traversed the seafloor at ∼0.5 kt and an altitude of 5–8 m and was navigated using the same LBL transponder network as the TCS. Alvin LBL navigation was merged with data from an RDI bottom-lock Doppler that resulted in positional accuracy of the submersible of ∼1–2 m. Alvin video footage was sub-sampled at 15 second intervals and analyzed identically to the TCS photographs based on a scheme developed by Fox et al. [1988]. The combination of ∼30,000 TCS photographs and Alvin video frames imaged ∼720,000 m2 of seafloor. DSL-120A side scan sonar swaths 1 km wide were collected in 2 north-south trending lines over the ridge axis, and 1 east-west line over the SOSC during the AHA-Nemo2 cruise of 2000 (Figure 4). It is difficult to evaluate the accuracy of estimates of collapse pit area based on side scan imagery, because none of our photographic images cover the entirety of an area that appears to be collapsed in the side scan imagery. We therefore chose to use the DSL-120A data primarily as a geologic context for the more detailed analyses possible with the photographic and hand specimen data.

3.2. Collapse Styles and Distributions

[10] Collapse pits in the photographs cover ∼93,600 m2, 13% of the total surveyed area, and are associated with lobate lavas in 98% of the images and sheet flows in the remaining ∼2% (Figures 4 and 5). Lava flow morphologies were classified identically to those in Kurras et al., [2000, Figure 4]. Nearly all the collapsed sheet flows photographed occurred within the ASTs or between the two limbs of the SOSC. No collapsed pillow flows were documented at any location along the ridge crest. The distribution of lobate, sheet, and pillow flows in our study area is as follows: lobate flows 70%, sheet flows 25%, pillows flows 3%, and transitional flow types 2%. These percentages are similar to results from other near-bottom surveys of fast-spreading MORs [Kurras et al., 2000; White et al., 2000]. We documented a broad size and volumetric range of collapse pit types and associated features, ranging from collapse pits a few centimeters deep and <1 m in diameter up to ∼25 m deep and hundreds of meters in length or width.

[11] We used the statistical F and Student's t tests [Swan and Sandilands, 1995] to evaluate the size distribution of collapse pits relative to distance from the two overlapping AST limbs. We find that at the 99% confidence level, the distribution of collapse pits larger and smaller than 2 m in diameter is not the same (Figure 5). Collapse pits <2 m in diameter are relatively evenly distributed out to ∼300 m from the ASTs, with only a slight peak in frequency near the ridge axes. The distribution of collapse pits >2 m in diameter peaks dramatically within <100 m from each axial trough. There is no statistically significant difference in the distribution of collapse pits north and south of the SOSC or on the Pacific versus the Cocos plate.

[12] The largest collapse pits we observed have been described in previous studies of collapse pits, lava pillars, and lineated sheet flows on the MAR, JdFR, EPR, and SEPR, and have been called “lava pond collapses,” a terminology we will adopt here (Figures 1e, 1f, 1g, and 1h) [e.g., Embley and Chadwick, 1994; Gregg and Chadwick, 1996; Fornari et al., 1998a; White et al., 2000; Fox et al., 2001]. Lava pond collapse pits range in size from 10 m to >100 m in diameter, and are often aligned parallel to the trend of the AST. Many occur within the AST itself in which case they often are as deep as the nearby AST walls, up to a maximum of ∼25 m deep. Large collapse pits are often ringed with stubby remnants of roof crust ranging in thickness from ∼2–10 cm. Lava pillars are ubiquitous features throughout the largest collapse pits, and often are topped by ragged roof crusts. In some locations lava pillars are aligned with the collapse pit margins, while in others they are more randomly distributed at the centers of the pits. Roof talus is often piled at the base of the collapse walls near the margins of lava pond pits. An interesting feature of lava pond collapse pits is their multistoried character in some locations. In TCS photographs, multistory collapse pits show 2–4 generations of lava roofs at different heights, all of which are collapsed (Figure 1).

[13] Collapse pits of progressively smaller sizes are distributed within the AST and on the ridge flanks. A second readily distinguishable morphologic style of pits are ∼2–10 m in diameter, and the shape of these pits differs from the lava pond collapse pits in that they are often sub-circular instead of ragged and broken (Figures 1c and 1d). These collapse pits are similar to descriptions of subaerial “skylights” [Greeley, 1987]; Haymon et al. [1993] use the skylight terminology to describe small collapse pits over tubes at 9°51′N. Skylight collapse pits at 9°37′N appear approximately equidimensional in depth and breadth, and are rarely associated with lava pillars. Roof crust remnants that have not collapsed range from 2–10 cm thick. Skylight pits sometimes contain piles of roof talus, though the base of the pits is often lined with sheet lavas and no talus is visible.

