ARTICLESPUBLISHED ONLINE: 25 JULY 2010 | DOI: 10.1038/NGEO922

Influence of high-latitude vegetation feedbacks onlate Palaeozoic glacial cyclesDaniel E. Horton1*, Christopher J. Poulsen1 and David Pollard2

Glaciation during the late Palaeozoic era (340–250 Myr ago) is thought to have been episodic, with multiple, often regional,ice-age intervals, each lasting less than 10 million years. Sedimentary deposits from these ice-age intervals exhibit cyclicaldepositional patterns, which have been attributed to orbitally driven glacial–interglacial cycles and resultant fluctuations inglobal sea level. Here we use a coupled general-circulation/biome/ice-sheet model to assess the conditions necessary forglacial–interglacial fluctuations. In our simulations, ice sheets appear at atmospheric pCO2 concentrations between 420 and840 ppmv. However, we are able to simulate ice-sheet fluctuations consistent with eustasy estimates and the distribution ofglacial deposits only when we include vegetation feedbacks from high-latitude ecosystem changes. We find that ice-sheetadvances follow the expansion of high-latitude tundra during insolation minima, whereas ice retreat is associated with theexpansion of barren land close to the edge of the ice sheets during periods of high insolation. We are unable to simulateglacial–interglacial cycles in the absence of a dynamic vegetation component. We therefore suggest that vegetation feedbacksdriven by orbital insolation variations are a crucial element of glacial–interglacial cyclicity.

The late Palaeozoic ice age (LPIA; 340–250Myr ago) hasconventionally been characterized as a time of protractedcontinental-scale glaciation1–4. This view was shaped by in-

terpretations of Northern Hemisphere transgression–regressionsequences (cyclothems) that required hundreds ofmetres of eustaticchange1–4, and was supported by early efforts to model Gondwana-land ice sheets5,6. Improved dating resolution of Gondwanan glacialdeposits7,8 and the development of geochemical proxy records9,10have led to the alternative view that continental-scale glaciation wasthe exception rather than the rule7,8. This view of the LPIA suggeststhat multiple distinct regional-scale glacial intervals (<10Myr)coincided with low (<1,000 ppmv) atmospheric pCO2 and thatintraglacial climate variability was driven by orbital insolationfluctuations that led to ice-sheet growth and retreat7,8.

The conclusion that late Palaeozoic glaciation was episodic andregional is based on the observation that Gondwanaland glaciogenicbasins are spatially disparate (located in Africa, Antarctica,Australia, India and South America) and temporally diverse7,8.This temporal heterogeneity indicates that the LPIA ice sheetswere typically not continental in scale7. This evidence, supportinglower-volume regional glaciation, has led to a re-analysis ofglacioeustatic interpretations. In a summary, 80% of all studies (64total) indicate that global sea-level changes ranged from 5-to-60mthroughout the LPIA, whereas the total range from all studies is5-to-250m, reflecting the difficulty inherent in making palaeo-eustasy calculations11. Uncertainties in palaeo-eustasy calculationsarise from a number of complicating factors, among them,accounting for changes in local/regional tectonics, predicting theevolution of shoreline geometry and estimating the gravitationaleffect of the redistribution of land-based ice and the solid Earth12.

Modelling studies have demonstrated that late Palaeozoicpalaeogeography and low pCO2 produce below-freezing summersurface temperatures that favour accumulation of ice sheets thatspan the Gondwana supercontinent5,6,13,14. Owing to feedbacksassociated with their height and high surface albedo, large ice sheetshave cold surface conditions that are not susceptible to changes

1Department of Geological Sciences, University of Michigan, Ann Arbor, Michigan 48109, USA, 2Earth and Environmental Systems Institute, ThePennsylvania State University, University Park, Pennsylvania 16802, USA. *e-mail: danethan@umich.edu.

in orbital insolation or reasonable increases in pCO2 (ref. 13).At higher pCO2, summer ablation outcompetes accumulation andonly small ice sheets are simulated. This presents a paradox;ice sheets of adequate size to account for inferred LPIA eustasyfluctuations are too stable, whereas smaller, orbitally sensitive icesheets do not represent sufficient volumes of water to accountfor observed eustatic changes13. In this study, we demonstratethat the incorporation of ecosystem feedbacks in a coupledgeneral circulation model (GCM)/ice-sheet model allows for thesimulation of ice sheets that are sufficiently large and sensitiveto orbital variations.

