Predictions of ice-sheet mass loss, and therefore predictions of global sea level rise, depend sensitively upon how ice-sheet motion is incorporated into numerical models. Using field observations and numerical modelling, this thesis demonstrates that two frequently overlooked processes are central to describing borehole observations of fast ice-sheet motion --- intermediate-scale (<25 m, ⪅2 km) interaction of ice motion with realistic or real bed topography, and modulation of these ice-motion patterns through a basal layer of temperate ice (much softer ice at the pressure-melting point). I first present a fibre-optic data set from a 1,043 m deep borehole drilled to the base of the fast-moving (>500 m a‾¹) marine-terminating Sermeq Kujalleq (Store Glacier) at the western margin of the Greenland Ice Sheet. This reveals hitherto unappreciated complexity in the processes behind fast ice-sheet motion. I observe substantial but isolated strain heating ~220 m beneath the surface within stiffer interglacial-phase ice where previously none was expected. Ice deformation within glacial-phase ice below 889 m is further observed to be strongly heterogeneous, with a possible high-strain interface demarcating the Last Glacial-Interglacial Transition. I also find a 73-m-thick temperate basal layer, notably thicker than the <10-m-thick temperate layer just 8.9 km away, unexplained by existing theory, and interpreted to be important for the glacier's fast motion.

To disentangle this observed complexity, I then model three isolated 3D domains from the Greenland Ice Sheet's western margin --- two from Sermeq Kujalleq and one from the land-terminating Isunnguata Sermia, all centred above a central borehole observation. By incorporating high-resolution realistic geostatistically simulated topography, I demonstrate that a layer of basal temperate ice with spatially highly variable thickness forms naturally in both marine- and land-terminating settings, alongside ice-motion patterns which are far more complex than previously considered. I show that temperate ice is expected to be vertically extensive in deep troughs, but to thin over bedrock highs. I further show that basal-slip rates are interconnected with this variability, reaching >90% or <5% of surface velocity dependent on setting. Last, I apply the assembled model to real high-resolution bed topography data produced by radio-echo sounding at Thwaites Glacier, Antarctica. This reveals a distinct pattern of ice motion controlled by rough topographic highs where basal slip rates are highly variable, and the landscape is predominantly erosive, with broader topographic basins where basal slip is high and uniform, and the landscape is predominantly depositional. This work further suggests the existence of basal temperate ice layer beneath Thwaites Glacier, at least at the rougher topographic highs. Overall, this thesis advances understanding of how ice sheets move, which may ultimately lead to improved parameterisations of ice-sheet motion for predictive models.