The Elaver Vallis channels provide a unique opportunity to develop calibrated hydrographs for a Martian flood. Groundwater filled Morella Crater until the enclosed lake overtopped the rim, confirming that the groundwater potentiometric surface exceeded 1771 m. The overtopping flow rapidly breached the crater wall, catastrophically drained the lake, and eroded a broad scabland and two main channels. Laser altimeter data were used to calculate a preflood lake volume of 2.216 × 1012 m3. Dam breach methods were used to develop hydrographs of discharge and lake stage over time. The peak discharge was in the range of 1.94 × 107 to 3.51 × 107 m3 s–1, depending on the underlying breach erosion scenario. Approximately 95% of the drainable lake volume discharged in 6.4 to 7.5 days. The 240 m deep southern channel entirely formed in 1 to 1.7 days, indicating a mean erosion rate of 0.10 to 0.17 m min–1. The volume of material eroded from the channels was 0.45 of the flood volume, but the actual eroded fraction was probably greater because distant channel reaches were obliterated by the postflood growth of Ganges Chasma. For comparison, flow calculations were performed using open-channel methods. Channel energy slopes had to be corrected for a postflood tectonic tilt in the regional surface. Comparing the hydrographs to the open-channel calculations reveals that the northern Elaver Vallis channel as seen today never flowed bank full or even at half its maximum depth because the implied flows would have exceeded the peak discharges.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Elaver Vallis begins at the eastern rim of Morella Crater; a 78 km wide Noachian crater located south of Ganges Chasma (Figure 1). Morella is cross-listed as crater 3086-095 in the revised Catalog of Large Martian Impact Craters [Barlow, 2003; U.S. Geological Survey (USGS), 2012]. The channel system was carved by catastrophic drainage of a Hesperian lake that formerly existed in the crater. The lake dimensions and the water volume drained can reasonably be reconstructed because of the availability of gridded data from the Mars Orbiter Laser Altimeter (MOLA), and because a small channel formed by overtopping was preserved on the crater rim. The flood occurred in late-Hesperian time because the Elaver channels eroded early Hesperian strata (unit Hpl3) of the Plateau Sequence [Scott and Tanaka, 1986; Witbeck et al., 1991].
 The lake level rose until the wall of Morella crater was overtopped and breached, and the lake catastrophically drained through the gap. I analyze the Elaver Vallis megaflood using dam breach methods to develop discharge and lake stage hydrographs. With insights gained from the discharge hydrographs, the flooding is also evaluated using traditional open-channel methods that have previously been used to study outflow channels.
 This paper is divided into five main parts. Section 2 examines the water source and flood volume by analyzing the dimensions and characteristics of Morella Crater and its paleolake. Section 3 presents a hydrologic analysis of the flood using dam breach methods and presents discharge and stage hydrographs. Section 4 examines the geomorphology of the channels eroded by the flood, and section 5 compares the hydrologic analysis in Section 3 to open-channel methods commonly used to estimate peak flood discharges on Mars. Section 6 evaluates the volume of material eroded from the channels.
1.1 Previous Work
 A terrestrial megaflood analogous to the Elaver Vallis flood produced the largest known Holocene flood. The caldera at Aniakchak Volcano, Alaska filled with ~3.7 × 109 m3 of water and eventually overflowed and breached. The peak discharge was estimated at 1 × 106 m3 s–1 [Waythomas et al., 1996; O'Connor and Beebee, 2009].
Burr et al.  presented numerous papers that focus on the extensive evidence for megaflooding on Earth and Mars. Head et al. , Manga , and Hanna and Phillips  showed that fossae like Cerberus Fossae could be capable of transmitting groundwater at high discharge rates. Manga  found that aquifer permeabilities similar to those of young basalts on Earth, i.e., 109 m2, could support discharges of 106 m3 s–1 as interpreted by Burr et al.  for Athabasca Vallis. Burr et al.  acknowledged their discharge estimate may be too large due to assuming bank-full flow. Manga  concludes that multiple floods may have been needed to create Athabasca. Harrison and Grimm  propose that hundreds of floods may have been needed to erode the large circum-Chryse channels. In addition, see the Supporting Information for a brief critique of alternative claims that Martian outflow channels were eroded by lavas.
Komatsu et al.  discussed the crater-lake source of Elaver Vallis. I have previously analyzed the Morella Crater-Elaver Vallis system with several coauthors, including Coleman et al. , Coleman and Dinwiddie , and Coleman and Baker . Those earlier works did not develop hydrographs for the Elaver flooding, but we estimated that the peak discharge may have reached 3 × 107 m3 s–1 and established that a high potentiometric surface in Ophir Planum led to the groundwater outflow that formed Elaver Vallis and two other channels. We also concluded that Ganges Chasma as seen today probably did not exist at the time of the Elaver Vallis flood; otherwise, the groundwater breakouts would have occurred in its depths instead of on the adjacent plateau [Coleman and Dinwiddie, 2007; Coleman and Baker, 2009]. Lucas et al.  also studied the geomorphology of Elaver and Allegheny Valles, and their relationships to Ganges Chasma. They estimated peak discharge rates in Elaver Vallis of 3–9 × 107 m3 s–1 (bank-full conditions) and what they interpreted to be a more realistic range of 6–9 × 106 m3 s–1 based on channel groove depths.
Walder and O'Connor  describe methods for developing hydrographs and predicting the peak discharge of floods caused by failure of natural and constructed earthen dams. Their methodology is used here to evaluate the Elaver Vallis flood. The dam breach literature is extensive. For a review of related theory see Fread and Harbaugh , Fread , and Singh .
2 Morella Crater and its Paleolake
2.1 Source of Water
 Elaver Vallis begins at a gap in the eastern rim of Morella Crater (Figures 1 and 2). This morphology reveals that a paleolake was present in the crater prior to the breach of the crater rim. The discharge that formed Elaver Vallis clearly flowed out of the crater rather than into it. This is verified by the mean energy slope of the channels, by the position of streamlined “tails” on the eastern end of midchannel “islands”, and by the existence and orientation of a cataract in the crater just west of the crater breach (Figure 3a). The energy slope or gradient is the slope of the energy line of a body of flowing water, referenced to any plane.