Climate/ice-sheet/vegetation response to orbital forcingTo evaluate the effects of changing vegetation on the LPIA, wecouple the BIOME4 ecosystem model to the GENESIS GCM/ice-sheet model used in previous late Palaeozoic simulations13,14 (seetheMethods section). GENESIS includes atmosphere, land-surface,mixed-ocean and sea-ice climate components. Experiments weredeveloped using an Early Permian (Sakmarian, about 290Myrago) palaeogeographic/palaeotopographic reconstruction15 andreduced solar luminosity in accordance with solar evolutionmodels16. Transient changes in the orbital parameter configurationare applied at 5 kyr intervals to simulate the spectrum oforbital variation observed throughout the past 10Myr (ref. 17).Atmospheric pCO2 concentrations of 420, 560 and 840 ppmv areprescribed consistent with proxy pCO2 estimates during LPIAglacial intervals9,10. To isolate the climatic effects of changingvegetation, two experiments were run for each pCO2 experimentwith (1) static land-surface vegetation or (2) the BIOME4dynamic ecosystem model.

BIOME4 ecosystem types are determined yearly, using GCMaverages ofmonthlymean temperature, sunshine and precipitation,as well as soil texture and pCO2 concentration18–20. BIOME4predicts 27 distinct modern ecosystems and has been shownto successfully simulate ecosystem distributions in the modern18

and at the Last Glacial Maximum20. The floral compositions of

572 NATURE GEOSCIENCE | VOL 3 | AUGUST 2010 | www.nature.com/naturegeoscience

© 2010 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO922 ARTICLESa

b

c e fd

c d e fIn

sola

tion

(W m

¬2 )

Ice volume (10

6 km3)

Ice

volu

me

(10

6 km

3 )

Sea-level equivalent (m)

Sea level equivalent (m)

420 ppmv

840 ppmv560 ppmv

240160800

Summer insolation

Time (kyr)

Time (kyr)

560 ppmv BIOME4 560 ppmv non-dynamic

0 40 80 120 160 200 240 280

57

47

38

28

19

9

0

540

520

500

480

400

420

440

460

20

0

40

60

80

100

Tropical forest

Warm temperate forest

Temperate forest

Boreal forest

Dry Woodland

Dry Shrubland

Desert/barren

Tundra

Land ice

24

20

16

12

8

4

0

237

47

0

95

142

190

Figure 1 | Time series of insolation, ice volume, sea level and terrestrial ecosystems. a, Southern Hemisphere (65–70◦ S) summer insolation (orange/greylines) and ice volume for the 560-ppmv-pCO2 simulations with (blue line) and without (red line) a dynamic ecosystem model. The grey line shows theactual insolation sequence used by the model, resulting from the asynchronous coupling scheme and its 5 kyr time step (see the Methods section). Thevertical grey bars correspond to the biome time-slices presented in c–f. Note that in the case with dynamic vegetation, ice-volume fluctuations typicallycorrespond with maxima and minima in insolation. b, Ice volume predicted for dynamic ecosystem experiments with 420, 540 and 840 ppmv pCO2.c–f, Early Permian (Sakmarian, about 290 Myr ago; ref. 15) biome predictions for four time intervals, 85, 95, 210 and 235 kyr. At the insolation minimum(c) the tundra ecosystem expands northward to 60◦ S. In medium insolation orbits (e) boreal forests dominate the high latitudes. During periods of highinsolation (d,f) ice-proximal barren lands expand.

late Palaeozoic ecosystems are not known in detail, but weredistinct in many respects from modern vegetation types21–25. Toaddress this issue we have eliminated grasslands as a potentialecosystem type, as they had not yet evolved26 (see SupplementaryDiscussion), while assuming that the remainder of late Palaeozoicecosystems/plant types were analogous to the physiology ofmodernecosystems/vegetation types that evolved in similar environmentalconditions. A similar approach to palaeovegetation reconstructionwas used to examine late Palaeozoic tropical climate–vegetationinteractions in low-latitude Pangaea27. For presentation purposes,BIOME4’s 27 ecosystem types have been condensed into eightmega-biomes and land-based ice in Fig. 1.

