Climate of the Past The initiation of modern soft and hard Snowball Earth climates in CCSM 4

Geochemical and geological evidence has suggested that several global-scale glaciation events occurred during the Neoproterozoic Era in the interval from 750– 580 million years ago. The initiation of these glaciations is thought to have been a consequence of the combined influence of a low level of atmospheric carbon dioxide concentration and an approximately 6 % weakening of solar luminosity. The latest version of the Community Climate System Model (CCSM4) is employed herein to explore the detailed combination of forcings required to trigger such extreme glaciation conditions under present-day circumstances of geography and topography. It is found that runaway glaciation occurs in the model under the following conditions: (1) an 8–9 % reduction in solar radiation with 286 ppmv CO 2 or (2) a 6 % reduction in solar radiation with 70–100 ppmv CO 2. These thresholds are moderately different from those found to be characteristic of the previously employd CCSM3 model reported recently inYang et al.(2012a,b), for which the respective critical points corresponded to a 10–10.5 % reduction in solar radiation with 286 ppmv CO 2 or a 6 % reduction in solar radiation with 17.5–20 ppmv CO 2. The most important reason for these differences is that the sea ice/snow albedo parameterization employed in CCSM4 is believed to be more realistic than that in CCSM3. Differences in cloud radiative forcings and ocean and atmosphere heat transports also influence the bifurcation points. These results are potentially very important, as they are to serve as control on further calculations which will be devoted to an investigation of the impact of continental configuration. We demonstrate that there exist “soft Snowball” Earth states, in which the fractional sea ice coverage reaches approximately 60–65 %, land masses in low latitudes are covered by perennial snow, and runaway glaciation does not develop. This is consistent with our previous results based upon CCSM3. Although our results cannot exclude the possibility of a “hard Snowball” solution, it is suggested that a “soft Snowball” solution for the Neoproterozoic remains entirely plausible.


Introduction
Based on the observations of glacial deposits on the continents at low latitudes, the associated extreme carbon isotope fluctuations, banded iron formations (BIFs) and post-glacial cap carbonates in glacial deposits of the Neoproterozoic era, it is generally thought that two significant and global-scale glaciations occurred at ∼716 Ma 5 and at ∼635 Ma, during which land based ice-sheets reached deep into tropical latitudes. However, it is still disputed as to that whether the entire ocean was covered by thick sea-ice, the so-called "Snowball" or "hard Snowball" Earth hypothesis (Kirschvink, 1992;Hoffman et al., 1998;Hoffman and Schrag, 2002), or tropical open-water oceans coexisted with low-latitude continental ice-sheets, the so-called "Slushball" or "soft 10 Snowball" Earth hypothesis (Hyde et al., 2000;Peltier et al., 2007;Allen and Etienne, 2008). Aside from these major variations upon what is generally agreed to have been extreme glacial conditions, additional suggestions to the nature of Neoproterozoic climate have included a state in which the global ocean was covered by sea-ice but the tropical sea-ice was thin enough to provide suitable refugia to explain the survival of 15 eukaryotic life (Pollard and Kasting, 2005;Warren and Brandt, 2006), and one in which the tropical ocean was covered by sea-ice but the polar regions were ice-free, the "high-obliquity" hypothesis (Williams, 2008, and references therein).
