The gradual discovery that late Neoproterozoic ice sheets extended to sea level near the equator poses a palaeoenvironmental conundrum. Was the Earth's orbital obliquity > 60° (making the tropics colder than the poles) for 4.0 billion years following the lunar-forming impact, or did climate cool globally for some reason to the point at which runaway ice-albedo feedback created a `snowball' Earth? The high-obliquity hypothesis does not account for major features of the Neoproterozoic glacial record such as the abrupt onsets and terminations of discrete glacial events, their close association with large (> 10‰) negative δ13C shifts in seawater proxies, the deposition of strange carbonate layers (`cap carbonates') globally during post-glacial sea-level rise, and the return of large sedimentary iron formations, after a 1.1 billion year hiatus, exclusively during glacial events. A snowball event, on the other hand, should begin and end abruptly, particularly at lower latitudes. It should last for millions of years, because outgassing must amass an intense greenhouse in order to overcome the ice albedo. A largely ice-covered ocean should become anoxic and reduced iron should be widely transported in solution and precipitated as iron formation wherever oxygenic photosynthesis occurred, or upon deglaciation. The intense greenhouse ensures a transient post-glacial regime of enhanced carbonate and silicate weathering, which should drive a flux of alkalinity that could quantitatively account for the world-wide occurrence of cap carbonates. The resulting high rates of carbonate sedimentation, coupled with the kinetic isotope effect of transferring the CO2 burden to the ocean, should drive down the δ13C of seawater, as is observed. If cap carbonates are the `smoke' of a snowball Earth, what was the `gun'? In proposing the original Neoproterozoic snowball Earth hypothesis, Joe Kirschvink postulated that an unusual preponderance of land masses in the middle and low latitudes, consistent with palaeomagnetic evidence, set the stage for snowball events by raising the planetary albedo. Others had pointed out that silicate weathering would most likely be enhanced if many continents were in the tropics, resulting in lower atmospheric CO2 and a colder climate. Negative δ13C shifts of 10–20‰ precede glaciation in many regions, giving rise to speculation that the climate was destabilized by a growing dependency on greenhouse methane, stemming ultimately from the same unusual continental distribution. Given the existing palaeomagnetic, geochemical and geological evidence for late Neoproterozoic climatic shocks without parallel in the Phanerozoic, it seems inevitable that the history of life was impacted, perhaps profoundly so.
In 1987, the Caltech biomagnetist and palaeomagnetist Joe Kirschvink gave undergraduate Dawn Sumner a rock sample to study for her senior thesis. The sample, collected by UCLA palaeontologist Bruce Runnegar, was a reddish, uncompacted, rhythmically laminated siltstone from the Elatina Formation, a late Neoproterozoic, glacial and periglacial unit widely exposed in the Flinders Ranges and elsewhere in South Australia (Preiss, 1987; Lemon and Gostin, 1990). The rhythmic laminations are interpreted to be lunar tidal bundles (Williams, 2000), implying a shallow marine depositional environment. Glacigenic deposits (diamictites and ice-rafted dropstones) occur in most sections of the Elatina Formation (Lemon and Gostin, 1990), but previous palaeomagnetic studies suggested that the siltstone was deposited close to the equator on the basis of unusually stable remnant magnetization carried by detrital haematite (Embleton and Williams, 1986). Kirschvink was skeptical that glaciers would ever reach sea level in the tropics, so he instructed Sumner to perform a fold test (McElhinny and McFadden, 2000) on soft-sediment folds (Williams, 1996) in Runnegar's sample. A positive fold test would prove that the natural remnant magnetization (NRM) is primary; a negative result would show that it is secondary, posing no constraint on the palaeolatitude of the Elatina glaciation. To Kirschvink's surprise, the fold test was positive (Sumner et al., 1987), as others subsequently confirmed (Schmidt et al., 1991; Schmidt and Williams, 1995; Sohl et al., 1999). A stratigraphically consistent, polarity reversal test (Sohl et al., 1999; see also Schmidt and Williams, 1995) confirmed the primary component of NRM in the Elatina Formation, while the existence of multiple reversals (Sohl et al., 1999) suggests that the Elatina glacial epoch lasted for several 105 to a few 106 years.
The Elatina results refocused attention on Neoproterozoic glaciations. A critical review of the stratigraphic, geochronological and palaeomagnetic constraints on virtually all late Neoproterozoic glacial deposits (LNGD) world-wide was recently published (Evans, 2000). A total of 16 regional-scale units, known to have formed near sea level, possess primary or near-primary NRM components giving at least `somewhat reliable' palaeolatitudes (Evans, 2000). Many were apparently deposited within 10° of the equator, and none was laid down at a palaeolatitude greater than 60° (Fig. 1). Increased non-dipole components in the Proterozoic geomagnetic field (Kent and Smethurst, 1998; Bloxham, 2000) would not greatly affect those conclusions (Evans, 2000) (Fig. 1). The observations are surprising and they argue that the Elatina result is no fluke.
