3.1. Fan Hydraulics and Sedimentology
 Terrestrial alluvial fans form from sediments ranging in dominant grain size from mud to coarse gravel and by flows ranging from debris flows to normal fluvial transport, both as channelized and as sheet flows. Sediment size and flow type affect both the fan morphometry as well as surface features. In previous discussion we compared the Martian fan morphometry to their terrestrial counterparts. Inferences about flow processes and dominant grain size are hindered by postdepositional degradation of the Martian fan surfaces, lack of direct information on sediment grain size, and the lower gravity of Martian fans. The gravitational effects are particularly difficult to assess. Gravel bed streams on Earth typically have gradients that are close to the threshold of motion for the dominant bed grain size [e.g., Howard, 1980]. Scaling analysis suggests that gravitational effects on fluvial gradients at the threshold of motion should be minor [Pieri, 1980; Komar, 1980]. However, for equivalent discharge, channel dimensions, and gradient, sediment loads in sand-bed channels should be about 50 percent greater on Mars than on Earth. However, sediment loads supplied from headwater channels may be smaller for equivalent discharges because headwater sediment entrainment depends on flow shear stress [e.g., Howard, 1994], which will be lower on Mars. Both of these factors suggest sand-bed fan channels on Mars should have lower gradients than on Earth for equivalent source area relief. Debris flows may require steeper gradients on Mars than Earth in order for shear stresses to be large enough to exceed the yield strength of the debris flow slurry.
 For equivalent source basin size, the Martian alluvial fans are closer in gradient to the steep fans of Death Valley dominated by coarse gravels and cobbles than the fine-grained (sand and silt) fans of coastal California (Figure 12). Given that fine-bed alluvial channels might be gentler on Mars than Earth, we tentatively conclude that the Martian Fans are dominated by gravelly sediment. The low concavity of the Martian fans is also similar to that of the coarse-grained terrestrial alluvial fans of the Basin and Range province (Figure 17). In addition, for a given average gradient in the source basin (Figure 15) and for a given source basin relief (Figure 16), both the Martian fans and Basin and Range fans have steeper gradients than the Noachian degraded craters and terrestrial low-relief alluvial surfaces, which is also consistent with a supply of coarse sediment.
 The mode of sediment transport (debris flow versus fluvial) is less certain. Terrestrial debris flow fans are generally both smaller and steeper than fluvial fans (Figure 13). The Martian fans, being both relatively large and low in gradient, fall within the field to terrestrial fluvial fans (Figure 13). On terrestrial alluvial fans with mixed debris flow and fluvial sedimentation, the debris flows commonly are most prevalent on the upper portions of the fan and fluvial on the lower parts [Hooke, 1967], presumably because of the greater mobility of fluvial flows. The large size of the Martian fans requires flows capable of traversing tens of kilometers before depositing all sediment. Terrestrial fluvial fans generally display wide, multiple-branching distributaries, which are also apparent on Martian fan K (Figure 6). Taken together, the large size and low gradient of the Martian fans along with bedforms typical of fluvial deposition lead us to favor a fluvial origin for the fans of this study.
3.2. Implications for Paleoclimate
 The alluvial fans of this study are not features that could have formed during a single event, such as a catastrophic landslide. Their construction must have taken many years. To gain a sense of the minimum time to emplace a fan, we consider fan A1, which has a surface area of ∼500 km2. If we assume, after an inspection of the contour map of this feature (Figure 4d), an arbitrary but reasonable average thickness of this fan as ∼100 m, we get a volume of 50 km3. For this exercise we use a report of a 860,000 m3 deposit emplaced on an alluvial fan in the White Mountains of California during a single event derived from a catchment of 17 km2 [Beaty, 1970, 1990], ten times smaller than the catchment of fan A1. If we simply scale the White Mountains deposit by the catchment we have a value of 8.6 × 10−3 km3, which, if this amount of material was added every year to the construction of fan A1, it would take ∼5800 years to form this fan. Of course this ignores the real lapse time between successive deposits necessitated by the need to regenerate a new supply of loose detritus in the catchment susceptible to transportation by flash flood flushing, which in the White Mountains case results in a deposition event of the magnitude reported by Beaty [1970, 1990] recurring on average of every ∼320 years. Recurrence rates for Martian fan deposits is unknown, but the implications of the terrestrial example is that Martian fans probably cannot form in less than a millennia and might reasonably be expected to at least take more on order of 100 millennia. If the precipitation and runoff inducing climate were intermittent, the period of fan growth on Mars could be much longer.
