Investigating the internal structure of glaciers and ice sheets using Ground Penetrating Radar
Delf, Richard John
Ice penetrating radar (IPR) is a key tool in understanding the internal geometry and nature of glaciers and ice sheets, and has widely been used to derive bed topography, map internal layers and understand the thermal state of the cryosphere. Modern glacier and ice-sheet models facilitate increased assimilation of observations of englacial structure, including glacier thermal state and internal-layer geometry, yet the products available from radar surveys are often under-utilised. This thesis presents the development and assessment of radar processing strategies to improve quantitative retrievals from commonly acquired radar data. The first major focus of this thesis centres on deriving englacial velocities from zero-offset IPR data. Water held within micro- and macro-scale pores in ice has a direct influence on radar velocity, and significantly reduces ice viscosity and hence impacts the long-term evolution of polythermal glaciers. Knowledge of the radar velocity field is essential to retrieve correct bed topography from depth conversion processing, yet bed topography is often estimated assuming constant velocity, and potential errors from lateral variations in the velocity field are neglected. Here I calculate the englacial radar velocity field from common offset IPR data collected on Von Postbreen, a polythermal glacier in Svalbard. I first extract the diffracted wavefield using local coherent stacking, then use the focusing metric of negative entropy to deduce a local migration velocity field from constant-velocity migration panels and produce a glacier-wide model of local radar velocity. I show that this velocity field is successful in differentiating between areas of cold and temperate ice and can detect lateral variations in radar velocity close to the glacier bed. The effects of this velocity field in both migration and depth-conversion of the bed reflection are shown to result in consistently lower ice depths across the glacier, indicating that diffraction focusing and velocity estimation are crucial in retrieving correct bed topography in the presence of temperate ice. For the thesis’ second major component I undertake an assessment of automated techniques for tracing and interpreting ice-sheet internal stratigraphy. Radar surveys across ice sheets typically measure numerous englacial layers that can be often be regarded as isochrones. Such layers are valuable for extrapolating age-depth relationships away from ice-core locations, reconstructing palaeoaccumulation variability, and investigating past ice-sheet dynamics. However, the use of englacial layers in Antarctica has been hampered by underdeveloped techniques for characterising layer continuity and geometry over large distances, with techniques developed independently and little opportunity for inter-comparison of results. In this paper, we present a methodology to assess the performance of automated layer-tracking and layer-dip-estimation algorithms through their ability to propagate a correct age-depth model. We use this to assess isochrone-tracking techniques applied to two test case datasets, selected from CreSIS MCoRDS data over Antarctica from a range of environments including low-dip, continuous layers and layers with terminations. We find that dip-estimation techniques are generally successful in tracking englacial dip but break down in the upper and lower regions of the ice sheet. The results of testing two previously published layer-tracking algorithms show that further development is required to attain a good constraint of age-depth relationship away from dated ice cores. I make the recommendation that auto-tracking techniques focus on improved linking of picked stratigraphy across signal disruptions to enable accurate determination of the Antarctic-wide age-depth structure. The final aspect of the thesis focuses on Finite-Difference Time-Domain (FDTD) modelling of IPR data. I present a sliced-3D approach to FDTD modelling, whereby a thin 3D domain is used to replicate modelling of full 3D polarisation while reducing computational cost. Sliced-3D modelling makes use of perfectly matched layer (PML) boundary conditions, and requires tuning of PML parameters to minimise non-physical reflections from the model-PML interface. I investigate the frequency dependence of PML parameters, and establish a relationship between complex frequency stretching parameters and effective wavelength. The resultant parameter choice is shown to minimise propagation errors in the context of a simple radioglaciological model, where 3D domains may be prohibitively large, and for a near-surface cross-borehole survey configuration, a case where full waveform inversion may typically be used.