Erosional features on a plateau in the Valles Marineris provide evidence that a lake filled Capri Chasma until it overflowed its eastern rim, carving two large spillover channels. The floodwaters surged into the adjacent lowlands of ancestral Ganges Chasma and eastern Eos Chasma. The channel floor elevation and depth of incision at Daga Vallis reveal that a 1200-m-deep lake water column was drained by the flooding. The width, depth, and steep energy slope (∼0.02) of Daga Vallis and the existence of several cataracts, including a 500-m-high dry falls, demonstrate the power of the floodwaters. We estimate a possible peak discharge rate of 1–6 × 108 m3·s−1. The catastrophic flows may have been triggered by the collapse of topographic barriers in eastern Coprates Chasma, providing a gateway for lake waters in the central Valles Marineris to pour eastward toward Capri Chasma and the lowlands beyond. These may have been among the earliest flows in Simud-Tiu Valles, contributing discharge to a possible sea in the northern plains of Mars.
 The outflow channels that emptied into the lowlands of Chryse Planitia provide important evidence that great quantities of water once flowed on the Martian surface (Figure 1). Some channels were created when the cryosphere ruptured and groundwater discharged from chaotic terrain or from pits that formed along major fault zones. Other circum-Chryse channels were carved by floods that issued directly from the Valles Marineris, which today is a system of canyons more than 3000 km long and up to 8 km deep. The ancestral canyons were probably smaller and more isolated. The channels of Simud Vallis and Tiu Vallis begin at the terminus of this longest chain of interconnected canyons on Mars. This drainage system begins in the Tharsis volcanic province, in the canyons of Noctis Labyrinthus, and extends eastward through the Valles Marineris canyons and beyond to the chaotic lowlands of Aurorae Chaos and Hydraotes Chaos (Figure 1). Simud and Tiu Valles begin north of Hydraotes Chaos and persist northward into the Chryse Planitia region bounded by latitude 24–30°N and longitude 28–38°W. These channels are 150 to 200 m deep and have floor elevations between −3950 and −4030 m; their distal elevations in Chryse Planitia provide insight about a possible northern ocean. The floor elevation range is 190–270 m below the Contact 2 ocean shoreline proposed by Ivanov and Head [2001, Plate 3] at an elevation of −3760 m, which suggests that there was no water body in the northern plains with a surface elevation higher than −4 km at the time of the initial flows in these distant channels. Subsequent flows in Simud-Tiu Valles may have created or enlarged a sea that rose above this elevation. Our interpretation is consistent with that of other researchers [Carr and Head, 2003] who likewise found no support for higher ocean shorelines. They concluded that the Vastitas Borealis Formation, which fills the polar basin north of Chryse Planitia [Kreslavsky and Head, 2002, Figure 2], may consist of sublimation residue from ponded floodwater effluents. Carr and Head  report an average elevation for the Vastitas Borealis Formation of −3903 ± 393 m. There is no topographic barrier between Chryse Planitia and the North Polar basin. Therefore, the disappearance of the distal Simud-Tiu channels could be explained by the presence of a former, probably at least partially frozen sea with a surface in this elevation range.
 Catastrophic drainage of lakes following the collapse of topographic barriers or ice-debris dams provides a plausible mechanism for the release of water from the Valles Marineris canyons, and for the formation of Simud and Tiu Valles. Thick sequences of layered deposits in many parts of the canyons may have formed within lakes [Carr, 1996]. Other evidence of Hesperian-aged canyon lakes has largely been obscured since the time of the floods by continued tectonic growth of the canyons, landslides, cratering events, possible volcanic activity on the canyon floors, and the creation of expanses of chaotic terrain in the eastern canyons [Lucchitta et al., 1994; Schultz, 1998; Chapman and Tanaka, 2001]. If spillover channels could be found on plateaus between canyons, that would provide evidence that lakes were not only present but actually overflowed the canyon rims.