Snowball initiation is a crucial aspect of the debate. Budyko (1969) and Sellers (1969) found, using simple Energy Balance Models, that when the sea-ice is forced to 20 advance to below approximately 30 • latitude, runaway glaciation would occur owing to a strong nonlinear ice albedo feedback and therefore a "soft Snowball" solution with open water continuing to exist at the equator would be unstable. In some general circulation models, various authors also have found this characteristic to prevail (e.g. Lewis et al., 2007;Voigt and Marotzke, 2009;. However, in other 25 models, it is found that a state with tropical open-water may be stable even if the seaice edge reaches as low as 10 • latitude (Chandler and Sohl, 2000;Baum and Crowley, 2001;Pierrehumbert et al., 2011;Abbot et al., 2011;Yang et al., 2011a,b). One of the 3 reasons for the absence of runaway glaciation in these models may be associated with the relatively low sea-ice and/or snow albedo employed, which restricts the strength of the albedo feedback (Pierrehumbert et al., 2011;Yang et al., 2011a). For instance, the sea-ice albedo in EBM/ice-sheet models has been set to be as low as 0.45 (Hyde et al., 2000;Peltier et al., 2004Peltier et al., , 2007Peltier, 2010, 2011), and the snow 5 albedo in CAM3 and CCSM3 is set to 0.66-0.78 (Pierrehumbert et al., 2011;Abbot et al., 2011;Yang et al., 2011a), both of which are smaller than observations, 0.47-0.52 for bare sea-ice, 0.55-0.66 for sea glacier and 0.75-0.87 for snow (Perovich, 1996;Warren et al., 2002;Warren and Brandt, 2006). As the sea-ice and/or snow albedo increases, it will clearly be easier to enter a globally ice-covered state (Lewis et al., 2003;Pierrehumbert et al., 2011;Yang et al., 2011a).  reported that, based upon the analyses performed using the ECHAM5/MPI-OM model, that a "hard Snowball" Earth would occur under a plausible level of CO 2 (278 ppmv) but with a high sea-ice albedo of 0.75 when the solar constant was reduced by 6 % as would be appropriate under Neoproterozoic conditions. 15 Furthermore, in previous models, the parameterization for sea-ice and snow albedo has been overly simplified (Briegleb and Light, 2007;Yang et al., 2011a). The dependences of surface albedo on the solar zenith angle, snow grain radius, cloud cover and the incidence of melt ponds have been neglected, which have significant effects on temporal and spatial variations of the surface albedo (Wiscombe and Warren, 1980; future climates. A relatively low resolution version suitable for paleoclimate studies, is employed in the present work. The atmosphere component is an atmospheric general circulation model with a spectral Eulerian core and a horizontal gird with a resolution of 3.75 • . In the vertical the atmosphere has 26 levels from the surface to approximately 2 hPa. The land component is the Community Land Model version 4 (CLM4), which uses the same horizontal resolution as the atmosphere component (Lawrence et al., 2011). The ocean component is based on the Parallel Ocean Program version 2 (Smith et al., 2010), which has a horizontal grid consisting of 100×116 points and a zonal resolution of 3.6 • . The meridional resolution is variable with a spacing of approximately 0.9 • around the equator. There are 60 depth layers in the vertical (vs. Compared to the previous version CCSM3, CCSM4 includes several significant developments and improvements, including a much improved representation of atmo-20 spheric deep convection, a new freeze-dry scheme for polar low clouds, an improved representation for surface runoff and frozen soil, an improved parameterization for near-surface oceanic eddies, new parameterizations for submesoscale mixed layer eddies and ocean overflows, and a new radiative transfer scheme for sea-ice/snow (Gent et al., 2011;Shields et al., 2011, and references therein). These changes improve the model representations of the frequency of the ENSO variability (Richter and Rasch, 2008;Neale et al., 2008), of polar cloud cover (Vavrus and Waliser, 2008), 5 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | of the penetration of the North Atlantic Meridional Overturning Circulation (Danabasoglu et al., 2010), of land water storage (Lawrence et al., 2011), and of the Arctic sea-ice concentration and its trend (Gent et al., 2011). For example, the ENSO power spectrum in CCSM4 is now comparable to the observed 2-7 yr period (Gent et al., 2011), whereas it is dominated by variability at ∼2 yr period in CCSM3 (Collins et al.,5 2006a). As a further example, the Antarctic Bottom Water (AABW) transport in CCSM4 is approximately 8 Sv, which is in much better agreement with observations and much weaker than in CCSM3 (∼16 Sv) .