At 420 ppmv pCO2, with dynamic vegetation, multiple ice-spreading centres form over southern Gondwana and coalesceinto a single supercontinental ice sheet with an average height inexcess of 2,900m. In non-dynamic vegetation simulations with560 ppmv pCO2, ice volume (Fig. 1a) and extent are significantlyreduced, with localized ice centres concentrated mainly along thePanthalassan coast. Simulations of both dynamic and non-dynamicvegetation at 840 ppmv pCO2 predict no significant terrestrial ice. Incomparison with these experiments that produce too much or too

little ice (Fig. 1b), the 560-ppmv-pCO2 simulation with dynamicvegetation compares well to observed locations of glaciogenicdeposits7 (Fig. 2). A single ice-spreading centre forms in southernAfrica and western Antarctica, whereas discrete spreading centresform in SouthAmerica coastal Australia and northern Siberia.

At pCO2 of 420 ppmv, with dynamic vegetation, ice sheetsrapidly grow to volumes that are sufficiently large to be insensitiveto variations in orbital forcing (Fig. 1b). In the 560-ppmv-pCO2simulation without dynamic vegetation, the growth and ablationof Gondwanan ice sheets in response to orbital forcing contributesless than 2m of glacioeustatic change (Fig. 1a). In contrast, withthe incorporation of dynamic vegetation, orbitally driven glacial–interglacial glacioeustatic fluctuations of ∼33m are simulated(Fig. 1a), which represent both expansion/contraction and thick-ening/thinning of Gondwanan ice sheets. Land-based ice volumeincreases when Southern Hemisphere high-latitude (65◦–70◦ S)summer (December, January and February) insolation is at a mini-mum (mean insolation∼415Wm−2), whereas ice volume typicallydecreases when summer insolation is at a maximum (mean insola-tion ∼520Wm−2; Fig. 1a). These insolation changes drive changesin ambient temperature that are amplified by vegetation feedbacks.

NATURE GEOSCIENCE | VOL 3 | AUGUST 2010 | www.nature.com/naturegeoscience 573© 2010 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO922

1 2 3 4

A

Ice height (km)

B

C

D

E

Figure 2 |Maximum simulated Southern Hemisphere ice extent in the560-ppmv-pCO2 dynamic vegetation simulation at 175 kyr. The location ofsimulated ice is in agreement with glacial evidence found in South America(Paraná, Chaco–Paraná, Paganzo, Calingasta–Uspallata, Sauce Grande, SanRafael, Teupel and Golondrina basins) (A), Africa (Congo, Haub, Tanzanian,Zambesi, Limpopo, Kalahari, Warmbad and Karoo basins) (B), the FalklandIslands (C), Antarctica (Heimefrontfjella and Victorialand basins and theEllsworth, Pensecola and central Transantarctic mountains) (D) andAustralia (Murray, Oaklands, Tasmanian, Bowen–Gunnedah, Sydney andTamworth belt basins) (E; ref. 7).

Orbitally driven vegetation feedbacksLower-than-average summer insolation facilitates expansion oftundra into areas previously dominated by boreal forests (Fig. 1c).In BIOME4, tundra replaces boreal forest ecosystem when thesum of growing-degree-day temperatures above 0 ◦C (GDD0) fallsbelow ∼800 and the net primary productivity decreases belowthe 140 gm−2 threshold18. The replacement of two-storey borealforest with single-storey lower canopy tundra allows snow coverto persist through the summer unmasked by trees, which increasessurface albedo and cools the periglacial environment. The neteffect of tundra expansion and lengthened summer snow coverageis a 20–30% increase in summer surface albedo (Fig. 3a), whichdecreases periglacial summer temperatures by 6–10 ◦C (Fig. 3b).Vegetation change and persistent snow coverage magnify theinsolation-induced cooling, decreasing summer snowmelt (Fig. 3c)and ultimately allowing land-based ice accumulation (Fig. 1a).