 Groundwater was the only plausible source of the water that filled Morella Crater because Elaver Vallis exits Morella at a single outlet and no inflow channels exist. Morella therefore represents a Hesperian lake basin that was fed and filled by groundwater, unlike other paleolakes in impact structures (e.g., Gusev Crater) that were fed by inflows of surface water. It was previously proposed that the inception of Ganges Cavus (Figure 1) ruptured the cryosphere, permitting confined groundwater to rush upward into the crater [Coleman et al., 2007]. Water discharged onto the adjacent crater floor at an elevation of ~1080 m (see Figure S2 in the Supporting Information).
 With several coauthors [Coleman et al., 2007], we have previously interpreted Ganges Cavus as the easternmost cavus of the Ophir Catenae structural complex (see online regional map at http://planetarynames.wr.usgs.gov/images/mc18_mola.pdf). Ophir Catenae is a long chain of pits, likely formed by dilational faulting [cf. Wyrick et al., 2004] that extends 750 km from eastern Candor Chasma to western Ganges Chasma. The en echelon segments of this pit chain are likely surface manifestations of a fault zone that is part of the larger Valles Marineris structural complex. Dilational faulting would have created highly anisotropic conditions, increasing the aquifer permeability within and parallel to the catena and enhancing the eastward flow of groundwater. It is possible that the upward groundwater flow to Morella Crater occurred during the inception of the structural features that ultimately formed Ganges Cavus, and that the deep subsidence feature itself did not fully form until after the flooding.
 Ganges Cavus is as deep as Ganges Chasma, an enormous canyon just north of Morella Crater. The southern rim of Ganges Chasma truncates and therefore postdates the Elaver channel. The western rim of Ganges Chasma also truncates Allegheny Vallis, another channel system that originates from a cavus in the Ophir Catenae structural system. Therefore, the cross-cutting relationships between the channels and the margins of Ganges Chasma demonstrate that the chasm continued to evolve, deepening and expanding after the epoch in which the channels formed [Coleman et al., 2007].
2.2 Lake Surface Elevations in Morella Crater
 If the rim of Morella Crater had been high enough, the paleolake surface would have risen until the pressure of the lake water column equaled the pressure in the subcryosphere aquifer system. The crater acted like an enormous standpipe that slowly filled with groundwater. However, the water level in the lake did not rise as high as the regional potentiometric surface because the crater rim was eventually overtopped. MOLA data show that the lake level rose at least as high as 1771–1786 m above the datum because a small channel at this elevation preserves high-water marks on the crater rim just south of the water gap (Figures 2-4). This small crossover channel was formed by the initial overtopping and preserved as a hanging valley while a deeper channel rapidly eroded the breach to form the water gap, capturing all remaining flow. The existence of the small, well preserved, crossover channel on a highly degraded crater rim confirms that the wall of Morella Crater was overtopped and eroded, not simply breached by lateral pressure from the lake water or from piping and erosion through the wall.
 MOLA data confirm that the northern rim of Morella Crater is in places a few tens of meters lower than the elevation of the crossover channel. There are no channels present at the low places, which suggests the northern crater rim could have been slightly lowered by erosion long after the Elaver flooding. However, the pristine appearance of the crossover channel suggests that postflood erosion and crater degradation effects have been minimal. Another plausible explanation is that the southward expansion of Ganges Chasma caused some post-Elaver subsidence of the northern rim of Morella. The chasm boundary is only ~5 km from the northern floor of Morella Crater.
 The overtopping elevation of Morella Crater is consistent with the likely sources of water that filled the crater. Both Allegheny and Walla Walla Valles were produced by groundwater discharges along the structural trend of Ophir Catenae, ~150 km west of Morella Crater and Ganges Cavus [Coleman et al., 2007]. Those groundwater discharges occurred at surface elevations ≥ 2525 m, more than 1400 m higher than the initial discharges from Ganges Cavus to the crater floor and 700 m higher than the overtopping elevation for Morella.
2.3 Water Volume Drained from Morella Crater
 Before analyzing the volume of the paleolake in Morella Crater, I considered whether degradation effects may have significantly changed the crater topography since the time of the Elaver Vallis flood. That discussion is contained in the Supporting Information. Morella has certainly been modified by degradation effects, but most of the degradation likely occurred in the Noachian Period long before the paleolake filled Morella Crater.
 Subtracting the channel floor elevation just east of the water gap (i.e., ~1251 m; Figure 2) from the elevation range for the overflow channel on the rim shows that a water column 520–535 m deep catastrophically drained through the breach in Morella Crater (Figure 3 and Figure S2). This range was determined using five MOLA profiles to evaluate the overflow elevation of the floor of the water gap—this elevation represents the base level to which the crater lake could drain. Two of these passes, 12145 and 17012, cross the eastern part of the water gap and show that the channel floor reaches an elevation of 1251 m, approximately 30 m higher than at the western end of the water gap (pass 19262). Contouring of gridded MOLA data [Goddard Space Flight Center (GSFC), 2006] shows what appears to be a high point, or topographic “saddle,” on the floor of the water gap just west of MOLA pass 17012. However, this “saddle” does not exist—it is an artifact of the MOLA grid contouring extrapolation routine caused by the absence of MOLA profiles at that part of the water gap.
 I calculated a drainable lake volume of 2.265 × 1012 m3 using GRIDVIEW  software and gridded MOLA data at a resolution of 1/128th degree [GSFC, 2006]. This is the crater volume between an approximate crater overtopping elevation of 1775 m and the maximum breach floor elevation of 1251 m in the outflow gap. This lake volume is well defined because most of the floor of Morella Crater lies at elevations <1251 m, and therefore the base of the drainable water column is planar except at the lake margins. A residual lake 171 m deep (1080 to 1251 m) would have remained after the flood (see Figures S1 and S2 in the Supporting Information). Even if Ganges Cavus existed at the time of the Elaver flood, its entire volume lies below the floor of the outflow breach and therefore could not drain. Water in the cavus would have remained as part of the residual lake. As the lake level dropped >520 m, the overlying pressure at Ganges Cavus would have diminished by ~1.9 MPa, causing an increase in the groundwater outflow rate. This water flowed up into the water column of the residual lake, which explains why the discharge did not erode a distinct channel from Ganges Cavus across the floor of Morella Crater toward the outlet. Subsequent outflow would have continued at declining rates until the potentiometric surface fell below the breach floor elevation of 1251 m. MOLA profiles show that the continuing discharge was not sufficient to erode a significant inner channel on the floor of the breach. As already noted, the breach floor to the east is ~30 m higher than the floor to the west, which would not be the case if erosive flows persisted for a significant time.