2. Daga and Columbia Valles
 In the eastern Valles Marineris, at the transition between canyons and chaotic lowlands, two large spillover channels known as Columbia Valles and Daga Vallis exist on the eastern margin of Capri Chasma (Figures 1 and 2) . At first glance these channels may appear to be relict hanging valleys carved by floods prior to canyon development. Closer inspection, however, reveals that the channels are hanging only at their downstream termini. Columbia Valles abruptly ends at an escarpment 2500 m above the floor of Ganges Chasma, and Daga Vallis is truncated by the rim of Eos Chasma, 3000 m above the canyon floor. The channel floors have relatively minor offsets at their source areas in eastern Capri Chasma, indicating that the eastern part of this chasma had already largely formed when the channels were born. Energy slopes for the channel floors reveal that flow was away from (not into) Capri Chasma. But our estimates for peak discharge in Daga Vallis provide the strongest evidence that the channels postdate Capri. Discharge rates >108 m3 s−1 (see Section 3) far exceed those expected in ancient valley networks and also exceed realistic estimates for most outflow channels. Discharges of this magnitude would be entirely consistent with catastrophic releases from canyon lakes.
 The upstream margins of Daga and Columbia Valles display similar high-water marks representing overflow elevations. We determined the elevations of eroded channel margins using MOLA altimetry data and THEMIS and MOC images to estimate the range of stage elevation and width of the initial floodwaters. The channel margin for Columbia Valles' overflow to Ganges Chasma (Figure 3) has an elevation of approximately 1150 m (10.124°S, 42.982°W). The Daga Vallis spillover toward eastern Eos Chasma (Figure 4) has a channel margin elevation of 1150 m on the rim of Eos Mensa (11.720°S, 42.700°W). Thus, the elevation of the paleolake surface in Capri Chasma was approximately 1150 m. This is a minimum elevation because continued subsidence in the Valles Marineris region may have somewhat lowered the terrain over billions of years.
 We also determined the highest points on the upstream floors of Daga and Columbia Valles. These elevations represent the paleolake surface elevations when flow in each channel ceased. The overflow “saddle” on the floor of western Daga Vallis has an elevation of −65 m (at 11.935°S, 42.926°W). Subtracting this elevation from the initial flood level (1150 m) reveals that a 1215-m-deep lake drained through the spillover channels. The overflow for Columbia Valles has an elevation of 413 m (at 10.066°S, 43.390°W). The Daga overflow is approximately 480 m deeper, indicating that flow and downcutting persisted in Daga Vallis longer than in Columbia Valles.
 We examined the morphology of Daga Vallis (Figure 4) and performed hydrologic analyses for several flood stages. The channel is ∼22 km wide where it exits Capri Chasma (Figure 5). The channel width increases downstream to >33 km where the terrain was scoured by the early, wide, and relatively shallow scabland flooding. As flooding progressed, a deeper channel system formed in the southern half of the scabland, which captured all of the flow and terminated overland flow in scoured terrain to the north.
 Three parallel channels lie within the main valley. The northern channel is a hanging valley, truncated by continuing flow and erosion in the central, deepest channel. The overflow level on the upstream floor of this channel has an elevation of 224 m (11.781°S, 42.772°W), which means that flow in the northern channel ceased when the lake surface elevation fell below 224 m. The southernmost channel (Figure 4) appears in THEMIS images to be a continuous valley from northwest to southeast. Gridded MOLA data suggest this valley is interrupted by high ground at the eastern end. However, the geometrically precise curve in the MOLA topography at 12.36°S, 42.05°W reveals the high ground to be a false artifact caused by smoothing of elevations in an area without MOLA tracks. The central channel leads down into a basin that terminates the channel system on the southeastern end. The channel floors display features characteristic of differential hydrodynamic erosion. Multiple terraces fringe the channel margins, and the floors are veneered with longitudinal ridges and streamlined mesas. Longitudinal ridges are most prominent on raised portions of the channel floor, which suggests they were eroded in a higher energy regime than the final ebbing flows that produced the smoother central channel. The eastern part of the central channel has a relatively flat energy slope.
