Decoding glacial landscapes using automated geomorphological mapping and machine learning

Melting of ice from polar ice sheets and mountain glaciers will be the largest contributor to 21st Century sea level rise, but uncertainties remain in projections of future rates and patterns of ice mass loss. The response of the cryosphere to past episodes of climatic change in Earth history provides an important analogue that can be used to help develop more robust predictions of future behaviour.

Landscapes in the Arctic, Antarctica, and mountainous regions provide a valuable record of historical glacial and fluvial erosive activity over a range of spatial and temporal scales (e.g., Rose et al., 2013; Paxman et al., 2021). This, in turn, can shed important insights into past ice extent and dynamics. However, owing to their inaccessibility, the landscape evolution and glacial history of many of these regions is poorly understood. With the recent acquisition of large subglacial topography datasets (e.g., MacGregor et al., 2021) and the development of high-resolution digital elevation models of exposed terrain (e.g., the 'ArcticDEM'), there are now significant opportunities for systematic analysis of regional- and continental-scale topography.

The aim of this project is to use automated techniques to map the morphology of subglacial and/or subaerial landscapes and in turn reconstruct patterns of erosion and past ice extent and dynamics. The student will build on recently developed methods such as continuous valley width measurement (Clubb et al., 2022) and the use of automated classification schemes to characterise subglacial environments (Jamieson et al., 2014). Geomorphological interpretations will be integrated with numerical ice sheet modelling and (where available) chronology from offshore sediment records to constrain past glacial and climatic conditions.

The project is multi-disciplinary, with opportunities for the student to develop expertise in landscape morphometric analysis, use of geostatistical techniques and machine learning, and numerical...