The glacial origin of the Elatina Formation was first recognized by Sir Douglas Mawson – the older, thicker and more localized, Surtian LNGD in the same region were recognized much earlier by Howchin (1908), following the first described LNGD in northern Norway (Reusch, 1891) – and he discovered its cap dolostone (Mawson, 1949b). Although conservative by nature (Sprigg, 1990), Mawson was the first (to our knowledge) to argue that late Neoproterozoic glaciation was global, with large ice sheets in the tropics (Mawson, 1949a). He went on to suggest that climatic amelioration paved the way for the first metazoa (Mawson, 1949a), which had been discovered by a Mawson protegé, Reg Sprigg, in the Ediacara Hills west of the Flinders Ranges (Sprigg, 1947). Ironically (for an Adelaide resident), Mawson was opposed to continental drift, and his argument for tropical ice sheets depended critically on the occurrence of LNGD in tropical Africa today (Mawson, 1949a). [To his credit, Mawson urged students to read his contemporary's book (Wegener, 1922) and Sprigg, for one, identified the Adelaidean succession as a pre-Mesozoic rifted continental margin (Sprigg, 1952).] Mawson's argument, however, collapsed in the plate tectonic revolution, with most subsequent workers attributing the extent of LNGD (Fig. 1) to rapid drift of different continents through polar regions at different times (Crawford and Daily, 1971; McElhinny et al., 1974; Crowell, 1983, 1999; Eyles, 1993).
Kirschvink was not the first to wrestle with the low-latitude Elatina results. George Williams (1975, 1993, 2000) had long advocated a large (> 54°) orbital obliquity, or tilt (the angle between the Earth's axes of rotation and orbit around the sun), to account for low-latitude glaciation in conjunction with inferred strong equatorial seasonality during the Elatina glaciation (Williams and Tonkin, 1985; Williams, 1998). In theory, obliquity could have varied chaotically within 60–90° following the early lunar-forming impact event, but a low obliquity, once achieved, is self-stabilizing through the gravitational interaction of the Moon and the Earth's equatorial bulge (Laskar et al., 1993). Williams (1993) postulates a sharp decrease in orbital obliquity after 600 Ma, with resultant moderation of the seasonal climate cycle giving rise to the metazoa (Williams, 1993). With an obliquity > 54°, mean annual temperatures are lower around the equator than at the poles (Williams, 1975; Hunt, 1982; Oglesby and Ogg, 1999; Jenkins, 2000), making glaciation more probable at lower latitudes (Williams, 1975). Summer insolation is so large in the polar regions (the sun staying high in the sky throughout the diurnal cycle) that surface temperatures over land areas might exceed the boiling point of water (Oglesby and Ogg, 1999). The tropics, although cooler overall, would have two hot seasons near equinox, when insolation would be indistinguishable from the hottest tropical seasons with low obliquity (Oglesby and Ogg, 1999). Indeed, hot summers and strong seasonality make glaciation problematic at any latitude with large obliquities (Hoffman and Maloof, 1999). It does, however, provide a ready but non-unique (Maloof et al., 2002) explanation for the existence of well-developed, periglacial, sand-wedge polygons in the Elatina Formation (Williams and Tonkin, 1985). These structures are most easily explained as resulting from large seasonal oscillations in soil temperature (Lachenbruch, 1962), which should not occur near the equator if the obliquity was close to its present value of 23.5°. On the other hand, a persistent large obliquity does not explain the characteristically abrupt onsets and terminations of LNGD (Hambrey and Harland, 1981; Preiss, 1987; Narbonne and Aitken, 1995; Hoffmann and Prave, 1996; Hoffman et al., 1998a). Nor does it account for the close association of LNGD with carbonates, because the `redistribution of the radiant energy balance to polar latitudes should also move the carbonate belts from equatorial latitudes to the poles, where the glaciers (in Williams' model) should not encounter them' (Kirschvink, 1992). The dynamics of the postulated obliquity reduction after 600 Ma remain conjectural (Williams, 1993; Ito et al., 1995; Néron de Surgy and Laskar, 1997; Williams et al., 1998; Pais et al., 1999).
Richard Sheldon (1984) postulated that orbiting ice rings episodically collapsed into the atmosphere, transiently shielding sunlight and giving rise to glaciers at low latitudes. The theory strives to explain iron formations and cap carbonates, but does not adequately account for low-latitude glaciation. Due to orbital obliquity, equatorial ice rings will cast shadows only on the winter hemisphere (see figure 2 in Sheldon, 1984). Glaciation depends critically on cool summers, not cold winters (Köppen and Wegener, 1924), as it is contingent on a fraction of the annual snowfall surviving the melting seasons.
Kirschvink responded to the positive Elatina fold test with a theory he first aired in 1989 (Maugh, 1989). It eventually appeared as an article, seven paragraphs long, embedded in a 1348-page book (Kirschvink, 1992). It postulated that conditions amenable to global glaciation were set up by an unusual preponderance of land masses within middle to low latitudes, a situation that has not been encountered at any subsequent time in Earth history. One effect of such a palaeogeography would be a substantially higher albedo in the subtropics, where clouds are least important (Kirschvink, 1992). Any resultant glaciation would further increase the Earth's albedo by lowering sea level, exposing continental shelves and inland seas (Kirschvink, 1992). Placing more continents in the tropics would also increase the silicate weathering rate (provided tectonic uplift was not mysteriously absent there), leading to a colder planet with lower atmospheric pCO2 (Marshall et al., 1988; Worsley and Kidder, 1991). In addition, the meridional heat transport by the Hadley cells would be weakened because tropical air masses would be drier on average with greater continentality. Oceanic heat transport might also be lessened, but this is less certain. These combined effects might lead to the growth of large ice caps, nucleated on islands or continents bordering the polar seas.
Assuming an albedo runaway did occur, the climate would be dominated by the dry atmosphere and the low heat capacity of the solid surface (Walker, 2001). It would be more like Mars (Leovy, 2001) than Earth as we know it, except that the greater atmospheric pressure would allow surface meltwater to exist. Diurnal and seasonal temperature oscillations would be strongly amplified at all latitudes because of weak lateral heat transfer and extreme `continentality' (Walker, 2001). Despite mean annual temperatures well below freezing everywhere, afternoon temperatures in the summer hemisphere would reach the melting point (Walker, 2001). Evaporation of transient melt water would contribute, along with sublimation, to maintain low levels of atmospheric water vapour, and glaciers would feed on daily updrafts of this moisture (Walker, 2001). The global mean thickness of sea ice depends strongly on sea-ice albedo (˜1.4 km for albedo 0.6) and meridional variability is a complex function of solar incidence, greenhouse forcing (see below), zonal albedo, ablation or precipitation, and equatorward flowage of warm basal ice (Warren et al., 2002).