In the first step of the present analyses, we run CCSM4 under the pre-industrial (PI) forcings: the solar constant is set to 1360.89 W m −2 and the carbon dioxide concen-

Results
We begin by comparing the results obtained using CCSM4 and CCSM3 in the preindustrial control experiments and in the 6 % reduced solar radiation experiments; 25 through these comparisons, we will establish the main differences between CCSM4 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | and CCSM3. We will next address the values for the hard Snowball Earth bifurcation points (above which runaway albedo feedback occurs) and the conditions for existence of a soft Snowball state in CCSM4. Figure 1 shows the surface temperature (T S ), sea-ice coverage, surface albedo, net cloud radiative forcing (CRF, shortwave plus longwave, at the top of the model) and 5 poleward atmospheric and oceanic energy transports under the pre-industrial forcings. The global-mean T S in CCSM4 is 285.6 K, 1.5 K lower than that in CCSM3, associated with low tropical and polar temperatures. In CCSM4, sea-ice albedo and snow albedo are 0.61-0.65 and 0.71-0.91, respectively, both of which are greater than those in CCSM3, 0.43-0.50 and 0.66-0.78 (Table 1). As shown in Figs. 1c and 2c, the com-10 bined sea-ice/snow albedo in CCSM4 is about 0-0.2 greater than in CCSM3. Consequently, the zonal-mean T S in CCSM4 in the North (South) Polar region is about 8 K (5 K) lower than in CCSM3. Over the equator the T S in CCSM4 is approximately 2 K lower than in CCSM3, which is mainly due to the stronger CRF in CCSM4 than in CCSM3, approximately −36 vs. −22 W m −2 (Fig. 1d). Atmospheric energy transport at North Atlantic Deep Water (NADW) is several hundred meters deeper in CCSM4 than in CCSM3, associated with the implementation of the new parameterization for density driven overflows documented in Danabasoglu et al. (2010). (2) The Antarctic Bottom Water (AABW) transport in CCSM4 is much weaker than in CCSM3 (4 vs. 16 Sv) mainly due to less brine rejection into the ocean in CCSM4 (Gent et al., 2011). On the 5 other hand, the NADW transport is enhanced by about 2 Sv in CCSM4. (3) The mass streanfunction in the region of the Antarctic Circumpolar Current (ACC) is significantly reduced from 44 to 18 Sv, partly due to the addition to the model of a parameterization for eddy-induced fluxes  and partly due to relatively weak zonal wind stress driving the ocean in CCSM4 (Gent et al., 2011). This aspect of the 10 southern ocean circulation is much closer accord with the observations of Orsi et al. (2002). (4) In CCSM4 the Subtropical-Tropical Cells (STCs, wind-driven circulations) are several sverdrups greater than in CCSM3. All of these results are qualitatively consistent with those in high-resolution versions of the CCSM4 and CCSM3 models described by Danabasoglu et al. (2011). 15 In response to a reduction of solar constant, the trends of the circulation pattern are similar between CCSM4 and CCSM3: NADW formation becomes weaker, AABW formation and the STC circulations become stronger (for further details, please see Yang et al., 2011b). However, in CCSM4, the clockwise circulation in the latitudes of the ACC is transformed to a robust counter-clockwise circulation, likely arising from the 20 added parameterization of oceanic near-surface eddies (Danabasoglu et al., 2008). For the atmospheric mass streamfunction, the strength and extent of the Hadley cells are very similar between CCSM3 and CCSM4 (not shown).
The analyses above demonstrate that between CCSM4 and CCSM3 the main differences are in sea-ice/snow albedo and cloud radiative forcings. In CCSM3, the low 25 sea-ice/snow albedo is used to compensate the overly small downward solar radiation at the surface caused by a bias due to too much low cloud in the polar regions (Collins et al., 2006a;Vavrus and Waliser, 2008). In CCSM4, this bias has been alleviated (Vavrus and Waliser, 2008) and the sea-ice/snow albedo now is in good agreement 8 with observations. Notably, there are significant differences in cloud radiative forcings between CCSM3 and CCSM4, which exerts important influences on the surface temperature and on sea-ice advance. From CCSM3 to CCSM4, the strength and extent of the ocean overturning circulations have also been modified significantly but the integrated oceanic energy transport shows slight changes (except in the region of the 5 ACC), which should be expected from the compensations among different regions. However, as the sea-ice approaches the tropics, the differences in atmospheric and oceanic heat transports between CCSM4 and CCSM3 are amplified as we will demonstrate in what follows.
