The Magnitude and Duration of Late Ordovician–Early Silurian Glaciation

See allHide authors and affiliations

Science  18 Feb 2011:
Vol. 331, Issue 6019, pp. 903-906
DOI: 10.1126/science.1200803


Understanding ancient climate changes is hampered by the inability to disentangle trends in ocean temperature from trends in continental ice volume. We used carbonate “clumped” isotope paleothermometry to constrain ocean temperatures, and thereby estimate ice volumes, through the Late Ordovician–Early Silurian glaciation. We find tropical ocean temperatures of 32° to 37°C except for short-lived cooling by ~5°C during the final Ordovician stage. Evidence for ice sheets spans much of the study interval, but the cooling pulse coincided with a glacial maximum during which ice volumes likely equaled or exceeded those of the last (Pleistocene) glacial maximum. This cooling also coincided with a large perturbation of the carbon cycle and the Late Ordovician mass extinction.

Earth history is punctuated by glacial episodes that vary widely in their magnitude and duration (1), as well as in their effects on global biodiversity (2). Far more is known about the most recent glacial age in the Pleistocene than about older glacial episodes. The Late Ordovician–Early Silurian glaciation of the southern supercontinent of Gondwana (Fig. 1A) is unusual because it occurred during a period when atmospheric partial pressure of CO2 (pco2) was generally higher [perhaps 8 to 16 times higher (3)] than today’s pco2, was short-lived compared to subsequent Gondwanan glaciations (1), and is the only glacial episode that appears to have coincided with a major mass extinction of marine life (4) (Fig. 1B). These observations have led to suggestions that the Late Ordovician–Early Silurian icehouse represents a climate mode distinct from more recent glaciations (5), but fundamental questions about the event are still largely unresolved: Estimates of peak ice sheet volume range from ~50 to more than 250 million km3 (6) (Fig. 1A), estimates of its duration range from 35 million years (1) to less than 1 million years (5) (Fig. 1B), and it is unclear how much sea surface temperatures (SSTs) cooled in the tropical environments that hosted much of Late Ordovician biodiversity (79).

Fig. 1

(A) South polar view of a simplified Late Ordovician paleogeographic reconstruction (27), indicating the positions of Late Ordovician to Early Silurian–aged glacial deposits in Gondwana (Gw, tan solid circles and areas) and of the Laurentian (La) localities sampled for this study (red diamonds). Ba indicates Baltica. Two possible reconstructions of the Gondwanan ice sheet (6) are shown: a minimal, discontinuous reconstruction (light blue shading) and a maximal continent-spanning reconstruction (dashed blue outline). (B) Hypotheses regarding the duration of the icehouse interval: short and sharp, restricted largely or entirely to the Hirnantian (Hirn.) stage lasting as little as 500,000 years (5), and protracted, with a peak in the Hirnantian interval (1, 28). Marine invertebrate genus diversity (black spindle) (29) declined both at the beginning and at the end of the Hirnantian stage. mya, million years ago.

These uncertainties persist because few climate proxies can be reliably applied to Paleozoic rocks. Oxygen isotope ratios (δ18O) in well-preserved marine carbonate and phosphate minerals provide a useful proxy and have been widely applied in Paleozoic climate reconstruction (5, 7, 8) but suffer a fundamental limitation: The δ18O value of a mineral is influenced by both the temperature and the isotopic composition of the water reservoir from which it precipitates (δ18Owater). Consequently, without independent constraints on one or the other of these variables, interpreting δ18O trends in the stratigraphic record poses an underdetermined problem. This issue has been addressed for the Pleistocene last glacial maximum (LGM)—for example, using sediment porewater profiles (10)—but remains largely unresolved for older glaciations.