 I addressed a potential uncertainty in the Morella lake volume related to material lost from the eastern crater rim by fluvial erosion. A promontory can be seen in Figures 2 and 3a on the crater floor just west of the breach outlet. If this promontory was part of the crater rim, which seems likely, then the crater volume NE, E, and SE of it represented rim materials eroded away rather than lake volume. This zone has a volume of 49.00 km3, or 4.900 × 1010 m3 in the elevation band of 1251 to 1775 m, the drainable portion of the volume. Subtracting this volume from the calculated lake volume yields a revised paleolake volume of 2.216 × 1012 m3, similar to Pleistocene Lake Missoula. The Mississippi River at high flood stage (30,000 m3 s–1) would require 855 days to fill this volume.
2.4 Lake Ice Thickness
 Ice would have formed on the Morella paleolake as it filled. It is useful to estimate how thick the ice might have become and whether it could have affected flow conditions at the breach. The most important factor to determine the ice thickness is the time needed to fill the lake. Unfortunately, the geometry of the groundwater discharge source is not known, and that is needed to evaluate the influx rate and fill time. Also needed is information or assumptions about the aquifer. Groundwater outflows from a confined aquifer over time will be highly nonlinear, as illustrated in the fissure flow model of Manga  for discharges at Cerberus Fossae, given a 40 km fissure length and 2 m width. For the conditions analyzed, the discharge rate dropped by a factor of 5 in only 2 h.
 There is an important difference in groundwater outflow geometry at Ganges Cavus vs. Cerberus Fossae. The floor of the cavus is < 20 km long east to west, so if the flow into Morella was from a deep fissure then it was less than half as long as the 40 km flow length at Cerberus Fossae. The possible fissure widths at Ganges Cavus are indeterminate. The reduced flow length plus insights from Manga's  analysis suggest that Morella Crater was very unlikely to fill in less than one year, and could have taken decades depending on the outflow geometry and conditions. What thickness of ice could have formed on the Morella paleolake in such time spans? Times to form various ice thicknesses can be estimated using equation (1), which solves the Stefan problem [Hobbs, 1974], i.e.,
where t is time (s), h is ice thickness (m), Ts is the annual average surface temperature, To is the water freezing point (depends on salinity), ρi is the density of ice (919 kg m–3 at –20 °C), κi is the thermal conductivity of ice (2.4 W m–1 K–1 at –20 °C), Li = 3.3 × 105 J kg–1 (the fusion heat of ice).
 Using a 50–55 K temperature difference between the freezing point of the lake water and the mean surface temperature at the time of the paleolake, ice 10 m thick could form in 3.7 to 4 years, and 20 years would be needed to reach a thickness of ~23 m. Equation (1) only provides upper limits for the rate of ice growth because it assumes no heat is added beneath the ice. The influx of relatively warm groundwater would have constrained ice growth over the southern part of the lake, and fluid convection would have thinned the ice elsewhere. Other factors could also have reduced the rate of growth of an ice cover, such as extreme groundwater salinity and the presence of entrained bubbles in the ice that would have lowered its bulk thermal conductivity. Therefore, even if the lake took 20 years to fill, its ice cover could be expected to be much less than 20 m thick. Such relatively thin ice would have been unlikely to form ice jams at the outlet because the breach in the crater wall was 5 km wide at the top. The rafting of thinner ice blocks toward the outlet could actually have enhanced the erosion of the breach.
3 Peak Discharge and Hydrographs for Outflow Flood From Crater
3.1 Gap Erosion Rate and Factors Influencing Peak Discharge
 Analysis of the outflows from Morella Crater is based on methods developed for terrestrial flooding caused by dam failures. The impoundment size and depth, rate at which a breach can form, breach shape, and whether the breach is cut to the base of the dam are factors that strongly influence the peak discharge rate and the final form of the discharge hydrograph. For impoundments that are very large with respect to final breach depth, the peak discharge is mainly a function of final breach geometry [O'Connor and Beebee, 2009]. In such cases the breach fully forms before significant drawdown of the lake occurs. As discussed below, this scenario applies to Morella Crater, but it is nonetheless useful to further consider why a breach could have formed rapidly in the rim of this crater.
 An extensive literature exists on dam breaches because of the inherent hazards to human lives and infrastructure. The process of breach formation is complex and is the subject of continuing research, based on factors related to embankment design, construction, and material types; primarily rock fill and earth fill of varying properties. For cohesive materials in dams the susceptibility to erosion is a key parameter, dependent on material types, degree of compaction, and moisture content [Morris et al., 2008]. Our understanding of the surface materials at Morella Crater and Elaver Vallis is limited to what can be interpreted from visible and infrared imagery, from inferences about surface mineralogy from other instruments, and from the dynamics of crater formation. MOC (Mars Orbiter Camera) image R0301227 reveals at least six rock layers in the upper wall of the water gap. Layering is not apparent in the lower walls, perhaps due to a veneer of talus material. Nighttime IR images show exceptionally bright banding in the upper walls, indicative of bedrock layers that are likely basalts. Observations made by the TES (Thermal Emission Spectrometer), THEMIS, and OMEGA instruments have shown that the surface of Mars is dominated by minerals common to basaltic rocks [Wyatt and McSween, 2006]. Likewise, Salvatore et al.  analyzed CRISM (Compact Reconnaissance Imaging Spectrometer for Mars) spectra of crater ejecta and bedrock exposures to report evidence of widespread Hesperian basalts in Chryse and Acidalia planitiae.
 Surface materials in the region, and especially for the crater rim, are likely to consist of poorly consolidated rubble that nonetheless is cohesive due to the presence of ground ice in the matrix. Basaltic units probably underlie the surface regolith, with the possibility of overturned stratigraphy as documented at Barringer Meteor Crater in Arizona [Kring, 2007]. An overturned rim sequence is a hallmark of impact craters. Ejected and overturned units exist more than 1 km outward from Barringer Crater, which has a diameter of ~1.2 km [Kring, 2007]. Severe disruption of the stratigraphy is expected to have occurred tens of kilometers outward from the much larger 78 km wide Morella Crater. Disrupted and overturned beds would have been more easily eroded than less altered bedrock, and more susceptible to flood scour and deep channel incision.
