The copyright line for this article was changed on 6 May 2015 after original online publication.
 Volcanic eruptions at mid-ocean ridges (MORs) control the permeability, internal structure, and architecture of oceanic crust, thus establishing the foundation for the evolution of the ocean basins. To better understand the emplacement of submarine lava flows at MORs, we have integrated submersible-based geologic mapping with remote sensing techniques to characterize the lava flow morphology within previously mapped lava flow fields produced during single eruptive episodes at the Galápagos Spreading Center (GSC). Detailed attributes describing the surface geometry and texture of the lava flows have been extracted from high-resolution sonar data and combined with georeferenced visual observations from submersible dives and camera tows; based on signatures contained in these data, a fuzzy logic-based classification algorithm categorized lava flow morphology as pillows, lobates, or sheets. The resulting digital thematic maps offer an unprecedented view of GSC lava morphology, collectively covering 77 km2 of ridge axis terrain at a resolution of 2 m × 2 m. Error assessments with independent visual reference data indicate approximately 90% agreement, comparable to subaerial classification studies. The digital lava morphology maps enable quantitative, spatially comprehensive measurements of the abundance and distribution of lava morphologies over large areas of seafloor and within individual eruptive units. A comparison of lava flow fields mapped at lower- and higher-magma-supply settings (95° and 92°W, respectively) indicates that effusion rates increase along with magma supply and independent of spreading rate at the GSC, although a complete range of eruptive behavior exists at each setting.
 Volcanic eruptions on mid-ocean ridges (MORs) are the fundamental building blocks of the upper oceanic crust, which covers ∼70% of Earth's surface. Although we have not been able to directly observe these eruptions due to the relative inaccessibility of the deep seafloor, studying the characteristics of the resulting lava flows provides valuable clues about the behavior of MOR eruptions: the volume and composition of lava flows enables estimates of the size, temperature, and geochemical heterogeneity within subsurface magma reservoirs, and the surface morphology of lava flows contains important information about effusion rates and emplacement processes. Unfortunately, establishing the linkage between MOR eruptions and volcanic morphology has been limited by the paucity of MOR eruptions that have been identified and mapped in detail [e.g., Chadwick et al., 1991; Embley et al., 1995; Chadwick et al., 1995; Embley et al., 1999; Sinton et al., 2002; Soule et al., 2007; Colman et al., 2012]. This lack of observations is a critical gap in our understanding of the crustal accretion processes at MORs.
 Submarine lava flows are typically categorized as having pillow, lobate, or sheet morphology [Perfit and Chadwick, 1998]. These end-members actually represent a continuum of morphologies with gradational, but significant, distinctions [Gregg and Smith, 2003]. Pillows are bulbous or spherical in form and generally have characteristic striations parallel to the direction of extrusion, indicating that the flow rate was low enough to allow grooves to freeze during emplacement, and a “bread crust” surface texture that indicates no integration with nearby pillows [Moore, 1975]. Pillow lavas tend to pile atop each other in a mass of interconnected branches to form steep-sided mounds or ridges; as pillow edifices grow upward, the newest pillows are erupted from the top of the edifice and flow downward, with large elongate pillows often serving as distributary tubes [Chadwick and Embley, 1994]. Lobate lava flows have been conceptualized as a transitional morphology between pillows and sheets. Lobate lavas can be broad and flat or small and bulbous (similar to subaerial pahoehoe lava) with a gentle, hummocky relief [Fox et al., 1988]. Lobates are extruded in a similar fashion to pillows but faster, prohibiting any striations on their surface. Lobate flows often host collapse features caused by post eruption draining of lava back into the eruptive vent or out through conduits once supply from the vent diminishes [Perfit and Chadwick, 1998]. Sheet flows are shallow, low-relief lava surfaces that may exhibit a variety of surface microtextures (e.g., lineated, ropy, whorly, folded, jumbled, hackly), depending on the underlying seafloor roughness or boundaries to the flow [Fox et al., 1988; Griffiths and Fink, 1992; Gregg and Fink, 1995]. Sheet flows are usually thin (<10 cm) and consist mostly of glass, but can also be several meters thick [Perfit and Chadwick, 1998]. Sheet flows within lava channels are generally thought to form at the highest effusion rates and flow velocities [Pinkerton and Sparks, 1978; Soule et al., 2005].
 Within the MOR neovolcanic zone, a corridor along the spreading axis where most eruptive activity is concentrated, the uppermost layer of oceanic crust is constructed as a series of overlapping lava flows, each with variable morphological properties that provide clues to the mechanisms and timescales of lava emplacement. Previous studies of MOR lava morphology [e.g., Ballard et al., 1979; Crane and Ballard1981; Fox et al., 1988; Kurras et al., 2000; Engels et al., 2003; Cann and Smith, 2005; Soule et al., 2005; Fundis et al., 2010; Yeo and Searle, 2013] have used photographic data from submersibles and deep-towed cameras along with multibeam bathymetry and/or side-scan backscatter imagery to examine lava flow morphologies. However, these phototransects cover <1% of the area in the neovolcanic zone—insufficient coverage for a representative view of lava morphology distribution. Sophisticated techniques are required to use sonar data in lieu of direct observations to classify lava flow morphology on the surrounding unexplored seafloor.
 Here, we describe the abundance and distribution of lava morphologies within two study areas at the western Galápagos Spreading Center (GSC), where the magma supply rate varies along axis due to the contribution of excess magma from the nearby Galápagos hot spot [Sinton et al., 2003]. This gradient in magma supply is manifested in the volcanic morphology of the GSC. White et al., 2008]. The separation of magma supply from spreading rate-induced stresses creates an ideal setting in which to examine the controls on emplacement of MOR lava flows. We have mapped the lava flow morphology using a machine-learning classification algorithm presented in McClinton et al. , where high-resolution, sonar-derived bathymetric maps, acoustic imagery, and near-bottom visual observations exist for a set of mapped GSC lava flow fields [Colman et al., 2012]. The resulting digital lava morphology maps offer an unprecedented view of the spatial distribution of lava morphology over large, contiguous areas of ridge axis terrain. We examine the abundance and distribution of lava morphologies between and within eruptive units as a means of ascertaining and comparing relative effusion rates and modes of emplacement. Finally, we compare these data to the catalog of identified and mapped MOR eruptions and discuss their implications.
2. Geologic Investigations of the GSC
 The intermediate-spreading GSC separates the Cocos and Nazca Plates in the eastern equatorial Pacific Ocean (Figure 1). Spreading rates vary from 48 mm yr−1 at 100°W to 61 mm yr−1 at 85°W [Argus et al., 2011]. Spreading rates are ∼53mm yr−1 at the Galápagos Transform, a 90 km long offset transform fault located near 91°W that separates the GSC into western and eastern sections [Mittelstaedt et al., 2012]. The southern tip of the Galápagos Transform is situated approximately 200 km north of the main Galápagos archipelago, the western extent of which is presumed to overlie the center of the Galápagos hot spot [Hooft et al., 2003; Villagomez et al., 2007]. The Galápagos hot spot has profoundly influenced the history and morphology of the GSC, which has apparently maintained its proximity to the hot spot through a series of southward ridge jumps [Hey et al., 1977; Wilson and Hey, 1995; Mittelstaedt et al., 2012].
