Quantifying the uncertainty in the Eurasian ice-sheet geometry at the Penultimate Glacial Maximum (Marine Isotope Stage 6)

Oliver Pollard, Natasha Barlow, Lauren Gregoire, Natalya Gomez, Víctor Cartelle, Jeremy C. Ely, Lachlan Astfalck

TL;DR

This paper tackles the significant uncertainty in reconstructing past ice sheet geometries, specifically for the Eurasian ice sheet during the Penultimate Glacial Maximum (PGM). By developing a novel Bayesian uncertainty quantification framework combined with a simple ice-sheet model, the authors generate a constrained ensemble of PGM ice sheet geometries, estimating a total volume of 48±8 m sea-level equivalent (SLE). This work provides crucial data for understanding past sea-level changes and offers a transferable methodology for other poorly constrained ice sheet histories.

What Problem Does This Paper Address?

Understanding past sea-level changes is critical for predicting future climate impacts. The Last Interglacial (LIG), a period warmer than today, saw significant sea-level rise, largely influenced by changes in polar ice sheets. However, accurately reconstructing these changes is complicated by "glacial isostatic adjustment" (GIA) – the Earth's crust responding to the massive weight of ice sheets. The GIA signal is driven by the timing and magnitude of ice sheet growth and decay. While reconstructions for the Last Deglaciation (leading up to the LGM) are relatively well-constrained, the ice sheet geometries from earlier glacial periods, like the Penultimate Glacial Maximum (PGM, preceding the LIG), are poorly understood due to a lack of detailed geological and chronological data. This uncertainty in PGM ice sheet geometry directly impacts our ability to model LIG sea-level changes accurately. The authors aim to reduce this uncertainty by creating new, constrained 3D reconstructions of the Eurasian ice sheet during the PGM.

Methods

The core of this research lies in a Bayesian uncertainty quantification framework applied to a simple, steady-state ice-sheet model called ICESHEET.

  • ICESHEET Model: This model is computationally efficient and requires minimal inputs: ice-sheet margins, regional topography, and basal shear stress. It generates physically plausible ice sheet geometries by calculating thickness profiles along flowlines.
  • Parameterised Shear-Stress Map: A key innovation is the development of a parameterised basal shear-stress map. This map incorporates geological information (sediment types, bedrock) and accounts for processes like ice streaming and cold-based ice. Instead of using a single value, the shear stress is represented by a range of values and associated uncertainties, allowing for a broader exploration of possible ice sheet configurations.
  • Bayesian Uncertainty Quantification (History Matching): To constrain the vast parameter space of the shear-stress map and other inputs, the authors employ a history matching approach. They first use existing reconstructions of the Last Deglaciation (GLAC-1D and ICE-6G) as "observations." ICESHEET is run with a large ensemble of parameter combinations, and Gaussian process emulators are trained to predict ice sheet volumes. Implausibility metrics are then used to identify parameter combinations that produce unrealistic ice sheet geometries (e.g., volumes inconsistent with the target reconstructions). This "NROY" (not ruled out yet) space of plausible parameters is then applied to the PGM reconstructions.
  • Glacial Isostatic Adjustment (GIA) Topography: The model incorporates GIA-driven topography changes, which are crucial for accurate ice sheet evolution, by using a simplified elastic lithosphere relaxing asthenosphere (ELRA) model.

Datasets

  • Last Deglaciation Ice Sheet Reconstructions (for calibration):
    • GLAC-1D (Tarasov et al., 2012)
    • ICE-6G (Peltier et al., 2015)
  • PGM Ice Sheet Margin Data (for input):
    • Compilation of empirical and modelled ice extent outlines for MIS 6 from Batchelor et al. (2019).
  • Topography Data:
    • RTopo-2 modern-day global topography.
  • GIA-adjusted topography provided with GLAC-1D and ICE-6G for Last Deglaciation simulations.

Key Insights and Ideas

  • The Power of Simple Models + Bayesian Framework: The research demonstrates that even a relatively simple ice-sheet model, when coupled with a robust Bayesian uncertainty quantification framework, can produce valuable and constrained results for poorly understood past ice sheets.
  • Shear Stress as a Primary Control: The study highlights the critical role of basal shear stress in determining ice sheet geometry. By parameterizing and exploring uncertainties in shear stress, the authors can effectively explore a wide range of plausible ice sheet configurations.
  • History Matching as a Constraint: Using well-established Last Deglaciation reconstructions as targets for history matching is an effective way to filter out unrealistic ice sheet geometries for earlier periods like the PGM, where direct constraints are scarce.
  • Regional Variability is Key: The Barents-Kara Sea region emerges as a key area of interest, showing both the largest mean ice volume and the highest volume uncertainty, underscoring the importance of regional analysis.