 Given the intensely fractured and poorly consolidated nature of crater rim material, I apply the methods of Walder and O'Connor  for the failure of earthen dams to estimate the peak discharge for Elaver Vallis and to develop hydrographs. Their parametric approach assumes that the cross-sectional shape of an evolving dam breach is trapezoidal and that the breach shape does not change with time. MOLA data confirm the breach in the rim of Morella Crater is trapezoidal and relatively flat-floored. Walder and O'Connor  reported that the breach erosion rate significantly affects peak discharge only if the dimensionless lake volume (initial lake volume divided by the cube of the lake depth) is less than about 104. In physical terms, a large value of means the lake has a relatively large volume for the given depth. For the paleolake in Morella Crater, > 1.5 × 104, and therefore exceeds the general criterion for presuming the breach would have fully formed before significant lowering of the lake occurred. Under these conditions the peak discharge approaches a physical limit represented by critical flow through a fully formed breach [Walder and O'Connor, 1997]. As will be discussed, there is good evidence that the channels formed rapidly.
3.2 Water Gap Geometry
 To develop hydrographs and a more accurate value of peak discharge, it is necessary to define the geometry of the breach in the crater wall. Three MOLA passes cross the gap in Morella crater. None cross the gap at an ideal location, and there is a hole in MOLA coverage at the center of the breach. Nonetheless, THEMIS images scaled using JMARS (JAVA Mission-Planning and Analysis for Remote Sensing) coordinates and MOLA data from the best of the passes, 19262 (Figure 3b), can together be used to reconstruct the elevation profile of the breach. MOLA pass 19262 intersects the southern bank of the water gap at right angles and thus yields a relatively accurate slope of 26° (Figure 5). The MOLA slope for the northern bank is less than the true slope because it cuts across the bank at an acute angle. Analysis of scaled THEMIS images shows that the relatively flat breach floor has a width of 2.32 km and the northern bank has a similar slope to that of the southern bank. Therefore, a slope of 26° was used for both sides of the crater breach. Farther east the breach profile loses its symmetry. The average elevation of the channel floor is 1227 m, varying from 1221 to 1249 m at the crossing of MOLA pass 19262. The full width of the channel at its upper margins is ~5 km in the vicinity of MOLA pass 19262. The symmetrical reconstructed profile and MOLA data that support it are shown in Figure 5. This cross-sectional flow area is more than four times larger than at Wallula Gap on the Columbia River, which served as a constriction point for the Pleistocene Missoula floods.
 The geometry of the cross-section in Figure 5 may have changed somewhat over the 3+ Gyr since the flood. There is evidence of a landslide in the Morella water gap (Figure 3a) but not where the breach geometry is defined. The breach profile has a classic trapezoidal shape, typical of many dam breaches on Earth. The sides of the water gap slope ~26° (θ), significantly less than the typical range of 35° ≤ θ ≤ 65° for observed side slopes in 16 terrestrial dam breaches [MacDonald and Langridge-Monopolis, 1984]. The breach at Aniakchak Caldera that spawned a Holocene megaflood has side slopes of 35°–40° [Burr et al., 2009, Plate 20].
3.3 Flood Hydrographs
 Hydrographs of the Elaver flood were developed using the defined breach parameters and the interpreted geometry, depth, and volume characteristics of the paleolake in Morella Crater. The very large dimensionless volume () computed for the former paleolake means the breach would likely have fully formed before significant drawdown occurred. Based on these conditions, I apply a mathematical model by Walder and O'Connor  to generate flood hydrographs. Variables in equations (2) and (3) are defined in Table 1.
Table 1. Parameter Definitions and Resulting Values From Analysis of Water Release From the Crater Lake
bf (at end of flood, height of breach floor relative to base of crater rim)
c1, c2 (numerical constants for breach shape; see discussion in section 3.3)
c1 = 0.405; c2 = 0.544
d (drop in lake level during flood)
Dc (height of overtopped crater rim relative to breach floor)
g (equatorial surface gravity on Mars)
3.71 m s–2
h (drop in lake depth during flood)
from 524 to 0 m
k (breach erosion rate for scenario where breach takes an entire day to erode)
0.364 m min–1
m (exponent related to lake shape)
~ 1.05 (this value matches model output to MOLA-derived volume-elevation curve) m3 s–1 (inflow recharge via Ganges Cavus was occurring but is presumed very small compared to flood discharge) model output in m3 s–1
Wi (initial lake level relative to base of crater rim)
θ (slope of breach sides)
 To properly calibrate the model, a volume-elevation curve was developed for Morella Crater. This curve shows the crater volume that exists below any given lake stage, excluding lake depths below the water gap overflow elevation of 1251 m. In other words, a lake stage height of zero meters corresponds to an elevation of 1251 m. Cumulative discharge from the dam breach model, subtracted from the lake volume, was then plotted on the volume-elevation curve (Figure 6).
Walder and O'Connor  also provide equations (not reproduced here) that address scenarios where more gradual breach erosion rates would be expected. In such cases the hydrograph takes the form of a rising limb that climaxes at a peak discharge, then diminishes over time as the lake drains. To ensure that a full range of plausible breach erosion scenarios has been considered, I also analyzed the Elaver Vallis flood using a linear erosion rate of 0.364 m min–1 that delays complete formation of the breach for one day. It is reasonable to incorporate a range of scenarios for slower channel incision east of Morella Crater, and to also allow for possible gravity scaling effects in the criterion for assuming rapid breach formation (i.e., > 10,000). However, the Morella paleolake exceeded this criterion by 50%.
 Equations (2) and (3) include numerical constants c1 and c2 that relate to breach shape. These constants incorporate hydraulic correction factors (cv and cs) for the velocity of approach and the submergence of broad-crested weirs [Fread, 1977]. Consistent with the approach by Walder and O'Connor , I have chosen values for these constants that neglect energy loss as the floodwater approaches the breach (cv = 1) and that assume there are no significant tailwater effects at the outlet (cs = 1).
 Hydrographs of discharge and lake stage are presented in Figures 7 and 8. The maximum possible discharge through a breach is critical flow with a specific energy equivalent to the head represented by the difference in height between the lake surface and the breach bottom [O'Connor and Beebee, 2009]. The peak discharge (Qp) for the Elaver Vallis flood was ~3.51 × 107 m3 s–1, which represents flow through the fully formed breach. If the breach took one day to fully form, the peak discharge would have been reduced to 1.94 × 107 m3 s–1 because the lake level would have fallen ~140 m by the time the breach fully formed. Approximately 95% of the drainable lake volume discharged in 6.4 to 7.5 days, depending on the breach erosion scenario. The Elaver Vallis channels that we see today were created in that time, with subsequent truncation by the southward growth of Ganges Chasma, chaos development on the channel floors, and modification by meteoritic bombardment.