In the next step in this series of analyses, we investigate the initiation processes of 10 a modern Snowball Earth in CCSM4. Sea-ice evolution is investigated in a first set of experiments with differing solar constants but fixed CO 2 concentration and in a second group of experiments with different CO 2 concentrations but fixed solar radiation. The results for these two sets of experiments are shown in Fig. 5a,b, respectively. Figure 6 shows the sea-ice evolution with different initial conditions under the same forcing of 15 94 % solar radiation and 100 or 120 ppmv CO 2 . In our pervious study with CCSM3 (Yang et al., 2011a,b), it was found that runaway glaciation occurs for a 10-10.5 % reduction in solar radiation with 286 ppmv CO 2 or a 6 % reduction in solar radiation with 17.5-20 ppmv CO 2 . In CCSM4, the current analyses show that the threshold is an 8-9 % reduction in solar radiation with 286 ppmv CO 2 or a 6 % reduction in solar radiation 20 with 70-100 ppmv CO 2 (Fig. 5). The main reason for the difference in the bifurcation points for the two models is connected with the higher sea-ice/snow albedo in CCSM4. Differences in cloud radiative forcings and poleward atmospheric and oceanic energy transports also affect the forcings required to trigger a hard Snowball Earth. Notably, when the sea-ice fraction reaches ∼45 %, it jumps abruptly to ∼60 % in ap-25 proximately 20 model years, which is due to the positive feedback associated with the Hadley cells. As the sea-ice fronts enter the tropics, the near-surface branch of the Hadley cells (trade winds) transports relatively cold air to the sea-ice fronts (Bendtsen, 2002) and meanwhile the strong stresses associated with the trade winds enhance 9 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the sea-ice flow velocity (Yang et al., 2011a,b), both of which amplify the equatorward expansion of the sea-ice. In other words, there exist no stable states between 45-60 % sea-ice coverage, consistent with the results in CCSM3 (see Fig. 3 of Yang et al., 2011a). Furthermore, in all of the experiments shown in Figs. 5 and 6, as sea-ice coverage reaches approximately 60-65 % and the corresponding sea-ice edge approaches 5 ∼10 • S and 30 • N latitudes (Fig. 7a,b), the growth rate of the sea-ice significantly decreases and in some cases the sea-ice margin remains stable near these latitudes without entering a hard Snowball Earth state. From Fig. 5, it is clear that this feature does not depend on the strength of the solar radiation or CO 2 forcing, and in Fig. 6, it is found that it is also independent of the initial conditions. These results imply that 10 there are internal processes which act to effectively prevent the runaway albedo feedback from occurring. In CCSM3, there is no indication of this phenomenon (Yang et al., 2011a,b), nor does it appear to exist in other models, such as the Fast Ocean Atmosphere Model (Poulsen and Jacob, 2004) or the fully-coupled ocean-atmosphere model ECHAM5/MPI-OM (Voigt and Marotzke, 2009;. It is clearly 15 important to know the cause of this distinguishing feature in CCSM4. Investigating the differences between CCSM4 and CCSM3 enables us to provide an answer. To this end, we have analysed the results from the case of 94 % insolation and 100 ppmv CO 2 in CCSM4 and from the case of 94 % insolation and 50 ppmv CO 2 in CCSM3. We choose these two experiments to compare because both enter a near- 20 Snowball state (Fig. 7) and the final equilibrium sea-ice coverages are similar. Although in this comparison the CO 2 forcing in CCSM4 is ∼2.66 W m −2 stronger (Collins et al., 2006b) than that in CCSM3, its influence over the sea-ice/snow would be compensated by the relatively high sea-ice/snow albedo in CCSM4. The sea-ice evolution during the integration period and the final equilibrium states are shown in Figs. 8 and 9, respec-25 tively. In these two cases, the global-mean sea-ice coverage and T S are very similar, approximately 60 % and 254 K, respectively, but the meridional distributions are quite different. In CCSM3 the sea-ice distribution is more symmetrical about the equator than it is in CCSM4; in CCSM4, the sea-ice is closer to the equator in the SH and further Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | from the equator in the NH. Moreover, in CCSM4 surface air temperature is higher in the NH and lower in the SH than in CCSM3. From Fig. 8, it is found that the sea-ice growth rate in the SH is much greater than in the NH in both CCSM3 and CCSM4, consistent with the results in ECHAM5/MPI-OM (Voigt and Marotzke, 2009). This is due to the factors that (1) the surface albedo over 5 snow/ice covered regions of the SH is greater than that over snow-free land areas of the NH and (2) the cloud radiative forcing in the SH is about 10 W m −2 stronger (negative, cooling effect) in the SH than that of the NH (Figs. 1d, 2d, and 9d), both of which enhance the surface cooling (Figs. 1a, 2a and 9a) and thereby support the sea-ice growth and expansion in the SH. Furthermore, in the early stage of evolution, the sea-10 ice growth in CCSM4 is much slower than in CCSM3, likely arising from the relatively weak cloud radiative forcing and stronger CO 2 radiative forcing over the oceans in CCSM4 as mentioned earlier. However, in the SH, as the sea-ice moves closer to the tropics, the high sea-ice/snow albedo in CCSM4 greatly enhances the albedo feedback so that the sea-ice extent abruptly increases from approximately 110 to 160 × 10 6 km 2 15 (Fig. 8a). Interestingly and importantly, as the sea-ice approaches to the equator in the SH, it does not across the equator owning to the increased cross-equatorial heat transport. There is a 4.2 PW transport of energy across the equator from the NH to the SH by atmospheric and oceanic processes (Fig. 9e,f); this transport averages to a net cooling of 16.8 W m −2 in the NH and a compensating warming of 16.8 W m −2 in the SH, which effectively prohibits the cross-equatorial expansion of the sea-ice.
In the NH, the expansion of the sea-ice in CCSM4 is much slower than in CCSM3 and the sea-ice fronts in CCSM4 are out of the domain of the Hadley cells during the entire integration. This is due primarily to the fact that in CCSM4 a greater fraction of land is snow-free in boreal summer and the surface is warm (0-15 • C, Fig. 7a), which would 25 support heat transports to the downstream oceans and inhibit the growth of sea-ice in these regions. Moreover, over the open water, in CCSM4 the cloud radiative forcing is about 6-10 W m −2 weaker than in CCSM3, and the surface air temperatures over the open oceans are ∼4 K higher than in CCSM3, which act to prevent the equatorward advance of the sea-ice. Therefore, it is the combination of atmosphere and ocean dynamics, cloud radiative forcing and less snow over land which suppresses the spread of the sea-ice and promote the existence of open-water oceans in CCSM4. Following the definition of a "soft Snowball" Earth in Yang et al. (2011a), i.e., sea-ice margins enter the tropics, 5 sea-ice coverage is higher than 50 % and less than 100 %, and a significant area of the tropical continents is covered by perennial snow, the CCSM4 solutions with 60-65 % sea-ice coverage under 100-140 ppmv CO 2 and 94 % solar radiation should be viewed as "soft Snowball" Earth states . Although the snow-covered land regions are quite limited in CCSM4 (Fig. 7a,b), it should keep in mind that these regions would 10 become covered by continental heavy ice-sheets on the long timescale of million years and these ice-sheets would "flow" into the remaining tropical continents, as discussed in Hyde et al. (2000), Pollard and Kasting (2004), Peltier et al. (2004Peltier et al. ( , 2007, and Peltier (2010, 2011). 15 In comparison with CCSM3, we find that CCSM4 is able to somewhat easily enter a globally ice-covered state. However, the requirements in CCSM4 for an 8-9 % reduced solar luminosity and 286 ppmv CO 2 or a 6 % reduced solar luminosity with 70-100 ppmv CO 2 , may be still not met by the expected conditions of the Neoproterozoic. For the solar luminosity, compare with the present-day level, the Neoproterozoic sun 20 was only about 5.0-6.7 % fainter at 635 million years ago (the time of the onset of the Marinoan glaciation) and about 5.3-7.4 % fainter at 715 million years ago (the time of the onset of the Sturtian glaciation) (Gough, 1981;Crowley and Baum, 1993).

