Richard Blaustein discusses a recent breakthrough in the numerical modelling of transient and permanent palaeo-Antarctic ice sheets
At the Eocene/Oligocene Transition (EOT), around 34 million years ago (Ma), a continental-scale ice sheet developed in Antarctica. Many scientists consider the EOT to be the most consequential transformation of the Earth system during the Cenozoic—the current geological era going back 65 million years (for example, DeConto & Pollard, 2003; Bagniewski et al., 2022). The vast ice sheet is believed to have had a major effect on global climate and sea level, marking the start of the current icehouse conditions (when ice is present on Earth) in which we still live. While the ice sheet likely diminished in the subsequent Miocene (23 – 5 Ma), the Antarctic has apparently been continuously glaciated ever since the EOT.
Long-standing discussions on the factors or forcing mechanisms controlling the growth of the Antarctic ice sheet at the EOT point towards the importance of a declining trend in atmospheric CO2 concentrations throughout the late Eocene to Oligocene. Additionally, variations in Earth’s orbit play a significant role, while the thermal isolation of the Antarctic continent by the Antarctic Circumpolar Current would have also strengthened the stability of the glaciation. Once established, such large-scale polar ice sheets are thought to persist because positive feedback mechanisms maintain the glaciation and the thresholds (or tipping points) for ice-sheet growth and decay are different—a process known as hysteresis (see box 1). Hysteresis attracts much interest in climate science because it features in diverse biomes and regions, including the Amazon rainforest, as well as in reconstructions of the past and in future scenarios for Antarctica, and helps to frame abrupt change on societal time scales. In their paper looking at the different trajectories for the future of Antarctic hysteresis in context of global warming, Garbe and colleagues (2020) write, “Each of these thresholds give rise to hysteresis behavior: that is, the currently observed ice-sheet configuration is not regained even if temperatures are reversed to present-day levels.” Hysteresis also encompasses the reverse: once Antarctic glaciation is established at a given CO2 level (along with other factors), deglaciation would occur at a higher CO2 level than when the glaciation took hold.
Box 1 | Hysteresis
Hysteresis is the dependence of the state of a system on its history. For ice sheets this means that the thresholds that enable glaciation are different from those that enable deglaciation. In the (grossly simplified) illustration here, which is adapted from Garbe et al., (Nature 2020; https://doi.org/10.1038/s41586-020-2727-5), once the threshold of (for example) CO2 concentrations and global temperature for ice-sheet formation is reached, the expansion becomes self-reinforcing and accelerates until a new equilibrium state is reached. If CO2 concentrations and global temperature then return to their previous levels, the ice sheet doesn’t immediately decay to its former state. Instead, even greater atmospheric CO2 concentrations and warmer global temperatures are required to establish large-scale melt—the glaciation and deglaciation thresholds are different.
While research has illuminated the factors that controlled Antarctic ice-sheet growth during the EOT, challenging questions persist. One looming riddle concerns isotopic and geological evidence (such as the presence of ice-rafted debris near Antarctica) for transient continental-scale glaciations earlier in the Eocene. Given our understanding of hysteresis, how then did we get large-scale, episodic glaciations? And how did glaciation finally take hold at the EOT?
Such transient glaciations had eluded climate modelling. However, Jonas Van Breedam and Philippe Huybrechts at the Vrije Universiteit Brussel, Belgium, together with Michel Crucifix at the Earth and Life Institute at UCLouvain, Belgium, have now modelled these specific palaeo enigmas (Van Breedam et al., 2022a, 2022b). Their results show that transient continental-scale glaciations can be reproduced computationally when specific thresholds for atmospheric CO2 concentrations, solar radiation, and Antarctic bedrock topography are met.
A novel modelling approach
Modelling glaciations is challenging because the extended timeframe for ice-sheet development and evolution must be matched with the rapid timeframe for atmospheric processes. “You face modelling issues every time you have to consider processes that involve very different timescales,” Michel Crucifix explains, highlighting that ice-sheet build-up takes tens of thousands of years, while the atmosphere dynamics typically occur over a matter of weeks, even days. “To get the best of both worlds, we need an ice-sheet model that has enough physics to be realistic, to grow and shrink with the right timescales, and we need an atmosphere model that has enough physics to capture the nonlinear impacts of the growth and decay of the ice sheet on precipitation, the jet stream and the like,” Crucifix says.