 The peak discharge rates calculated for Elaver Vallis exceed the peak discharge of ~1.7 × 107 m3 s–1 estimated for maximum Missoula Flood stages in the Spokane Valley [O'Connor and Baker, 1992]. Figure 9 compares the released volume and peak discharge rate for the Elaver Vallis flood to more than 120 terrestrial floods from natural and rock material dams, including three of the largest known megafloods on Earth.
 Although the crater breach sides have an observed slope of 26°, as a test of the sensitivity of the results to breach shape the peak discharge was also calculated assuming steeper side slopes of 40°. The increase in θ changes the r value from 4.43 to 5.34 and results in a calculated peak discharge of 3.40 × 107 m3 s–1, only 3% less than calculated with the observed Morella breach geometry. This result supports the conclusion of Walder and O'Connor  that peak discharge rates are only weakly sensitive to variations in breach side slopes. Therefore, even if some postflood reduction in breach side slopes has occurred due to mass movements, the impact on the peak discharge for Elaver Vallis would be small.
 The computed hydrographs represent discharges from Morella Crater through 10 days of the flood, during which the discharge diminished by more than two orders of magnitude. The initial flows were broad scabland flows prior to channel incision. These flows would have transported large fractions of suspended materials stripped from scablands and incipient channels, which means the actual flood discharge volumes over the plains would have been greater than the relatively clearwater flows from the crater breach. Channel incision would have been rapid, and it is possible to evaluate incision rates by noting the time when flood levels in the northern channel fell below the floor of the southern channel, at the hanging valley where the southern channel begins (Figures 2 and 3a). Three MOLA passes, 17157, 17012, and 12145 cross the bench of the hanging valley and indicate high points on the valley floor of 1474, 1460, and 1463 m, respectively. The track of pass 17012 is shown in Figure 3a. Elevation 1460 m corresponds to a drop in the crater lake stage of 310 to 315 m, allowing for some slope in the floodwater surface from the lake to the channel intersection. According to the lake stage hydrographs in Figure 7, lake depths fell 310–315 m within 1 to 1.7 days of the start of the flood, which means the entire southern channel was eroded in that time. Depths for the southern channel range from 150 to 240 m. Therefore, the mean incision rate for its deepest reaches was a remarkable 0.10 to 0.17 m min–1. Because flow in the southern channel lasted only 1 to 1.7 days (depending on erosion scenario), for the remainder of the flood the hydrograph in Figure 8 also represents the discharge hydrograph for the northern channel. Flow conditions in the northern channel are further evaluated in section 5. I now examine the morphology of the channels eroded by the crater breach flood.
4 Geomorphology of the Elaver Vallis Channels
 The morphology of the Elaver channels is consistent with megaflood features identified on Earth. These features include scabland topography, streamlined hills, residual uplands separating channels, low sinuosity, exhumation of resistant preflood topography, longitudinal grooves, hanging valleys, cataracts, scarps indicating high-water marks, and scour marks and moats around flow obstacles. The fact that the margins of the initial overland flows outside the incised channels remain visible billions of years after the flood shows that resurfacing processes have not been great enough to remove high-water marks, and dust veneers have remained thin enough that the IR thermal properties of eroded surfaces retain good contrast (Figures 1 and 10).
 The initial outflows from the breach of Morella Crater produced broad, 75 km wide scabland flooding (Figure 10). The large region of inundation is not surprising given the extreme peak discharge estimates and the fact that a flood hundreds of meters deep issued from the eastern end of the breach, which rapidly thinned and spread over the plains to the east. Just beyond the crater breach the overland flow began to erode two large channels separated by residual uplands. The northern channel is 350–450 m deep, while the shallower one to the south is 150–240 m deep and forms a long, abandoned meander with hanging valleys at both ends. The presence of this hanging channel (Figures 1 and 2) is consistent with the hydrograph and confirms that the northern channel eroded faster. MOLA pass 17012 (Figure 3a) crosses the first juncture of the southern and northern channels and clearly shows the hanging valley topography, where the floor of the southern channel is 200 m higher than the northern channel floor. Present topography suggests the initial overland flow may have extended beyond the southern channel (see lower left corner of Figure 10), but this flow would have rapidly ceased after the two main channels began to form.
 Longitudinal ridges were carved by hydrodynamic erosion on the deeper channel floors, and streamlined hills occur on the downstream side of craters (Figures 10 and 11). These teardrop-shaped “islands” were produced by a combination of enhanced erosion on the upstream side and deposition and reduced erosion on the lee side of the crater. Multiple MOLA passes (e.g., 14698, 16509, 13717, 15352) cross the northern channel west of chaos B where the channel floor has been well preserved. Individual MOLA shots do not consistently hit crests and valley bottoms of the longitudinal ridges, but give enough detail to show these ridges vary in height from 10 to 35 m. A 9 km wide mesa with longitudinal ridges on its upper surface stands 100–150 m higher than the floor of the northern channel (Figure 2). This round mesa appears to have been a Noachian crater that was overrun and deeply filled by flood basalts, then covered by younger basalts and later exhumed and shaped by differential fluvial erosion [Coleman, 2012].
Table 2. Open-Channel Flow Calculations for the Northern Channel of Elaver Vallis
Flow width and mean depth for the northern channel determined using channel cross-section in MOLA pass 17276; maximum flow depths were greater.
Mean energy slope of the northern channel for the 17 km reach between MOLA passes 13390 and 10661.
Flow speeds were calculated using the approach of Komar  and a range of terrestrial Manning “η” values from 0.045 to 0.055. Values of the Chézy friction coefficient (Cf) were obtained using these “η” values and the estimated range of flow depths [Carr, 1996].
 Multiple chaos occur in the deepest portions of both the northern and southern channels (Figures 1, 10, and 11). Their presence on the floors of these channels cannot be a coincidence and indicates their formation probably relates to channel incision, which thins or removes the cryospheric seal that confines the underlying aquifers. I have described similar flood-triggered chaos on the floor of Ravi Vallis that probably contributed significant groundwater discharges to that channel [Coleman, 2005]. In Figure 11, interior blocks of the chaos at far left retain longitudinal ridges on their tops, which suggests this chaos mostly formed after the channel floor was incised to its present depth. Another small chaos without interior blocks is at far upper left below the figure's scale bar. Interior blocks of chaos B are almost nonexistent because the chaos probably began forming while the Elaver flood ensued and thus experienced extensive erosion. Chaos A in the southern channel also appears to have been initiated by overburden removal. However, the southern channel junction with the northern channel is a hanging valley, therefore any groundwater discharges from chaos A must have ceased before flow stopped in the northern channel.