Rapid ablation of land-based ice in the 560-ppmv-pCO2dynamic vegetation simulation occurs when Southern Hemispheresummer insolation is at a maximum (Fig. 1a). Warm summertemperatures facilitate the expansion of barren lands along theice margins (Fig. 1d). BIOME4 simulates barren lands when thenumber of GDD0 exceeds 800 and there is insufficient soil moistureto support boreal forests18. The expansion of non-vegetated landcover leads to a drawdown of available soil moisture (Fig. 3d)resulting in reduced summer latent heat flux and increasedsensible heating in the ice-proximal environment (Fig. 3e). Thisanomalously high sensible heat flux increases summer temperaturesat ice frontal margins by up to 4 ◦C (Fig. 3f) and ultimatelyleads to melting. Although this vegetation/ice-ablation feedbacksuccessfully leads to deglaciation during Southern Hemispheresummer insolation maxima 95, 235 and 255 kyr ago, significant iceloss is not simulated 175 kyr ago (Fig. 1a). This lack of ablation isattributed to a 5% decrease in barren land area, a 4.5% increasein ice volume and a 7% increase in ice area, compared with the95 kyr iteration (Fig. 4). Owing to an increase in local albedo anda decrease in summer sensible heat flux, these differences lead to

lower high-latitude temperatures and a reduced melt susceptibilityto orbitally driven insolation changes. This missed deglaciationunderscores the sensitivity of nonlinear systems to stochasticelements and demonstrates that subtle variations in ice extent, localclimate and ecological factors can cause glacial and deglacial cyclesto diverge from orbital pacing.

Comparison with the LPIA geologic recordOur results demonstrate that the pCO2 window allowing for largeice sheets that are sensitive to orbital fluctuations is relatively narrow(420–840 ppmv). This range of atmospheric pCO2 is consistent withcalculations based on the isotopic composition of soil carbonatesduring periods of late Palaeozoic glaciation9,10. Although ourmodel simulates ice in most late Palaeozoic glaciogenic basins7, itdoes not predict ice in northern South America, northern Africa,the Arabian Peninsula, India or western Australia. It is likelythat the lack of ice in these regions is due to our specificationof palaeogeography and palaeotopography15. (1) The Sakmarianreconstruction represents one snapshot in time, and does notaccount for the evolution of palaeogeography (for example, driftingof continents) or palaeotopography (for example, orogenic events)throughout the late Palaeozoic. (2) Climate modelling studieshave shown that uncertainties in late Palaeozoic palaeogeographicreconstructions are substantial enough to alter the climate28 andtherefore may influence the location of ice accumulation. (3) Thelocation and elevation of plateaux and alpine regions in the latePalaeozoic are not well constrained. The existence of relativelylow high-latitude plateaux that would not violate any geologicalconstraints could explain the existence of ice sheets where ourmodel failed to simulate them. To illustrate this point, we carriedout further experiments, in which plateaux of modest elevations(that is, 1,000–1,500m) were added to areas of the Australiancontinent on which glaciogenic sedimentation has been observed29.The addition of these plateaux led to the growth of small ice sheets(∼2×106 km3) in areas that were previously ice free.

Glacioeustatic fluctuations of ∼33m represent a significantproportion of the 5–60m range of observed LPIA eustasychange. Owing to the (static) palaeogeography and (conservative)palaeotopography employed in our simulations, the magnitudeof this glacial component should be considered a lower estimate.An increase in land area near the austral pole or the addition ofhighlands could increase the glacial contribution to total eustasy.For example, the Australian plateau ice-growth experimentsincrease the potential glacioeustatic change by∼5m.

Palaeovegetation studies of low-latitude depositional systemsreport dynamic ecosystem fluctuations throughout the LPIA(refs 21–24). These ecological changes are interpreted to correspondto Southern Hemisphere glacial advance and retreat. Studiesof high-latitude palaeovegetation demonstrate an interchangebetween tundra and forest ecosystems, which is interpreted to bedriven by periodic climate deterioration and amelioration25. Theseobserved palaeofloral fluctuations are consistent with the LPIABIOME4 simulations of dynamic vegetation. Ecosystem variabilityin the 560-ppmv-pCO2 dynamic vegetation simulation is regionallycomplex, but generally demonstrates aridification in low latitudescoincident with decreasing ice volume and variable dominanceof boreal forest, tundra and barren lands in the periglacialenvironment coincident with glacial advance and retreat (Figs 1b–eand 4). High-latitude periglacial vegetation fluctuations correspondto changes in orbital forcing and indicate no hysteresis behaviour.