So, the researchers developed and employed the intermediary Gaussian Process Emulator CLISEMv1.0 model that couples the HadSM3 climate model with the AISMPALEO dynamical ice-sheet model (Van Breedam et al., 2021) for both the pre-EOT and EOT glaciations. Crucifix describes the emulator as a sort of ‘meta-modelling’ strategy that statistically analyses the atmosphere data for the data points that the ice-sheet model can employ, while Jonas Van Breedam, whose PhD research included work on creating the Gaussian Process Emulator, explains that the Emulator is “a statistical tool that grasps the climatic output when you do a large number of climate model runs for a wide variety in the forcing parameters”. Doing so, the Emulator can estimate the climate for any input combination in the forcing parameters without the need to run the climate model again. This reduces the need for multiple model runs, which are computationally expensive, and helps fill any gaps in the data needed for palaeo-Antarctic modelling.
Once set-up, the Emulator allows the exchange of information with the ice-sheet model in a computationally very efficient way, “an additional advantage for paleoclimatic modelling for computational limits that are challenging”, Van Breedam adds.
Van Breedam and colleagues applied this methodology for conditions relevant to the pre-EOT, for the period between 38 Ma to 32 Ma. The researchers used as model inputs a synthesis of recent CO2 estimates, orbital parameters, and two sets of bedrock topographic reconstructions with different maximum and minimum elevations. They conducted one hundred HadSM3 climate model runs with an atmospheric CO2 range of 550 to 1,150 parts per million by volume (ppmv).
transient continental-scale glaciations could have occurred at about 37 Ma and 35 Ma, with key external forcings being atmospheric CO2 levels and planetary orbital parameters
The model results show that transient continental-scale glaciations could have occurred at about 37 Ma and 35 Ma, with key external forcings being atmospheric CO2 levels and planetary orbital parameters. Specifically, a pre-condition for large-scale glaciations is a low summer insolation because the amount of solar radiation reaching Antarctica during the southern hemisphere summer determines whether or not ice is melting there. This condition, in combination with low atmospheric CO2 concentrations, could have been achieved at about 37 Ma, leading to Antarctic ice-sheet expansion. Importantly, due to the movement of Earth’s perihelion (the point closest to the Sun) along the orbit and continuous variations in the obliquity of Earth’s orbit, summer insolation rose to a level that would have halted the glaciation within ten thousand years (ka). Continental-scale growth of the ice sheet is only possible if the eccentricity of Earth’s orbit remains small for a very long time, offsetting the shorter orbital variations. Eccentricity minima occur roughly every 100 ka, but times of sustained minima occur less frequently, mostly every 400 ka, with even longer eccentricity minima occurring every 2,400 ka. With low eccentricity, sustained low levels of solar radiation can then trigger a robust continental-scale glaciation, provided the atmospheric CO2 concentration is low enough: “The orbitals and the CO2 are forcing agents that have similar effect so that when you combine them the effect can be stronger,” Van Breedam says.
The results also show that the threshold at which the atmospheric CO2 levels can trigger a glaciation depends on the bedrock topography—higher bedrock topography enables ice-sheet expansion under a lower CO2 threshold because ice forms more easily at higher, cooler altitudes. This implies that during pre-EOT times, large-scale glaciations would have been more likely to be enabled if the Antarctic had high bedrock topography. The decline of the ice sheets then occurred when “CO2 values exceeded the bedrock-dependent CO2 threshold” (Van Breedam et al., 2022a). Overall, the transient glaciations suggest a strong hysteresis effect leading to the growth of ice sheets, because it was found that the forcing required to melt the ice sheet needs to be significantly higher than the forcing required to induce the ice-sheet growth.
The researchers also employed their modelling technique to the EOT itself, with the results presented by Van Breedam at the European Geosciences Union (EGU) meeting in 2022 (Van Breedam et al. 2022b). The results show that the bedrock topography again had a large influence on the hysteresis effect. The researchers also investigated the effect of isostacy, whereby the ice sheet’s weight deflects the bedrock beneath, lowering potential ice-sheet height and effecting temperature (the elevation feedback). They found isostacy to be prominent negative feedback that slows ice-sheet growth, while enhancing hysteresis (that is, widening the glaciation-deglaciation threshold difference).