 As the Elaver Vallis flood progressed the northern and southern channels continued to deepen, with the northern one deepening faster. Flow from the south was directed to a convergence with the northern channel by high ground in the vicinity of 9.35°S, 49.10°W. Here the floor of the southern channel is >150 m higher than the northern channel floor. Downstream from the convergence point is a 12 to 14 km long channel reach where all flow occurred in a single deep channel. In the vicinity of 9.07°S, 48.93°W the northern channel diverges around high-standing ground, an “island” 23 km long, beyond which the flow again converged to a single channel (Figure 11, upper right). The channel reach north of the “island” is 100–200 m deeper than the reach on the southern side, and therefore had a longer duration of flow. I had difficulty defining the energy slope for the eastern reaches of Elaver Vallis until discovering a 60 m high cataract across the northern channel at longitude 49.21°W. The cataract is relatively degraded, perhaps from waning flood deposition. The headwall appears to have migrated upstream a few km after originating in the extreme hydrodynamic environ where flows from the northern and southern channels converged.
 Elaver Vallis abruptly terminates at the southern rim of Ganges Chasma, showing that the Elaver flood occurred before the formation of Ganges Chasma as we see it today (Figure 12). The chasma margin continued to grow southward after the fluvial episode. The steep-walled canyon at the terminus was likely formed by mass-wasting processes, not by fluvial erosion. Evidence for mass wasting includes the abrupt discontinuity in the thalweg elevation profile, similarity of the terminal canyon to analogous valleys not associated with channels, and the fact that landslide deposits of positive relief exist on the canyon side slope and floor below the channel terminus (Figure 12). Also, as shown in the Supporting Information (Figure S3), the longitudinal ridges on the channel floor are cleanly crosscut by the head of the canyon in a pattern that shows headward migration by mass wasting. The fact that the canyon is relatively deep, 1200 m at MOLA pass 18156, and aligned with the channel suggests that the flood incision of Elaver Vallis cut through resistant surface units on the plateau, exposing softer units below and guiding the axis of sapping to align with the channel axis.
5 Open-Channel Flow Analysis
 Now that plausible hydrographs and peak discharges have been modeled for a flood of known total flow volume, I can also evaluate flow conditions using methods that traditionally have been applied to floods on Mars. Open-channel methods have previously been used to analyze Martian paleofloods, but have inherent uncertainties because the maximum flow depths achieved and the channel geometry at that time are difficult to reconstruct. Extreme upper limits for discharge have been estimated by assuming bank-full flows occurred in the channels we observe today, a condition unlikely to have existed in most channels. Having good estimates of discharge over time in the northern channel we can evaluate more realistic flow depths in Elaver Vallis.
 Flow in open channels can be evaluated using various methods, such as the Chézy equation, the modified Manning equation, or other methods. Regardless of the theoretical advantages of one method over others, they yield comparable results when judiciously applied. For a review of theory and analyses for floods on Mars, see Baker , Komar , Wilson et al. , and Kleinhans . I apply a generalized form of the Chézy and Manning equations as described by Carr  to estimate flow speeds and discharges for the northern Elaver channel. The equations include factors for flow resistance, e.g., the Chézy resistance coefficient (Cf) and the Manning roughness coefficient (η). We have only limited close-up views of Martian channel surfaces, and these have been altered over eons by bombardment and aeolian activity. A plausible range of roughness coefficients must be used to address the uncertainties. Tables of descriptions and values to estimate Manning's “η” were published by Chow , Dunne and Leopold , Dingman , and others. I used a range of Manning's “η” values from 0.045 to 0.055, considering that the channels were eroded in basaltic bedrock, with closely spaced joints where macroturbulence and hydrodynamic erosion would produce rough surfaces. Also, channel floor roughness height is substantial with longitudinal ridges 10–35 m high that are concentrated in the deeper parts of the channels. The chosen range of Manning's “η” also considers insights from discussions of channel roughness estimates for Mars by Komar , Baker , and Wilson et al. .
 While analyzing energy slopes with gridded MOLA data, I found the presence of a regional tectonic surface slope toward the chasma. This slope, which formed after the Elaver Vallis flood, was found by drawing topographic profiles up to 100 km long through the upper eroded margins of the Elaver channels. The regional dip can clearly be seen in a MOLA profile in Figure S4 (Supporting Information). The true dip of the regional slope varies from 1.5–2.5 × 10–3 depending on where profiles are drawn. The line of strike for the regional slope is approximately azimuth 55°, with the maximum dip to the NNW approximately along azimuth 325°. This regional slope must be considered to evaluate and correct channel energy slopes and to estimate the volume of material eroded from the channels. I used a mean regional slope of 2 × 10–3 to correct the channel energy slopes.
 Thirty MOLA profiles were used to develop an elevation profile along the thalweg for the northern channel (Figure 13a). The thalweg represents the line of lowest elevation along a valley. The left side of the profile begins partway along the channel at longitude 49.91°W, the crossing of MOLA pass 12975, and extends 95 km to the east. Excluding the easternmost data point, which crosses the terminal canyon formed by mass wasting, the profile suggests a mean energy slope of 2 × 10–3. However, the energy slope is locally variable and does not include the necessary correction for the regional tectonic slope. East of MOLA pass 12975, the channel can be viewed as three line segments with different orientations. From west to east these line segments trend along azimuths 78°, 63°, and 87°, with 90° representing due east. Therefore, the regional slope correction for each segment varies.
 Figure 13a shows two plots: MOLA elevations, and corrected elevations that compensate for the effects of the regional surface slope. The total correction increases with distance from starting point, from west to east. The correction reduces but does not eliminate the reverse energy slope seen in the eastern reach at distances of 75 to 85 km. This area of reverse slope occurs east of the juncture where the southern channel merged with the northern channel, producing extreme hydrodynamic flow conditions. As noted earlier, a degraded cataract 60 m high extends across the channel just upstream from the channel convergence, and its location is apparent given the vertical exaggeration in the thalweg plot.