Climate physicists originally assumed that no ice-albedo catastrophe ever actually occurred because it would be permanent: the high planetary albedo would be irreversible. A saviour exists, however, and Kirschvink (1992) identified it as the buildup of an intense atmospheric CO2 greenhouse through the action of plate tectonics in driving the long-term carbon cycle (Walker et al., 1981; Caldeira and Kasting, 1992; Kirschvink, 1992). On a snowball Earth, volcanoes would continue to pump CO2 into the atmosphere (and ocean), but the sinks for CO2– silicate weathering and photosynthesis – would be largely eliminated (Kirschvink, 1992). Even if CO2 ice precipitated at the poles in winter, it would likely sublimate away again in summer (Walker, 2001). CO2 levels would inexorably rise and surface temperatures would follow, most rapidly at first and more slowly later on (Fig. 7) due to the non-linear relation between CO2 concentration and the resultant greenhouse forcing (Caldeira and Kasting, 1992). With rising surface temperatures, sea ice thins but ground ice sheets expand in some areas due to a stronger hydrological cycle. If CO2 outgassing rates were broadly similar to today (a reasonable assumption for 600–700 Ma), then the time needed to build up the estimated 0.12 bar CO2 required to begin permanent melting at the equator, assuming a planetary albedo of 0.6, would be a few million years (Caldeira and Kasting, 1992; Crowley et al., 2001). This estimate (Fig. 7), while subject to large uncertainties, is of the same order as the estimated time-scale of LNGD from palaeomagnetic (Sohl et al., 1999) and stratigraphic (Hoffman et al., 1998a) evidence. Once the tropical ocean begins to open up perennially, deglaciation proceeds rapidly due to reverse ice-albedo feedback (Caldeira and Kasting, 1992; Crowley et al., 2001). The fall in planetary albedo occurs far faster than the excess atmospheric CO2 can be consumed by silicate weathering, with the result that a transient heat wave must follow the ice retreat (Caldeira and Kasting, 1992; Kirschvink, 1992). Kirschvink (1992) concluded that a cold planet with many tropical continents and large polar sea-ice caps `would be a rather unstable situation, with the potential for fluctuating rapidly between the “ice house” and “greenhouse” states'.
Testing the snowball hypothesis
Kirschvink (1992) highlighted three implications of his theory which `might lend themselves to geological tests'. First, `glacial units should be more or less synchronous Harland, 1964)'. Second, `strata from widely separate areas which preserve a record of these [rapid] climatic fluctuations [between “ice house” and “greenhouse” states] might bear an overall similarity in lithologic character, which would be a result of the global scale of the climatic fluctuations'. Third, `the presence of floating pack ice should reduce evaporation, act to decouple oceanic currents from wind patterns and, by inhibiting oceanic to atmospheric exchange of O2, would enable the oceanic bottom waters to stagnate and become anoxic. Over time, ferrous iron generated at the mid-oceanic ridges or leached from the bottom sediments would build up in solution and, when circulation became reestablished toward the end of the glacial period, the iron would oxidize to form a “last-gasp” blanket of banded iron formation in upwelling areas.' Here is how these tests turned out.
Test 1: carbon isotopes
Direct radiometric testing of predicted glacial synchroneity has not proved possible (Table 1) because of an apparent dearth of suitable material for dating (Evans, 2000). Low diversity and turnover rates of microfloral and protistan remains pose severe limits on biostratigraphic resolution until after the major Neoproterozoic glacial events (Knoll, 2000). Regardless, a sea change in favour of synchronous glaciation occurred in the 1990s, and carbon isotope stratigraphy was principally responsible.
Table 1. Radiometric age constraints on Neoproterozoic glacial events
The history of carbon isotopic variation (expressed as δ13C, the difference in per mil of 13C/12C with respect to standard VPDB) in marine carbonates and covarying organic matter is a robust, but ambiguous, record of biogeochemical cycling (Summons & Hayes, 1992). Despite the fact that significant spatial heterogeneity in δ13C is dynamically maintained within the ocean (Kroopnick, 1985), large secular changes in δ13C at individual sites are globally correlated (Veizer et al., 1999; Saltzman et al., 2000), reflecting the residence time of carbon ∼ 150 times the ocean mixing time (Kump and Arthur, 1999). The δ13C of carbonate rocks is relatively immune from large diagenetic changes because pore fluids are effectively buffered by the large rock carbon reservoir (Veizer et al., 1999).
