4.1 Flow-Related Features
4.1.1 Broad Channels
 The five broad channels we have described here (Figure 5) are morphologically similar to wide outflow channel systems observed on Mars [e.g., Baker and Kochel, 1978] and Venus [e.g., Head et al., 1991] (though we note the absence of an equally wide lunar counterpart), which are characterized by low channel sinuosity, streamlined bedforms, terraced margins, and longitudinal grooves [Baker and Kochel, 1979; Baker et al., 1992; Leverington, 2007]. Early workers also drew comparisons between the examples on Mars and analogous features in the Channeled Scabland on Earth, which are thought to have formed during a rapid, voluminous discharge of water during the Pleistocene [Baker and Milton, 1974]. Martian outflow channels were thus assigned a diluvial origin due to the swift release of large volumes of groundwater [e.g., Masursky et al., 1977; Coleman, 2003; Manga, 2004].
 Liquid water is not stable at the surface temperature and pressure on Mercury, and so we discount an aqueous fluid as the responsible agent for the broad channels we observed. We also discount channel formation via erosion by impact melt, which would lack the necessary flux, volume, and spatial distribution required to form the expansive channel network seen on Mercury.
 Given the proximity of the outflow channels in our study to the northern volcanic plains and the circum-Caloris plains, we suggest that these features may have been produced, at least in part, by lava erosion. There is a large body of literature in which the erosive capacity of lava has been invoked as a mechanism for the formation of channelized landforms on terrestrial planets, including the Moon [e.g., Greeley, 1971; Hulme, 1973; Williams et al., 2000; Hurwitz et al., 2012], Venus [e.g., Komatsu et al., 1992, 1993], Mars [e.g., Leverington, 2004; Williams et al., 2005], and Io [Williams et al., 2001b; Schenk and Williams, 2004]. Such lava erosion would have led to streamlined erosional residuals, to which the elongate kipukas in Mercury's valles may correspond. Lava erosion is accomplished via a combination of thermal and mechanical processes, with the capacity of lava to incise through substrata enhanced by factors such as higher temperatures, lower viscosity, and more turbulent flow conditions [e.g., Leverington, 2007; Hurwitz et al., 2012].
 There is strong geochemical evidence that much of Mercury's crust is mafic to ultramafic in composition [Nittler et al., 2011; Stockstill-Cahill et al., 2012; Weider et al., 2012] (see section 3.2). The low SiO2 content, high liquidus temperatures, and high densities of these lithologies when molten, relative to more evolved melts, means that the factors listed above will be all the more relevant, especially for ultramafic, komatiite-like compositions. Taken together, therefore, the morphological observations we document and the geochemical results reported by others provide a compelling case for the above-ground movement of, and mechanical and/or thermal erosion by, voluminous, high-temperature, low-viscosity mafic to ultramafic lavas.
 Lavas mechanically erode substrata by physically degrading and removing material, whereas thermal erosion is accomplished by the ablation and melting of country rock [e.g., Williams et al., 2001a; Leverington, 2007]. Consolidated substrata, such as solidified basalts, resist mechanical erosion to a greater extent than less cohesive material (e.g., impact regolith), whereas hotter surfaces are thermally eroded in less time than that required for cooler material [Hurwitz et al., 2010]. Hurwitz et al.,  calculated and discussed possible erosion rates for Angkor Vallis, on the assumption that the channel depth observed today represents the maximum extent of incision. They concluded that thermal erosion was likely the dominant regime in which it formed, with the time taken to incise the channel differing by an order of magnitude (30–300 days) depending on the lava viscosity and channel slope (see their Table 3). (These workers also suggested that mechanical erosion may have contributed to initial channel formation in locations where unconsolidated substrata were present; although the depth of regolith is unknown and undoubtedly varies across the area, the round, softened forms of both channel kipukas and the margins of the coalesced depressions attest to at least some depth of surficial, loosely consolidated material.)