 After correcting for the effects of regional tilt the mean energy slope increases from ~2.0 × 10–3 to 2.8 × 10–3. Without the correction the mean slope would have an error of almost 30%. For application to the open-channel flow equations I determined a mean energy slope of 2.23 × 10–3 for the 17 km long reach located upstream from the cataract. This reach is relatively straight with a consistent slope measured using four MOLA passes (Figure 13a). Although the Elaver flood was in gradual recession, the calculations here assume quasi-uniform flow conditions were present in this 17 km reach. The energy slope is greatly reduced and in places reversed for the 30 km reach downstream from the cataract.
 It is challenging to select a cross-section for flow calculations because the Elaver Vallis channels are complex, with two main channels that diverge and converge and the presence of multiple chaos and many craters on the channel floors. MOLA pass 17276 (Figures 11 and 13) crosses the 17 km reach upstream of the cataract that is relatively straight and has uniform width and slope without chaos development. The topographic profile for pass 17276 is plotted in Figure 13b. The MOLA pass is not quite perpendicular to the channel axis, differing by 5°. This adds an error of less than ½% in the channel flow widths.
 The channel cross-section in Figure 13b was used to estimate flow speeds and discharge rates for the northern channel. The cross-section shows the southern channel as well, but flow in it was not modeled. Four different water surface elevations were analyzed using open-channel flow methods. These elevations are shown as dashed lines in Figure 13b and represent mean stage elevations because the post-flood regional surface slope has been added. The calculation results are given in Table 2, and these were compared to the discharge hydrograph to gain insights about the timing and plausibility of the open-channel discharge rates.
 Several observations come to mind after comparing open-channel results to the discharge hydrographs in Figure 8. None of the flood stages represent bank-full flow, a scenario which has been used for other channels on Mars to estimate upper limits for discharges. Various researchers have suggested that assuming bank-full flow would likely overestimate the calculated discharge [Wilson et al., 2004]. The 1500 m stage fills the northern channel to only three-fourths of its maximum depth but still results in unrealistic discharges that exceed by 20–50% the hydrograph peak of 3.51 × 107 m3 s–1. The discharge would be even greater if concurrent flow in the southern channel were included. There is no evidence that the channels existed prior to the breach of Morella Crater, so the initial flows would not have been confined to channels and would have caused broad scabland flooding, consistent with the observed geomorphology.
 The 1400 m stage yields lower but still unrealistically high discharges for the northern channel of 1.8–2.2 × 107 m3 s–1. The hydrographs in Figure 8 show that discharges of that magnitude occurred during the first 10–24 h, which is when the southern channel was forming. Therefore, discharge would need to be added to represent the concurrent southern channel flow and residual overland flow, and the combined flows would likely have exceeded the hydrograph peak discharges. It is also possible that, 24 h into the flood, the northern channel at the location of Figure 13b (85 km east of Morella Crater), may not yet have eroded to its full depth and therefore the present geometry would not be proper for calculating discharges for that time.
 The 1300 m stage yields plausible discharges of 2.3–2.8 × 106 m3 s–1, which in both breach erosion scenarios occurred on the third day of the flood when the southern channel was dry and the northern channel conducted all the flow. This stage represents a mean flow depth of 55 m and maximum flow depths of 90–100 m. Likewise, discharges for the 1250 m stage are plausible and represent a mean flow depth of 33 m and maximum flow depth of 50 m. The hydrographs suggest the calculated discharges would have occurred on the 4th or 5th day of the flood, depending on the scenario.
 These results support the view that assuming outflow channels ran bank full with their present geometry would lead to large overestimates of the peak discharge, especially for scenarios where no preflood channels existed. There may be scenarios identified for other channels on Mars where preexisting channels were filled by basin-overtopping flows and then bank-full or greater flows would have been possible. Other hydraulic calculations similarly informed by the discharge hydrographs could be done with respect to bottom shear stress, flow power, Froude number, etc.
6 Eroded Material Transport
 The volume of the eroded channel system was determined using GRIDVIEW  software. The region of fluvial erosion was first outlined using the highest elevations along the upper channel margins for both the northern and southern channels. This was done using gridded MOLA data and individual MOLA passes. The channel “cavity” volume was then determined using the GRIDVIEW area and volume calculation tool. The outer margins of overland flooding were first outlined, encompassing the deeper incised channels. A reference “plane” was then fitted to the defined channel margins. The result is a reference plane that dips slightly to the NNE in the paleoflow direction. The dip would be greater but is partly offset by the regional surface slope to the NNW. The cavity volume below this tilted reference plane is 963 km3, or 9.63 × 1011 m3. The NNE dip of the reference plane differs from the NNW regional tilt because this plane is fitted to the channel margin scarps, the elevations of which reflect both the regional tilt and the terrain energy slope that directed the initial overland flooding.
 To further refine the channel volume, the 9.0 km3 volume of three sizeable, unnamed craters was subtracted from the total, yielding an eroded volume is 954 km3. Chaos volumes below the channel floors were not removed because it is unclear how much of that volume was eroded and transported during the Elaver flooding, during discharge from the chaos themselves, or how much resulted from postflood subsidence. Also, the 49 km3 of material eroded from the easternmost crater rim (section 2.3) was added to the eroded volume, yielding a total of 1003 km3. Given the total flood volume of 2.216 × 1012 m3, on average, each cubic meter of floodwater eroded and transported 0.45 m3 of rock, sediments, and ground ice.
 The method applied gives approximate results. Greater accuracy in the eroded volume could be obtained with the same method by dividing the channel system into numerous segments. Some inherent uncertainties would remain, such as the possibility that broad valleys with elevations below the reference plane existed before the Elaver flood. On the other hand, the transported volumetric fraction was probably greater than calculated here because only part of Elaver Vallis has been preserved. After the flood, the channel was truncated by the southward expansion of Ganges Chasma. Materials eroded from now absent reaches of Elaver Vallis would have further increased the fraction of transported material, but probably not greatly because most of the associated scabland erosion and all of the southern channel erosion took place in preserved reaches. Only the most distant reaches of the northern channel are missing. The large transported fraction suggests the possibility that some eroded material may have included ground ice or even surface ice, which if present would have become incorporated in the flow, adding to the liquid fraction. The large volume of eroded material also suggests the flood had a very high fraction of wash load, which refers to the finest part of the suspended load in a flood. Wash load does not reduce the energy of the flood and can add to the flow power [Carr, 1996]. Wash load was undoubtedly highest during flood onset when flow power and erosion were greatest.