[14] The smallest collapse pits observed in the survey are <2 m in diameter and usually occur within a single lobate lobe either within the AST or on the ridge flanks (Figures 1a and 1b). These small collapse pits have been called lobate “blisters” by Wilkins et al. [1991] and Fornari et al. [1998a], and we adopt this terminology for our study. Blister pits differ from lava pond and skylight pits in that individual blister pits do not appear to be connected to each other in the subsurface. From Alvin video, it is often possible to glimpse roofed-over tube-like cavities connecting lava pond and skylight pits in the subsurface, but blister pits appear to be attached to the substrate where the junction of the collapse roof and the underlying lava is visible. Attached roof crust remnants range in thickness from 1–<5 cm thick. Blister pits are often remarkably circular, and it is common to see the roof talus from blister pits arrayed with jigsaw-fit accuracy at the base of the blister collapse pits.

3.3. Collapse Roof Samples

[15] Samples of collapse pit roofs for each style of collapse pit described above were collected within lobate and sheet flows using Alvin (Figure 2). The hand specimens we collected consist of lava pit roof crusts that are ∼2–10 cm thick. A number of physical characteristics were noted on these samples either singly or in combination: (1) glassy, gently convex exterior surfaces with smooth to breadcrust textures, (2) generally concave interior surfaces with a dull to waxy sheen, (3) interior surfaces with a concentration of microvesicles, and more rarely, crystallites of silicate minerals, (4) glassy rinds on the interior surfaces less than a few millimeters thick that contrast with the thick vitreous rinds on the exterior of the roof crust, (5) drip-like protrusions up to ∼4 cm in length that either thin downward from the underside of the crust interior, or bulge to form a bulbous node at the tip of the drip, and (6) flange-like to cuspate septa ∼1–5 mm thick, roughly polygonal in plan view, that in some cases are interpreted to be preserved walls of bubbles on the crust interior.

4. Discussion

[16] We explain the suite of volcanic collapse features described above and the localities in which they form via a comprehensive multistage physical model. Although lava pond, skylight, and blister pits have a range of characteristic morphologies and associated collapse features, we think they represent easily identifiable examples of a spectrum of collapse styles formed by similar processes. Incorporated in our model is information gained from previous qualitative explanations for collapse formation, observations of analogous collapse features in Hawaii, and laboratory and theoretical modeling of the behavior of lava erupted in the deep ocean. Our model initiates at the onset of an eruption.

4.1. Stage 1: Eruption of Lava and Initial Cooling Behavior

[17] To generate ponded lavas thick enough to create lava pillars as tall as ∼20 m high [Gregg et al., 2000], it is likely that lava is erupted into a topographically confined area relatively rapidly [Gregg and Chadwick, 1996]. In the context of the AST, barriers to drainaway of lava might be the walls of the AST, linear alignments of coalesced pillars (ramparts) [Embley and Chadwick, 1994], previous locations of drained out lava ponds, or flow levees. On the basis of in situ measurements [Fox et al., 2001] and reconstructions [Gregg and Chadwick, 1996; Fornari et al., 1998a; Gregg et al., 2000] of eruptive events, many authors infer that the initial stage of lava emplacement and inflation may be extremely short. In one location on Axial Seamount inflation has been documented to occur over periods as short as ∼72 min, with bursts of rapid inflation (32 cm/min) lasting just minutes [Fox et al., 2001]. Volumetric effusion rates inferred during the inflationary stage at the JdFR Rumbleometer site are ∼2.7 m3/s [Fox et al., 2001]. Unfortunately, it is not possible to infer eruption rates of the lava as it spreads out in a thin layer prior to ponding and inflation, but the rates recorded by the Rumbleometer [Fox et al., 2001] give an approximate lower bound for preinflation effusion rates in an intermediate-rate seafloor spreading environment.

[18] As soon as lava is erupted onto the seafloor, it begins to interact in a complex way with the substrate and the surrounding mass of cold seawater. The behavior of the cooling mass of lava can be modeled based on temperature contrasts [Griffiths and Fink, 1992] and has been observed and described in detail in a shallow-water setting off the coast of Hawaii [Moore et al., 1973; Moore, 1975; Tribble, 1991]. Convective cooling is inferred to be the primary mechanism by which lava cools in contact with seawater [Griffiths and Fink, 1992; Gregg and Fink, 1995; Gregg and Fink, 2000]. Thermal data for subaerial Hawaiian flows [Hon et al., 1994] show that the crust below the water/lava interface is likely to be composed of three distinct layers: (1) A brittle outer layer that has low tensile strength relative to the deforming forces of the lava below, as evidenced by common fracturing. (2) A middle layer of solidified crust with temperatures between ∼800–1070°C that behaves viscoelastically and gives strength to the flow. (3) A highly viscous inner layer of lava that acts as a boundary with the molten core of the flow. The brittle layer and viscoelastic layer thicken non-linearly with continued cooling of the lava, contributing to the tensile strength of the crust [Hon et al., 1994].