The conventional explanation for Neoproterozoic carbon isotopic oscillations and cap carbonates is oceanographic (Knoll et al., 1986, 1996; Kaufman et al., 1991, 1997; Derry et al., 1992; Grotzinger and Knoll, 1995; Grotzinger and James, 2000). The idea is roughly the following. The ocean was physically stratified for long periods pending glaciation, but switched to a turnover mode during and particularly after glacial events. In stratified mode, the biological pump (descent of organic particles from the surface) drove the δ13C of dissolved inorganic carbon (DIC) in the surface waters to higher values. Partial remineralization of the organic `rain' in conjunction with bacterial sulphate reduction in the water column produced deep water laden with 13C-depleted DIC. With prolonged stratification, the contrast in δ13C values between the surface and deep water DIC reservoirs (2–3‰ in the present ocean, Kroopnick, 1985) increased to > 10‰, consistent with the preglacial high δ13C values (Fig. 8). When turnover began, anoxic, alkalinity-laden, deep water upwelled to the surface, releasing CO2 and precipitating cap carbonates with low δ13C values. Like the snowball hypothesis, the ocean `turnover' model (Grotzinger and Knoll, 1995; Knoll et al., 1996) seeks to explain many of the salient observations, but the basic premise is faulty. First, it is physically difficult if not impossible to shut down the overturning circulation for periods > 1.5 kyr (Zhang et al., 2001). Second, organic productivity would crash if the nu;trient-rich upwelling flux were removed (Hotinski et al., 2001). Consequently, stratification would not raise the δ13C in surface waters to 5–10‰, as is observed. Third, cap carbonates formed during post-glacial sea-level rise (see below), when the injection of glacial meltwater would suppressed not enhance, ocean turnover. Kennedy (1996) attributes cap carbonates and their low δ13C values simply to post-glacial sea-level rise, but his model also depends on an 8–10‰ difference in δ13C across the thermocline, compared with ˜ 2.5‰ for Mesozoic–Cenozoic oceans.
In 1998, we invoked the snowball hypothesis, then in eclipse, as a possible explanation for the δ13C profiles (Figs 8 and 9) of carbonates bounding glacial horizons in northern Namibia (Table 1). Hoffman et al. (1998b) gave three, mutually compatible explanations for the 12C enrichment observed in cap carbonates. First, organic productivity on snowball Earth would be greatly reduced for 106–107 years, which should cause δ13C values to shift towards the hydrothermal CO2 input (−6 ± 1‰ VPDB, Des Marais and Moore, 1984) at mid-ocean ridges (Kump, 1991). Second, if carbonate sedimentation rates globally were very high, due to the alkalinity flux from weathering in the greenhouse transient, the organic fraction of the burial flux would be negatively effected. Consequently, carbonate would have low δ13C values even if organic productivity had fully recovered from the snowball event (Hoffman and Schrag, 1999). Third, the transfer of atmospheric CO2 to bicarbonate (via the weathering cycle) involves an ˜ 8‰ isotopic fractionation favouring 13C. In transferring > 98% of the built-up atmospheric CO2 to the ocean and ultimately the sedimentary reservoir, the kinetic isotope effect (Rayleigh distillation) will drive the atmospheric CO2 reservoir to ever lower δ13C values (Hoffman et al., 1998b). To the degree that atmospheric CO2 remains a major source of DIC, this would be reflected by cap-carbonate δ13C values that decline with time (Fig. 8). However, the nadir in δ13C (Fig. 9) does necessarily correspond to the completion of atmospheric CO2 drawdown, as Kennedy et al. (2001b) argue, but reflects the net effect of many factors, including those already mentioned, as well as methane hydrate destabilization during deglaciation (Kaufman et al., 1997; Kennedy et al., 2001a; but see Hoffman et al., 2002a) and dolomite–calcite isotopic equilibrium.
`Cap dolostones' (Fig. 3) are the transgressive systems tracts (Van Wagoner et al., 1988) of depositional sequences (`cap-carbonate sequences') associated with post-glacial sea-level rise. They typically grade across a marine flooding surface (Fig. 3) into deeper water limestone or shale, which gradually shoals upward through carbonate or siliciclastic lutites and arenites to an exposure surface 100s of metres above the base of the sequence (Figs 4a and 9). The great thickness and deep-water aspect of many cap-carbonate sequences (von der Borch et al., 1988; Narbonne and Aitken, 1995; Dalrymple and Narbonne, 1996; Hoffman et al., 1998b) is anomalous within their overall stratigraphic context (Fig. 9). This is commonly ascribed to post-glacial sea-level rise, but long-term tectonic subsidence is required to preserve the sequence from isostatically driven erosion, because the glacio-eustatic cycle alone creates no permanent accommodation space. On the Otavi carbonate platform in northern Namibia, an estimated background tectonic subsidence rate equivalent to ˜ 0.065 mm yr−1 of isostatically adjusted carbonate (Halverson et al., 2002) implies a timescale on the order of 5 million years to create the accommodation space occupied by the younger cap carbonate sequence (Fig. 9), which is 300–400 m thick. Hoffman et al. (1998a) suggested that that the snowball event occupied the vast majority of this time, when the average sedimentation rate on the platform was very low. By the end of the event, substantial net subsidence had occurred, although masked at that time by relative sea-level fall. Upon deglaciation, sea level rose and chemical and/or detrital sediments rapidly filled in the accommodation space provided, making up for `lost time', so to speak. High sedimentation rates in cap dolostones and overlying limestones are consistent with their low δ13C values (see above) and with their petrological characteristics (see below).