 Previous work [e.g., Canales et al., 2002; Detrick et al., 2002; Sinton et al., 2003; Cushman et al., 2004; Blacic, 2004; White et al., 2008] documented systematic variations in the characteristics of the western GSC (97°–91°W) with proximity to the hot spot (Figure 2). The ridge transitions from a rift valley to an axial high as the hot spot is approached from the west, roughly mirroring changes in morphology along the eastern GSC [Sinton et al., 2003; Christie et al., 2005]. Crustal structure has been determined from multichannel seismic data [Detrick et al., 2002]. The data show that crustal thickness increases from west to east, reaching nearly 8 km near 91°30′W, with the thickest crust coincident with the shallowest ridge depths [Canales et al., 2002]. Axial magma chambers (AMCs) were seismically imaged 2–3 km below the seafloor between 92°42′-91°W; the AMCs appear as deeper, discontinuous features from 92°42–94°18′W before disappearing altogether farther west [Blacic, 2004]. The disappearance of the seismically imaged AMCs coincides with a significant increase in the number of volcanic cones and seamounts′, in addition to other variations in volcanic morphology [White et al., 2008].
 In March–April 2010, we conducted a geologic mapping and sampling campaign, the Galápagos Ridge Undersea Volcanic Eruptions Expedition (GRUVEE) in two detailed study areas (DSAs): a hot spot-distal, low magma supply area at 95°W (DSA1) and a relatively high magma supply area at 92°W near peak hot spot influence (DSA2) (Figure 3). The DSAs were centered on the ridge axis approximately 330 km apart. Over this distance, the average magma supply to the crust (calculated as spreading rate × crustal thickness after Sinton et al. ) increases by roughly one third, from 0.3 × 106 m3/yr/km in DSA1 to 0.4 × 106 m3/yr/km in DSA2, while spreading rates remain nearly constant at 55 and 57 mm yr−1, respectively [Colman et al. 2012]. Using the Sentry autonomous underwater vehicle (AUV), the Alvin submersible, the TowCam photo sled, and with acoustic backscatter imagery from the DSL-120A side-scan sonar system obtained in 2006 [White et al., 2008], we mapped and sampled GSC lava flow fields and examined fine-scale hydrothermal and tectonic features to investigate the effects of the differential magma supply on the size and character of GSC volcanic eruptions.
3.1. GRUVEE Near-Bottom Data Acquisition
 The AUV Sentry was equipped with a Reson 7125 400 kHz multibeam sonar system. Sentry mapped approximately 31 km2 in DSA1 and 47 km2 in DSA2 in 17 separate deployments. Operating along preprogrammed navigation tracks at an average survey altitude of 65 m, we found the optimal sonar swath to be 250–300 m wide. The AUV multibeam data were processed using the MB-System open source software package [Caress and Chayes, 1996]. Bathymetry was gridded using the footprint-weighted-mean algorithm within the mbgrid module, in which depth values for each grid cell are calculated as weighted averages of the area of each beam falling within the grid cell. The algorithm generates an initial low-resolution grid that provides seafloor slope information, uses this grid to enable the correct representation of beam footprints on steep slopes, and then finally produces the full-resolution grid. The density of bathymetric soundings supported gridding at 1 m × 1 m; the vertical precision of the bathymetry is 15–20 cm after applying tidal corrections. Minimum curvature spline interpolation was applied to fill gaps less than three grid cells wide nearest of two defined grid cells. The AUV Sentry bathymetry is the highest-resolution bathymetric data collected along the GSC to date (available through the Marine Geoscience Data System at www.marine-geo.org). Fine-scale volcanic features (e.g., hydrothermal chimneys, inflation and collapse features, lava channels) and morphological transitions observed during Ultra Short Baseline- or Long Baseline-navigated Alvin and TowCam dives can clearly be seen in the final bathymetric maps. Acoustic backscatter imagery (2 m pixel resolution) from the DSL-120A side-scan sonar system was also available for DSA2. Backscatter swaths were 1000–1250 m wide and centered on the ridge axis. Processing details are outlined in White et al. .
 High-resolution sonar data from Sentry and the DSL-120A (where available) provided targets for further investigation with Alvin and TowCam. Thirteen Alvin dives were conducted in each DSA, averaging 3.5 km length per dive. Approximately, 160 h of digital video and >90,000 digital photos were recorded using Alvin and 300 rock samples were collected for petrologic analyses. The deep-towed camera sled TowCam [Fornari, 2003] was deployed 17 times, 11 in DSA1 and six in DSA2, averaging 2.5 km length per dive. The camera system was towed at an altitude of 5–7 m above the seafloor at a speed of ∼0.5 kts, taking down-looking 3.5 megapixel digital photographs of regions 3–5 m across at 10 s intervals. These images were transmitted to the ship by fiber-optic cable, enabling real-time interpretation and tagging of imagery; approximately 25,000 digital photos were collected. TowCam also retrieved a total of 47 glass samples via wax cores.
3.2. Submersible-Based Geologic Mapping
 High-resolution bathymetric and extensive observational data, combined with results from petrologic analyses, enabled Colman et al.  to distinguish 18 separate lava flow fields within the GRUVEE study areas (Figure 4). Colman et al.  designate flow fields as units of genetically related basalts, as suggested by petrologic similarities and the spatial correlation of inferred eruptive vent(s), following the criteria established by Chadwick and Embley  and Sinton et al. . Some of these flow fields were likely emplaced as multiple outpourings of lava [e.g., Wadge, 1978; Self et al., 1998], but each flow field is interpreted to be the aggregate of lavas from a single eruptive episode.
 The initial flow field boundaries were established by compiling Alvin and TowCam observations of lava flow morphology, flow direction, and sediment thickness. Sedimentation rates are relatively high along the GSC (1 mm/20–30 years; [Mitchell, 1998]), making sediment cover a useful indicator of relative age relationships between volcanic features. Later, these boundaries were refined using near-bottom photographs and videos, high-resolution Sentry bathymetry and DSL-120A backscatter imagery, sample descriptions, and volcanic glass compositions. Flow field length and areal extent were determined from the flow field boundaries.