4.2. Stage 2: Entrapment of Seawater Within the Lava Flow Generates a Vapor Phase Fluid

[19] The lava drips and septa on the undersides of roof samples (Figure 2), as well as instances of lava stalactites, [Gregg et al., 2000] lead us and Perfit et al. [2003] to infer the presence of a vapor phase beneath the lava crust during collapse formation. The drips measure up to 4 cm long and are extremely delicate with pointed or bulbous terminations (Figure 2) as if produced by low-viscosity lava. In a few locations, drops or coiled trails from these drips have been seen on the underlying lava surfaces. The drips are typically glassy and contain tiny crystallites suggesting that this part of the lava was not instantly quenched by cold seawater. Laboratory experiments have shown that ∼1150°C lava in contact with seawater heated to ∼7°C will cool to glass in <1 s [Griffiths and Fink, 1992; Gregg and Fornari, 1998]. The highest recorded temperatures of seawater from hydrothermal vents are ∼450°C [Gregg et al., 2000; Von Damm, 2000], a full 320°C below the glass transition temperature for basaltic lava at 730°C [e.g., Grosfils et al., 2000]. At temperatures less than ∼1070°C, basaltic lava is a solid [Hon et al., 1994] and could not deform plastically to form the drips we observe in hand sample. It is therefore clear that not only do the drips observed on roof crusts have to form within a low viscosity medium, but they are also unlikely to have formed in contact with seawater, even hydrothermal vent fluids. We hypothesize that a vapor phase fluid heated to magmatic temperatures is the medium within which lava drips form. Recent studies of the microscopic crystals and encrustations on the underside of the roof crusts confirm that they were formed at magmatic temperatures and suggest deposition from a vapor phase [Perfit et al., 2003].

[20] Other authors have invoked a vapor phase at depths of ∼2500 m to explain drips, gas cavities, and limu o pele [Fornari et al., 1998a; Perfit and Chadwick, 1998; Clague et al., 2000; Gregg et al., 2000; Maicher and White, 2001; Clague and Davis, 2002], however there has been no consensus as to whether the vapor phase results from the exsolution of magmatic volatiles or entrapment and boiling of seawater. Clague et al. [2000] and Clague and Davis [2002] present a model for the formation of limu o pele at depths as great as 4200 m on the Northern Hawaiian Arch via exsolution of CO2 from alkalic lavas and argue that it is unlikely that steam can be produced in the deep ocean due to high pressures. They suggest that if significant quantities of CO2 exsolved from MORB magma at depth, CO2 might remain trapped in a frothy layer at the top of the axial magma chamber (AMC). If such a frothy magma is extruded from the eruptive fissure located at the ridge axis, it might create cm-sized CO2 vapor cavities in the lava pond that forms in the AST. However, voided vapor pockets (collapse features) of comparable size are commonly seen in lavas within and well outside of the AST. The vapor pockets we observe in off-axis lava could not have come from excess magmatic CO2 gas in the first stages of an eruption, since that gas would all have been dispersed within the AST. The low vesicle concentrations in recovered EPR samples (generally less than 2 volume %) and lack of evidence for any frothy lavas or explosive volcanism along this segment of the EPR do not support the hypothesis that accumulated CO2 is a significant factor in the formation of the vapor pockets and collapse features we describe here. In addition, the low CO2 concentrations (∼200–300 ppm) measured in our EPR MORB by Le Roux et al. [2002] would only provide enough exsolved gas from a 2 m thick flow to produce a 1 cm-thick vapor cavity if all of the CO2 coalesced at the top of the flow.

[21] Experimentally derived phase diagrams for NaCl-water mixtures demonstrate that water vapor can form in seawater heated to temperatures above ∼400°C at 25 MPa pressure [Sourirajan and Kennedy, 1962; Berndt and Seyfriend, 1997]. Since seawater is the most voluminous potential source for vapor phase fluids in lavas that form collapse features, we hypothesize that it is the primary gas involved in the formation of vapor pockets and collapse features. To raise seawater temperatures to magmatic levels, seawater must be forced into contact with lava (Figure 6a). Maicher and White [2001] describe three scenarios in which water could become trapped in contact with extruding lava: (1) Projection of lava off an overhanging ledge or steep slope resulting in incorporation of water within the lava before the lava touches down. (2) Lava flows over water-filled gaps in talus. (3) Lava overrides or burrows into sediment, or heated wet sediment rises into lava. If a portion of seawater-saturated substrate is overrun by lava, and the weakest confining point of the trapped substrate water is the overriding lava, then where the buoyancy force of the heated water exceeds the drag force exerted by melt viscosity, the lava will be penetrated by a rising bubble of water vapor [Maicher and White, 2001]. Calculations of the heat flux from a 1200°C lava flow into 0°C seawater show that in 30s, 1.9 × 104 kJ/m2 of heat will pass through the base of the flow, and 1.9 × 105 kJ/m2 in 50 min. Because the base of the lava flow in contact with the surrounding seawater and substrate rocks rapidly cools to half the difference in their temperatures [Jaeger, 1968], the water touching the flow will not be heated above ∼500°C. At ∼480°C and 25 MPa pressure seawater transforms to a 2-phase liquid/vapor, has a bulk density of only 160 kg/m3 compared with a bulk density of basalt of ∼2670 kg/m3, and is thus highly buoyant [Berndt and Seyfried, 1997]. The rising bubble of water vapor can penetrate a viscoelastic crust or a partially solidified crust that has cracks in it [Maicher and White, 2001]. In some places these heated water vapor bubbles could penetrate the lower flow boundary and become the foci for the growth of lava pillars [Ballard et al., 1979; Francheteau et al., 1979; Gregg et al., 2000] if there is a sufficient flux of heated seawater through the lava flow (Figure 6a).

