1. Introduction
Early efforts to represent the earth’s climate with energy balance models (EBMs) uncovered the disconcerting possibility that a relatively small decrease in the solar output might lead to a catastrophic global glaciation—the result of a runaway ice albedo feedback (e.g., Budyko 1969; North 1975; Lindzen and Farrell 1977). Although the issue remains controversial (e.g., Kerr 2000; Fairchild and Kennedy 2007; Allen and Etienne 2008), assorted lines of geological evidence appear to indicate that earth passed through several episodes of complete, or near-complete, glaciation during the Proterozoic (e.g., Kirschvink 1992; Hoffman et al. 1998; Hoffman and Li 2008). Follow-up integrations of more-comprehensive global climate models have also found climate states with a global or near-global glaciation, though they typically require larger reductions in the solar output than the earlier calculations suggested (e.g., Crowley and Baum 1993; Jenkins and Smith 1999; Crowley et al. 2001; Voigt and Marotzke 2009).
To our knowledge, the factors controlling snowball earth have never been presented in terms of a formal feedback analysis and doing so provides an opportunity to demonstrate several basic properties of feedbacks. Applying this analysis to the original zonal mean energy balance climate models, the physical mechanism of the runaway glaciation can be clearly and simply demonstrated. The strength of the feedback is shown to equal the ratio of competing stabilizing and destabilizing tendencies on the global energy balance or, equivalently, competing tendencies on the local energy budget at the advancing ice line. The phenomenon of a snowball earth is a simple illustration of how climate sensitivity and feedback strength can change as a function of the mean climate state, which is an issue of some relevance for future climate predictions. Moreover, although there are obvious caveats because of the simplifying assumptions of the models, the instability is also an interesting example of a climate “tipping point.”
The analytical solutions for the simple energy balance models permit feedback strengths to be calculated even for the unstable equilibrium climates. Doing so gives the somewhat counterintuitive but explainable result that the ice albedo can—under some conditions—behave as a negative feedback on global mean temperature. The cause is the peculiar physics of the small ice cap instability (e.g., North 1975), and that of a previously unreported counterpart at low latitudes.
2. Analysis
To a good approximation, S(x) may be represented as S(x) = 1 + s2P2(x), where s2 = −0.482 and P2 is the second Legendre polynomial: P2 = 1/2(3x2 − 1) (e.g., Chylek and Coakley (1975); Fig. 1a). We adopt parameter values from Lindzen and Farrell (1977): A = 211.1 W m−2; B = 1.55 W m−2 °C−1. Note that the unit of T is degrees Celsius. We allow Q to vary in the vicinity of the modern-day value, which Lindzen and Farrell took to be Q0 = 1336 W m−2.
a. Budyko-style energy balance models
1) Traditional analysis
An outline of the solution is briefly given here for the clarity of presentation, but it follows previous studies (e.g., Lindzen and Farrell 1977).
We show in the next section that a formal analysis of the ice albedo feedback provides a simple poof of the slope-stability theorem and that it gives physical insight into the cause of the instability.
2) Feedback analysis from the ice-line perspective
The instability results from the variation of albedo with changing climate state (as represented by xs,
Catastrophe occurs in the limit f → 1. Equation (13) demonstrates that, provided there is some poleward heat transport (i.e., C ≠ 0), this instability must be present for all parameter values: since S′ tends to 0 as xs nears the equator (Fig. 1a), at some latitude f must exceed 1. The slope stability theorem also follows directly from Eq. (12): for fx < 1 (i.e., stable equilibria), Δxs/ΔQ > 0; for fx > 1 (i.e., unstable equilibria), Δxs/ΔQ < 0. This behavior is shown in Fig. 2b.
3) What is the physical explanation of the instability?
The snowball instability is inevitable in this climate model simply because of the geometry of a sphere. The rate at which the local insolation increases (or in other words, the restoring warming tendency described earlier) diminishes as the ice line latitude moves equatorward (i.e., Fig. 1a) while the destabilizing effect of the local divergence of heat flux increases. As the equilibrium ice line descends to lower and lower latitudes, it becomes easier and easier for a perturbed ice line to advance. Thus, the strengthening of this positive albedo feedback as the ice line advances reflects a robust property of the climate system; therefore, it is likely to hold in more sophisticated models. We note that Lindzen and Farrell (1977, 1980), Poulsen et al. (2001), and others have explored how including dynamical circulation regimes such as the Hadley cell or additional heat transport processes, such as ocean circulation, can modify this picture and we broach this further in the discussion.
4) What is the dependency of the instability on physical parameters?