Van Breedam emphasizes that thresholds are much more pronounced when the ice–albedo feedback is adequately considered. When tundra vegetation is replaced by ice, the amount of reflected solar radiation is increased, and this has a significant effect on the surface temperatures with local cooling. Previous studies of long-term climate–ice sheet feedbacks did not properly account for this feedback, Van Breedam says. In his EGU presentation, he offered the novel quantification that when an ice sheet attains 30 % coverage of Antarctica, there is a rapid increase in ice formation, underscoring the nonlinear impact of the ice–albedo feedback.
Crucifix emphasizes that the findings from the pre-EOT continental glaciation study “basically change our sense of what a tipping point is” because tipping point hysteresis has been understood as “you cross a threshold, and you get an all or nothing phenomenon.” However, the team’s work shows that this is not the case. Crucifix explains, “We know there is hysteresis behavior in ice sheets because there are good physical reasons — ice–albedo feedback, the elevation feedback and all the like. But here, we have a proper ice-sheet model that has hysteresis behavior, but we are outside of equilibria because we have transient dynamics, and it is no longer an all or nothing phenomenon.”
Alexander Robinson, a climate scientist and cryosphere modeller at the Complutense University of Madrid, Spain, and not affiliated with the study, feels the researchers’ Emulator innovation is very auspicious for modelling past ice-sheet dynamics. “I think the Gaussian Emulator does it in a way that is robust and much more effective than the past method we had that was just linearly interpolating between [climate] snap shots,” Robinson says. He adds that modellers looking at the future stability of Antarctica will also gain from Van Breedam, Huybrechts, and Crucifix’s findings and technique, especially because it matches geological evidence. “If we are going to have an ice-sheet model and run it for the future, we need to be confident that it is able to show us what happened in the past in a robust way,” Robinson says. “The more pieces of the puzzle of the past that the model does well gives us more confidence that we are doing things right in the future.”
Richard is a science, environmental, and legal journalist based in Washington, DC, USA.
This report is based on work presented in Van Breedam et al. (Earth Planet. Sci. Lett. 586, 117532, 2020) and at the 2022 European Geosciences Union meeting (Van Breedam, J., et al., EGU General Assembly 2022, EGU22-1501), as well as discussions with Jonas Van Breedam at the Vrije Universiteit Brussel, Belgium, Michel Crucifix at the Earth and Life Institute at UCLouvain, Belgium, and with Alexander Robinson at the Complutense University of Madrid, Spain.
- Bagniewski, W. et al. (2022) Paleoclimatic tipping points and abrupt transitions: An application of advanced time series analysis methods. EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12501; https://doi.org/10.5194/egusphere-egu22-12501
- DeConto, R. &Pollard, D. (2003) Rapid Cenozoic glaciation of Antarctica induced by declining atmospheric CO2. Nature 421, 245–249; https://doi.org/10.1038/nature01290
- Garbe, J. et al. (2020) The hysteresis of the Antarctic Icesheet. Nature 585, 538–544; https://doi.org/10.1038/s41586-020-2727-5
- Staal, A. et al. (2020) Hysteresis of tropical forests in the 21st century. Nature Communications 11, 4978; https://doi.org/10.1038/s41467-020-18728-7
- Van Breedam, J., Huybrechts, P. & Crucifix, M. (2021) A Gaussian process emulator for simulating icesheet–climate interactions on a multi-million-year timescale: CLISEMv1.0. Geoscientific Model Development 14, 6373–6401; https://doi.org/10.5194/gmd-14-6373-2021
- Van Breedam, J., Huybrechts, P. & Crucifix, M. (2022a) Modelling evidence for late Eocene Antarctic glaciations. Earth and Planetary Science Letters 586, 117532; https://doi.org/10.1016/j.epsl.2022.117532
- Van Breedam, J., Huybrechts, P. & Crucifix, M. (2022b) Hysteresis and orbital pacing of the early Cenozoic Antarctic icesheet. EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-1501; https://doi.org/10.5194/egusphere-egu22-1501