Details are in the caption following the image
Figure 6a shows an arbitrary cross section of a lava flow as it is extruded over a water-filled substrate and traps seawater. Time t0 shows the water-saturated substrate prior to the onset of the eruption. Time t1 shows the seafloor as hot lava flows over it, trapping water as it falls off a cliff (panel 1a), overrides a pond of water-saturated sediments (panel 1b), or flows over a fissure or other gap in the substrate (panel 1c). Water under the flow is instantly vaporized to steam at magmatic temperatures with a ∼20 times increase in volume. Vapor ascends upward through the flow as long as the buoyant force of the confined steam exceeds the viscous drag force of the flow. Zones of enhanced substrate water flux, for example over preexisting hydrothermal vents or fissures, may become the loci of lava pillars. See text for a more detailed discussion of the processes shown in this figure.

[22] We think that it would be difficult for seawater to infiltrate a flow via the flow's upper surface. Observations by Haymon et al. [1993] and Embley et al. [1995] of diffuse venting of hydrothermal fluids through pristine smooth lobate crusts on the EPR and JdFR lead us to infer that microscale cracks may form in some lava crusts as thermal shrinkage of the basalt occurs in contact with seawater [Fornari et al., 1998b]. However, Hon et al.'s [1994] thermal data for Hawaiian lava flows show that it is unlikely that most surface-initiated cracks could penetrate deep enough into the flow for seawater to come in contact with and be trapped and heated by the molten interior of the flow. Hon et al.'s [1994] brittle outermost layer 1 may fracture as the flow below it deforms, but the viscoelastic layer 2 and the highly viscous inner layer 3 will not fracture brittley, and thus will shield the molten core of the flow from contact with seawater. Seawater in contact with the upper surfaces of a flow is confined only by the overburden pressure of seawater, which is not strong enough to overcome the viscous drag force and tensile strength of Hon et al.'s [1994] layers 2 and 3 in order to reach the flow interior. Even if seawater infiltrates the cracks in Hon et al.'s [1994] brittle layer 1 and comes in contact with layer 2, seawater heated to vapor by layer 2 would immediately begin to ascend buoyantly and escape back up through the same cracks which by which it initially came in contact with layer 2. Seawater heated to vapor by the ∼800–1070°C layer 2 would not be hot enough to keep MORB lavas molten and thus could not form a gas pocket hot enough for the formation of lava drips.

[23] The common presence of pipe vesicles and spiracles where lava crosses wet ground and in ancient submarine lava flows, such as those in Troodos [Schmincke and Bednarz, 1990], is clear evidence that water does in fact enter lava flows from beneath the flow. Pipe vesicles are elongate and rise vertically ∼10 cm above the base of a flow, usually ending in a microvesicular patch or vesicle cylinder [Waters, 1960]. Each pipe is mm-cm in diameter and the pipes are often bent at the top into the direction of the flow of the lava. The origin of pipe vesicles is widely considered to be the result of volatiles rising buoyantly through liquid lava after a flow overruns moist ground [Tomkeieff, 1983], or in this case, a seawater-saturated substrate. Pipe vesicles can only form where steam rises through stiff, near solid lava, since if the lava is a low-viscosity liquid the vesicles will be distorted by the flow or the passage of the bubble will not be recorded. The height that pipe vesicles reach above the base of the flow may simply be the thickness of the relatively viscous, partially cooled lava at the base of the flow. Once the steam reaches the hot, low-viscosity lava in the middle of the flow (∼10 cm from the flow margin), the steam continues to rise and heat, but its path is no longer recorded in the lava until it reaches the top of the flow.