5) Feedback analysis from the global temperature perspective
The magnitude of a feedback within a system can depend on the variable or field of interest (e.g., Roe 2009). This can be illustrated by recasting the EBM system to solve for global mean temperature instead of ice line latitude. This makes the problem closer to the normal definition of the climate sensitivity to a radiative perturbation (e.g., Charney et al. 1979; Knutti and Hegerl 2008; Roe 2009).
b. Diffusive energy balance models
1) Traditional analysis
North (1975) demonstrated that an accurate analytical approximation to Eqs. (1) and (26) could be obtained using hypergeometrical functions and matching boundary conditions at the ice line. North (1975), Cahalan and North (1979), Shen and North (1999), and others have studied the stability properties of these solutions, analyzing the time-dependent behavior of perturbations away from the derived equilibrium solutions.
Figure 3a reproduces the original analytical solutions derived by North (1975), using his chosen parameter set (which are slightly different from those used up to this point in this paper). From the slope of xs versus Q, it is clear that stable climates do not exist equatorward of xs ≈ 0.6. In addition, there is also a striking phenomenon poleward of xs ≈ 0.95, the so-called small ice cap instability (e.g., North 1984): beyond some latitude, the slope of xs versus Q turns negative, implying that the polar ice cap can only be stable if it extends past some finite latitude. The reasons for this behavior has been analyzed in detail in simple systems (e.g., Lindzen and Farrell 1977; North 1984), though its presence in more complete climate models is still discussed (e.g., Crowley et al. 1994; Lee and North 1995; Langen and Alexeev 2004; Rose and Marshall 2009; Enderton and Marshall 2009).
2) Feedback analysis
A simple alternative to the time-dependent analyses cited earlier is to calculate the feedback strengths by direct substitution of the analytical solutions provided in North (1975) into Eqs. (18), (23), and (25). Figure 3d shows both fx and fT; fx behaves as expected—it lies between 0 and 1 in the stable ice line regime and exceeds 1 for unstable ice line regimes. The behavior of fT is more interesting. It goes through two singularities and actually becomes negative near the equator and near the pole.
The cause of this peculiar behavior is related to the small ice cap instability and, as it turns out, there is a directly analogous counterpart near the equator. The explanation closely follows arguments in Lindzen and Farrell (1977) for the small ice cap instability and is illustrated schematically in Fig. 3. Three curves are shown for equilibrium climate states using the Budyko-style approximation for ∇ · F, but using different values for the ice line albedo [(i) αs = α1; (ii) αs = 0.5 × (α1 + α2) as has been used up to now; and (iii) αs = α2]. Recall that these curves give pairs of (xs, Q) that are equilibrium solutions of the model equations, and that the stability of these equilibrium states can be judged from whether dxs/dQ > 0 (stable) or dxs/dQ < 0 (unstable).
The small ice cap instability can be understood by considering the intersection of these curves with xs = 1. Imagine starting with an ice-free earth and high Q (point A1 in Fig. 4). If Q is now gradually lowered, then the system moves toward point A2. As soon as any ice forms on the planet though, the solution trajectory must jump from A2 to A3 because of the discontinuity in albedo (i.e., α = αs at this point). In other words, the introduction of any ice at all means somewhat counterintuitively, that the solar constant is no longer large enough to maintain the ice at that latitude in equilibrium.
As pointed out by Lindzen and Farrell (1977), in the Budyko-style EBM the nonlocal nature of the heat transport means the discontinuity is confined to xs = 1. For North-style diffusion however, the influence of the albedo discontinuity leads to a boundary layer that extends into the domain with a characteristic length scale equal to
There is a directly analogous discontinuity at the equator. Start with an ice-covered earth and low Q (point B1). If Q is now gradually increased, then the system moves along the path from B1 to B2. But again, as soon as any ice-fee areas emerge, the solution trajectory must jump to B3. Following the same reasoning as before, Δ
In summary, imagine a global temperature increase from an unspecified cause. For most values of xs, this causes a retreat of the ice line, amplifying the original warming [Fig. 3c and Eq. (20)]. However, in the vicinity of the equator and pole, the discontinuity in albedo exerts a stronger control on the system dynamics, and the warming is in fact associated with an advance of the ice line. This damps the original warming and so the feedback is negative. Although this only occurs here in equilibrium climate states that are unstable, it is an exotic illustration of the point that if the dominant physical processes change as a function of mean climate state, then the magnitude and even the sign of the feedback can vary (e.g., Roe 2009).