We used carbonate clumped isotope paleothermometry (11, 12) to constrain the precipitation temperatures of a suite of Late Ordovician–Early Silurian carbonates. This approach measures the state of ordering of heavy isotopes (Δ47) in carbonate minerals and is independent of the isotopic composition of water from which the minerals grew. Combination of this approach with conventional carbonate-water oxygen isotope paleothermometry thus provides a means for untangling trends in reservoir composition from those in temperature. To minimize the effects of burial diagenesis, we examined exceptionally fossiliferous and well-studied successions in the U.S. midcontinent and on Anticosti Island, Québec, Canada (13) (Fig. 1A and fig. S1), which have experienced relatively little sedimentary burial. To assay the quality of our proxy measurements, we sampled fossils from several taxonomic groups and surrounding sediments from a broad range of lithotypes and across a large range of preservation states to characterize diagenesis and vital effects (13) (figs. S2 and S3).

There is no evidence of a systematic burial overprint on Δ47 values in our data set. Similar ranges are recorded in the Vauréal Formation and the Jupiter Formation (Fig. 2A) (12) despite ~500 m of intervening strata in the Anticosti Basin (fig. S1). The highest Δ47 values (lowest temperatures) on Anticosti are not observed in the stratigraphically highest samples but rather in samples from the middle of the stratigraphic column. To confirm that Δ47 does capture known postdepositional thermal gradients, we sampled micritic carbonates from the Vauréal Formation that are intruded by a Jurassic-aged dike. Δ47 values immediately adjacent to the dike are the lowest in our entire data set, corresponding to precipitation temperatures exceeding 230°C, but inferred temperatures fall to 39°C within 14 m of the dike-country rock contact (fig. S4).

Fig. 2

(A) Δ47 values of all samples examined for this study, keyed to taxonomic group and preservation state, and plotted against stratigraphic position. (B) PC1 of log-transformed Mn, Fe, and Sr concentrations for a large subset (n = 52) of the samples shown in (A). PC1 explains 86% of variation in trace metal composition and receives strong positive loading from Mn and Fe and weak negative loading from Sr. Dashed lines indicate PC1 and Δ47 cutoffs for inclusion (shaded region) in paleoclimate reconstructions; H, Hirnantian; Rhudd, Rhuddanian; Aer, Aeronian. Ages within stages are interpolated on the basis of stratigraphic position.

Δ47 values of skeletal carbonates range from 0.631 to 0.501 (Fig. 2 and table S1), corresponding to a temperature range from 28° to 64°C. This range implies a mixture of plausibly primary and diagenetically altered phases, the latter being typically depleted in both Δ47 and δ18O (fig. S5). Because diagenetic recrystallization of calcite tends to deplete Sr and enrich Mn and Fe (14, 15), we evaluated concentrations of these metals in a large and representative subset of our samples (13) (figs. S6 to S8). Most samples fall within the “well-preserved” compositional range identified by previous studies (fig. S6), but there is a strong relationship between the first principal component (PC1) of trace metal concentrations and Δ47: High PC1 values (low Sr and high Mn and Fe, fig. S7) are associated with higher precipitation temperatures and in many cases with textural evidence of diagenetic alteration (Fig. 2B). The highest Δ47 value and lowest PC1 value associated with texturally altered samples are 0.589 and –0.285, respectively. We therefore excluded all samples that fall outside this range from our reconstructions of ocean temperature and chemistry (Fig. 2B). We included samples that were not evaluated for trace metal concentration but that have Δ47 values higher than 0.589, but similar trends result if these samples are also excluded (fig. S9). Even the best-preserved samples likely contain small amounts of dispersed recrystallized phases that cannot easily be avoided given the large sample sizes required for adequately precise measurement of Δ47. Our temperature reconstructions should therefore be viewed as maximal SST estimates, and we base trend lines on the lowest-temperature samples from each stratigraphic interval.