4.3. Stage 3: Ascent of Vapor and Creation of Lava Drips and Septa

[24] Once seawater has been trapped within an inflating and moving lava flow, it will rise buoyantly through the flow due to density contrasts between the water vapor and the basalt (Figures 6a and 6b. The following history for the evolution of seawater as it is heated to magmatic temperatures is based on Berndt and Seyfried [1997, Figure 3] which is a phase diagram for NaCl-water at 30 MPa, constructed on the basis of a range of experimental results. (1) As the temperature rises from zero to 400°C, seawater remains a single phase liquid, and its density drops from 1000 kg/m3 to 700 kg/m3. (2) At 400°C, a vapor phase starts to form within the liquid. (3) By 480°C the 2-phase liquid-vapor is composed of 7% brine with a salinity of 55% NaCl, and 93% vapor, composed of almost pure water. The bulk density is now only 160 kg/m3. (4) At 480°C, the liquid-vapor mixture is transformed to one of vapor and solid halite. (5) At 500°C, the bulk density of the vapor-solid is 115 kg/m3. (6) The mixture of vapor-halite persists to 680°C, when it is transformed back to liquid-vapor. (7) Increased heating allows more NaCl to be dissolved in the vapor, until at ∼900°C seawater becomes a single-phase vapor. (8) At 1000°C, the bulk density of the vapor is 52 kg/m3. Thus during heating of seawater to magmatic temperatures, the vapor phase generated experiences approximately a 20 times increase in volume. The bulk density of basalt is ∼2670 kg/m3, so it is clear that heated seawater will rise buoyantly through the molten core of a basalt flow due to density contrasts at all stages in its evolution from liquid seawater to seawater vapor at magmatic temperatures.

Details are in the caption following the image
See text for a more detailed discussion of the processes shown in this figure. At time t2, vaporized substrate seawater at magmatic temperatures has entered the lava flow interior and begins to ascend buoyantly through the flow as the flow lengthens and inflates. The upper and lower margins of the flow begin to cool inward. At time t3, the upper and lower margins of the lava flow continue to thicken and the flow continues to inflate. Vapor bubbles begin to coalesce in thin gaseous layers at magmatic temperatures at the base of the upper crust of the lava flow, inhibiting further thickening of the crust. Low viscosity lava begins to form drips, septa, and stalactites in the vapor cavity. At time t4, the part of the upper flow crust that is still in contact with molten lava continues to thicken, as does the lower margin of the flow. The lava flow has inflated to its full height. Vapor bubbles have nearly all coalesced into large vapor cavities below the upper solidified crust of the flow. See text for a more detailed discussion of the processes shown in this figure.

[25] Both sheet and lobate flows may inflate after emplacement (Figure 6b) as long as the continued injection of lava into the flows' interiors does not overcome the yield strength of the flow crust [Gregg and Chadwick, 1996; Chadwick et al., 1999; Gregg et al., 2000]. Vapor bubble ascent through a flow probably initiates immediately after the flow overruns its substrate (Figure 6a) prior to inflation, since the base of a flow will continue to cool and thicken during inflation and eventually prevent injection of water vapor through the basal crust (Figure 6b). Vapor bubbles have an initial lava-derived heat of ∼500°C, but as they rise through the cooling outer margin of the flow (occasionally leaving pipe vesicles in their wake), they will eventually be heated to magmatic temperatures by the molten interior of the flow. Using Stoke's Law calculations, the rise rate of bubbles of varying sizes through the flow can be determined. For a density contrast between the basalt and seawater of 2500 kg/m3 and a lava viscosity of 100 Pa s, a bubble with a radius of 5 mm will rise through a 2 m thick flow in 24 min, and a bubble with a radius of 2 cm will rise through the same flow in 1.5 min. The Rumbleometer flow at the Axial Seamount site is recorded to have inflated 3.5 m in 72 min [Chadwick, 2003]. Water trapped by the Rumbleometer flow may have ascended to the surface of the flow at approximately the same rate as the flow itself inflated. In instances where flow inflation is less rapid, water vapor probably overtakes the inflation rate of the flow and accumulates below the upper surface of the flow anywhere from a few minutes to an hour after entrainment, depending on the thickness of the flow (Figure 6b). In the opposite scenario, Moore et al. [1971] described an alignment of decimeter-scale gas cavities ∼2 m below the top of a pillow flow unit exposed on Mt. Etna. In this instance, the flow must have cooled quickly enough that the viscous drag of the melt overcame the buoyancy of the water vapor bubbles, freezing them mid-flow.

[26] After ascending through the flow, water vapor bubbles become trapped at the upper flow crust where they coalesce into an expanding gas-filled cavity (Figure 6b) unless they can escape through cracks in the surface crust or during a breakout of lava [Maicher and White, 2001]. Heat flux calculations derived from Fowler [1990] indicate that a 2 m thick flow heating 20 kg/m2 (∼2 cm) of liquid water below the flow from 0°C–500°C–1000°C as it is entrained and rises could yield a 4 cm thickness of water vapor after 30 s, 12 cm after 5 min, and 40 cm after 50 min. As soon as the vapor bubbles coalesce below the upper crust of the flow and separate the molten interior from the flow crust, they inhibit the continued growth of the roof crust (Figure 6b). Using theoretical constraints on basalt crystallization on the seafloor developed by Griffiths and Fink, [1992], we calculate that 1 cm of crust will form in 30 s, 3 cm in 5 min, and 10 cm in 50 min. Thus during bubble ascent in a flow inflating to 2 m thick, between 1–10 cm of roof crust is likely to develop. However, after the bubbles have ascended all the way to the surface of the flow, the roof crust will maintain a constant thickness over the vapor cavities throughout the continuing evolution of the flow. The results of the roof crust thickness calculations based on theoretical constraints match the roof thicknesses that we observe in hand specimens (Figure 2).