 Daga Vallis also has several dry falls along its course, including a spectacular 500-m-high dry falls cataract near its terminus (Figures 4 and 6) . Below the dry falls is a 25-km-wide depression that likely formed through a combination of plunge pool erosion beneath the falls and large-scale genesis of chaotic terrain in the northern part of the depression. Some scattered knobs appear on the floor of this basin. MOLA track ap10672L crosses the basin from north to south and shows that elevation varies only 38 m over 18 km. This relatively smooth floor may have been formed by waning-stage fluvial deposition of sediments and post-fluvial aeolian activity. The basin was formed during and likely triggered by the Daga Vallis flooding [cf. Coleman, 2005]. The northwest margin of the basin truncates the scabland that was eroded by the earliest flooding (Figures 4 and 6), but the central channel that was carved by the final Daga flows is not offset where it intersects the basin. Streamlined mesas exist east, west, and south of the basin. The orientation of the dry falls and these mesas further shows that flow was eastward (away from) Capri Chasma.
 The presence of two large spillover channels (Daga and Columbia Valles) 140 km apart at essentially the same elevation provides evidence that the channels formed at the same time and that a massive influx of water had entered and filled ancestral Capri Chasma. If the lake had filled slowly, the eventual overtopping of the canyon rim would have been more likely to erode a single drainage channel because fluvial erosion would have begun and concentrated at one location. It is an interesting historical point that the presence of multiple spillways at equal elevations provided key support for J Harlen Bretz in the great “Channeled Scabland” debate [Baker, 1981].
 If our interpretation is correct, the valley that forms easternmost Eos Chasma (Figure 1) did not exist during these floods, because if it had existed, the flooding would have proceeded through that valley instead of carving new channels at the rim of Capri Chasma. It is also possible that a topographic barrier existed between the ancestral Capri and Eos Chasmata at the time of the flooding. The termini of both Daga and Columbia Valles are hanging valleys, indicating that post-fluvial canyon growth continued where the channels intersected Ganges Chasma (Figure 3) and eastern Eos Chasma (Figure 6).
 What were the sources of the floodwaters that filled and overtopped Capri Chasma? This canyon and western Eos Chasma to the south both contain chaotic terrain that may have been formed by groundwater breakouts. But we do not know the ancestral size of these canyons at the time of the flooding or whether chaotic terrain had formed within them at that time. We favor an alternative explanation for the overtopping flows—that a topographic barrier between ancestral Capri Chasma and Coprates Chasma (Figure 1) may have been breached, catastrophically draining canyon lakes to the west in Coprates Chasma and possibly in other interconnected canyons (i.e., Melas, Candor, and Ophir Chasmata). Over time, multiple floods may have issued from this region. Several locations in eastern Coprates Chasma are relatively narrow today (Figure 1) and may indicate where floods could have been triggered by the collapse of canyon walls, releasing lakewaters that had been ponded in interconnected canyons. Indeed, if a new gateway in eastern Coprates Chasma was the primary source for the flooding that issued from Capri Chasma, the hydrographs of those floods may have resembled those hypothesized for the Missoula paleofloods, in which floodwaters rose slowly to a crest, peaked, and then fell rapidly. Baker  reported field evidence that the Missoula floods receded rapidly because bed forms such as giant current ripples were not washed out during a prolonged recession of high-energy floodwater. Therefore, unlike outflow channels sourced by groundwater (e.g., Ravi, Mangala, Allegheny, Walla Walla, and Athabasca Valles), those at the eastern end of Capri Chasma may have briefly carried flows of great depth in the paleo-channels we observe today.
3. Hydrologic Calculations
 We evaluate the flow speed and discharge through Daga Vallis to eastern Eos Chasma under several different scenarios to illustrate the possible magnitude of this flooding (Figure 5). We made independent estimates using the methods of Komar , Kleinhans , and Wilson et al. . Our calculations assume quasi-uniform flow conditions were present for at least part of the time under each hydrologic scenario. We first analyzed the early overland flooding (wide but relatively shallow) prior to channel incision, with a flood stage at the canyon overflow elevation of 1150 m. We then estimated peak flow in Daga Vallis assuming the channel system we now observe carried flooding to approximately half its present depth. We agree with Wilson et al.  that the magnitude of paleoflooding in most Martian channels would likely be overestimated using analyses that assume bank-full flow in the channels observed today. Finally, to evaluate the possible magnitude of late-stage flows, we also analyzed flow in the central (deepest) channel. Results of our calculations are shown in Table 1.