 Of course, the ability of lava to shape Mercury's broad channels would have been augmented by the presence of any preexisting linear depressions. Fassett et al.  described a set of troughs and grooves radial to the Caloris basin within the geological unit termed the Van Eyck Formation [McCauley et al., 1981], which is regarded as terrain sculpted by the Caloris basin impact on the basis of similarity to terrain around large lunar basins. Of the broad channels we describe in this paper, all are part of the Van Eyck Formation, and all but one (Paestum Vallis) are radial to Caloris; indeed, two are featured in the study by Fassett et al.  (Angkor and Timgad Valles, though the valles were not named as such at the time; see their Figure 7a).
 These observations, then, provide support for the formation of the broad channels by lavas that first flooded, and then modified the shape of, preexisting topographic depressions. Other radial troughs in the Van Eyck Formation of similar lengths and distances to Caloris feature scalloped margins in contrast to the regular edges of the broad channels, and they are not smooth floored, do not feature elongate kipukas aligned parallel to the trough axes, and do not form an interconnected network of smooth plains. Therefore, although we attribute the morphological characteristics of these channels to erosion by lava, the dimensions of preexisting impact-sculpted troughs and grooves likely played a controlling role in final channel morphometry in at least four cases.
 Definitive channel depths are difficult to estimate. The depths we quote in Table 1 are valid as absolutes only if total drainage of lavas within the valles occurred. Otherwise, the channel floors we observe represent the upper surfaces of cooling crusts formed by lavas that continued to flow at depth, potentially for some time. It is not possible to determine with certainty whether or not the channels were entirely drained (though the lack of benches along the channel walls suggests not), or to evaluate the initial depths of the precursory impact-sculpted troughs. We note, however, that a 15-km-wide trough to the northeast of Caloris (located at 41°N, 185°E), within the Van Eyck Formation but largely unfilled by Caloris exterior plains, has an approximate depth of ~2 km, substantially deeper than the depths for the valles we report in Table 1 and suggestive of a considerably greater depth than appears today.
 The majority of kipukas in our study area (Figure 6) are likely due to impact processes, reflecting portions of central peaks, peak rings, crater rims, and ejecta partially buried by volcanism, and this partial burial applies equally to kipukas within the channels, as remnants of original sculpted terrain. Streamlined forms within Martian outflow channels are often teardrop shaped, the result of hydraulic modification of obstacles in the channels that divert the flow, with the tapered point indicating the downstream direction [e.g., Baker and Milton, 1974; Melosh, 2011]. The examples within Mercury's broad channels are not characteristically teardrop in planform and so do not unequivocally indicate the flow direction within each channel. The flow direction instead is given, to within a 180° ambiguity, by the long axes of the channels themselves. Selecting between the two possible directions of flow through the valles is challenging, given that expansive (and possible source) regions of smooth plains are present both to the northwest and southeast of the flow assemblage. The splay-like pattern of kipukas at the southeastern end of Angkor Vallis (described in section 3.1.1) could represent a widening of the channel as lava encountered and flowed around a topographic barrier in the form of the Kofi basin rim, an interpretation that invokes a flow direction from northwest to southeast. We acknowledge that without clear flow indicators, however, this interpretation remains tentative.
 Finally, it is worth comparing the morphology and spatial locations of outflow channels on other bodies with those on Mercury. Broad channels on the innermost planet are not as well preserved as their counterparts on Mars and Venus, which may be the result of a higher impact flux on Mercury than those on other terrestrial planets [e.g., Cintala, 1992]. Venusian outflow channels can originate in large topographic depressions [Baker et al., 1992], and Martian channels often head in chaotic terrain [Leverington, 2009]. Few candidate source depressions are evident for Mercury's broad channels, but as each channel links flooded craters we cannot discount the possibility that distinct, early source vents were subsequently buried. Large channels on Mars and Venus also commonly exhibit complex patterns of braided and anastomosing reaches [Masursky, 1973; Komatsu et al., 1993], landforms that are largely absent from the valles we describe. In fluvial systems on Earth, such patterns arise from local increases in river slope or discharge [Schumm, 1985]; if such a relation is applicable to the channels on Mercury, we might infer that their slopes and discharges remained generally steady throughout their formation.