Locally, cap dolostones are overlain by limestone buildups composed largely of sea-floor cements (Peryt et al., 1990; Aitken, 1991; Grotzinger and Knoll, 1995; Kennedy, 1996; Soffer, 1998; James et al., 2001). The cements consist of upward-fanning, prismatic crystals pseudomorphosed by calcite (Figs 4f and 5e). The square-tipped, pseudo-hexagonal habit of the crystals indicates that the primary phase was aragonite. Although individual crystal fans are small (< 1 cm), they aggregate into reef-like masses (pseudo-stromatolites) that range from metres to decametres in scale (Fig. 5a,d). The cements did not precipitate slowly in the absence of particulate sediment, but grew rapidly in a race against micritic burial (Figs 4f and 5d,e). The cementstone buildups are up to 50 m thick, but eventually they give way to flaggy, allodapic, pink limestone with large-scale hummocky cross-bedding (Maieberg Formation, Namibia), or to black, organic-rich shale (Sheepbed Formation, Canada). The cements imply a high degree of carbonate oversaturation (Grotzinger and Knoll, 1995; Sumner, in press) and the dendritic-prismatic growth habit suggests rapid, diffusion-limited, growth rates (Lasaga, 1998). The presence of dissolved Fe(II) may have inhibited carbonate nucleation in anoxic waters below the mixed layer, thereby promoting the growth of crystal fans on the sea floor (Sumner and Grotzinger, 1996) while the micritic component nucleated in oxic surface waters. Deep-water anoxia is inevitable if the ocean was ice-covered for long periods (see below) and is consistent with dolomitization of cap dolostones on the sea floor (Baker and Kastner, 1981) and with the co-occurrence of sea-floor barite precipitates (Fig. 4e). The localization of these cementstone buildups at ancient shelf breaks (Soffer, 1998; James et al., 2001) suggests that their occurrence may be dictated by upwellings.
Differences among cap-carbonate sequences have elicited debate as to temporal vs. spatial control (Kennedy et al. 1998; Brasier and Shields, 2000). Some examples (Fig. 3d,e) lack transgressive systems tracts and their isotopic profiles also appear base truncated – the descending phase of the δ13C anomaly is missing (Hoffman et al., 2001). This could result from undersaturation of the local water mass until after the post-glacial sea-level rise was complete, bearing in mind the large alkalinity input required to maintain saturation if atmospheric pCO2 rose to high levels (Fig. 10) during the hypothesized snowball event. Base truncation appears more typical of `Sturtian' (Table 1) cap-carbonate sequences (Kennedy et al., 1998), and could mean that the glacial ocean was less well buffered by carbonate dissolution and/or that less carbonate was available for weathering during deglaciation (see below). While actively subsiding areas (Kimberley, Adelaide, Cordillera, Otavi) have thick (200–1200 m) post-glacial sequences (Plumb, 1981; von der Borch et al., 1988; Christie-Blick et al., 1995; Narbonne and Aitken, 1995; Hoffman et al., 1998a), the Jbéliat cap-carbonate sequence in the cratonic Taoudeni basin (Deynoux and Trompette, 1976) is highly condensed (Figs 3 and 11). After isostatic adjustments, minimal accommodation space remained after post-glacial sea-level rise because the background tectonic subsidence rate was very low. The cap carbonate built rapidly to sea level, resulting in the formation of spectacular supratidal tepee structures (Fig. 5g) and associated, early diagenetic (phreatic), barite crusts (Fig. 5f). We speculate the latter may have formed in a meteoric-marine groundwater mixing zone beneath a surface exposed to the hypothesized transient greenhouse atmosphere. Where the sequence expands to fill palaeovalleys (Fig. 11), the transgressive nature of the basal cap dolostone which drapes the palaeotopography is apparent. The Jbéliat example is also interesting inasmuch as the glacial deposits (continental tillites and outwash) preceded the hypothesized snowball event, which is apparently represented by a regolith horizon (Fig. 11) with well-developed polygonal sand wedges (Deynoux, 1982, 1985). This horizon separates the glacial deposits from the cap dolostone, and marks a period when ablation exceeded precipitation in the region. Cap carbonates also thin basinward: the Maieberg Formation is 300–400 m thick on the Otavi platform (Fig. 9), but its equivalent (correlated isotopically) thins to < 25 m within 30 km downslope from the shelf break (Hoffman et al., 1998b; Kennedy et al., 1998).
We are not burdened by an overabundance of explanations for cap carbonates. Fairchild (1993) suggested that they were precipitated from glacial meltwaters saturated with alkalinity due to interaction with carbonate-rich glacial debris and rock powder. This does not adequately explain the presence of cap carbonates in areas where the glacial deposits have little carbonate, for example the Kimberleys in Australia (Plumb, 1981). Grotzinger and Knoll (1995) proposed that cap carbonates record the turnover of a previously stratified ocean (see above), with upwelling of alkalinity-laden deepwater driving precipitation of 13C-depleted carbonates. This model dovetails with Kirschvink's (1992) post-glacial interpretation of Neoproterozoic BIF (Klein and Beukes, 1993), but the premise is questionable for reasons already given. Kennedy (1996) related cap carbonates to post-glacial flooding of continental shelves and seas, analogous to the `coral reef' hypothesis for the Quaternary (Berger, 1982), and Kennedy et al. (2001a) hypothesized that cap carbonates are byproducts of methane released from permafrost on continental shelves globally during post-glacial flooding. But methane oxidized aerobically in the water column produces CO2, causing carbonate dissolution; only methane oxidized anaerobically in conjunction with bacterial sulphate reduction in pore waters yields bicarbonate (Boetius et al., 2000). If very large amounts of methane were released rapidly (Kennedy et al., 2001a), sulphate limitaton would likely prevent substantially increased rates of anaerobic methane oxidation. Moreover, the characteristic spatial variability in δ13Ccarb (− 40 to 0‰) observed in Phanerozoic methane-seep carbonates (Kauffman et al., 1996) is not found in cap carbonates. Tube-like structures purportedly conduits for methane release (Kennedy et al., 2001a) are strictly internal to cap carbonates (Fig. 5b): they do not tap the conjectured methane source rocks (Hoffman et al., 2002a). They occur only within microbial boundstones (Fig. 5c), even in cap dolostones deposited directly on metamorphic or granitic rocks (Cloud et al., 1974; Wright et al., 1978), unlikely to be prolific sources of methane.