3.3. Computer-Aided Lava Morphology Mapping
 The meter-scale Sentry and DSL-120A sonar data contain information about the surface geometry, acoustics, and texture of the seafloor and collectively cover ∼77 km2 in the GRUVEE study areas. In comparison, Alvin and TowCam covered <1% of this area by strategic transects of the ridge axis; this is an effective strategy for mapping geological contacts, but does not provide an adequate view of lava flow morphology distribution. This obvious gap in the scale of seafloor observations can be overcome using advanced remote sensing techniques, which provide innovative tools for characterizing lava flow surfaces. Similar to using airborne light detection and ranging or laser altimetry data for mapping subaerial lava flows [e.g., Mazzarini et al., 2005, 2007; Favalli et al., 2010a, 2010b; Deardorff and Cashman, 2012], detailed attributes that describe the geometry and texture of submarine lava flow surfaces can be extracted from high-resolution sonar data. However, limited observational data points require additional, unconventional techniques to interpolate between observation points and extrapolate them over the surrounding unexplored seafloor; classification algorithms have previously been used to bridge the gap between sparse observations at MORs [e.g., Stewart et al., 1994; Hurst and Karson, 2004; Meyer and White, 2007].
McClinton et al.  incorporated GRUVEE observational and sonar data with a classification algorithm to map lava flow morphologies along the GSC ridge axis in DSA2. The classifier, known as the adaptive neuro-fuzzy inference system (ANFIS) [Jang, 1993], incorporated geometric properties extracted from Sentry's Reson 7125 multibeam bathymetry and acoustic and texture attributes extracted from DSL-120A side-scan backscatter through point-relational, second-order statistical measures derived from gray-level cooccurrence matrices (Figure 5). These sonar signatures were extracted where georeferenced Alvin and TowCam photographs clearly indicated a single morphology type (pillow, lobate, or sheet lava and fault/fissure scarp). The sonar signatures were then compiled and subjected to pattern analysis by the ANFIS classifier, which constructed a database of rules that described membership in each morphological category. Other locations that match those identified by the classifier as indicative of a particular lava morphology type were subsequently mapped by the ANFIS.
 We also used this methodology to produce a lava morphology map for DSA1, where DSL-120A backscatter imagery was not available, by using the sonar amplitude from the Sentry Reson 7125 sonar as the seafloor texture data. The raw beam amplitude was filtered through two iterations of a 3 m × 3 m Gaussian mean low-pass filter to reduce the inherent “speckle,” because the Gaussian mean preserves the frequency content better than a simple boxcar mean filter [Sauter and Parson, 1994; Caress and Chayes, 1996]. The backscatter swaths were then mosaicked together with a weighted mean algorithm that prioritized data points by the apparent grazing angle of the sonar; the nadir, near-nadir, and far-field of the swath received the lowest priority, whereas data between grazing angles of 15°–45° received the highest priority. These angles were chosen to mimic the incident angles from side-scan sonars, thus producing a synthetic backscatter signal. Like the bathymetric data, the Sentry backscatter swaths were gridded at 1 m × 1 m with a two-dimensional thin plate spline algorithm applied to interpolate gaps within three grid cells of two defined grid cells. The complete ANFIS lava morphology classification methodology was then conducted in DSA1, substituting Sentry multibeam amplitude data for DSL-120A side-scan backscatter imagery.
3.4. Evaluating the Accuracy of Digital Lava Morphology Maps
 An accuracy assessment is a critical part of any digital classification to evaluate the reproducibility of the result and its agreement with the ground reference data from direct observations. Previous interpretations of MOR lava morphology have been based on manual photointerpretation; although generally accepted as correct, and commonly used as a regionally representative sample, this method typically lacks verification. Automated classifiers are more appropriate for regional-scale or complete flow-field-scale studies, not only because they provide more spatially complete sampling, but also because they include a quantifiable uncertainty.
 To quantitatively assess the accuracy of the digital lava morphology maps, we compared the points of ANFIS-classified lava flow morphology to independent sets of ground reference imagery that were not used during each ANFIS training phase. The most common way to represent the raw results of an accuracy assessment is in the form of an error matrix, a square matrix that compares the colocated pixels assigned to each morphological class by the classifier to the actual morphology, defined by some type of ground reference data. The diagonal elements of the error matrix represent the pixels that have been correctly classified; we used these values to calculate class-specific and overall accuracy [e.g., Congalton, 1991]. Misclassification between categories was explicitly calculated from the matrix and is presented as omission error (errors of exclusion) and commission error (errors of inclusion) [e.g., Story and Congalton, 1986].
 In addition to these descriptive statistics, kappa analysis [Cohen, 1960], a standard analytical statistical technique, was also performed on each error matrix. The result of kappa analysis is a coefficient that describes the agreement between the digital classification and the actual terrain; a value of 1.0 indicates perfect agreement beyond chance, whereas 0.0 indicates no agreement. The kappa coefficient is a better representation of the general quality of a digital classification because it is calculated using the entire error matrix and removes the effects caused by differences in sample size [Rosenfield and Fitzpartrick-Lins, 1986]. In addition, kappa analysis provides an important metric for evaluating the relative accuracy between the products of digital classifications [e.g., Foody, 2004].
 Error matrices and calculated descriptive and analytical statistics are presented with each lava morphology map. Accuracy is well over 80% for most categories. This level of accuracy is comparable to recent digital classifications of subaerial terrains that employed multispectral or hyperspectral imagery and/or high-resolution topographic data [e.g., Lu et al., 2004; Buddenbaum et al., 2005; Stow et al., 2007; Bork and Su, 2007; Qiu, 2008]. Although synthetic, multibeam-derived backscatter was substituted for side-scan backscatter to produce a lava morphology map for DSA1, calculated kappa coefficients are very similar for both GSC lava morphology maps (0.79 for DSA1, 0.84 for DSA2). The close similarity in kappa values provides no reason to interpret the digital lava morphology maps differently, which enables the comparisons of lava morphologies between DSAs that form the foundation of this paper.
4.1. DSA1 (95°05′-94°45′W): Lava Flow Fields at Lower-Magma Supply
 ANFIS classification of Sentry multibeam bathymetry and backscatter produced a map of fine-scale lava morphology for a total of 28.4 km2 of axial terrain in DSA1. The lava morphology map completely or partially covers six of the eight eruptive units defined in this study area by Colman et al.  (Table 1); the Tortuga, Pinguino, and Buho seamounts and portions of the Frijoles and Del Norte flow fields were not covered by Sentry sonar surveys and therefore could not be included in the ANFIS lava morphology classification. Overall, DSA1 is dominated by pillow lavas, which cover 76.8% (21.9 km2) of the terrain. Due to the limited extent of sheet lavas in DSA1, we chose to combine the lobate and sheet lava classes, which together covered 16.2% (4.6 km2) of the study area. The remaining 6.9% (1.9 km2) was classified as tectonized terrain. An accuracy assessment (Table 2) indicated an overall agreement of 87.3%; combining the lobate and sheet lava classes improved accuracy from 79.1%. An overall kappa value of 0.79 was calculated from the DSA1 error matrix; this value is classified as “substantial” and “excellent” agreement by Landis and Koch  and Fleiss et al. , respectively.