4.1.2 Narrow Channels
 The five narrow channels we describe (Figure 7) are morphologically similar to sinuous rilles documented on the Moon [e.g., Hurwitz et al., 2012] and Mars [e.g., Byrne et al., 2012a], as well as the “canali” on Venus [e.g. Komatsu et al., 1992, 1993]. Though there is some debate as to the fluid agent responsible for Martian rille formation [Bleacher et al., 2010; Murray et al., 2010], several workers concluded that sinuous rilles on the Moon formed by thermal erosion of the surface by high-effusion-rate lava flows [Hulme, 1973; Mouginis-Mark et al., 2008; Williams et al., 2000; Hurwitz et al., 2012], with pooling and subsequent drainage of lava a contributing factor, at least in the case of Vallis Schroteri [Garry and Bleacher, 2011]. More exotic compositions have been suggested as responsible for Venusian canali [e.g., Baker et al., 1992]. Therefore, a volcanological origin for the narrow channels on Mercury is not only plausible but consistent with their geological context and with analogous features elsewhere.
 No distinct source areas for these channels are evident. Three narrow channels link flooded craters and so do not have defined points of origin; the two examples that join the flow assemblage at one end show no depressions at their other, poorly defined ends that would correspond to source vents. In the absence of local vents, then, the narrow channels may have been formed by small pulses of lava that escaped the confines of broad channels and impact craters and that flowed in the downhill direction through lows in the intercrater plains, in some cases entering neighboring craters and helping to fill them, and in at least two instances extending some distance outside the main assemblage. Indeed, as these two channels do not have specific ends but instead appear to fade into muted portions of the surrounding plains, they may have helped to facilitate flooding of these plains by channeling of lavas flowing through the main flow feature assemblage.
 Should the streamlined kipukas within narrow channels represent erosional residuals, however, as they do in the broad channels, lava erosion may have also shaped these channels. The general morphometry of the narrow channels differs from those of the broad channels (Table 1), which may reflect different starting conditions. In the absence of preexisting linear depressions, such as those of impact origin we infer for the broad channels, and with no evidence of any tectonic landform having controlled the course of the narrow channels we describe, erosion of substrata by flows channelized by the hummocky intercrater plains could have exacerbated narrow channel formation. On the basis of analysis of similar channels elsewhere on Mercury, this erosion would likely have been almost entirely thermal in nature [Hurwitz et al., 2013].
4.1.3 Flooded Impact Features
 Those craters and basins that form part of the interconnected channel network (Figure 8) feature one or more breaches of their perimeters, through which lava entered before ponding. Flows would have exploited and widened existing gaps in basin and crater rims, or forced already weakened portions of rim walls to yield, or both. Some craters may have been flooded, either partially or completely, from flows entering at one point, before suffering a further breach in their perimeter elsewhere and so facilitating the onward flow of lava. Depending on the depth of volcanic infill within a given crater, impact-related structures such as central peaks or peak rings may have been partially or completely buried.
 The wrinkle ridges present within many flooded basins are similar to those distributed across the northern and other smooth plains regions on Mercury [Strom et al., 1975; Head et al., 2011], which formed due to shortening of surface units through folding above a blind reverse fault [e.g., Watters, 1988]. The ridges in our study area probably reflect a combination of global stresses imparted by the contraction of Mercury as its interior cooled [cf., Hauck et al., 2004] and basin-localized compression in response to volcanic-fill-induced subsidence [Watters et al., 2012], with local processes contributing to the orientation of those structures with basin-circumferential orientations.
4.1.4 Coalesced Depressions
 On the basis of their irregular, non-circular outlines, together with the lack of prominent raised rims about their perimeters, we discount an impact origin for the four depressions we have described (Figure 9). Instead, their locations within smooth plains units and their proximity to the channel landforms are consistent with a volcanic origin.