[27] We infer that lava drips, stalactites, and bubble wall septa form as soon as small vapor bubbles begin to coalesce below the upper flow crust and form larger cavities (Figure 6b). Prior to coalescing, bubbles will accumulate below the upper flow crust. In some instances, fragile bubble skins are preserved in hand-sample specimens as nearly complete bubbles or as fragments of bubble walls known as septa (Figure 2). As a vertical bubble wall ruptures, the lower portion of the dangling bubble wall end will subside gravitationally into the molten flow below it and be reincorporated, leaving no trace. However, the portion of the bubble wall attached to the upper part of the flow is not likely to rebound upward against the pull of gravity and be reincorporated into the molten skin of lava below the cooled crust, but will instead hang pendulously downward into the expanding gas cavity. Narrow fingers of downward extending lava may serve as preferential loci of drip activity for molten lava coating the interior of the upper flow crust (Figure 2). Because MORB is undersaturated with respect to water [Dixon and Stolper, 1995; LeRoux et al., 2002], water vapor may dissolve into the magma from the vapor bubble. If this occurs, locally the melting point of the basalt would be lowered, enhancing its ability to remain molten and form easily dripped material. At the polygonal juncture of several large vapor bubbles, a bubble wall may stretch vertically as the gas cavity thickens. The stretching bubble wall will thin at its center and may eventually rupture to form a pointy lava stalactite. These are often observed in hand sample.

[28] There are several points of the vapor ascent scenarios we have outlined above that are supported by observations of collapse pit morphology and vesicle segregation zones in subaerial settings, although the primary source of the gas is undoubtedly different.

[29] 1. Many small collapse pits appear to be localized at the apex of a lobate lobe. Although this part of the lobe is the area that is geometrically least supported by bounding walls, it is also the likely location of gas bubble pockets prior to lava drainaway, as the buoyant vapor will rise to the highest point of its enclosure. A similar phenomenon has been observed by Hon et al. [1994] in collapsed areas of Hawaiian lava flows. They observe that if no avenue of escape is available through the upper lava flow crust, gas accumulates just beneath the upper crust of the flow and inhibits further crystallization along that interface. Therefore not only are the centers of lobate lobes and large sheet flows physically unsupported by bounding walls or lava pillars, but they are also likely to have thinner crust than adjacent areas that were in constant contact with the molten core of the flow. This thin crust is likely to be more susceptible to collapse than areas that did not experience inhibited crystallization.

[30] 2. Similar scenarios to the one we describe have been invoked to explain gas cavities and vesicle accumulation zones in subaerial Hawaiian lavas, though the gases need not be water vapor in subaerial environments [Walker, 1991; Hon et al., 1994]. Shelly pahoehoe flows are notorious for zones of vesicle accumulation near their surface crusts that make them susceptible to collapse when walked on. In addition, pahoehoe lobes have often been found to contain gas cavities beneath their upper crusts, without the frothy vesicle accumulation zones typical of vent-proximal shelly pahoehoe (Sinton and Dye, personal communication, 2002). The collapse pit shapes in lobate pahoehoe are morphologically identical to those found in lobate flows at 2500–3000 m water depth on the EPR, implying a similar genetic origin.

[31] 3. Walker [1992] describes vesicle segregation layers in subaerially exposed pillows found at 2500–3000 m depth. He notes that in some locations, the vesicle accumulation zones are sufficient to make the outer pillow skins fall off. He also describes a prevalence of pipe vesicles in some pillow units.