Table 1. Names, Estimated Sizes, and Relative Abundance of Lava Morphologies for Lava Flow Fields in DSA1a
Mapped Length (km)
Mapped Area (km2)
Length and areal extent are from Colman et al. .
Table 2. Error Matrix and Calculated Accuracy Measures for DSA1 Lava Morphology Map
Omission Error (%)
Commission Error (%)
Class Accuracy (%)
Overall accuracy: 87.26%.
Overall kappa coefficient: 0.7887.
 The Buho flow field was mapped in the northern half of the axial graben in westernmost DSA1 through multiple Sentry and Alvin deployments (Figure 6). Buho is the largest mapped unit in DSA1 and consists of a large volcanic cone (2.6 km diameter; ∼250 m height) flanked by mound fields (individual mounds are 100–200 m diameter, <60 m height) to the east and west. The entire flow field stretches ∼7 km at a trend slightly oblique to the GSC axis. Lava flow morphology within the Buho field consists mostly of pillows (>86%) with a minor amount of lobates. Dome-shaped mounds are exclusively pillows, whereas lava morphology at flat-topped mounds transitions from flanks of large or elongate pillows to low-relief lobates at the summits. Many mounds have coalesced into linear and arcuate ridges, trending subparallel or oblique to the spreading axis. The outermost mounds appear more tectonized than the inner mounds. The large flat-topped volcanic cone within Buho is not covered by the lava morphology map, but Alvin observations indicate that its flanks are constructed of elongate pillows, lobates, and talus, with no obvious flow contacts observed. The summit of the cone has a 2–5 m-scale hummocky topography of pillows and lobates with two large collapse pits (∼400 m diameter; 70–80 m depth); the walls of the pits expose broken pillows and lobates. Multiple fissures oriented 270–310° were observed at the summit and match the trends of features observed in the mound fields.
 The eastern edge of Buho overprints the graben-centered Pulgar flow field (Figure 7), which is composed of another cluster of mounds. The Pulgar mounds (300–400 m diameter; 60–80 m height) are broader than those within Buho and are predominantly constructed of pillows (∼84%). Lobate flows were observed at several summits and are also mapped by the ANFIS classifier. Small smooth-sided pillows, possibly representing late-stage eruptive activity, were observed in several areas between lobes. Greater sediment thickness was observed throughout Pulgar, and the percentage of tectonized terrain is higher in Pulgar (8.5%) than Buho (5.8%), suggesting Pulgar is older.
 Just to the east of Pulgar is the Dragón flow field, an irregular, arcuate cluster of overlapping mounds that hugs the southern graben wall. Dragón has roughly the same areal extent as the Pulgar and the Buho mound fields (excluding the central Buho cone), and is also dominated by pillow lavas (88.5%). The Dragón mounds are the steepest in DSA1; flank slopes average 36° and consist of highly elongate pillows that transition to lobates at the flatter summits of several mounds. Dragón has been heavily tectonized (9% of flow unit area). Faults and fissures trending roughly 290° cut some mounds; fault scarps expose truncated pillow lavas. Moderate to heavy sediment cover and extensive faulting suggest Dragón may be the oldest mapped flow field in DSA1.
 In the central portion of DSA1, most of the graben floor is covered by the low-relief Del Norte flow field (Figure 8). Del Norte consists of a large amount of lobate and sheet lavas (45.5%) with abundant inflationary features and collapses. The flow field forms a platform that fills the width of the graben, extending west to pond against Dragón, where highstand lava crusts are visible ∼10 m above the current elevation of the platform; and extending to the east, where it is locally overlain by the younger Frijoles complex. Several channels and tubes distributed lavas up to 1.6 km south from source vents located at the northern edge of the axial graben. Lava rises/pressure plateaus throughout the flow field hint at lava ponds beneath, perhaps indicating these areas were previously bathymetric lows in which Del Norte lava accumulated [e.g., Appelgate and Embley, 1992; Walker, 1991] Collapsed areas (3–4 m depth) within the flow indicate flow thickness and reveal a stratigraphic sequence of basal sheet lavas overlain by lobates (and pillows in some areas); some collapses contain lava pillars and bathtub rings. This progression suggests the Del Norte eruption was began as a fast and voluminous extrusion of sheet flows, which transitioned into lobate flows that inflated and then collapsed. As effusion rates waned and the eruption localized to point-source vents, pillow mounds formed locally around the vents. A southern inner graben fault appears to postdate the emplacement of the flow field, but at the northern inner graben fault, Del Norte lavas change morphology and drape the scarp in some places, suggesting at least a portion of the fault scarp predates the emplacement of the flow field.
 The Del Norte flow field is partially overprinted by Frijoles, likely the youngest mapped flow field in DSA1. The Frijoles form a discontinuous, 10.6 km long chain of elongate mounds (85.1% pillow lavas). Individual Frijoles mounds have a characteristic “bean” shape, are broad (500–600 m diameter) and relatively low relief (20–30 m height), with steep flanks and flat summits where lobate flows and hydrothermal staining are prevalent. By area, the Frijoles mounds are the largest individual mounds mapped in DSA1. Sentry bathymetry reveal that some Frijoles mounds are superposed by small pillow mounds atop a previously flat summit of lobate flows, suggesting waning late-stage effusion rates or second-stage eruptive activity. Relatively light sediment cover was observed throughout Frijoles, including direct observations of sediment contrasts between Frijoles and Del Norte lavas; the unit also has the lowest amount of tectonized terrain (1.95%) among all of the mapped DSA1 units, and Frijoles mounds overprint the main inner graben fault that dissects other units. Sentry bathymetry and the ANFIS lava morphology map cover less than 40% of the total inferred extent of the Frijoles unit, but Alvin observations and ship-collected bathymetry do not suggest any major deviations from these morphological characteristics.
 The Altares flow field is the easternmost mapped eruptive unit in DSA1 (Figure 9). Altares consists of a cluster of hummocky mounds (250–600 m diameter), forming an elongate, axis-subparallel ridge that narrows from west to east. The mounds are predominantly pillow lavas (91.7%) with lobates found near the summits of several larger mounds. Minimal ornamentation was observed on Altares lavas. At the center of the flow field, an extremely steep, 100 m tall constructional spire of elongate pillow lavas sits atop the largest mound. Moderate sediment cover and the abundance of tectonized terrain (5.49%) within Altares suggest the unit is similar in age to the Buho field.