 Some depressions elsewhere on Mercury have been identified as sites of pyroclastic activity, on the basis of morphological and spectral characteristics. These vents often show evidence of coalescence; are located atop low, broad rises; and feature a halo of high-reflectance material that has a distinctively steeper or “redder” spectral slope over visible to near-infrared wavelengths than is typical for Mercury [Kerber et al., 2009, 2011]. Whereas the depressions in our study area have scalloped outlines (and thus probably experienced coalescence), there is no attendant rise, bright halo, or spectral contrast between the depression and its surroundings. If the features we describe here were source vents, any erupted material was likely not pyroclastic in nature, but effusive.
 Even so, there is no obvious morphological evidence for associated lava flows, by way of leveed margins, partially collapsed lava tubes, or chains of rootless vents characteristic of effusive volcanism [cf. Bleacher et al., 2007], even in targeted MDIS NAC data, so alternative interpretations for these landforms must be considered. The presence of a smaller, circular pit within the large depression to the northwest in Figure 8a resembles nesting of calderas as is observed on Mars, for example, within the summit caldera complexes of the Tharsis Montes [e.g., Byrne et al., 2012a]. No associated structural evidence of caldera collapse, e.g., a peripheral fault zone, is observed around this or any other depression, though such coherent structures need not necessarily form to facilitate the collapse of overlying strata into subsurface voids. Should these depressions correspond to magma chambers, however, it is unlikely that they alone hosted the volume of material that fed the voluminous flows in this region.
 Alternatively, the depressions may originally have been overlapping impact craters that were then filled by flows, serving to pond lava that eventually drained through subsurface lava tubes. Their crater rims could have been erased by the sustained flow of lavas, though it is not obvious why so few such depressions occur in a region replete with impact features. A final possibility is that these features correspond to lava rise pits (depressions formed by the inflation of a lava flow around an area that experienced little to no inflation [Walker, 1991]), though their size, orders of magnitude larger than those on Earth, may preclude this interpretation.
 In any case, the lack of any apparent lava flows emanating from the depressions suggests collapse without related surface volcanism [e.g., Walker, 1975]. Seven other depressions outside our study area, each located within one of five impact craters or basins, were described by Gillis-Davis et al. . Termed “pit craters,” these landforms have attributes similar to the depressions we describe: they are irregularly shaped, have no observable ejecta, and are rimless with steep sides. Pit craters also have dimensions similar to those we report in Table 2 (see Table 1 in Gillis-Davis et al. ). These authors interpreted the pit craters as the result of collapse into underlying drained magma chambers.
 Under a scenario by which the depressions of this study formed by the lateral withdrawal of magma from shallow magma chambers through subsurface lava tubes, then, their flat floors formed by the infill of country rock that slumped into the pits after the evacuation of the tubes (such that the depressions are essentially very large collapse pits or calderas), with this slumping contributing to the formation of their scalloped margins. The faint fluting along the perimeter of the northwest depression in Figure 8a may correspond to small-scale failure of an upper stratum of poorly consolidated regolith. This scenario renders the depressions late-stage features that formed after surface flow within the channels had ceased, accounting for why they remain visible today and were not buried by subsequent lavas.
4.1.5 Graded Terrain
 The even grading of the intercrater plains material from its characteristic hummocky texture to a more muted surface expression (Figure 10) implies covering to some depth by volcanic material, with this depth a function of flow volume and preexisting topography. MLA gridded data show that the intercrater plains have local, shallow gradients today (Figures 11 and 12), and so it is possible that at least some of their lower-lying portions were inundated by lavas during channel formation, and the degree of surface tempering would correlate directly with the depth of cover.
 Under this interpretation, then, the inundation of the surrounding plains implies that at least some volume of eruptive materials was not constrained to the channels (broad or narrow), but instead flowed overland in a flood lava mode, perhaps reflecting temporary increases in eruptive flux. Such widespread inundation agrees with the interpretation of Head et al.  for the emplacement of the northern plains as a whole. Impact basins and craters within the intercrater plains would have been filled through breaches in their perimeters by sheets of lava without the development of a channel network, but without a sufficient eruptive volume to bury these features entirely. (We regard the smooth deposits within craters that clearly postdate volcanic activity in this area as likely to be ponded impact melt, on the basis of the intact rims and relatively fresh appearances of these features.) As short-lived increases in eruptive flux waned, the flood-mode emplacement of lavas would have ceased, with lava drain-back serving to channelize these flows once more (similar to the scenario suggested by Jaeger et al.  for flows within the Athabasca Valles outflow channel system on Mars).
