The Antarctic Ice Sheet is currently losing mass. In large part, this is due to elevated flux of ice across grounding lines in West Antarctica, without the changes in accumulation over the ice sheet required to balance this. Over the last three decades, it has become apparent that the dynamics of these ice streams are highly sensitive to changes at their ocean margins, including to the floating ice shelves that extend out in front of many of them. So much so, in fact, that the dynamic imbalance of the ice sheet as a whole can be largely explained by a decrease in the amount of ice shelf buttressing in West Antarctica over recent decades, driven by enhanced sub-shelf melting due to incursions of warmer ocean waters onto the continental shelf. We have a good mechanical understanding of this process thanks to our treatment of marine ice sheet dynamics as a theory of viscous flow.
However, the processes of material failure, e.g. brittle fracturing, are not included in this theory but are important in determining boundary conditions and material properties at the ice sheet margin. For example, fracturing processes are responsible for the calving of ice, which determines the position of the margin itself, and the presence of fractures alters the ability of ice to bear and transmit membrane stresses.
This thesis deals with certain questions regarding the fracturing of ice, and, more generally, the sensitivity of ice stream dynamics to changes in material properties near the margin. I develop new methods for the extraction of crevasse data from satellite radar imagery, use the derived datasets to investigate changes to the structural integrity of ice shelves in the Amundsen Sea Embayment along with ice speed datasets from remote sensing imagery. I develop the capabilities of the BISICLES ice sheet model to solve inverse problems, and perform sensitivity analysis and diagnostic modelling, to investigate the sensitivity of ice dynamics to buttressing by landfast sea ice, and investigate links between observations of fracturing and isotropic ‘damage’ - a scalar field aiming to characterise how fractures change the large-scale material properties of the ice.
In chapter 2, I develop deep learning-based computer vision methods for the extraction of crevasses and calving fronts from Sentinel-1 synthetic aperture radar (SAR) imagery of the Thwaites Glacier Ice Tongue. I examine the concurrency of structural change and large fluctuations in ice speed in this area, measured using observations of ice speed derived from Sentinel-1 SAR data. By solving inverse problems for the softness of the ice shelf given observations of ice speed, I show that the quantity of rifts matches well the timeseries of ice softness in the shear margin - showing a link between the two and indicating that crevassing is dynamically important for the region. I show that the increases in crevassing and acceleration of the ice tongue reverse on timescales of years, indicating that any feedback between the two is not the dominant cause of change in the region.
In Chapter 3, I develop the crevasse-mapping procedure of chapter 2 into a process that can be used to generate pan-continental mosaics of crevasses on grounded and floating ice. This involves the use of deep learning methods similar to those of chapter 2, and the new ‘parallel structure filtering’ algorithm. I go on develop a method for quantifying structural change on ice shelves by measuring (roughly speaking) the local density of fractures. I show, in line with other literature, that there have been structural changes in parts of the Pine Island, Thwaites and Crosson Ice Shelves that contribute to ice shelf buttressing.
Chapter 4 is a bit of a detour from the theme of crevasses. Instead, through a case study on the Larsen-B Embayment, I look at whether landfast sea ice can provide buttressing to glaciers/ice shelves in the way that ice shelves provide buttressing to ice streams. I present satellite data showing the evacuation of landfast sea ice from the Larsen-B Embayment at the start of 2022, the subsequent disintegration of the Hektoria/Green/Evans and Crane Ice Shelves, and the acceleration by ∼ 100 m/yr of Hektoria, Green and Crane Glaciers. I use diagnostic modelling and sensitivity analysis, with a representation of the landfast sea ice as a thin meteoric ice shelf, to suggest that the observed changes to the glaciers are not due to a loss of direct mechanical buttressing by the landfast sea ice.
In Chapter 5, I focus on the process of solving inverse problems for the control fields representing slipperiness at the base of grounded ice, and the material softness of the ice from observations of ice speed using BISICLES. This is an ill-posed problem suffering underdeterminedness and poor conditioning. I introduce two methods of using the crevasse data generated using the methods described in chapter 3, along with data on surface strain rates, as prior information to help with this, and apply it to a case study focused on Pine Island Glacier. I show that, by doing so, one can arrive at softness fields that resemble the crevasse patterns seen on floating ice without harming the solution misfit. However, I also show that the use of crevasse data on grounded ice does not result in more plausible-looking solutions, indicating that the crevasses here do not dominate our uncertainty in the rheology of ice.