Reconstructed temperatures are nearly indistinguishable from each other and consistent with a narrow range from 32° to 37°C throughout most of the ~20 million years covered by our time series (Fig. 3A). SSTs in this range rarely occur in the modern tropics, but a variety of proxies record similar temperatures during Mesozoic–early Cenozoic greenhouse intervals (1618), during which atmospheric pco2 is inferred in some reconstructions to have exceeded five times present atmospheric levels (3). Temperatures below this range (28° to 31°C) occur only in samples from the Laframboise Member of the Ellis Bay Formation on Anticosti Island. This Hirnantian-aged unit (13) records a major drop in sea level (19, 20) and a large positive carbon isotope excursion (19); both are recognized globally in other sedimentary successions (4, 20). The best preserved of these successions also record exceptionally enriched δ18O values [–2 per mil (‰) to 0‰ Vienna Pee Dee belemnite (VPDB)] during Hirnantian time (8, 20, 21), as do our Laframboise Member samples (Fig. 3B and table S1). δ18O values preceding and following the Hirnantian peak are lighter but still enriched relative to the ~–5‰ (VPDB) baseline values that characterize the beginning and end of the time series (Fig. 3B). Well-preserved brachiopods and trilobites could not be extracted from the Laframboise Member, and hence our data from the Hirnantian maximum are derived exclusively from rugose corals. Δ47 is not known to be subject to disequilibrium vital effects among modern taxa (11), but such effects cannot be ruled out and there remains uncertainty regarding the possibility of a vital effect on δ18O in rugose corals (13). However, our Hirnantian δ18O values are similar to those recorded by brachiopods in contemporaneous sections (5, 8, 20, 21), and the Hirnantian excursion can be defined by using only rugose corals (fig. S10) and thus cannot be explained by systematic differences between rugose corals and other taxa.

Fig. 3

Symbols as in Fig. 2; color indicates provenance: purple, Upper Mississippi Valley; green, Cincinnati Arch; orange, Anticosti Island. Solid symbols indicate samples selected on the basis of both trace metal concentration and ∆47 criteria; open symbols indicate samples selected based only on ∆47 value. Error bars on individual samples (13) are ± 1 SE (δ18O errors are smaller than symbols) and reflect analytical precision for samples analyzed once and reproducibility for samples analyzed multiple times (table S1). Samples marked by asterisks are the lowest-temperature samples in their respective time intervals and are the basis for trendlines. (A) ∆47-derived near-surface ocean temperature trend for the early Katian to late Aeronian interval. (B) δ18O (VPDB) trend over the same interval. (C) Relative contributions of temperature and δ18Owater to changes in δ18O (∆δ18O) between successive time intervals. Bars are scaled to the magnitude of ∆δ18O, and color proportion is scaled to the relative contribution of temperature change (red) and change in the oxygen isotopic composition of seawater (blue) to ∆δ18O. (D) δ18Owater (VSMOW) trend. Dotted lines indicate δ18Owater value during the Pleistocene LGM (10) and expected δ18Owater value for an ice-free world.

Our Δ47 measurements place independent constraints on how much of the temporal variation in δ18O can be explained by temperature changes, with the remainder attributable to changes in the isotopic composition of seawater. For much of the study interval, δ18O variation is driven almost entirely by changes in δ18Owater (Fig. 3C); only during Hirnantian time can changes in temperature account for a substantial proportion of this variation.

δ18Owater estimates (Fig. 3D) fall between –1‰ Vienna standard mean ocean water (VSMOW), the value expected for an ice-free world (22), and 1‰, the value of LGM seawater (10), for most of the study interval. However, δ18Owater estimates exceed 2‰ during the Hirnantian glacial maximum. Assuming (i) the δ18O trend reflects changes in mean ocean water, (ii) the δ18O of all surface reservoirs combined has been unchanged from the Late Ordovician to the recent (13), and (iii) the δ18O of glacial ice was comparable to the LGM, these values imply that continental ice volumes during the Hirnantian maximum substantially exceeded those of the LGM (fig. S11). The mean isotopic composition of Late Ordovician glacial ice cannot be directly measured, but inferred Hirnantian ice volumes exceed those of the LGM for any mean ice value heavier than –60‰, approaching the most depleted values observed in the present day (fig. S11).

