Antarctic tipping: Just how rapid was the last retreat of the Antarctic ice sheet?

Year
2024
Primary Supervisor
Dr Jochen Voss <J.Voss@leeds.ac.uk>
Institution
University of Leeds (School of Mathematics)
Possibility of becoming a CASE project.
No
Academic Supervisors
Louise Sime <lsim@bas.ac.uk> (British Antarctic Survey)
Research Themes
, ,
Research Keywords
, , , , ,
Relevant Degree Courses
, , ,

This project will use data from and modelling of past climates to better understand how recent sea ice losses will affect the Antarctic ice sheet, and thus future sea levels.

Summary

Ice mass loss from the Antarctic ice sheet is a major driver for sea level rise.  For this reason, it is important to understand the ice sheet changes.  Recently, a vast amount of Antarctic sea ice has been lost, and it is unknown how fast the Antarctic ice sheet (AIS) will melt after this.  We propose to study the AIS during the Last Interglacial (LIG, a period in the past where temperatures were 1-2ºC warmer than preindustrial) when about 50% of sea ice was lost, to explore future melting of the AIS and to develop projections for future sea level rise.

The shape and extent of the AIS during the LIG is a matter of huge importance and controversy, with different studies proposing a wide range of reconstructions.  While there is little or no direct data available about the shape of the AIS, data extracted from Antarctic ice cores can be used to constrain the shape of the AIS during this period.

The analysis consists of the following steps.

  • Isotope enabled climate simulations will be used to understand how the shape of the AIS affects isotopic measurements.  By performing a carefully chosen selection of model runs, we can explore which ice sheet shapes are compatible with the observed isotope concentrations, and which ones are not.
  • Climate model runs are computationally very expensive, and it would take prohibitively long to perform enough model runs to explore the full space of possible shapes.  Instead, a statistical emulator will be used to approximate the response of the climate model to the AIS shape.
  • Statistical analysis, using Bayesian methods, will be used together with the emulator and the known uncertainties in the isotope measurements to determine the distribution of possible shapes of the AIS, given the observed isotope values.
  • The time evolution of the isotope values is known, so the analysis from the previous step can be repeated for a range of times before and during the LIG to understand how the AIS changed during this period.  A simple time-series model will be used to describe the change of the AIS in time as a stochastic process.

Domingo et al. (2020) demonstrated that an emulator-based approach can enable outputs from isotope-enabled models, alongside isotope measurements from ice cores.  The article shows that this requires a careful parameterization of ice sheet morphologies and a thorough approach to the characterization of uncertainty. They demonstrated that the major source of uncertainties that can be tackled via the construction of suitable emulators, arises from the impossibility of running the simulator on a high number of morphologies. Additional uncertainties stem from the fact that, both in reality and in simulations, the isotopic response in the ice is dependent not only on the ice sheet morphology but also on other parameters such as evaporation conditions, transport pathway effects, and Arctic sea ice variations (Sime et al., 2009; Holloway et al. 2016; and Goursaud et al. 2021).

Objectives

Tackling the problem requires that we tackle the following objectives:

  • Compile a list of all isotopic measurements from all Antarctic cores, and carefully analyse depth-to-age and measurement uncertainties.
  • Define the range of AIS shapes relevant for the time period under consideration, and develop an appropriate parametrization of these shapes.
  • Run isotope-enabled climate model simulations for selected AIS shapes. Consider how to incorporate the impacts of sea ice and other lesser controls on the isotopes. 
  • Fit a statistical emulator to the simulation results, and perform the Bayesian analysis required to match the data with the ice core data.

Potential for High Impact

In recent years, West Antarctic Ice Sheet ice mass loss has accelerated in response to ocean warming and changes in ocean circulation. The contributions of AIS mass loss to sea level rise have increased from 0.15 to 0.46 mm/year between 1992 and 2017, accelerating to 0.30 to 0.49 mm/year during 2012-2017 (IPCC AR6).   A full collapse of the marine-based portion of the West Antarctic Ice Sheet  would raise global mean sea level by over 3 m. In addition, the destabilisation of marine based portions of the East Antarctic Ice Sheet in the Wilkes Basin could lead to further rise of about 3 m in global sea level. The likely timescales associated with these critical AIS changes are currently unknown. Information from past warm periods, where the AIS underwent past rapid retreats, is thus essential for constraining the time-scale of these longer-term (>100 year) projections of future ice mass loss – and the rate and amount of associated global mean sea level rise.

Training

The student will work under the supervision of the two supervisors, Jochen Voss and Louise Sime.  The project is mostly based at the University of Leeds, but it is anticipated that the student will spend part of the time at the British Antarctic Survey in Cambridge.  Training for the required statistical and computational skills includes the following courses:

  1. APTS will provide training in statistics.
  2. One week National Centre for Atmospheric Sciences (NCAS) Unified Model Introduction Course.
  3. Additional bespoke water tracer and isotope modelling training by the BAS code developers, Louise C. Sime and Alison McLaren, as required. 
  4. Summer schools. Various of these are provided by supervisors, and which may include EU summer schools. 
  5. The university provides training for PhD students

Entry requirements

Candidates should have a background in either statistics or applied mathematics, and should have basic knowledge about computational statistics, e.g. about Monte Carlo methods.  Experience with R or Python would be desirable.

References