3. Discussion
In essence, the analysis presented here recasts existing solutions for simple energy balance models into the language of feedback analysis. In doing so, the physical cause of the snowball earth instability can be clearly and simply laid out. From the perspective of the global energy balance, the strength of the feedback is determined by the competition between the stabilizing tendency of the outgoing longwave radiation and the destabilizing tendency of less radiation being absorbed as the planet brightens. From the perspective of the ice line, the feedback is the ratio of changes in local insolation and in the divergence of the poleward heat flux as xs changes.
Our analysis enables derivation of simple expressions for the strength of the albedo feedback as a function of mean climate state and choice of climate parameters. One principal control is, of course, the spherical geometry of the earth, which, at least within the strictures of these simple models, makes the instability inevitable at some latitude. In the case of the Budyko-style model, the latitude of the ice line instability also depends on a simple nondimensional combination of model parameters.
We have investigated the apparently strange result that, for diffusive parameterizations of heat flux, the ice albedo can even act as a negative feedback (i.e., have a stabilizing effect) on global temperature variations. It happens here only for climate states that are unstable because of the very tight coupling assumed between the ice line and temperature in the energy balance model. However, the result that global mean temperature might have a minimum at a nonzero ice-line latitudes because of the albedo discontinuity is quite physical. It remains to be explored whether this negative ice albedo feedback is just a curiosity of these particular models, or if it can help explain the occurrence of equilibrium “slush ball” states (i.e., an ice-free equatorial band) found in some climate models (e.g., Hyde et al. 2000; Crowley et al. 2001) and that have been argued to be more consistent with geological evidence (Allen and Etienne 2008). Another useful diagnostic is suggested by the results in sections 2a(4) and 2a(3). When the overall climate is stable, it is because an equatorward advance of the ice line causes a net warming at the ice line. This is likely a very general result. Studying the energy budget response to an ice line perturbation in models that exhibit slush-ball states would elucidate which terms are responsible for that warming, and perhaps therefore explain the differences from models that do not exhibit slush-ball states.
The very concept of a feedback implicitly partitions the system into a reference state and a set of physical “feedback” processes (e.g., Roe 2009). In this context, having an ice albedo feedback means introducing a process that allows the albedo to vary with a climate state. A straightforward lesson that also applies to more complex systems is that the effect of adding this process depends on which part of the system is of interest. In this simple case studied here, the feedback strength is different for the global mean temperature and for the ice line.
Our representations of the feedbacks by ratios of derivatives illustrate the general feature of feedbacks; whereas in the simplified physical system considered in this paper, the derivatives were taken with respect to the spatial variable xs, the primes could more generally indicate derivatives taken with respect to other climate variables, such as circulation pattern, atmospheric composition, among others.
The zonal mean, annual mean EBMs presented here are obviously highly idealized representations of the real world. Severe approximations have been made in their derivation, not the least of which are the absence of clouds and a seasonal cycle, and these approximations render the albedo feedback as being substantially larger than is inferred from GCMs for the modern climate (e.g., Soden and Held 2006). It would be of interest to diagnose ice albedo feedbacks within GCMs as the solar constant is reduced (following, e.g., the methods of Soden and Held 2006), and to evaluate if the feedback strength varies in ways that are consistent with the predictions from (16).
Some studies have suggested that there might be a “stability ledge” because of the effects of the Hadley cell (Lindzen and Farrell 1977); some climate model results suggests that ocean transports (Poulsen et al. 2001) or latent heat fluxes (Poulsen 2003) can act to inhibit a complete glaciation. These processes could, in principal, be cast as additional feedbacks in the energy budget. To first order, the net effect on the climate is given by the sum of the individual feedback factors and so isolating just the ice albedo feedback provides a guide for how strong those negative feedbacks have to be to create a stable equilibrium (i.e., the sum must be less than one).
Recent advances in feedback analysis permit the full spatial structure of climate feedbacks to be calculated (e.g., Soden et al. 2008), and it can even include ocean heat uptake (Gregory and Forster 2008). A full feedback diagnosis of the simulations from more complicated models such as Voigt and Marotzke (2009) would permit the relative importance of individual processes in these models (and the uncertainties in them) to be propagated through the system dynamics. One important and robust expectation is that uncertainties in physical process (and in model parameterizations of them) lead to large uncertainties in the system response in the vicinity of f = 1 because of the strong amplification that is occurring (e.g., Roe 2009).
The snowball earth phenomenon illustrates how localized physical processes can have a global effect. Here, strong model assumptions control how something happening at one particular latitude—the albedo changing because of an ice line advance—acts to affect the global mean climate. In nature other important feedbacks are also localized, such as the strong negative feedback of subtropical stratus decks (e.g., Sanderson et al. 2008), or the high-latitude deep-ocean heat uptake (e.g., Gregory and Forster 2008; Winton et al. 2010; Baker and Roe 2009). Perhaps one important way forward for improving both global and regional climate predictions will be to better understand how these regional processes combine to give the full, global system response.