Once we realized that the snowball hypothesis could explain the δ13C anomalies associated with LNGD, we took as a test that it should also account for cap carbonates. Recall the deglaciation of a snowball Earth, sea ice and glaciers disappearing rapidly due to reverse ice-albedo feedback in an intense greenhouse with > 10% CO2 (Caldeira and Kasting, 1992). As surface temperatures soar, carbonic (and initially sulphuric) acid rain beats down on a landscape dominated by frost-shattered rock and rock powder from glacial action, not to mention millions of years' worth of unaltered volcanic material (Hoffman et al., 1998b; Hoffman and Schrag, 2000). Intense chemical (and physical) weathering ensues. Initially, where former carbonate platforms are exposed, carbonate weathering dominates. Later, after sea level has risen, silicate weathering becomes proportionally more important, and over time, draws down the atmospheric CO2 until a new steady state is reached (Hoffman and Schrag, 2000). The time scale of the greenhouse transient is unknown, but is conservatively estimated to be in the range 104–106 years [not < 104 years as suggested by Kennedy et al. (2001b)]. The alkalinity flux driving cap carbonate deposition during post-glacial sea level rise (i.e. below `maximum flooding' in Figs 3, 9) came from carbonate weathering; silicate weathering is too slow, given a deglaciation time of <104 years (Kennedy et al., 2001b).
Large inputs of alkalinity are required to maintain seawater saturation if atmospheric pCO2 greatly increased in a snowball event (Fig. 10). The hydrothermal dominance of the snowball ocean will drive down seawater pH unless it is buffered by dissolution of carbonate, supplied for example by glacial action. Buffering would enable cap carbonates to form immediately upon deglaciation. This is consistent with the inference (see above) that many cap carbonates formed during post-glacial transgression (Fig. 3) and with the absence of post-glacial karst beneath cap carbonates. The deglaciation of a snowball Earth involves the rapid generation and mixing of warm, oxic, surface water with cold, saline deepwater, the latter dominated hydrothermally and the former by melting and runoff that is alkalinity-laden from carbonate weathering. The internal stratigraphy of complex cap carbonates may reflect, in addition to changing water depth, the mixing of contrasting water masses (Hoffman et al., 2002).
The prerequisites for iron formation are three (Drever, 1974; Holland, 1984): (1) ocean anoxia to allow extensive transport of dissolved Fe(II); (2) an H2S:Fe(II) flux ratio < 2, so that not all Fe(II) is scavenged by sulphide (FeS2) burial, and (3) a local oxidant to drive the precipitation of an iron-formation precursor. Before the first atmospheric oxidation event around 2.4 Ga (Karhu and Holland, 1996), deep waters were anoxic and riverine sulphate input was low (Walker and Brimblecombe, 1985; Canfield et al., 2000; Farquhar et al., 2000). After 2.4 Ga, deepwater anoxia arguably persisted, but not until 1.8 Ga did oxidative weathering drive sulphate reduction in the ocean to levels sufficient to bury all Fe(II) as FeS2 (Canfield, 1998). The reappearance of iron formation in LNGD, after a 1.5-billion-year absence, is consistent with ocean anoxia due to stagnation beneath near-global sea-ice cover (Kirschvink, 1992), and also with an attenuated flux of reduced sulphur due to the frozen surface (Canfield and Raiswell, 1999).
Kirschvink (1992) and his early supporters (Klein and Beukes, 1993) suggested that Neoproterozoic iron formations were deposited at the end of snowball events, when renewed thermohaline circulation would have driven Fe(II)-rich deepwater into the oxic zone in upwelling areas. This model has been criticized on the grounds that iron formation occurs stratigraphically within or below, rather than above, certain LNGD (Williams and Schmidt, 2000). However, glacial erosion at low latitudes will itself be strongly biased toward the end of a snowball event, when the hydrological cycle has strengthened (Fig. 7). Alternatively, iron formation may have been deposited beneath the tropical ice pack if sea ice was thin (< 20 m) or discontinuous (McKay, 2000; but see Warren et al., 2002). Oxygenic photosynthesis would drive precipitation, given an anoxic ocean rich in dissolved iron. The giant Rapitan iron formations in the northern Canadian Cordillera, for example, were deposited close to the equator (Park, 1997) and are overlain by < 600 m of glacial diamictite and outwash (Yeo, 1981). [Young (1988) attributes the Rapitan iron formation to regional (but cryptic), rift-related, hydrothermal activity, but this is not supported by its trace-element geochemistry (Klein and Beukes, 1993; Graf et al., 1994) and ignores the complete absence of volcanics in the Rapitan belt and the age distribution of iron formations globally (Fig. 12).]
The snowball hypothesis has properly been questioned and the sharpest questions to date concern the terrestrial glacial regime, the isotopic evolution of snowball seawater and the fate of eukaryotic organisms. These and other questions highlight the central difficulty with the hypothesis, which is our limited conception of a snowball event itself. With postulated conditions lying far outside familiar parameter space, there is danger that the hypothesis we question is a caricature of a snowball Earth.