Key Results

  • PGM Eurasian Ice Sheet Volume: The reconstructed PGM Eurasian ice-sheet volume is estimated at 48±8 m sea-level equivalent (SLE). This is significantly larger (2-3.5 times) than estimated volumes for the Last Glacial Maximum (LGM).
  • Regional Volume Breakdown:
    • Barents-Kara Sea: 24±8 m SLE (largest volume and uncertainty).
  • Fennoscandia: 19±6 m SLE.
  • British-Irish sector: 1.7±0.2 m SLE (smallest volume).
  • History Matching Effectiveness: The history matching procedure successfully reduced the parameter space by 58%, leading to a more constrained ensemble of PGM ice sheet volumes.
  • Model Tendency: The study identifies a tendency for the ICESHEET model, when driven by a parameterized shear-stress map, to overestimate ice sheet thickness. The history matching process helps to correct for this bias.
  • Impact of GIA Topography: Deformed topography due to GIA can increase the total ice volume by approximately 9±4 m SLE.

Limitations and Future Work

  • Model Simplifications: ICESHEET does not explicitly model dynamic ice-sheet processes or climate forcing, which are important for defining spatial variations in ice geometry.
  • Synchronous Maximum Assumption: The current reconstructions focus on a single, synchronous maximum ice sheet extent, whereas actual glacial periods likely involved more complex, asynchronous changes.
  • Data Scarcity for PGM: The primary limitation remains the scarcity of direct chronological and geomorphological data for the PGM, making direct validation challenging.
  • Future Work:
    • Applying the methodology to other regions (e.g., Laurentide) or time periods with uncertain ice sheet histories.
  • Developing reconstructions of ice extent and thickness over a full deglacial cycle to better inform GIA models and RSL change predictions.
  • Improving the accounting for observation metric uncertainty by using original ensembles from target reconstructions.
  • Exploring the impact of different target metrics (e.g., maximum thickness) on the resulting ice sheet geometries.

Summary updated

Abstract

The North Sea Last Interglacial sea level is sensitive to the fingerprint of mass loss from polar ice sheets. However, the signal is complicated by the influence of glacial isostatic adjustment driven by Penultimate Glacial Period ice-sheet changes, and yet these ice-sheet geometries remain significantly uncertain. Here, we produce new reconstructions of the Eurasian ice sheet during the Penultimate Glacial Maximum (PGM) by employing large ensemble experiments from a simple ice-sheet model that depends solely on basal shear stress, ice extent, and topography. To explore the range of uncertainty in possible ice geometries, we use a parameterised shear-stress map as input that has been developed to incorporate bedrock characteristics and the influence of ice-sheet basal processes. We perform Bayesian uncertainty quantification, utilising Gaussian process emulation, to calibrate against global ice-sheet reconstructions of the Last Deglaciation and rule out combinations of input parameters that produce unrealistic ice sheets. The refined parameter space is then applied to the PGM to create an ensemble of constrained 3D Eurasian ice-sheet geometries. Our reconstructed PGM Eurasian ice-sheet volume is 48±8 m sea-level equivalent (SLE). We find that the Barents–Kara Sea region displays both the largest mean volume and volume uncertainty of 24±8 m SLE while the British–Irish sector volume of 1.7±0.2 m SLE is the smallest. Our new workflow may be applied to other locations and periods where ice-sheet histories have limited empirical data.
BibTeX
@inproceedings{pollard2023quantifying,
  author = {Oliver Pollard and Natasha Barlow and Lauren Gregoire and Natalya Gomez and Víctor Cartelle and Jeremy C. Ely and Lachlan Astfalck},
  title = {{Quantifying the uncertainty in the Eurasian ice-sheet geometry at the Penultimate Glacial Maximum (Marine Isotope Stage 6)}},
  booktitle = {˜The œcryosphere},
  year = {2023},
  doi = {10.5194/tc-17-4751-2023},
}
Figure