 The linear patterns visible in parts of the graded terrain are consistent with this interpretation of erosion by flood mode lavas, via the removal of some small mounds and knolls, and the aggradation of lava behind other such features, resulting in ridges and furrows that partially channelized (and therefore indicate the direction of) the overland flows. The furrows occur close to the margins of the valles but then disappear, a pattern that might reflect a reduction in the erosive capacity of the lavas as they moved progressively beyond the confines of these channels.
4.1.6 Apparent Topographic Anomaly
 The general bimodal distribution of elevations in our study area, with channels and flooded craters occupying lower elevations than those of the surrounding intercrater plains material, supports our interpretation that lavas flooded and shaped, through mechanical and/or thermal erosion, existing depressions (impact craters and impact-sculpted terrain). That those portions of intercrater plains with muted textures are also low-lying, and so would have been susceptible to inundation by lavas not constrained by the channels, bolsters this interpretation.
 However, the topographic rises along Angkor and Timgad Valles shown in Figure 11 are not an anticipated consequence of surface lava flow. A constructional component to relief within lava channels has been reported on Earth, in situations where solidified crust accumulates at sites of channel constriction [Lipman and Banks, 1987; Bailey et al., 2006] even as a consistently downhill gradient beneath the crust ensures that flow continues. Such constrictions are often temporary, however, removed by continued flow of lava from upstream, and are usually at the meter scale. Although the channels on Mercury may narrow somewhat at the locations where the rises appear to crest, they are still on the order of 10 km wide at these points. Local reverse gradients have been observed in outflow channels elsewhere (e.g., along the course of Kasei Valles [Robinson and Tanaka, 1990]), and the amplitudes of the rises we show here are substantially less than their wavelengths (the aspect ratio of the largest amplitude rise in the Figure 11b profile corresponds to an uphill grade of 1:40). Nonetheless, such a gradient would have presented a substantial impediment to the uphill flow of even high volumes of low-viscosity lavas.
 There is a growing catalog of topographic anomalies on Mercury that attest to long-wavelength changes in topography since the end of the Late Heavy Bombardment ~3.8 Ga [Solomon et al., 2012]. The northern and southern portions of the 1640-km-diameter Caloris basin floor are elevated, in some parts to levels above the basin rim, and an east–west-oriented trough crosses its center [Oberst et al., 2010; Preusker et al., 2011; Zuber et al., 2012]. This long-wavelength topography does not appear to be the result of volcanic construction, and it shows no evidence of control by the myriad tectonic structures within the basin [Byrne et al., 2012b]. A broad region some 1000 km across in Mercury's smooth northern plains rises ~1.5 km above the surrounding terrain and includes on its flanks craters flooded by plains units whose floors are tilted away from the rise center at angles similar to those of the flank slopes themselves [Klimczak et al., 2012; Zuber et al., 2012]. Such tilted craters strongly suggest that uplift of the rise occurred after plains emplacement. Finally, extensive contractional tectonic systems that bound regions of high-standing terrain on Mercury show evidence of a contribution to long-wavelength topographic relief at a scale of several hundred kilometers, on the basis of tilted crater floors proximal to these systems [Byrne et al., 2012c].
 Several candidate mechanisms for such deformation have been suggested. In some models of mantle convection in Mercury's comparatively thin mantle (relative to the other terrestrial planets), for instance, dynamic topography is predicted to display amplitudes and wavelengths similar to the observations described above [King, 2008]. Lithospheric folding, in response to horizontal shortening driven by global contraction, might also account for at least some measure of topographic change on Mercury [Dombard et al., 2001]. Although the shortest of the wavelengths of the topographic variations in the channels, at several tens of kilometers, are substantially shorter than those documented elsewhere or predicted by earlier studies, the longest-wavelength variations along the two valles are at a comparable scale.