4.2. DSA2 (92°01′-91°50′W): Lava Flow Fields at Higher-Magma Supply
 ANFIS classification of Sentry bathymetry and DSL-120A side-scan backscatter imagery produced a 26 km2 map of lava flow morphologies [McClinton et al., 2012]. The DSA2 lava morphology map is contiguous along over 20 km of GSC axial terrain and covers the majority of the 11 eruptive units mapped in the study area by Colman et al.  (Table 3). Pillow lavas are the most abundant lava morphology in DSA2 at 47.1% (12.33 km2) of the terrain; the remainder of DSA2 terrain consists of 31.6% (8.3 km2) lobate lavas, 11.7% (3.1 km2) sheet lavas, and 9.6% (2.5 km2) tectonized terrain. An independent check on the accuracy of the DSA2 classifier indicated an overall agreement of 88.4% (Table 4). A kappa coefficient of agreement value of 0.84 signifies “near perfect” and “excellent” agreement according to Landis and Koch  and Fleiss et al. , respectively. The axial portions of several flow fields (Iguana, Cocodrilo, Cobija) have been mostly overprinted by more recent lava, complicating the analysis of morphologic characteristics for these lava flow fields.
Table 3. Names, Estimated Sizes, and Relative Abundance of Lava Morphologies for Lava Flow Fields in DSA2a
Mapped Length (km)
Mapped Area (km2)
Length and areal extent are from Colman et al. .
Lobo del Mar
Table 4. Error Matrix and Calculated Accuracy Measures for DSA2 Lava Morphology Map
Omission Error (%)
Commission Error (%)
Class Accuracy (%)
Overall accuracy: 89.72%.
Overall kappa coefficient: 0.8418.
 Lagarto, the westernmost unit mapped in DSA2, is a composite flow field of primarily lobate (53.3%) and pillow (32%) lavas (Figure 10). Multiple flow units and vent locations are inferred based on observations of lava morphologies and flow direction indicators. A zone of 2–3 m deep collapses near the northern graben wall is floored by sheet lavas and rubble; the general southward flow direction of these lavas implies a source somewhere to the north, perhaps outside of the current graben. A graben-centered eruptive fissure produced low-relief lavas that flowed southward; a row of pressure ridges with flow-transverse medial cracks indicates inflation due to the southern graben wall impeding the advance of the flows [Appelgate and Embley, 1992]. These pressure ridges serve as excellent indicators of initial flow direction and morphology (and by extension, flow rate) [Cas and Wright, 1988]. The uplifted and rotated flanks of the pressure ridges are jumbled sheet flows that are partially overprinted by lobates at their bases. These structures suggest that this portion of the Lagarto flow field was emplaced as flat-lying sheet flows that were embayed by the southern graben fault scarp, inflated to some degree, and then overlapped by lobate lavas as effusion rates waned. Several additional vents outside and to the south of the current graben produced a series of low-relief flows that display contrasting flow directions and are difficult to differentiate. Colman et al.  interpreted these separate flow units to be co-eruptive, based on similar petrology and sediment cover.
 Just to the east of Lagarto, the Lobo del Mar flow field extends nearly 1.8 km along axis. The flow field is primarily lobate lavas (42.2%), but also contains one of the highest abundances of sheet lavas (17.6%) among the mapped DSA2 units. The source of Lobo del Mar lavas is a series of vents aligned along an eruptive fissure inside the axial graben. The western and eastern portions of the flow field are composed of low-relief lobate and sheet lavas. The central section of the flow field includes an extensive zone of collapses and channels. The collapses are up to 10 m deep with abundant lava pillars, and expose a sequence of basal pillow lavas, overlain by sheet flows and topped with lobates. This is the only area in either DSA that displays evidence of both waxing and waning eruption phases [e.g., Wadge, 1978]. The relatively rough morphology of hackly sheet flows within the channels and collapses led the ANFIS to classify these areas as pillow lavas, but the features are obvious in the bathymetry. The channels extend east and west inside the graben and into a partially collapsed lava lake (not covered by the lava morphology map but visible in bathymetry). Lobo del Mar and Lagarto have similar abundances of tectonized terrain (7.2 and 7.9%, respectively), suggesting similar ages of deposition; however sediment cover was not sufficiently different to assign relative ages.
 A series of lava flow fields was mapped in the central region of DSA2 (Figures 11 and 12). Cobija extends nearly 6 km along and nearly 1 km to the north and south of the axis. The near-axis portion of this lava flow field has presumably been buried by more recent eruptions. Cobija consists of low-relief lobate lava flows (48.5%) with the highest percentage of sheet lavas (25.2%) among all mapped DSA2 flow fields. The flow field contains numerous features indicative of lava flow inflation [Hon et al., 1994]. Whaleback-shaped tumuli with flow-parallel axial clefts are abundant throughout Cobija; their axis-perpendicular orientation and arrangement in sinuous “tumulus trains” suggests the trace of lava tubes beneath [e.g., Walker, 1991; Appelgate and Embley, 1992]. Preliminary, paleomagnetic intensity analyses of Cobija samples indicate an age of 400 ± 30 years before present [Bowles et al., 2011], making Cobija the oldest mapped DSA2 flow field.
 Moving toward the axial summit graben, we find two younger units, Iguana and Cocodrilo. Substantial portions of these flow fields have been buried by the products of other eruptions. Iguana is mostly lobates (49%) with sheet flows (12.9%) and some pillow mounds. Cocodrilo extends nearly 10 km along axis and is predominantly pillow lavas (57.8%); however, the western end of Cocodrilo in central DSA2 contains more lobate and sheet lavas than at its eastern end, which primarily consist of higher-relief pillow mounds. Observations of sediment cover suggest that Cocodrilo is younger than Iguana, although no numerical ages have been determined.
 The Calór flow field forms a ridge of mounds and low-relief flows extending ∼8.5 km along axis. The flow field has been strongly overprinted and extensively tectonized. Overall, the unit is mostly pillows (58.7%) and lobates (25.8%). Some sheet flows were mapped and observed at the southern margin of the western section, where bathymetry and acoustic backscatter seem to indicate a “spillway” of low-relief flows; abundant inflationary features and collapses south of this region suggest that Calór may extend further off-axis. The eastern section of Calór has been buried by later eruptions but, like the western section, consists of lobate flows that are often superposed by low-relief mounds of pillow lava; this distribution of lava types suggests waning effusion rates. Observed sediment thickness in Calór is less than in Cocodrilo and Iguana, and preliminary paleomagnetic intensity analyses of Calór samples indicate an age of <200 years before present [Bowles et al., 2011]. Several active high-temperature hydrothermal chimneys were observed in eastern Calór.
 The Empanada, a large low-relief seamount (∼1.7 km diameter; <100 m height) with associated low-relief flows, overprints the central portion of Calór. The Empanada seamount is the largest edifice within DSA2 and is located near the center of the 92° segment. The flanks of the seamount are constructed of many hummocky pillow lava terraces. The center of the cone has been cut by graben faults, exposing truncated pillow lavas with no indication of lava ponding against either fault scarp. Within the summit graben, broad lobates and sheet lavas appear in a chaotic distribution with no predominant flow direction or pattern; the summit also features several large tumuli, collapse features floored with sheet lavas, and a small ∼15 m deep pit crater. The cone overprints an earlier sequence of low-relief lobate and sheet flows that host several large collapses with lava pillars and bathtub rings, another indication of effusion rates that were initially high and waned over time. Overall, the majority of the Empanada flow field consists of pillows (53.9%) and lobates (32.6%).