Climate models suggest that an ice-albedo runaway might occur very rapidly (Hyde et al., 2000), freezing the tropical oceans before ice sheets had time to develop on low-latitude continents (Hoffman et al., 1998b; Pollard and Kasting, 2001). During a snowball event, mean temperatures slowly rise (Fig. 7) in response to atmospheric CO2 and this will cause an exponential increase in saturation vapour pressures. Moisture sources include subaerial volcanic outgassing, sublimation of tropical sea ice (Warren et al., 2002), and evaporation of transient melt ponds (Walker, 2001) and polynyas within the tropical ice pack (Kirschvink, 1992). Given large seasonal and diurnal fluctuations in surface temperature (Walker, 2001), water vapour would be advected to topographic highs on summer afternoons and moisture would condense there forming glaciers (Walker, 2001). Due to the weak hydrological cycle, snow accumulation rates would be extremely low, but a positive feedback would develop between surface elevation and snow accumulation if glaciers thickened. Over sufficient time, these glaciers must become wet-based and flow, possibly in localized ice streams or as episodic ice surges (Bentley, 1987; MacAyeal, 1993). The observed zonal distribution of LNGD (Fig. 1) might broadly reflect the zonal pattern of precipitation minus ablation on the snowball Earth. On the regional scale, the importance of topography, particularly coastal topography, in capturing the limited precipitation is not inconsistent with the common observation that thicker glacial deposits occur in tectonically more active areas and times (Young, 1995). They commonly overlie preglacial strata with angular unconformity, consistent with a prolonged hiatus (with ongoing block rotation) between the ice-albedo runaway and the onset of glacigenic sedimentation (Fig. 7). Where LNGD are absent, without evidence for removal by erosion, or where their position is occupied by aeolianite (Deynoux et al., 1989; Moussine-Pouchkine and Bertrand-Sarfati, 1997; Williams, 1998; 5Fig. 5a), ablation evidently exceeded precipitation over the full glacial cycle.
Would seawater change?
The elemental and isotopic composition of seawater should change in a snowball event. Decimated organic productivity and hydrothermal dominance should cause δ13C and 87Sr/86Sr, respectively, to fall (Kump, 1991; Jacobsen and Kaufman, 1999). The trajectories of the changes depend on the C and Sr residence times, and are sensitive to buffering by carbonate dissolution (Jacobsen and Kaufman, 1999) in response to submarine volcanic exhalations of CO2 and other acids. Consequently, few δ13C or 87Sr/86Sr values have been reported for the glacial ocean because primary synglacial carbonates are rare. Kennedy et al. (2001b) report positive δ13C values from purported examples in Australia, Namibia and California, whereas Xiao et al. (2001) find that thin calcareous crusts in the Tereeken glacials (Table 1) of NW China have δ13C close to − 5‰.
Cap carbonates have been substituted as proxies for seawater 87Sr/86Sr during snowball events (Jacobsen and Kaufman, 1999; Kennedy et al., 2001b), justified by the Sr residence time of ∼ 3.5 Myr in the present ocean. Neoproterozoic Sr isotope data are not very reliable because strontium is a mobile trace element and most carbonates of that age are totally recrystallized (Derry et al., 1992). Nevertheless, the `least altered' 87Sr/86Sr ratios in cap limestones (dolomites have very low Sr concentrations) are not significantly different from preglacial values (Jacobsen and Kaufman, 1999; Kennedy et al., 2001b). This may be interpreted in two ways. Either the 87Sr/86Sr ratio of seawater changed little because the glaciation was short-lived (Jacobsen and Kaufman, 1999), contrary to palaeomagnetic (Sohl et al., 1999) and subsidence (Hoffman et al., 1998a) data, or else the ratio was buffered by carbonate dissolution during the hypothesized snowball event and enhanced weathering in the greenhouse aftermath (Hoffman et al., 1998b). If atmospheric pCO2 rose from a preglacial value of 0.0003 bar to 0.12 bar over the course of a snowball event (Caldeira and Kasting, 1992), then the Ca ion concentration of seawater would have to increase five-fold simply to maintain carbonate saturation (Fig. 10). This enormous Ca input would be accompanied by a flux of isotopically conservative Sr from carbonate dissolution and radiogenic Sr from silicate weathering. These inputs might greatly reduce or even eliminate the predicted lowering of 87Sr/86Sr during a snowball event of any given duration (Jacobsen and Kaufman, 1999). The large magnitude of the resulting seawater Sr reservoir, plus the rapid weathering of carbonates and young volcanics, would damp any increase in seawater 87Sr/86Sr due to silicate weathering, consistent with available data (Jacobsen and Kaufman, 1999; Kennedy et al., 2001b). Contrary to Kennedy et al. (2001b), the transient weathering flux is not limited to the time-scale of the sea-level rise (< 104 years) but continued long after as indicated by the negative δ13C anomaly (Fig. 9) which encompasses the entire transgressive–regressive post-glacial sequence. Accordingly, 87Sr/86Sr ratios in cap carbonates are broadly similar to preglacial values but should rise there-after due to silicate weathering, as is observed (Shields et al., 1997).
Assuming snowball events occurred, what refugia ensured the survival of eukaryotic plankton, and early metazoans if they existed? How did the climate shocks entering and exiting snowball events impact their evolution? A variety of refugia would have existed and their relative isolation and selective stresses could have contributed to evolutionary diversification. They include brine channels in new sea ice (Thomas and Dieckmann, 2002), polynyas in the tropical ice pack (Kirschvink, 1992; Lemke, 2001), tidal cracks along ice-grounding lines (Gaidos et al., 1999), transient meltwater ponds (Walker, 2001), beneath clear sea or lake ice < 20 m thick (McKay, 2000), and shallow hot springs around volcanic islands (Hoffman and Schrag, 2000). Snowball seawater would be laden with nutrients due to hydrothermal activity and limited organic productivity (Kirschvink et al., 2000) particularly in the younger Neoproterozoic glaciations when iron formation, a major sink for phophorus, was rare (see above). Enhanced productivity and burial of organic matter when snowball oceans were uncorked would have released oxygen to the atmosphere (Kirschvink et al., 2000; Walter et al., 2000), consistent with evidence of increased environmental oxygenation roughly coincident with the Palaeoproterozoic (Prasad and Roscoe, 1996; Rye and Holland, 1998; Farquhar et al., 2000; Murakami et al., 2001) and Neoproterozoic (Des Marais et al., 1992; Logan et al., 1995; Canfield and Teske, 1996) snowball eras (Fig. 12). The isotopic signal of such organic-burial events might be muted if they were accompanied by a high carbonate burial flux, driven by weathering as we have argued. The appearance of particular Ediacaran fossils has long been attributed to a rise in free oxygen (Runnegar, 1982; Knoll and Carroll, 1999). Speculatively, snowball events may have acted not only as an environmental filter on the evolution of life, but also as a biogeochemical pump that permanently changed the environment itself.