Table 1. Hydrologic Flow Calculations for Daga Vallis, Mars
Except where indicated otherwise, flow velocities were calculated using the approach of Komar  and a range of terrestrial Manning n values from 0.04 to 0.06. Values of the Chézy friction coefficient (Cf) were obtained using these n values and the estimated range of flow depths [Carr, 1996]. For example, given a flow depth of 30 m and n = 0.05, Cf = 0.008.
Gradient estimated using present-day terrain slope on Eos Mensa parallel to and south of Daga Vallis (139 m/42,000 m = 0.0033).
Gradient estimated using present-day thalweg slope in western half of Daga Vallis (410 m/19,500 m = 0.021).
Velocity estimated using the Darcy-Weisbach equation with friction factor f = 0.40 derived with equation (13) of Kleinhans .
Velocity estimated using the Darcy-Weisbach equation with friction factor f = 0.024 derived from equation 10 of Wilson et al.  for boulder bed channels.
Gradient estimated using 30-km-long thalveg segment centered on profile B-B' (470 m/30,000 m = 0.016).
Velocity estimated using the Darcy-Weisbach equation with friction factor f = 0.29 derived from equation (13) of Kleinhans .
Velocity estimated using the Darcy-Weisbach equation with friction factor f = 0.039 derived from equation (10) of Wilson et al.  for boulder bed channels.
Earliest overland flow prior to channel incision (stage = 1150 m, Profile A-A')
 We estimate initial scabland flows in the range of 4–35 × 106 m3·s−1 using flow depths of 30–90 m. Peak flows with the present-day channel system flowing half full may have attained a discharge rate of 4–6 × 108 m3·s−1 given the steep thalweg gradient of 0.021. Similar results are obtained with the Darcy-Weisbach equation with a friction factor derived for boulder-dominated channels [Wilson et al., 2004, equation (10)], and assuming that 84 percent of channel bed clasts are smaller than 0.3 to 1.7 m. This clast size distribution is consistent with our view that beds of deeply incised channels on Mars likely consist of eroded basalt bedrock. The approach recommended by Kleinhans  yields significantly slower velocities and smaller discharges for both peak flows and late-stage flows than either of the other two methods (Table 1). We estimate late-stage flow in the deepest channel by presuming a maximum flow depth of 100 m (mean depth = 70 m). Using various methods (Table 1) we estimate flows in the range of 3–9 × 106 m3·s−1.
 Our interpretation of a former lake in Capri Chasma and associated spillover channels is consistent with geochemical evidence for past aqueous activity in the Valles Marineris canyons. Gendrin et al.  present several interpretations for the origin of abundant sulfates in the canyons (including Capri Chasma) and associated chaotic terrain, including alteration of mafic minerals by acidic precipitation, evaporation of standing water bodies, or by the seepage of hydrothermal brines. The sulfates exist over a wide range of elevations, which suggests that sulfate sediments may have formed in shallow water bodies during the growth of the canyons.
 The discharges in Daga and Columbia Valles may have been among the earliest flows in Simud-Tiu Valles, contributing discharge to a possible sea in the northern plains of Mars. The present-day channel complex of Simud-Tiu Valles was carved by the last large flows that discharged into Chryse Planitia. This is logical in the context of regional hydrology because the catastrophic draining of lakes in the main canyons of Valles Marineris would have rapidly and permanently lowered the groundwater elevations in the region east of Tharsis, thus inhibiting the recurrence of large outflows in other regional channels such as Maja, Kasei, Shalbatana, and Ravi Valles. The rapid draining of deep canyon lakes would have created large groundwater overpressures beneath canyon walls and floors, leading to instability and potentially triggering landslides and chaos formation with secondary groundwater breakouts. Fluvial incision of channels east of Capri Chasma and Eos Chasma may have penetrated the cryosphere, triggering additional chaos formation and groundwater release. This may explain why we see extensive chaotic terrain in the eastern Valles Marineris lowlands (e.g., Aurorae and Hydraotes Chaos).
 We appreciate the thoughtful reviews by Lionel Wilson (Lancaster University), Keith Harrison, Stuart Stothoff, and Gary Walter (all with Southwest Research Institute). This article was prepared by an employee of the U. S. NRC on his own time apart from regular duties.