 The Niños lava flow field forms a ∼6 km-long discontinuous ridge of mounds and low-relief flows. The flow field is 63.8% pillow lava and consists of mounds 100–300 m in diameter, forming a ridge 300–500 m wide, centered within the inner axial graben. Lobate lava flows (21.8%) are found at the summits of most mounds, along with inactive and active clusters of hydrothermal chimneys. The distal parts of the unit have a higher proportion of lobate and sheet lavas, including sheet flows that have been interpreted as the earliest phase of the Niños eruption [McClinton et al., 2011; Colman et al., 2012]. Fault scarps up to 25 m high dissect the ridge, but light sediment cover with hydrothermal staining was observed throughout Niños. Preliminary paleomagnetic intensity analyses of Niños samples indicate it is the youngest eruptive unit mapped within DSA2 (40 ± 30 years before present; [Bowles et al., 2011]).
 Two eruptive units in eastern DSA2 have the highest proportion of pillow lavas and are the smallest among the mapped DSA2 units (Figure 13). Gusanos is composed of two small hummocky ridges (950 and 800 m in length; 150 m width, 20–30 m height). The ridges are almost entirely pillow lavas (93.2%), with a minor amount of lobates at the summits of some individual mounds. In between and at either end of the Gusanos ridges are the three Dulces mounds (200 m diameter, 30–40 m height) of almost exclusively pillow lavas (91.4%). Dulces lavas have the highest SiO2 content among DSA2 flow fields and individual pillows on Dulces mounds are among the largest observed within DSA2. Light sediment cover on both Gusanos and Dulces suggest they were emplaced relatively recently. The neighboring Cocodrilo flow field has thicker sediment cover and both Gusanos and Dulces overprint some faults that cross-cut Cocodrilo lavas, indicating these are younger than Cocodrilo.
 This paper is one of the few to systematically examine the abundance and spatial distribution of lava flow morphologies within individual eruptive units at a MOR. Previous studies [e.g., Fox et al., 1988; Kurras et al., 2000; Engels et al., 2003; Soule et al., 2005; Fundis et al., 2010; White et al., 2008] used near-bottom photographs to manually tabulate the relative abundance of lava morphology at subsegment-scale portions of different MORs; these studies relied on data from towed camera systems that may not have provided a representative sampling of lava morphology distribution due to the limited extent of direct visual observations of deep seafloor. The ANFIS lava morphology maps provide a new method for examining volcanic emplacement processes over large contiguous areas of seafloor, thus enabling spatially comprehensive measurements of the abundance and distribution of the type of lavas emplaced; this type of data is fundamental for understanding how the upper crust forms at MORs. Furthermore, the combination of regional lava morphology analysis with detailed geologic mapping of specific eruptive units allows us to evaluate how lava emplacement varies within and between eruptive episodes. We use the overall relative abundance of lava morphologies as a means of comparing effusion rates between eruptive units, and use their local spatial distribution to interpret emplacement processes, based on empirical observations and the modeled relationship between effusion rate and lava morphology. In addition, the distribution of lava flow morphologies can be used to indicate spatial and temporal variations in emplacement parameters. However, prior to discussing the implications of the lava morphology maps in the context of effusion rates, it is important to discuss the potential influences of other variables.
5.1. Controls on Lava Morphology at the GSC
 Unlike subaerial lava flows, the morphology of submarine lava flows is profoundly affected by the development of a stiff surface crust of quenched glass due to the temperature gradient between seawater and lava. The growth of this crust is controlled by competing rates of heat loss and advection [e.g., Griffiths and Fink, 1992; Gregg and Fink, 2000; Sakimoto and Gregg, 2001]. If cooling exceeds advection, the surface crust will rapidly thicken and impede lateral movement, resulting in a pile of short pillowed flows; if advection exceeds cooling, weak surface crusts will buckle or fold in response to the movement of lava beneath, allowing flows to travel farther and crust over at flow margins [Fink and Griffiths, 1990]. For deep submarine lava flows (>1500 m), cooling rates remain relatively constant due to the significant thermal gradient between near-freezing seawater and the extruding lava [Fink and Griffiths, 1990; Griffiths and Fink, 1992]; thus, the dominant parameter controlling flow morphology becomes the timescale of advection, which essentially describes a volumetric flux of mass and thermal energy that is controlled by the viscosity of the lava, the slope of the underlying topography, and local flow rates [e.g., Bonatti and Harrison, 1988; Gregg and Fink, 1995; Gregg and Smith, 2003].
 Through laboratory simulations with polyethylene glycol wax analogues to submarine lava flows, Gregg and Fink [1995, 2000] showed that increasing the underlying slope had an effect similar to increasing the effusion rate, to the extent that both increase local flow velocities. The modeled relationship predicted that sheet flows would progressively become more common at steeper slopes. However, Gregg and Smith  observed the opposite relationship at Puna Ridge, the submarine extension of Kilauea Volcano's Eastern Rift Zone, with sheet flows forming at shallow slopes (<15°) and pillows and lobates preferentially formed at high slopes due to a stronger gravitational influence which acts to break the flow apart and thus decrease local flow rates.
 Using the Sentry bathymetry, we extracted slope values for each grid cell of the GSC lava morphology maps (Figure 14). This set of data consists of nearly 14 million separate data points (7.1 million in DSA1; 6.5 million in DSA2) that describe the distribution of lava flow morphologies as a function of the underlying slope, which could alternately represent either the slope on which the flow morphology formed or the slope formed by the flow morphology itself. Nearly 85% of the terrain in each of the DSAs has a slope of <30°. Pillowed flows are ubiquitous throughout both DSAs, but are most common at higher slopes. In DSA1, slopes above 15° consist exclusively of pillow lavas and rubble; lobate and sheet lavas become increasingly abundant as slope decreases and are the most abundant flow morphologies at slopes <5° (recall that sheets and lobates were mapped as a single morphology in DSA1). In DSA2, where we hypothesize average eruption rates are higher, sheet lavas are also exclusive to slopes <15°, while lobates are found at slopes up to 30°; only pillow lavas and rubble are found on slopes >30°.