Acknowledgments
We are grateful to Steve Warren for his helpful comments and suggestions; to Gerald North, Dorian Schuyler Abbot, and one other reviewer; and to Anthony Broccoli, the editor.
REFERENCES
Abramowitz, M., and I. A. Stegun, 1965: Handbook of Mathematical Functions: With Formulas, Graphs, and Mathematical Tables. Dover Publications, 1046 pp.
Allen, P. A., and J. L. Etienne, 2008: A sedimentary challenge to Snowball Earth. Nat. Geosci., 1 , 817–825.
Baker, M. B., and G. H. Roe, 2009: The shape of things to come: Why is climate change so predictable? J. Climate, 22 , 4574–4589.
Budyko, M. I., 1969: The effect of solar radiation variations on the climate of the Earth. Tellus, 5 , 611–619.
Cahalan, R. F., and G. R. North, 1979: A stability theorem for energy-balance climate models. J. Atmos. Sci., 36 , 1178–1188.
Charney, J., and Coauthors, 1979: Carbon Dioxide and Climate: A Scientific Assessment. National Academy of Sciences, 34 pp.
Chylek, P., and J. A. Coakley, 1975: Analytical analysis of a Budyko-type climate model. J. Atmos. Sci., 32 , 675–679.
Crowley, T. J., and S. K. Baum, 1993: Effect of decreased solar luminosity on late Precambrian ice extent. J. Geophys. Res., 98 , 16723–16732.
Crowley, T. J., K-J. Yip, and S. K. Baurn, 1994: Snowline instability in a general circulation model: Application to Carboniferous glaciation. Climate Dyn., 10 , 363–376.
Crowley, T. J., W. T. Hyde, and W. R. Peltier, 2001: CO2 levels required for deglaciation of a “near-snowball” Earth. Geophys. Res. Lett., 28 , 283–286.
Drazin, P. G., and D. H. Griffel, 1977: On the branching structure of diffusive climatological models. J. Atmos. Sci., 34 , 1858–1867.
Enderton, D., and J. Marshall, 2009: Explorations of atmosphere–ocean–ice climates on an aquaplanet and their meridional energy transports. J. Atmos. Sci., 22 , 1593–1611.
Fairchild, I. J., and M. J. Kennedy, 2007: Neoproterozoic glaciation in the Earth System. J. Geol. Soc. London, 164 , 895–921.
Ghil, M., 1976: Climate stability for a Sellers-type model. J. Atmos. Sci., 33 , 3–20.
Gregory, J. M., and P. M. Forster, 2008: Transient climate response estimated from radiative forcing and observed temperature change. J. Geophys. Res., 113 , D23105. doi:10.1029/2008JD010405.
Held, I. M., and M. Suarez, 1974: Simple albedo feedback models of the icecaps. Tellus, 36 , 613–629.
Hoffman, P. F., and Z-X. Li, 2008: A palaeogeographic context for Neoproterozoic glaciation. Palaeogeogr. Palaeoclimatol. Palaeoecol., 277 , 158–172.
Hoffman, P. F., A. J. Kaufman, G. P. Halverson, and D. P. Schrag, 1998: A Neoproterozoic snowball Earth. Science, 281 , 1342–1346.
Hyde, W. T., T. J. Crowley, S. K. Baum, and W. R. Peltier, 2000: Neoproterozoic ‘snowball Earth’ simulations with a coupled climate/ice-sheet model. Nature, 405 , 425–429.
Jenkins, G. S., and S. R. Smith, 1999: GCM simulations of snowball earth conditions during the Late Proterozoic. Geophys. Res. Lett., 26 , 2263–2266.
Kerr, R. A., 2000: An appealing snowball earth that’s still hard to swallow. Science, 287 , 1734–1736.
Kirschvink, J. L., 1992: Late Proterozoic low-latitude global glaciation: The snowball earth. The Proterozoic Biosphere: A Multidisciplinary Study. J. W. Schopf and C. Klein, Eds., Cambridge University Press, 51–52.
Knutti, R., and G. C. Hegerl, 2008: The equilibrium sensitivity of the Earth’s temperature to radiation changes. Nat. Geosci., 1 , 735–743.
Langen, P. L., and V. A. Alexeev, 2004: Multiple equilibira and asymmetric climates in the CCM3 coupled to an oceanic mixed layer with thermodynamic sea ice. Geophys. Res. Lett., 31 , L04201. doi:10.1029/2003GL019039.