Vegetation is king?Although orbitally driven ecosystem feedbacks can accountfor a considerable portion of the estimated sea-level fluctu-ations, it is likely that other feedback mechanisms not con-sidered here, involving greenhouse gases, ocean circulation

574 NATURE GEOSCIENCE | VOL 3 | AUGUST 2010 | www.nature.com/naturegeoscience

© 2010 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO922 ARTICLES

Temperature (°C)

Temperature (°C)Soil moisture (%) Sensible heat flux (W m¬2)

Albedo (%) Snow fraction (%)

a

d e

b c

f

¬30 ¬10 0 10 30

¬30 ¬10 0 10 30 ¬24 ¬16 0 16 24

0¬10¬8 ¬6 ¬4 ¬2 10 20 30 40 50

¬4 ¬2 0 2 4

85 kyr 85 kyr 85 kyr

95 kyr 95 kyr 95 kyr

Figure 3 | Climatic anomalies at the summer insolation minimum and maximum. The effects of ecosystem change at the insolation minimum (85 kyr;Fig. 1c) on surface albedo (a), surface air temperature (b) and snow fraction (c) and at the insolation maximum (95 kyr; Fig. 1d) on soil moisture (d),surface sensible heat flux (e) and surface air temperature (f). Southern Hemisphere summer anomalies are calculated as the difference from the averageconditions over one complete orbital cycle (85–165 kyr).

Ice volume (10

6 km3)

560 ppmv BIOME4

0 80 160 240

Ecos

yste

m c

over

age

(%)

Boreal forestTundraBarren land

100

80

60

40

20

0

24201612840

Time (kyr)

Figure 4 | Time series of global ice volume and ecosystem coverage variability from 65◦ to 70◦ S. Dominant ice-proximal ecosystem type varies with andamplifies insolation-driven temperature change. Ice volume increases when tundra expands and decreases when barren land expands.

NATURE GEOSCIENCE | VOL 3 | AUGUST 2010 | www.nature.com/naturegeoscience 575© 2010 Macmillan Publishers Limited. All rights reserved.

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO922

and/or dust, contributed to LPIA glacial–interglacial fluctuations.Plio-Pleistocene glacial–interglacial cycles, for example, were ac-companied by changes in pCO2 concentrations of up to∼100 ppmv(ref. 30). A previous study of Last Glacial Maximum climate notesthat whereas pCO2 has a larger influence on global temperature,vegetation feedbacks can dominate the temperature response onlocalized scales31. To quantify the local high-latitude surface heatresponse to changes in pCO2 during the late Palaeozoic, we rananother experiment (using orbital/biome conditions from the 90to 95 kyr time slice, Fig. 1a) in which pCO2 was increased from560 to 660 ppmv, but vegetation was not allowed to change. Thelocal surface heat response to increased pCO2 ranged from 0 to1.5Wm−2 near the ice-sheet margin. In comparison, the transitionfrom boreal forest to barren land at the same time yielded a 6 to12Wm−2 increase in surface heating near the ice margin. These re-sults indicate that variations in greenhouse gas concentrations con-tributed to orbitally paced insolation amplification, but probablyplayed a secondary role to high-latitude ecosystem change in drivingLPIA glacial–interglacial cycles. This conclusion is supported bythe observation that Plio-Pleistocene greenhouse gas variationsmostly lagged glacial–interglacial fluctuations30. The contributionsof ocean circulation changes and dust flux to late Palaeozoic glacial–interglacial changes are unresolved. Previous ocean–atmospherestudies suggest that the ocean responds dynamically to orbitalcycles, although the surface heat response to these changes istypically small (for example, ref. 32). The net radiative effect ofincreased dust flux during glacial periods is uncertain, with surfacedust deposited on ice and snow decreasing the albedo, and airbornedust both reducing the amount of insolation reaching Earth’ssurface and altering the absorption/reflection properties of clouds33.