The length of both eras, 100–200 Myr, is of the time-scale of global palaeogeographical reorganizations due to plate tectonics. Kirschvink (1992) proposed that a preponderance of land masses in the middle and low latitudes set the stage for snowball events, consistent with existing palaeomagnetic data for LNGD (Fig. 1). This is broadly in line with global reconstructions for 750 Ma (Fig. 7), prior to the `Sturtian' (Table 1) glaciation(s), but `Varanger–Marinoan' (Table 1) palaeogeography is quite uncertain, primarily because of poor age control. Reconstructions in which land masses are centred on the (south) pole (e.g. Hyde et al., 2000; Baum and Crowley, 2001; Crowley et al., 2001) hinge on Laurentia's migration to high latitudes by 577 Ma (Torsvik et al., 1996), after the last snowball event (Table 1), a migration which could have ended the snowball era rather than being a precondition for it (Hyde et al., 2000). Placing continents at tangent to the tropics results in a colder Earth for a variety of reasons. It raises the surface albedo (Kirschvink, 1992) and it enhances silicate weathering, unless tectonic activity is mysteriously absent there (Marshall et al., 1988; Worsley and Kidder, 1991; Schrag et al., 2002). Weathering rates are greater on small continents because they are wetter than large ones. The Neoproterozoic snowball era was a period of continental dispersal, involving the breakup of supercontinent Rodinia and the aggregation of megacontinent Gondwana (Hoffman, 1999). Polar ice caps may reduce global weathering rates and mitigate against snowball events if large land areas exist at high latitudes (Schrag et al., 2002), as in the Phanerozoic. An absence of high-latitude continents and a preponderance of tropical and subtropical land masses would be an unusual situation in Earth history, and might give rise to polar sea-ice caps large enough to threaten runaway ice-albedo feedback (Kirschvink, 1992; Schrag et al., 2002).
Clues regarding the triggering of individual snowball events may be found in preglacial δ13C records (Figs 8 and 9). Snowball events generally follow long stages (> 107 years) of high δ13C (> 5‰ PDB in shallow marine carbonates) (Kaufman et al., 1997; Brasier and Shields, 2000; Walter et al., 2000; Halverson et al., 2002). This likely signifies that organic matter was disproportionately represented in the global carbon burial flux (forg ≥ 0.4). We speculate (Schrag et al., 2002) that a preponderance of middle- and low-latitude continents would produce this effect. Nutrient transport patterns will cause marine biological productivity to be focused in the congested middle and low latitudes where high rates of organic production will drive some basins to become anoxic (Schrag et al., 2002). Anoxic bottom waters cause high C:P ratios in the organic burial flux (Van Cappellen and Ingall, 1994; Colman and Holland, 2000), which greatly increases phosphorus availability for recycling, which in turn allows high sustained rates of organic production and burial.
The appeal of the snowball hypothesis (Kirschvink, 1992) is that it provides credible explanations for many previously enigmatic features of Neoproterozoic Earth history. It explains: (1) the widespread distribution of LNGD on virtually every continent; (2) the palaeomagnetic evidence that glacial ice lines reached sea level close to the equator for long periods; (3) the stratigraphic evidence that glacial events began and ended abruptly; (4) the reappearance of iron formations, exclusively within glacial units, after an absence of 1.2 billion years; (5) the world-wide occurrence of cap carbonates with unusual features, resting sharply above successive LNGD; (6) the existence of very large positive and negative δ13C anomalies, before and after each glacial event, respectively. The alternative `loophole' model (Hyde et al., 2000; Runnegar, 2000) compromises many of these explanations (Schrag and Hoffman, 2001), and is therefore less attractive. As the observations have no parallel in the Phanerozoic, it should not be surprising that the events responsible for them have no Phanerozoic counterparts.
Following Kirschvink (1992), we speculate that a preponderance of continents in middle to low latitudes created conditions favourable for snowball events. Such conditions would be rarely met in Earth history overall, but could persist for intervals long enough to engender repeated snowball events, as is observed. A possible trigger for individual snowball events, suggested by carbon isotopic data in many areas (Halverson et al., 2002), arises counter-intuitively from a dependence on greenhouse methane, engendered by the same unusual continental distribution (Schrag et al., 2002).
We are grateful to A. Kröner and M. Schidlowski for inviting this paper, which stems from a 10-year study of Neoproterozoic successions, principally in Namibia, funded by the Tectonics, Earth System History, and Arctic programmes of the National Science Foundation (USA), and the Astrobiology Institute of the National Aeronautics and Space Administration (NASA). The Canadian Institute for Advanced Research, the Geological Surveys of Namibia and Morocco, the Norwegian Polar Institute, and Harvard University also supported this work. We thank G. P. Halverson, Hu Guowei, A. J. Kaufman, A. C. Maloof, A. R. Prave, G. Soffer, A. van Dusen and many others for field contributions, and G. P. Halverson and A. van Dusen for the new isotopic data presented here. We thank F. A. Carselti, J. L. Kirschvink, J. C. G. Walker, and S. G. Warren for comments and preprints.