 An extended period of fan development is also suggested by the history of fan deposition in Holden crater, where fan development both preceded and followed the time period of flows from Uzboi Valles through the crater. Although a single, geologically short-lived event of complicated history could be proposed to produce such a situation, similar terrestrial scenarios of fan evolution typically require millennia to millions of years.
 The feature of the Martian alluvial fans that most distinguishes them from terrestrial counterparts is their geographic restriction, both planet-wide and within craters. Fan-hosting craters have been found only within a narrow latitudinal belt, and only in three widely separated crater clusters within that belt (Figure 1). Within individual craters the fans almost universally originate from erosion of deeply incised alcoves in the crater walls. Most of the craters have alcove incision and fan deposition along at limited sites on the crater walls (e.g., Figures 4, 10, and 11), although a few support fans along half or more of the interior crater wall (e.g., Figure 5). Possible reasons for geographic isolation include climatic factors, variations in crater wall lithology, unique physiography of fan-producing crater walls, and local triggering mechanisms, such as effects of nearby impacts. We discount the latter two mechanisms as realistic causes. Although craters hosting fans are limited to steep, deep craters of late Noachian age, we have no evidence to suggest that such craters are restricted to the three planetary locations that we have found fans. Similarly, within craters the crater walls producing alcoves and fans appear not to be universally associated with particularly steep or high locations on the crater walls, although this may be a factor in fan location in Crater A (Figure 4). As noted above, the fans appear to have formed over an extended time period, so that it is unlikely that a local event such as a nearby impact or earthquake would have such long-lasting effects on fan formation.
 Variations in lithology are a possible contributor to the clustered pattern of fan development. Bolides impacting, for example, onto the margins of preexisting basins might have significant circumferential variations in wall lithology ranging, perhaps from fractured igneous rocks to loose sediment. One suggestive situation is the development of isolated fans on the two sides of the crater wall separating two impact basins (Fans E1 and E2 in Figure 11).
 We view climatic factors as a potentially strong control on the geography of fan development. Simulations of precipitation on Mars using global climate models show strong geographic control of location [Colaprete et al., 2004]. The fan cluster at 30°W is located on the divide between Argyre and the eastern Valles Marineris chaos and outflow channels, including the Uzboi channel system that was active during the time period of fan development. The cluster at 290°W is on the flanks of the Hellas basin that may have hosted a deep, ice-covered lake [Moore and Wilhelms, 2001], and the central cluster at 335°W is at a topographic highpoint of the cratered highlands. The restriction of fan source areas to alcoves on the upper crater walls and the lack of apparent precipitation and fluvial incision on the fan surfaces may reflect microclimatic and orographic controls. Precipitation in mountainous terrain is strongly concentrated on local highs (e.g., mountain peaks in the Basin and Range province typically receive >400 mm/yr, whereas basins often receive less than 100 mm/yr [Prudic et al., 1995; Harrill and Prudic, 1998]). Thus upper crater walls should be favored for precipitation either as rain or snow. However, we find no evidence for universal association of fans with particular azimuths on the crater walls, which would be a strong indication of climatic control.
 The restriction of erosion and, possibly, runoff production, to specific alcoves on the crater walls may also reflect a positive feedback between alcove formation and local microclimate and hydrology. Steep basins enhance local mountain winds: updrafts in the afternoon and downvalley winds at night. These might interact with precipitation. In addition, for equivalent precipitation or snowmelt, runoff might be enhanced in the alcoves due to steep topography and exposed bedrock. Deeply incised basins are also sheltered from sunlight during low sun incident angles (e.g., winter and times of high obliquity), allowing the alcoves to cold-trap thick snow covers.