Discussion and conclusions
In so far as the concentration of carbon dioxide is concerned, it is poorly known. Prior to the onset of the Neoproterozoic ice ages, it must have been high, ∼12 times 25 (3360 ppmv) the preindustrial level to compensate the effect of a relatively faint sun in order to keep the Neoproterozoic temperature close to present (Pierrehumbert et al., Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2011), while after the Neoproterozoic era, at the beginning of the Cambrian period (∼542 Ma), it is suggested to be ∼15 times the preindustrial level (Berner, 2006). During the Neoproterozoic ice ages, several models have been employed to examine the CO 2 level. Donnadieu et al. (2004) reported that, based on results of a climategeochemical model, the atmospheric CO 2 amount would decrease from an initial level 5 of 1830 ppmv (assumed) to a low level of 250 ppmv by enhanced continental weathering associated with changes in run-off and eruptions of fresh basaltic provinces as a result of the break-up of Rodinia supercontinent. However, Peltier et al. (2007) and Liu and Peltier (2011) have argued that, based on the results of a carbon-cycle-coupled climate model, oceanic dissolved organic carbon would be converted into inorganic car-10 bon as the system cooled and oxygen was drawn down into the Neoproterozoic ocean, and then the inorganic carbon released into atmosphere (see also Rothman et al., 2003). The remineralization of organic carbon might well prohibit the atmospheric CO 2 concentration from dropping to the very low level required for global glaciation to ocean. Because these two models are highly parameterized, however, there exists uncertainty 15 in the results. More comprehensive models employing these physical processes need to be developed.
In CCSM4, it is found that the maximal stable sea-ice cover is approximately 65 % (Fig. 6), smaller than in CCSM3 (76 %, Yang et al., 2011a) but larger than in ECHAM5/MPI-OM (55 %, Voigt and Marotzke, 2009;. The main rea-20 son for this divergence is the different sea-ice/snow albedos employed in these different models. The sea-ice albedo is 0.43-0.50 in CCSM3, 0.61-0.65 in CCSM4, and 0.55-0.75 in ECHAM5/MPI-OM. Increasing the sea-ice/snow albedo will lead to a smaller value for the maximal stable sea-ice cover (Yang et al., 2011a).
In Pierrehumbert et al. (2011) and Abbot et al. (2011), it was argued that low sea-25 ice/snow albedo is a necessary condition for the existence of a soft Snowball solution, which is to some degree in contrast to the results obtained using CCSM3 or CCSM4. It is found in fact that, after including an active ocean, a soft Snowball state exists in circumstance in which the sea-ice albedo may be as high as 0.60 in CCSM3 (Yang 13 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | et al., 2011a) or 0.61-0.65 in CCSM4 and the snow albedo may be as high as 0.71-0.91 in CCSM4. The atmosphere and ocean dynamics aid and abet the persistence of open-water oceans at low latitudes even under extremely cold global conditions. In the previous analyses presented in Yang et al. (2011a), it was proposed that as the sea-ice moves to low latitudes, the area coverage of melt ponds over the sea-ice would increase due to the fact of that the solar insolation in low latitudes is much greater than in high latitudes. In CCSM4 for both reproducing the preindustrial climate and simulating the Snowball Earth climates, it is found that the melt pond fraction in the summer is less than 10 %, much less than observations (as high as 40 % in the Arctic, Perovich et al., 2002), and its trend is unclear (not shown). In CCSM4, the influence 10 of the melt ponds remains highly parameterized (Holland et al., 2011); more complete and realistic representation of the physical processes should be incorporated into the model.
As a summary, the issue of the radiative forcings required to trigger a modern "hard Snowball" or a modern "soft Snowball" have been investigated using the coupled atmosphere-ocean model CCSM4. Owing to the uncertainties in model parameterizations, the absence of land ice-sheets and the modern geography used in the simulations, we cannot exclude the possibility of a "hard Snowball" Earth. More importantly, however, our results confirm that a "soft Snowball" Earth solution is an equally tenable solution to the Neoproterozoic climate puzzle. simulations with a coupled climate/ice-sheet model, Nature, 405, 425-429, 2000. 3, 4, 12 16