Previous studies using either GCMs without dynamic ice-sheet models34,35 or Earth models of intermediate complexitywith36,37 and without38 dynamic ice-sheet models have exploredthe effects of vegetation on Quaternary climates. These studiesdemonstrate that vegetation feedbacks are important for glacialinception (for example, ref. 37) and suggest that the advanceand retreat of Northern Hemisphere boreal forests may haveamplified the temperature response to orbital insolation changethrough vegetation–snow–albedo feedbacks34–38. Using a coupledGCM/biome/ice-sheetmodel, this study demonstrates that orbitallyforced vegetation feedbacks amplify the ice-sheet response toboth insolation minima and maxima, and thus permit long-termglacial–interglacial fluctuations. So far, the simulation of LPIAand Plio-Pleistocene glacial–interglacial cycles has been a majorchallenge for dynamic climate/ice-sheet models13,39. The omissionof high-latitude vegetation feedbacksmight be an important reason.

MethodsThe simulations presented in this study use the Genesis version 3.0 atmosphericGCM synchronously coupled to the BIOME4 ecosystemmodel and asynchronouslycoupled to a three-dimensional dynamic ice-sheet model. The GenesisGCM includes dynamic atmosphere, land-surface, sea-ice and slab-oceancomponents40,41. The atmospheric GCM has a spectral resolution of T31(3.75◦×3.75◦) and 18 vertical levels. The land surface has a 2◦×2◦ grid spacing.BIOME4 is a coupled carbon and water flux model that predicts global steady-statevegetation distribution, structure and biogeochemistry at a 2◦×2◦ resolution18–20.The thermomechanical ice-sheet model is based on the vertically integratedcontinuity equation for ice mass and predicts ice-geometry evolution throughsurface-mass-balance, basal-melting and ice-flow calculations on a 1◦×2◦ surfacegrid42. Orbitally driven climate changes were simulated using an asynchronoustransient ice-sheet/atmosphere coupling scheme13, which consists of alternatingshort (15 yr) GCM–biome integrations and long (5 kyr) ice-sheet integrations.Between the integration of each component, updated boundary conditions(that is, meteorological information to the ice-sheet model or ice geometry tothe GCM–biome model) are passed from one model component to the other,and orbital parameters are updated. Each transient experiment was run overmultiple orbital cycles and represents 275 kyr or 55 iterations. A full discussionof individual model components and the coupling scheme can be found in theSupplementary Methods.

Received 30 May 2010; accepted 24 June 2010; published online25 July 2010

References1. Crowell, J. C. Gondwana glaciation, cyclothems, continental positioning, and

climate change. Am. J. Sci. 278, 1345–1372 (1978).2. Veevers, J. J. & Powell, C. M. Late Paleozoic glacial episodes in Gondwanaland

reflected in transgressive-regressive depositional sequences in Euramerica.Geol. Soc. Am. Bull. 98, 475–487 (1987).

3. Heckel, P. H. Sea-level curve for Pennsylvanian eustatic marinetransgressive-regressive depositional cycles along midcontinent outcropbelt, North America. Geology 14, 330–334 (1986).

4. Wanless, H. R. & Shepard, F. P. Sea level and climatic changes related to latePaleozoic cycles. Geol. Soc. Am. Bull. 40, 17–50 (1936).

5. Crowley, T. J. & Baum, S. K. Modeling late Paleozoic glaciation. Geology 20,507–510 (1992).

6. Hyde, W. T., Crowley, T. J., Tarasov, L. & Peltier, W. R. The Pangean ice age:Studies with a coupled climate-ice sheet model. Clim. Dyn. 15, 619–629 (1999).

7. Isbell, J. L., Miller, M. F., Wolfe, K. L. & Lenaker, P. A. in Extreme DepositionalEnvironments: Mega End Members in Geologic Time: Timing of Late PaleozoicGlaciation in Gondwana: Was Glaciation Responsible for the Developmentof Northern Hemisphere Cyclothems? (eds Chan, M. A. & Archer, A. W.)(Geol. Soc. Am. Spec. Pap., Vol. 370, 2003).