4.2 Flow Analysis
 The results of our flow analysis for Angkor Vallis are given in Table 3, with calculated values for Reynolds number, Re, lava friction coefficient, λ, and flow velocity, u, tabulated with corresponding values for slope, ψ, and flow thickness, h. We found that u for parameters appropriate to komatiite lava at 1360 °C range from <1 m s-1 to almost 60 m s-1 and depend on flow thickness and slope, whereas tholeiitic basalt at 1150 °C with the same range of thickness and slope had velocities of between 0.5 and 39 m s-1. These end-member values likely bracket the actual span of flow velocities for the channelized flows described here; some combinations of flow thickness and channel floor slope yield values of u comparable to flows on Earth (e.g., the 1801 Kaupulehu lava flow of Hualalai volcano reported by Baloga et al. ) and those within Athabasca Valles [Jaeger et al., 2010]. Of nine sets of variables for komatiite lava, all yielded derived values for Re in excess of 2000, the threshold above which flows are considered to be turbulent, whereas only five of nine variable sets for tholeiite lava satisfied this condition for turbulent flow. Should the lava morphologies we describe be the result of turbulent flow (see section 3.2), those Re values less than 2000 may not correspond to probable flow thicknesses and channel floor slopes.
 Calculated values for lava flow rate, Q, lie between 2.2 × 104 and 1.7 × 108 m3 s-1 for komatiite, and between 1.4 × 104 and 1.1 × 108 m3 s-1 for tholeiite. If the estimated flow rates for these lavas are representative of their effusion rates, then the lower Q values correspond to effusion rates only an order of magnitude greater than those postulated for historic, large-volume eruptions on Earth, as well as those for large lava flows on Mars [e.g., Baloga and Glaze, 2008; Glaze et al., 2009]. Moreover, the highest Q value we calculate (1.7 × 108 m3 s-1) is within the effusion rate limit for hypothesized large dikes thought to underlie major graben systems on Mars, e.g., Cerberus Fossae [Jaeger et al., 2010].
 From our calculated u and Q values, cross-sectional areas of 0.028, 0.28, and 2.8 km2 for Angkor Vallis (see Tables 1 and 3), and the estimated fill volume in Kofi basin of 1.5 × 104 km3, komatiitic lavas would have taken between 22 years and ~1 day to fill the basin, depending on slope and flow thickness. Corresponding fill times for tholeiitic basaltic lavas range from 34 years to less than 2 days (Table 4). These are end-member values. A 1-m-thick komatiite flow would solidify long before filling Kofi basin, whereas a fill time of less than 2 days requires the rapid evacuation from depth of enormous volumes of lava. However, a tholeiitic flow 100 m thick, or a 10-m-thick komatiite, would fill Kofi basin in months (with thicker komatiite flows reducing that time to several tens of days). These latter fill times are in line with the thermal erosion rates calculated by Hurwitz et al.  in their study of the formation of this same broad channel (see section 4.1.1).
 It should be stressed that our analysis has inherent limitations. For instance, the channel cross-sectional areas we used are functions of assumed flow thickness (for a fixed channel width). These figures may not reflect Angkor Vallis' original carrying capacity and may thus over-represent actual fill times if that capacity were greater. Moreover, our analysis considers only one broad channel out of a population of five. If each channel were shaped during the same eruptive event, then the effusion rate from the presumably now-buried source(s) of lava would have been substantially higher than that suggested by the flow rates calculated for this individual vallis. Nonetheless, the mid-range fill times we report are consistent with those of Head et al. , who concluded, on the basis of geological and geochemical observations, that the lavas responsible for the northern plains were emplaced rapidly, relative to more typical eruptions on other inner planets, and in large volumes, the result of extensive partial melting of Mercury's mantle and ensuing high effusion rates and widespread volcanism in a flood lava mode.