δ18Owater trends suggest multiple episodes of moderate glaciation and deglaciation throughout the mid-late Katian interval, with little evidence of substantial ice sheets before this time (Fig. 3D). The most enriched δ18Owater values before the Hirnantian peak come from the sub-Laframboise Ellis Bay Formation, the age of which has been controversial (13). Assigning this unit an early Hirnantian rather than latest Katian age would restrict δ18Owater values higher than 1‰ to Hirnantian time (fig. S12), but multiple mid-late Katian samples > 0‰ still indicate the development of substantial pre-Hirnantian ice sheets, at least transiently. Relatively high δ18Owater values also occur in latest Hirnantian samples from the lowermost Becscie Formation, which records a sharp rise in sea level and waning of the Hirnantian carbon isotope excursion. These observations reveal that latest Ordovician sea level rise represents only partial deglaciation of Gondwana. δ18Owater values consistent with moderate ice sheets persist for several million years, returning to near –1‰ by the end of the Aeronian Stage (Fig. 3D).

Our results imply that initial glaciation of Gondwana occurred with little or no cooling of the tropical oceans, that tropical SSTs exceeded the present-day range except during the Hirnantian glacial maximum, and that they warmed rapidly after the Hirnantian maximum despite the persistence of substantial continental ice volumes for several million years. This contrasts with previous work using classical oxygen isotope paleothermometry on conodont apatite from Anticosti and elsewhere (7) that reconstructed temperatures in the modern SST range for much of the Late Ordovician–Early Silurian except for cooling to ~24°C below and above the Laframboise Member (7). These estimates assumed a constant δ18Owater of –1.0‰; substituting our δ18Owater values from the same units raises inferred temperatures by, on average, 8°C. Recent revision of the phosphate-water oxygen isotope fractionation equation (23) further suggests that all conodont-derived temperature estimates should be revised upward, bringing them into the range that we observe for the Late Ordovician–Early Silurian.

We cannot rule out the possibility that the trends we observe are influenced by changes in the basin hydrology of the Taconic Foreland, but, if they accurately reflect global trends in the tropical oceans, they imply a nonlinear relationship between tropical ocean temperatures and continental ice volumes (fig. S13A). This contrasts with expectations from climate simulations using a modern continental configuration and from proxy records of the past 60 million years (13) (fig. S13, B and C). Furthermore, coexistence of substantial south polar ice sheets with tropical SSTs regionally in excess of 30°C implies a steeper meridional temperature gradient than during other major glacial episodes (12, 24). Minor glaciations inferred to have occurred under high CO2 conditions in the late Mesozoic–early Cenozoic (16, 25) may have exhibited similar gradients but were comparatively short-lived. Both of these observations could plausibly be explained by nonlinear changes in the intensity of oceanic meridional overturning circulation (26), similar to those previously invoked to explain changes in the behavior of the Hirnantian carbon cycle (4, 5, 20). Although speculative, some support for this hypothesis is provided by the coincidence of our observed cooling pulse with the globally recognized Hirnantian positive carbon isotope excursion (5, 19, 20).

Lastly, by demonstrating that tropical cooling was largely limited to the Hirnantian Stage, our results support hypotheses linking the two-pulsed nature of the Late Ordovician mass extinction to rapid climate changes at the beginning and end of this interval (4, 20).

Supporting Online Material

Materials and Methods

Figs. S1 to S13

Table S1


References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. We thank T. Raub, M. Rohrssen, and B. Gaines for assistance with field and lab work; D. Boulet and Société des établissements de plein air du Québec (SEPAQ) Anticosti for permission to work in Anticosti National Park; and B. Hunda for supplying samples. This work was funded by an Agouron Institute award to W.W.F. and D.A.F. and NSF Division of Earth Sciences awards to W.W.F. and J.M.E.
View Abstract

Navigate This Article