8. Fielding, C. R., Frank, T. D. & Isbell, J. I. in Resolving the Late PaleozoicIce Age in Time and Space: The Late Paleozoic Ice Age—a Review of CurrentUnderstanding and Synthesis of Global Climate Patterns (eds Fielding, C. R.,Frank, T. D. & Isbell, J. I.) (Geol. Soc. Am. Spec. Pap., Vol. 441, 2008).

9. Ekart, D. D., Cerling, T. E., Montañez, I. P. & Tabor, N. J. A 400 millionyear carbon isotope record of pedogenic carbonate: Implications forpaleoatmospheric carbon dioxide. Am. J. Sci. 299, 805–827 (1999).

10. Montañez, I. P. et al. CO2-forced climate and vegetation instability during thelate Paleozoic deglaciation. Science 315, 8–91 (2007).

11. Rygel, M. C., Fielding, C. R., Frank, T. D. & Birgenheier, L. P. The magnitudeof late Paleozoic glacioeustatic fluctuations: A synthesis. J. Sedim. Res. 78,500–511 (2008).

12. Kopp, R. E., Simons, F. J., Mitrovica, J. X., Maloof, A. C. & Oppenheimer, M.Probabilistic assessment of sea level during the last interglacial stage. Nature462, 863–867 (2009).

13. Horton, D. E. & Poulsen, C. J. Paradox of late Paleozoic glacioeustasy. Geology37, 715–718 (2009).

14. Horton, D. E., Poulsen, C. J. & Pollard, D. Orbital and CO2 forcing of latePaleozoic ice sheets. Geophys. Res. Lett. 34, L19708 (2007).

15. Ziegler, A. M., Hulver, M. L. & Rowley, D. B. in Late Glacial and PostglacialEnvironmental Changes-Quaternary, Carboniferous-Permian and Proterozoic:Permian World Topography and Climate (ed. Martini, I. P.) (Oxford Univ.Press, 1997).

16. Gough, D. O. Solar interior structure and luminosity variations. Sol. Phys. 74,21–34 (1981).

17. Berger, A. & Loutre, M. F. Insolation values for the climate of the last 10 millionyears. Quat. Sci. Rev. 10, 297–317 (1991).

18. Kaplan, J. O. et al. Climate change and Arctic ecosystems: 2. Modeling,paleodata-model comparisons, and future projections. J. Geophy. Res. 108,8171–8187 (2003).

19. Haxeltine, A. & Prentice, I. C. BIOME3: An equilibrium terrestrial biospheremodel based on ecophysiological constraints, resource availability, andcompetition among plant functional types. Glob. Biogeochem. Cycles 10,693–709 (1996).

20. Harrison, S. P. & Prentice, C. I. Climate and CO2 controls on global vegetationdistribution at the last glacial maximum: Analysis based on palaeovegetationdata, biome modelling and palaeoclimate simulations. Glob. Change Biol. 9,983–1004 (2003).

21. DiMichele, W. A., Pfefferkorn, H. W. & Gastaldo, R. A. Response of latecarboniferous and early Permian plant communities to climate change.Annu. Rev. Earth Planet. Sci. 29, 461–487 (2001).

22. DiMichele, W. A., Montañez, I. P., Poulsen, C. J. & Tabor, N. J. Climate andvegetational regimes shifts in the late Paleozoic ice age earth. Geobiology 7,200–226 (2009).

23. Falcon-Lang, H. J. Pennsylvanian tropical rain forests responded toglacial–interglacial rhythms. Geology 32, 689–692 (2004).

24. Falcon-Lang, H. J. et al. Incised channel fills containing conifers indicate thatseasonally dry vegetation dominated Pennsylvanian tropical lowlands. Geology37, 923–926 (2009).

25. Retallack, G. J. Carboniferous fossil plants and soils of an early tundraecosystem. Palaios 14, 324–336 (1999).

26. Strömberg, C. A. E. Decoupled taxonomic radiation and ecological expansionof open-habitat grasses in the Cenozoic of North America. Proc. Natl Acad. .Sci.102, 11980–11984 (2005).

27. Poulsen, C. J., Pollard, D., Montañez, I. P. & Rowley, D. Late Paleozoic tropicalclimate response to Gondwanan deglaciation. Geology 35, 771–774 (2007).

