| Summary: | PhD Deforming subglacial tills have a significant influence on the dynamics of many glaciers and ice sheets; however, due to their inaccessibility and spatial/temporal heterogeneity, laws defining their behaviour and rheology are still contentious. A lack of analytical and theoretical continuity between exposed relict tills, those under active ice and physical and numerical models using artificial analogues is partly responsible for this. Particle fabric, the 3D orientation of individual particles, could provide a quantitative link between such approaches; however, inconsistencies and weaknesses in data collection, presentation and interpretation have led to conflicting laws governing particle dynamics and therefore subglacial till behaviour. X-ray μCT provides 3D volumetric density maps of till samples at μm-scale, allowing for the extraction of true 3D properties for large particle populations (n>5000). Typically, such investigations have used sorted or artificial sands locked in resin to simplify particle identification; natural sediments however, are compositionally and lithologically heterogenous and require a supervised approach. Machine-learning protocols are presented and tested alongside a novel method which quantifies the best possible representation of particles within a sample. A mean accuracy of 85% is achieved. By applying these protocols to samples taken from a variety of active and relict glacier/ice sheet margins, a large database of particle properties (n>280k), including orientation, shape, size, 3D position and other experimental metrics has been created. Particle fabrics generated using X-ray μCT are much weaker and subtler than those obtained through other methods; therefore a detailed investigation into presentation, statistical significance and contextual interpretation of fabric data is conducted. The role of particle properties, particularly size and shape is shown to be an important controller of fabric in tills and must be carefully considered. By applying targeted analysis ...
|
|---|