4.4. Stage 4: Pressure-Induced Syneruptive Collapse of the Flow Carapace and Drainaway of Lavas

[32] For the delicate drip features described in the previous section to be preserved after the eruption that formed them, they must cool from magmatic temperatures without being reincorporated into the lava, broken, or overprinted. While the lava flow is molten and the surface crust of the flow is resting on a molten lava core, the pressure inside the lava flow will be the same as that in the ocean outside the lava flow, 25–30 MPa for the EPR. The lava flow will inflate or contract to equalize pressures since it is molten and can still flow. This implies that the pressure inside a gas cavity within the flow will also be the same as in the surrounding ocean. As the flow starts to cool from its margins inward, the volume of a given gas cavity becomes fixed as the flow solidifies. As the vapor phase cools through the glass transition temperature at 730°C [Grosfils et al., 2000], drips, stalactites, and septa solidify in situ. The temperature of the vapor within the gas cavity cools synchronously with the lava surrounding it, while the volume of the gas cavity stays constant, so the pressure inside the vapor pocket must begin to drop. If the gas cavity remains rigid, the pressure inside the gas cavity at 500°C is 16 MPa, causing a pressure difference across the outer crust of the flow and the interior of the gas cavity of 9–14 MPa. At 200°C the pressure inside the gas cavity is only 1.5 MPa, so the pressure differential can increase to 28.5 MPa. Depending on the geometry of the gas cavity and the thickness of the solidified crust above it, the pressure differentials caused by cooling of the vapor phase may be enough to cause inward collapse of the surface crust of the flow, flooding the vapor cavity with cold seawater (Figure 6c). Under these circumstances the delicate drip features we observe in hand sample would be preserved along the margins of the uncollapsed portion of the roof or serendipitously in the roof talus.

Details are in the caption following the image
See text for a more detailed discussion of the processes shown in this figure. At time t5a, the cooling lava flow induces a pressure differential across the top of the vapor cavity that may cause the uppermost roof crust to fracture, flooding the vapor cavity interiors. Alternatively, as seen at time t5b, the molten interior of the lava flow may subside in areas where the upper roof crust is not adequately supported by bounding walls or lava pillars, causing the roof to collapse due to gravitational effects. Roof material may be reincorporated into the flow. At time t6, hours to years after the onset of the initial eruption, larger expanses of roof crust may collapse due to local seismicity, the destabilizing motion of lava from subsequent eruptions, or hydrovolcanic explosions. Roof talus accumulates at the base of the collapse pits. See text for a more detailed discussion of the processes shown in this figure.

[33] During vapor ascent and coalescence within the molten flow, a ponded lava flow may also experience inflation as new lava is continually injected into its core (Figure 6b). Inflation causes the surface crust of the lava to rise up as it floats on the vertically thickening molten interior. At some point during the eruptive sequence, it is likely that drainaway of the flow core will take place. Drainaway may initiate due to a variety of factors, including reduced rates of lava supply to the flow that result in drainback of lava into the primary eruptive fissure, or pressure release due to a breakout along the flow margin that drains lava away from the lava pond. We use the term “drainaway” to describe both mechanisms. Drainaway of lava removes the buoyant force that supported the upper lava crust. In localities like the base of the AST where extensive lava ponds and lakes are common, or in large lava tubes on the flanks of the ridge crest, it is unlikely that wide expanses of inflated lava carapace will resist gravitational forces if not supported from below by pillars, bounding walls or ramparts. As the level of molten lava within an inflated lava pond or tube begins to descend, the roof floating on the surface of the lava will subside with it. At the margins of the flow where the roof is attached to bounding walls or lava pillars, the roof may have sufficient tensile strength and elasticity to bend downward slightly with the descending lava without immediately breaking. It is likely that vapor accumulation pockets will localize in these temporarily uplifted regions of roof crust at the margins of the subsiding flow [Chadwick, 2003], a scenario consistent with observations of drips preserved on the undersides of lava pillar selvages [Gregg et al., 2000]. As the flow level continues to drop, the roof will eventually break along the flow margins (Figure 6c), flooding the vapor cavities with seawater that preserves the drips, and creating characteristic lava pillar selvages [Chadwick, 2003]. The foundered roof may be reincorporated into the molten flow interior or rest on the flow top until it is deposited on the floor of the lava pond (Figure 6c). The process described above may be repeated many times during the descent of a lava flow during lava drainaway. The total time for inflation and deflation of a ponded lava probably varies considerably, though Fox et al. [2001] recently recorded a total time as low as ∼2.5 h.

4.5. Stage 5: Posteruptive Collapse and Reoccupation

[34] The roof collapse described above occurs as a direct result of the decrease in pressure within a gas cavity during cooling and/or due to the drainaway of the buoyant molten lava flow that supports the lava crusts from inside the flow. However, in the posteruptive period ranging from hours to years, continued collapse of standing lava roofs may initiate gravitationally or by ground motion related to seismicity at the ridge crest, as well as the destabilizing force imposed by subsequent lava flows or hypothesized hydromagmatic explosions (Figure 6c) [Haymon et al., 1993]. Subsequent eruptions along the ridge crest are likely to reoccupy cavities in collapse pits and tubes. Multiple eruptive cycles are probably responsible for the multistory collapse observed within the AST and just outside the axial trough [Kurras et al., 2000]. It is likely that repeated flooding of lava ponds by separate eruptions can produce several layers of collapse roofs at different heights along the AST walls. The AST will experience the eruption of lava and subsequent flow collapse far more frequently than the flank regions. This results in a much more frequent cycle of collapse pit generation at the base of the AST, while small distal collapse pits on the ridge flanks will tend to get infilled with lava or sediments over time [Fornari et al., 1998a].