576 NATURE GEOSCIENCE | VOL 3 | AUGUST 2010 | www.nature.com/naturegeoscience

© 2010 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO922 ARTICLES28. Fluteau, F., Besse, J., Broutin, J. & Ramstein, G. The late Permian climate.

What can be inferred from climate modelling concerning Pangea scenariosand Hercynian range altitude? Palaeogeogr. Palaeoclimatol. Palaeoecol. 167,39–71 (2001).

29. Eyles, N., Mory, A. J. & Backhouse, J. Carboniferous-Permianpalynostratigraphy of west Australian marine rift basins: Resolvingtectonic and eustatic controls during Gondwanan glaciations.Palaeogeogr. Palaeoclimatol. Palaeoecol. 184, 305–319 (2002).

30. Stott, L., Timmermann, A. & Thunell, R. Southern hemisphere and deep-seawarming led deglacial atmospheric CO2 rise and tropical warming. Science 318,435–438 (2007).

31. Jahn, A., Claussen, M., Ganopolski, A. & Brovkin, V. Quantifying the effect ofvegetation dynamics on the climate of the last glacial maximum. Clim. Past 1,1–7 (2005).

32. Lee, S. & Poulsen, C. J. Tropical Pacific climate response to obliquity forcing inthe Pleistocene. Paleoceanography 20, PA4010 (2005).

33. Bar-Or, R., Erlick, C. & Gildor, H. The role of dust in glacial–interglacial cycles.Quat. Sci. Rev. 27, 201–208 (2008).

34. Gallimore, R. G. & Kutzbach, J. E. Role of orbitally induced changes in tundraarea on the onset of glaciation. Nature 381, 503–505 (1996).

35. Crucifix, M. & Hewitt, C. D. Impacts of vegetation changes on the dynamics ofthe atmosphere at the last glacial maximum. Clim. Dyn. 25, 447–459 (2005).

36. Meissner, K. J., Weaver, A. J., Matthews, H. D. & Cox, P. M. The role of landsurface dynamics in glacial inception: A study with the UVic Earth SystemModel. Clim. Dyn. 21, 515–537 (2003).

37. Kubatzki, C., Clauusen, M., Calov, R. & Ganopolski, A. Sensitivity ofthe last glacial inception to initial and surface conditions. Clim. Dyn. 27,333–344 (2006).

38. Claussen, M., Fohlmeister, J., Ganopolski, A. & Brovkin, V. Vegetationdynamics amplifies precessional forcing. Geophys. Res. Lett. 33, L09709 (2006).

39. Charbit, S., Ritz, C., Philippon, G., Peyaund, V. & Kageyama, M. Numericalreconstructions of the Northern Hemisphere ice sheets through the lastglacial–interglacial cycle. Clim. Past 3, 15–37 (2007).

40. Thompson, S. L. & Pollard, D. Greenland and Antarctic mass balances forpresent and doubled atmospheric CO2 from GENESIS version-2 global climatemodel. J. Clim. 10, 871–900 (1997).

41. Pollard, D. & Thompson, S. L. Use of a land-surface-transfer scheme (LSX)in a global climate model: The response to doubling stomatal resistance.Glob. Planet. Change 10, 129–161 (1995).

42. Deconto, R. M. & Pollard, D. A coupled climate-ice sheet modelingapproach to the early Cenozoic history of the Antarctic ice sheet.Palaeogeogr. Palaeoclimatol. Palaeoecol. 198, 39–52 (2003).

AcknowledgementsThis manuscript benefited from comments by T. Torsvik and the members of theUniversity of Michigan Palaeoclimate Simulation Lab. In addition, we thank W.DiMichelle and I. Montañez for conversations motivating this work. D.E.H. and C.J.P.were supported by NSF grant SGPP-0544760.

Author contributionsD.E.H. and C.J.P. designed the experiments, interpreted the results and co-wrotethe manuscript. D.P. provided and/or coupled the model components andinterpreted the results.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper on www.nature.com/naturegeoscience. Reprints and permissionsinformation is available online at http://npg.nature.com/reprintsandpermissions.Correspondence and requests formaterials should be addressed toD.E.H.

NATURE GEOSCIENCE | VOL 3 | AUGUST 2010 | www.nature.com/naturegeoscience 577© 2010 Macmillan Publishers Limited. All rights reserved.