7 Data Summary and Discussion of Uncertainty
 Table 3 summarizes the hydrologic data and calculation results for this study. The hydrographs in Figure 8 represent the first calibrated hydrographs for a Martian flood. The volume of the Morella Crater paleolake was obtained using MOLA profiles and gridded data. This volume is quite accurate because its basal elevation is not the crater floor, but rather a planar surface representing the top of a postflood residual lake, and the maximum lake stage is based on a MOLA pass that crossed an overflow channel high on the crater rim. Most of the postimpact degradation of Morella likely occurred in Noachian time, therefore the present-day crater geometry should reasonably approximate the paleolake volume. The preservation of a large breach in the crater wall permits analysis of its geometry and cross-sectional flow area using MOLA data and high-resolution images. MOLA passes narrowly define the floor elevation of the breach. Equations developed for terrestrial dam breaches are readily applied, incorporating the Martian surface gravity. As shown in section 3.3, changes in the breach side slopes over time would have had little effect on peak discharge. I developed a volume elevation curve using MOLA data to confirm that the modeled discharge through the breach closely matched the decline in lake stage during the flood. Formation of a thick ice layer on the paleolake would have been inhibited by the influx of groundwater at Ganges Cavus. Therefore, ice jams would not have been expected to form at the outflow breach and would not have influenced the peak discharge and hydrographs.
Table 3. Paleolake and Channel Data and Hydrologic Calculation Results
Elevation of high point on floor of water gap, from MOLA pass 12202 (also represents approximate lake surface elevation after flood)
Elevation of overflow channel on the rim of Morella Crater
Elevation of groundwater potentiometric surface for aquifer beneath Ganges Cavus
> 1771 m
Elevation of groundwater potentiometric surface west of Morella Crater at source of Walla Walla Vallis (10.06°S, 54.41°W)
> 2525 m
Area of Morella Crater paleolake at 1775 m contour
Volume of Morella Crater below 1775 m contour
Volume of Morella Crater and Ganges Cavus below 1251 m contour
Range of drainable lake depth in Morella Crater
≥ 520; ≤ 535 m
Total volume of flood from breach of Morella Crater rim (corrected)
2.216 × 1012 m3
2.197–2.269 × 1012 m3
Peak discharge rate during Elaver Vallis flood (rapid breach formation)
3.51 × 107 m3 s–1
Peak discharge rate using hypothetical larger breach side slope (40°)
3.40 × 107 m3 s−1
(r = 5.34)
Peak discharge rate during flood (breach forms in 1 day)
1.94 × 107 m3 s–1
Range of Manning's “η” values for open channel flow analysis
0.045 to 0.055
Mean energy slope for 17 km reach of northern channel between MOLA passes 13390 and 10661
Corrected volume of channel system below a reference plane that reflects a post-flood regional slope to the NNW of ~2 × 10–3
1.003 × 1012 m3
Mean volume fraction of eroded material per cubic meter of floodwater
Incision rate for southern channel during its flow duration
0.10 to 0.17 m min–1
 The dimensionless volume for the Morella paleolake is so large that the breach should have fully formed before significant drawdown of the lake occurred. An extreme peak discharge, consistent with rapid breach erosion, is indicated by the great width of the eroded scabland east of Morella Crater, by the incision of two large channels that begin at a common point just beyond the breach, and by overtopping and severe erosion of the rim of a large crater 25 km east of the breach. Rapid erosion is further supported by the fact that the breach cut through the poorly consolidated materials that form a crater rim. To ensure the peak discharge has been determined within narrow limits, calculations were also done assuming the breach took a full day to erode (Figures 7 and 8). In that scenario the peak discharge would be reduced by 45% but would still exceed the estimated discharges for great megafloods on Earth (Figure 9).
 Although there is uncertainty about the duration of residual flow in the deeper northern channel, it is certain that a single flood of short duration eroded the southern channel. The southern channel is a hanging valley, which confirms that it formed early in the flood when discharge rates through the crater breach were highest.
 The Elaver Vallis channels were eroded by a flood of known volume and demonstrate that a single megaflood spawned by the breach of a crater lake can rapidly erode deep channels on Mars. Morella Crater slowly filled with groundwater until the rim was overtopped. The overflow level confirms the groundwater potentiometric surface stood at an elevation >1771 m. As the lake overtopped the rim, the outflow rapidly eroded and breached the crater wall, drained the crater lake, and eroded the Elaver Vallis channels. The volume of material eroded from the channels was at least 45% of the total flood volume. The transported volumetric fraction probably exceeded 45% because distant reaches of Elaver Vallis were obliterated by the southward expansion of Ganges Chasma.
 Using dam breach methods, hydrographs were developed of discharge and lake stage over time. These are the first calibrated hydrographs for a Martian flood. The peak discharge was ~3.51 × 107 m3 s–1, which represents flow through the fully formed breach. If the breach took one day to fully form, the peak discharge would have been reduced to 1.94 × 107 m3 s–1. Approximately 95% of the drainable lake volume discharged in 6.4 to 7.5 days, depending on the breach erosion scenario. For comparison, peak discharges were calculated for four different flood stages in the northern channel using open-channel methods that traditionally have been used to evaluate floods on Mars. Channel energy slopes first had to be corrected for a postflood change in regional surface slopes, possibly caused by crustal stresses in proximity to the rim of Ganges Chasma. The analyses show that the northern channel of Elaver Vallis as we see it today never flowed bank full or even at half its maximum depth because the implied flows would have exceeded peak discharges from the crater breach.
 The hydrographs reveal that the flood stage fell below the floor of the southern channel in 1 to 1.7 days, therefore the southern channel entirely formed in that time. The mean incision rate for the deepest reaches of that channel was a remarkable 0.10 to 0.17 m min–1. After flooding ceased in the southern channel the northern channel carried the remaining flood discharge. Therefore, after 1.7 days the discharge hydrograph represents the receding flood hydrograph for the northern channel.
 The confirmation of an elevated groundwater potentiometric surface, along with the abrupt truncation of Elaver Vallis and Allegheny Vallis at the rim of Ganges Chasma, reveals that this chasma could not have existed with its present size and depth when these channels formed. Otherwise, high groundwater pressures would have been relieved by breakouts from the walls or floor of this chasma, rather than by groundwater discharges high on the plateau.
 I thank Keith Harrison (SWRI) and an anonymous reviewer for their thoughtful comments that helped to substantially improve this manuscript. I also thank Cynthia Dinwiddie (SWRI) for contributing to early estimates of peak discharge for Elaver Vallis that we published in 2007.