5. Conclusions

[35] We present a multistage physical model that explains the formation of lava pond, skylight, and blister collapse pits and associated collapse features at a fast-spreading MOR. Our model is based on laboratory and theoretical models for the behavior of MORB lavas erupted on the seafloor and detailed observations of collapse features at 9°37′N on the EPR and attempts to explain each phase in the growth and foundering of lava crusts. On the basis of observations of delicate drip structures on the undersides of collapse roof talus that could not form in contact with cold seawater (Figure 2), we infer that most deep-water inflated lavas contain gas pockets between the molten core of the flow and the upper crust of the flow. The key points of our model are summarized below:

[36] The first three stages of our model describe the incorporation of seawater vapor into a lava flow erupted on the seafloor. Stage 1: Lava erupted at the base of the AST is distributed according to a balance between effusion rate at the vent source and cooling rate of the lava in contact with seawater. Lobate lava flows are most likely to generate collapse pits due to their low effective viscosities and high temperatures, as well as their susceptibility to ponding and subsequent inflation of the flow carapace. Stage 2: At the basal contact of the flow with the seawater-filled porous substrate, seawater trapped by the advancing lava flow flashes to vapor with a 20 times increase in volume (Figure 6a). Vapor is entrained directly in the flow or penetrates the basal flow contact due to the buoyancy force of heated water vapor confined by the molten lava. Where the buoyancy pressure of a trapped vapor pocket overcomes the viscous drag force of the lava flow, vapor will begin to ascend through the flow. Stage 3: Bubbles of water vapor ascend through the flow during flow inflation and concentrate at the base of the upper flow crust, coalescing to form elongate gas cavities that inhibit further crustal growth (Figure 6b). As bubble walls rupture, lava drips from the upper crust into the hot gas cavities to form individual drips, septa, and stalactites.

[37] The last two stages of our model specifically address collapse formation. Stage 4: As a lava flow cools and solidifies, gas cavities trapped at the base of the upper flow crust maintain a constant volume with decreasing pressure. Eventually, the pressure gradient between the upper flow crust and the gas cavity may be great enough to cause collapse of the crust, depending on the geometry of the gas cavity and the thickness of the roof crust (Figure 6c). During episodes of lava drainaway, where the roof is attached to bounding walls, ramparts, or lava pillars, the roof bends viscoelastically as the supporting hydrostatic force of the lava flow interior is removed. Lava pillar junctions thus temporarily become high points in the lava carapace, where vapor pockets can concentrate [Chadwick, 2003]. As the lava continues to subside, the roof eventually fails, flooding the gas cavity with seawater and allowing lava drips to be preserved (Figure 6c). Where lava roofs are supported by bounding walls, ramparts, or lava pillars, they may remain standing though their interiors are flooded with seawater. Stage 5: In the posteruptive period ranging from hours to years, the lava roofs that remained standing after the eruption may collapse due to seismicity, lava movement, or hydrovolcanic explosions (Figure 6c). Collapse pits may frequently be reoccupied during subsequent eruptive episodes, resulting in complex inverted stratigraphy. Repeated eruption and collapse events within the AST result in a predominance of large and extensive collapse features within the AST and fewer, smaller collapse features on the ridge flanks.

[38] Our evidence for the dynamic interactions between lava and seawater leads us to infer that many aspects of the submarine eruptive process may need to be re-evaluated in detail. For example, it is conceivable that seawater trapped beneath a thin layer of fresh lava might flash to steam explosively, causing hydrovolcanic explosions even at depths of 2500–3000 m along the ridge crest. The addition of large numbers of vapor pockets to a flowing lava mass may cause changes to the rheology and morphologic behavior of the flow that have previously gone unexamined. At the 9°37′N EPR site, collapse features are a pervasive component of the seafloor architecture, and their clear association with vapor pockets heated by the extruding flow suggest that vaporized seawater is a critical element in many seafloor eruptions.

Acknowledgments

[39] We thank the crews of the R/V Atlantis, DSV Alvin, R/V Melville, and the WHOI Deep Submergence group for their assistance in collecting these data. We are grateful to Gregory Kurras and Del Bohnenstiehl for shipboard data collection and fruitful scientific discussions, to Dan Scheirer and Paul Johnson for help with sonar data acquisition and processing. We would like to thank Bill Chadwick, Tracy Gregg, Dave Clague, Jackie Dixon, Jim Natland, and Bill White for thoughtful reviews of early versions of this manuscript. We also thank Matt Smith for his help with this project. This research was supported by grants from the National Science Foundation, Grants: OCE-9986874 (M. Edwards), OCE-0138088, OCE-9402360, OCE-9403773 (M. Perfit), OCE-9100503 (M. Perfit and D. Fornari), OCE-9408904, OCE-9912072 (D. Fornari). J. Cann was supported in part by internal funds from the Woods Hole Oceanographic Institution. This is SOEST contribution 6203, and HIGP contribution 1294.