Lee, W. H., and G. R. North, 1995: Small ice cap instability in the presence of fluctuations. Climate Dyn., 11 , 242–246.
Lindzen, R. S., 1990: Dynamics in Atmospheric Physics. Cambridge University Press, 320 pp.
Lindzen, R. S., and B. Farrell, 1977: Some realistic modifications of simple climate models. J. Atmos. Sci., 34 , 1487–1501.
Lindzen, R. S., and B. Farrell, 1980: The role of polar regions in global climate, and the parameterization of global heat transport. Mon. Wea. Rev., 108 , 2064–2079.
North, G. R., 1975: Theory of energy-balance climate models. J. Atmos. Sci., 32 , 2033–2043.
North, G. R., 1984: The small ice cap instability in diffusive climate models. J. Atmos. Sci., 41 , 3390–3395.
North, G. R., 1990: Multiple solutions in energy balance climate models. Global Planet. Change, 2 , 225–235.
Poulsen, C. J., 2003: Absence of a runaway ice-albedo feedback in the Neoproterozoic. Geology, 31 , 473–476.
Poulsen, C. J., R. T. Pierrehumbert, and R. L. Jacob, 2001: Impact of ocean dynamics on the simulation of the Neoproterozoic snowball Earth. Geophys. Res. Lett., 28 , 1575–1578.
Roe, G. H., 2009: Feedbacks, timescales, and seeing red. Annu. Rev. Earth Planet. Sci., 37 , 93–115.
Rose, B. E. J., and J. Marshall, 2009: Ocean heat transport, sea ice, and multiple climate states: Insights from energy balance models. J. Atmos. Sci., 66 , 2828–2843.
Sanderson, B. M., C. Piani, W. Ingram, D. A. Stone, and M. R. Allen, 2008: Towards constraining climate sensitivity by linear analysis of feedback patterns in thousands of perturbed-physics GCM simulations. Climate Dyn., 30 , 175–190.
Sellers, W. D., 1969: A global climatic model based on the energy balance of the earth-atmosphere system. J. Appl. Meteor., 8 , 392–400.
Shen, S., and G. R. North, 1999: A simple proof of the slope stability theorem for energy balance climate models. Can. Appl. Math Quart., 7 , 203–215.
Soden, B. J., and I. M. Held, 2006: An assessment of climate feedbacks in coupled ocean–atmosphere models. J. Climate, 19 , 3354–3360.
Soden, B. J., I. M. Held, R. Colman, K. M. Shell, J. T. Kiehl, and C. A. Shields, 2008: Quantifying climate feedbacks using radiative kernels. J. Climate, 21 , 3504–3520.
Su, C. H., and D. Y. Hsieh, 1976: Stability of the Budyko climate model. J. Atmos. Sci., 33 , 2273–2275.
Voigt, A., and J. Marotzke, 2009: The transition from the present-day climate to a modern Snowball Earth. Climate Dyn., doi:10.1007/s00382-009-0633-5, in press.
Winton, M., K. Takahashi, and I. M. Held, 2010: Importance of ocean heat uptake efficacy to transient climate change. J. Climate, 23 , 2333–2344.
(a) Normalized insolation distribution S(x) as a function of latitude. The normalization is such that 
Citation: Journal of Climate 23, 17; 10.1175/2010JCLI3545.1
Properties relating to the ice line instability in the Budyko model. (a) Equilibrium ice line as a function of insolation relative to modern. Following Lindzen and Farrell (1977), Q0 = 1336 W m−2: regions with positive slope are stable equilibria, negative slopes are unstable equilibria; (b) albedo feedback factors fx and fT. Only regions with f < 1 are stable equilibria.
Citation: Journal of Climate 23, 17; 10.1175/2010JCLI3545.1
Properties of solutions to the North diffusive EBM. (a) The ice line as a function of Q/Q0, using North (1975) analytical solutions and parameters. The thin lines show turning points. (b) Global mean temperature vs solar constant for the same solution. (c) Global mean temperature vs ice line for the same solution. (d) Albedo feedback factors fx and fT. The thin lines confirm that the feedbacks exceed 1 at the latitude of the turning points in (a). Note that fT becomes negative near the equator and the pole. Refer to text for an explanation.
Citation: Journal of Climate 23, 17; 10.1175/2010JCLI3545.1
Schematic explanation of small ice cap instability, and the regions of negative fT feedback, extending the arguments of Lindzen and Farrell (1977). Refer to text for details.
Citation: Journal of Climate 23, 17; 10.1175/2010JCLI3545.1