 The localized nature of the alcove erosion responsible for fans is possibly suggestive of erosion due to preferential groundwater emergence within the alcoves. Such a source of runoff has been suggested for the smaller and more recent gully systems on Mars [Malin and Edgett, 2000b]. However, the physiography of the alcoves and fans is not supportive of groundwater sources. The rims of the relatively undegraded craters supporting fans, being local topographic highs, are unlikely sources for large quantities of groundwater. Several of the fans extend from alcoves eroded into the septa rims between adjacent impacts (Fans M1, E1 and E2 in Figure 11). In particular, the alcoves supplying Fans E1 and E2 abut against each other at a narrow divide. It is also questionable that groundwater could supply discharge at a rate sufficient to transport coarse sediment and create sediment-transporting flows that would extend across tens of kilometers of fan surface.
 The lack of fan head trenching (exclusive of the fans in Holden), the fairly constant shapes and gradients, and the absence of changes in deposition centers on the fans of this study indicate that the last episodes of deposition occurred under a hydrological regime that was similar to that of its immediate predecessors. In other words, there is no evidence for a gradual decline in the final hydrological regime. By contrast, many if not most terrestrial fan systems show fan head trenching and translocation of deposition to the toe of the fans in response to changes from glacial to interglacial climate [e.g., Bull, 1991]. There is no further fluvial modification of fan surfaces of even a modest scale. These fans formed in a climate that very abruptly ended at least with respect to its ability to generate precipitation and runoff, something that is not seen on Earth. Equally noteworthy is the absence of evidence for antecedent fans of the slope, size and isolation of those of this study. Sediment deposits within mid-Noachian degraded craters are gentler (Figures 15 and 16), more concave (Figure 12), and derive from widespread dissection of the crater walls rather than incision of localized alcoves, resulting in planar rather than fan-shaped deposits. Indeed, some of the fans of this study have feeder valleys that incise preexisting smooth and presumably sediment-filled basins as they travel from their catchments (e.g., Bakhuysen, Holden), which otherwise exhibit no nonfan dissection. Taken together, this may imply that the climate preceding the era of the study fans was also not conducive to generation of steep, large, and isolated alluvial fans.
 A climate that suddenly stops supporting fan formation and may have just as suddenly commenced seems unlikely to be the consequence of a gradual decline in Mars' ability to support an atmosphere-surface hydrologic cycle [e.g., Pollack et al., 1987; Squyres and Kasting, 1994]. Perhaps better candidate climates are those ushered in by “cataclysmic” events that induce excursions from some steady state. If Mars had already evolved to a “steady state” climate that disfavored a hydrologic cycle conducive to the formation of this study's fans by the time we speculate that they were formed (at approximately the Noachian-Hesperian boundary), potential “cataclysmic” events that might induce sudden climate perturbations that have been dated to this time are outflow channels [e.g., Grant, 1987, 2000], large-scale volcanic eruptions [e.g., Scott and Carr, 1978; Tanaka, 1986], and large impacts [e.g., Tanaka, 1986; Hartmann and Neukum, 2001]. The ability of outflow channel flooding to induce a period of precipitation and runoff was called into question by Moore et al. . Several studies have proposed that the release of volatiles by large-scale volcanic events can bring on a hydrologic-cycle-conducive environment; however, these events are usually ascribed to a time well prior to the Hesperian [e.g., Phillips et al., 2001].
 Carr  and, more recently, Segura et al.  have proposed that large impacts could induce a period of precipitation (and/or ground ice melt) and runoff. Even if individual impact-induced precipitation and runoff episodes do not persist long enough to form the fans of this study, the accumulation of the effects of many such events could. We searched for evidence (i.e., partially buried craters) of long (∼106 year scale) hiatuses in fan growth but saw none. This, however, does not mean that hiatuses did not occur, as the last episode of fan deposits in combination with subsequent mantling could easily mask any such evidence. Also, the Segura et al.  hypothesis, as it was originally presented, required impact events much larger than those that we have evidence took place at the Noachian-Hesperian transition. Recent modeling by this group [e.g., Colaprete et al., 2004], however, indicates that impacts in the range of those seen to have taken place during the time of fan formation could produce several years to several decades of precipitation and runoff over a regional area. So, while the impact-induced climate change hypothesis looks promising, it does not explain why there was a long hiatus prior to the era of fan formation, as large impacts occurred throughout the Noachian.