Day 4

Detailed paper information

Back to list

Paper title Automated extraction of high resolution damage data over the Antarctic ice sheet
Authors
  1. Trystan Surawy-Stepney University of Leeds, Leeds, UK Speaker
  2. Anna E. Hogg University of Leeds, Leeds, UK
  3. Stephen L Cornford Swansea University
Form of presentation Poster
Topics
  • C1. AI and Data Analytics
    • C1.07 ML4Earth: Machine Learning for Earth Sciences
Abstract text Understanding how regions of ice sheet damage are changing, and how their presence alters the physics of glaciers and ice shelves, is important in determining the future evolution of the Antarctic ice sheet. Ice dynamic processes are responsible for almost all (98%) of present day ice mass loss in Antarctica (Slater et al 2021), with ice fracturing and damage now known to play an important role in this process (Lehrmitte et al, 2021). Though progress has been made, damage processes are not well integrated into realistic (as opposed to highly idealized) ice sheet models, and quantitative observations of damage are sparse.

In this study we use a UNet (similar to Lai et al 2020) to automatically map crevasse-type features over the whole Antarctic coastline, using the full archive of synthetic aperture radar (SAR) imagery acquired by Sentinel-1. SAR data is well suited to the task of damage detection as acquisitions are light- and weather-independent, and C-band radar can penetrate 1−10m into the snow-pack, depending on its composition, revealing the presence in snow-bridged crevasses. Our small version of UNet, trained on a sparse dataset of linear features, provides a pixel-level damage score for each Sentinel-1 acquisition. From this we produce an Antarctic-wide map of damage every 6 days, at 50m resolution. This dataset is used to measure the changing structural properties of both the grounded ice sheet, and floating ice shelves of some of the largest glaciers in the world.

Due to the slow rate of change of the Antarctic ice sheet, simulations of its evolution over century timescales can be sensitive to errors in the prescribed initial conditions. We use our observations of damage to provide a more robust estimate of the initial state of the Antarctic ice sheet using the BISICLES ice sheet model. This type of model requires both an initial ice geometry, which can be observed directly, and model parameters: basal slipperiness C(x,y) and effective viscosity μ(x,y), which cannot. Both C(x,y) and μ(x,y) are typically found by solving an inverse problem, which is undetermined. We use the damage observations to regularize the inverse problem by providing constraints on μ(x,y). This represents a step change in reducing the under-determinedness of the inverse problem, giving us higher confidence in the initial conditions provided for simulations of the ice sheet as a whole.


[1] Lai, C.-Y., Kingslake, J., Wearing, M. G., Chen, P.-H. C., Gentine, P., Li, H., Spergel, J. J., and van Wessem, J. M.: Vulnerability of Antarc-tica’s ice shelves to meltwater-driven fracture, Nature, 584, 574–578, 2020.
[2] Lhermitte, S., Sun, S., Shuman, C., Wouters, B., Pattyn, F., Wuite, J., Berthier, E., and Nagler, T.: Damage accelerates ice shelf in-stability and mass loss in Amundsen Sea Embayment, Proceedings of the National Academy of Sciences, 117, 24 735–24 741,https://doi.org/10.1073/pnas.1912890117, 2020.
[3] Slater, T., Lawrence, I. R., Otosaka, I. N., Shepherd, A., Gourmelen, N., Jakob, L., Tepes, P., Gilbert, L., and Nienow, P.: Earth’s ice imbalance, The Cryosphere, 15, 233–246, 2021.