 Aside from any potential differences in vent effusion rates between DSAs, higher slopes apparently disrupt local flow rates to the extent that a morphological transition occurs. A slope of 15° appears to be a threshold above which sheet flows are not formed in either DSA. Lobate lavas are the most abundant morphology at slopes <5° in both DSAs and appear on slopes up to 30° in DSA2, at which point the terrain probably becomes too steep for lava to advance as a single coherent flow, and a transition to pillows occurs. This may be similar to subaerial lava flowing over Hawaiian pali (steep cliffs). As lavas flow over the pali, and onto the flat surface at the base of the pali, their morphologies are strongly affected; some flows return to their original morphology while the morphology of other flows is permanently altered [e.g., Wolfe et al., 1987; Mattox et al., 1993; Hon et al., 1994].
 Laboratory simulations have also shown that increasing lava viscosity has an effect similar to decreasing the effusion rate [e.g., Fink and Griffiths, 1990; Griffiths and Fink, 1992; Gregg and Fink, 1995, 2000]. The viscosity of submarine basaltic lavas is a combined function of melt composition (including water content), eruption temperature, and crystal content [Bottinga and Weil, 1972; Shaw, 1972]. Colman et al.  calculated melt viscosity values for rock samples collected in both DSAs using the method of Giordano et al. ; we have averaged viscosity values for each flow field and compare these average viscosities to lava morphology (Figure 15). This comparison does not account for variations in lava viscosities throughout each flow field possibly arising from variations in crystal content or any syneruptive, small-scale viscosity variations related to lava emplacement.
 In both DSAs, there are variations in lava flow morphology that are independent of melt viscosity. In DSA1, the Del Norte flow field, which is largely composed of sheet and lobate lavas, displays nearly identical average viscosity values as Buho and Pulgar lavas, which are predominantly pillows; estimated phenocryst contents of hand samples from Del Norte are <5–15% plagioclase-phyric and 1–2% olivine-phyric, samples from Buho are nearly aphyric, and samples from Pulgar are 10–15% plagioclase-phyric and <2% olivine-phyric [Colman et al., 2012]. Outside of the Del Norte flow field, melt viscosities in DSA1 are nearly uniform, possibly reflecting a lack of fractionation processes at lower-magma supply. In DSA2, higher volatile content related to increased proximity to the Galapagos hot spot may increase the effects of lava viscosity on the resulting flow morphology. The effects of water content can be illustrated by comparing Lagarto lavas (0.49–0.52 wt % H2O; Colman et al. ), which exhibit the lowest melt viscosity and are primarily sheets and lobates, and Dulces lavas (1.09–1.35 wt % H2O; Colman et al. ), which have the highest melt viscosity and are almost exclusively pillows; samples from both of these flow fields are nearly aphyric. Melt lenses have been seismically imaged at 2–3 km depth beneath DSA2 [Blacic, 2004], and fractionation processes provide a mechanism for the higher and more variable melt viscosity values for DSA2 lavas. Although there appears to be a trend of increasing pillow lava abundance with increasing melt viscosity in DSA2, the complete range of flow morphologies were observed and mapped for lava flow fields with similar melt viscosities: Cobija and Gusanos lavas have nearly identical average melt viscosity values but vastly different flow morphologies, and the same is true for Lobo del Mar and Calór lavas. Given the distribution of lava flow morphologies over the range of estimated melt viscosities, we conclude that flow morphologies within our mapped eruptive units are influenced to a greater extent by flow rates (influenced by the effusion rate from the vent and local topography) than by melt viscosity.
5.2. Effect of Magma Supply on Lava Morphology
 With effusion rate established as the primary control of lava morphology, the relative abundance of lava morphologies within each DSA can be used to indicate different average effusion rates between low- and high-magma-supply study areas (Figure 16). In DSA1, pillows are 77% of mapped lava morphology, and probably more abundant if the Del Norte flows are as atypical of low-magma-supply environments as most of the literature suggests. Excluding the Del Norte eruptive unit, pillows make up 87% of lava morphology in DSA1. In DSA2, pillows compose less than half of the mapped area, with an increased abundance of lobate and sheet lavas (31.6 and 11.7%, respectively). The widespread presence of sheet and lobate lavas implies that eruptions with relatively high effusion rates are common in DSA2.
 Changes in the abundance of lava flow morphologies between study areas are clearly evident in the contrasting volcanic morphology within each DSA. In general, slow, periodic, or episodic effusion rates build structures with high aspect ratios, preferentially increasing in height rather than diameter, whereas faster effusion rates build structures with low aspect ratios, increasing in diameter rather than height [Fink et al., 1993]. Eruptive edifices in lower-magma-supply DSA1 have aspect ratios (height/diameter) between 0.12and 0.24; however, these aspect ratios do not necessarily constrain flank slopes, which are typically between 15° and 25° but can range up to ∼40° for some flat-topped cones and mounds. This high-relief terrain is due to low effusion rates that slow the advance rates and result in lava flows piling atop each other, forming relatively steep-sided structures. In contrast, eruptive edifices in higher-magma-supply DSA2 have aspect ratios <0.10 with flank slopes typically <15°. Higher effusion rates in DSA2 allow lavas to advance from eruptive vents as thin flows, which results in a much lower-relief terrain.
 Outside of these correlations, the complete spectrum of eruption styles may occur within segments at any overall magma supply, demonstrating that local factors controlling individual eruptions may vary strongly, and that limited lava morphology mapping may not capture the true nature of variability in volcanic behavior in the system. For example, the Del Norte flow field in lower-magma-supply DSA1 (estimated volume of 0.14 km3; Colman et al. ) is analogous to DSA2 flow fields in its form and emplacement, displaying abundant lava flow inflation and collapse features and indications of flow localization. Higher effusion rates in DSA2 are apparently sufficient to maintain long fissure eruptions, but the Frijoles unit in DSA1 is the longest fissure eruption in either study area (10.6 km). In DSA2, the Dulces and Gusanos flow fields are atypical compared to other GSC eruptions at higher-magma-supply segments [Colman et al., 2012; White et al., 2008]. Both units are small-volume (<0.005 km3) eruptions of almost entirely pillow lavas. Dulces lavas are basaltic andesite with the highest melt viscosity values in either DSA (∼340 Pa s; Colman et al. ); the rheological properties of Dulces lavas thus influenced their flow morphology. At 340 Pa s, the maximum effusion rate for the formation of pillows is about 10 m3/s [Gregg and Fink, 1995], resulting in a minimum emplacement time of ∼18 h for each of the Dulces mounds. Viscosity values modeled for the basaltic Gusanos lavas are similar to other DSA2 lavas; however; this implies a low effusion rate (0.1–1 m3/s; [Gregg and Fink, 1995]) for the Gusanos eruption that is inconsistent with other DSA2 eruptions.
 Relatively few MOR eruptions have been identified, mapped, and sampled; even fewer include estimates of lava flow morphology abundance. The available lava morphology information shows a correlation with spreading rate of the ridge [e.g. Bonatti and Harrison, 1988]; this relationship predicts more sheet flows at higher spreading rates and more pillow lavas at lower spreading rates. Perfit and Chadwick  argued that the supply of magma from AMCs is a primary influence, but others [e.g., Sinton and Detrick, 1992; Phipps-Morgan and Chen, 1993; Rubin and Sinton, 2007] have suggested that AMCs are spreading-rate-dependent features themselves, with faster-spreading ridges supporting shallow, steady-state AMCs.
 Overall lava morphology abundances at the GSC fit within the global trend. Relative abundances at the intermediate-spreading GSC plot between the slow-spreading Mid-Atlantic Ridge (MAR) and the fast-spreading East Pacific Rise with respect to both spreading rate and magma supply (Figure 17). However, there are significant variations in lava morphology abundance between lower-magma-supply DSA1 and higher-magma-supply DSA2. Our GSC lava morphology maps support the idea that effusion rates increase with magma supply even at a nearly constant spreading rate. This relationship implies that the rate of supply of magma to a ridge is the pivotal factor in the construction of oceanic crust, affecting the style of volcanic eruptions, the mode of lava emplacement, and ultimately, controlling the physical properties of the upper ocean crust. Furthermore, this predicts that similar volcanic morphology may be displayed by MORs with different spreading rates but comparable magma supply rates.
5.3. Distribution of Lava Morphologies: Indicators of Eruption and Emplacement Processes
 In DSA2, magma ascent and effusion rates are apparently high enough to maintain fissure eruptions >5 km in length (e.g., Ninos, Calór, Empanada, Cobija)[Colman et al., 2012]. The initial products of these eruptions are low-relief sheet and lobate lava flows that often extend up to 1 km off axis. These low-relief flows contain abundant lava inflation features, commonly have pillowed flow fronts enclosing lobate flow fields and are overprinted by vent-proximal pillow mounds that are aligned along the trace of the original fissure. Collapses within the low-relief flows often expose layers of sheet lavas. We interpret this progression of flow facies as the result of gradually waning effusion rates and increasing flow localization, which has previously been inferred from visual observations of lava flow morphology at other MORs [e.g., Ballard and van Andel, 1977, Chadwick and Embley, 1994; Embley et al., 1999; Fundis et al., 2010; Caress et al., 2012]. Flow localization has been frequently observed during subaerial basaltic fissure eruptions in Iceland [e.g., Thorarinsson et al., 1973; Wadge, 1978; Harris et al., 2001] and Hawai'i [e.g., Richter et al., 1973; Swanson et al., 1979; Wolfe et al., 1987; Lockwood et al., 1987] as extrusion along most of a fissure system ceases and flow focuses to point-source vents that build cones or shields. The localization process has been linked to both variations in effusion rate [e.g., Wadge, 1978] and flow restrictions arising from cooling and solidification within the dike and eruptive conduit [e.g., Delaney and Pollard, 1982].
 We assume that flow focusing also occurs in lower-magma-supply DSA1, although the process is less obvious from lava flow morphology. Eruptions either start at lower effusion rates, or begin at higher effusion rates but proceed to erupt sufficient lava at lower effusion rates to overprint the initial higher effusion rate flows. DSA1 eruptions likely result in a swift transition from line- to point-source eruptions as conduit widths rapidly decrease. In this scenario, any initial high-flow-rate extrusions of lobates or sheets are presumably overprinted by pillow mounds or ridges (e.g., Buho, Pulgar, Dragón, Altares). Eruptive volumes are larger for DSA1 flow fields [Colman et al., 2012]; the large flat-topped volcanic cones in DSA1 are therefore indicative of long-term, steady-rate eruptions. Assuming a constant effusion rate of 0.5–1 m3/s from a point-source vent [Gregg and Fink, 1995] yields emplacement times of ∼1.9–3.8 years for the cone within the Buho flow field, 1.1–2.2 years for the Pinguino cone, and 0.75–1.5 years for the Tortuga cone. Applying the same estimate of effusion rate to mound fields within DSA1 yields emplacement times of ∼0.29–0.58 years for Dragón, 0.38–0.76 years for Pulgar, and 0.32–0.63 years for Altares, assuming all mounds were emplaced simultaneously.
 Digital lava flow morphology maps produced by ANFIS classification of sonar-derived geometric, acoustic, and texture attributes of seafloor, and calibrated with observational data from submersible dives and camera tows, have enabled spatially comprehensive measurements of the lava flow morphology along the GSC. We have examined the lava morphology maps to better understand the controls on the emplacement of a series of previously mapped lava flow fields.
 (1) At the GSC, lava flow morphology primarily varies in response to spatial and temporal variations in effusion rate from eruptive vents and topography-driven variations in local flow rates. Although there are lava flow fields in which viscosity may have affected flow rates (e.g., Dulces, Lagarto), most of the variations in lava flow morphology are independent of variations in modeled viscosity values. Each lava morphology is also distributed over a range of bathymetric slopes; however, there are slope thresholds that influence local flow rates thus resulting in morphological transitions. In lower-magma-supply DSA1, a 15° gradient marks a transition from mixed pillows, sheets, and lobates to exclusively pillows comprising steeper slopes; in higher-magma-supply DSA2, no sheet flows are found above 15°, and no lobates are found above 30°.
 (2) The increased abundance of lobate and sheet lavas in DSA2 compared to DSA1, along with the presence of lava lakes and channels, imply eruptions with higher average effusion rates at higher-magma-supply rates. The hot spot-distal, lower-magma-supply DSA1 (95°W) contains ∼77% pillow lavas, ∼16% lobate and sheet lavas, and ∼7% tectonized terrain. Our observations suggest eruptions in DSA1 are predominantly low effusion rate events that quickly focus to point-source vents, producing high-relief pillow lava structures at various scales. The higher-magma-supply, hot spot-proximal DSA2 (92°W) is composed of ∼47% pillow lavas, ∼32% lobate lavas, ∼12% sheet lavas, and ∼9% tectonized terrain. Effusion rates at 92°W are high enough to maintain >5 km long fissure eruptions in which lobates are emplaced as inflated lava flow fields, and flow localization builds lines of small pillow mounds along the eruptive vents on the spreading axis as the eruption proceeds.
 (3) The GSC lava morphology maps provide data points that, for the first time, extend the catalog of MOR lava morphology observations independent of spreading rates. Overall differences in lava morphology abundance (and by inference, average effusion rates) between DSA1 and DSA2 are directly correlated with a 30% change in magma supply but a negligible change in spreading rate, implying that magma supply is a primary control on the nature of volcanic eruptions at MORs.
 The authors wish to acknowledge the crew of the R/V Atlantis and Alvin, Sentry, and TowCam teams for their assistance with data acquisition at sea, in addition to the GRUVEE science team for their advice and feedback throughout the course of this project. We also thank Debbie Smith and Eric Mittelstaedt for thoughtful reviews that improved this manuscript. This research was supported by National Science Foundation grant OCE08-49711.