Patterns and drivers of ice dynamic variability on the Antarctic Peninsula from satellite data and deep learning

In the period of satellite observations, the Antarctic Ice Sheet has undergone significant changes, both in terms of the atmosphere and ocean climate it experiences and its ice dynamics. This is particularly true of the Antarctic Peninsula, the most northerly and warmest region of Antarctica, where major ice shelves have collapsed, glaciers have retreated and accelerated, and ice mass loss has contributed to global sea-level rise. Despite observations of these changes, the Antarctic Peninsula Ice Sheet remains challenging to study, because of its pace of change, extreme topography and severe weather. In this thesis, I leverage the large volumes of data generated by the latest generation of remote sensing missions, combined with newly developing techniques of artificial intelligence and deep learning, to investigate the patterns, changes, and drivers of ice dynamics in the Antarctic Peninsula Ice Sheet.
The work of this thesis begins with a study of the short-term variability in ice dynamics of the glaciers of the west Antarctic Peninsula coastline. Using Sentinel-1 synthetic aperture radar to measure ice velocity for 105 glaciers and a Bayesian recursive smoother to post-process data, I show that seasonal ice flow speed-up is widespread in this region. Peak seasonal ice speed variability is 288.3 ± 41.3 m yr-1 or 22.3 ± 3.2%, with an average value of 12.4 ± 4.2%. I use data from a regional climate model and ocean physics reanalysis to show that peak ice speeds coincide with ocean thermal forcing and meltwater runoff, demonstrating that these glaciers can respond to forcings in the ice-ocean-atmosphere system on seasonal timescales. I conclude this chapter by showing that seasonal ice flow fluctuations must be accounted for in ice mass balance calculations.
Following on, I extend the time-series of ice velocity using additional measurements from optical and radar imagery, to study the rapid acceleration and retreat of Cadman Glacier on the Peninsula’s west coast. My observations show that this glacier had been stable prior to a sudden acceleration of 1.47 ± 0.6 km yr-1 from November 2018 to December 2019, followed by the total collapse of the glacier’s 8 km floating ice tongue in March 2021. This acceleration led to state of significant dynamic imbalance with an ice discharge increase of 28.1 ± 11.4 % and a dynamic thinning on grounded ice of 20.1 ± 2.6 m yr-1. With oceanographic and bathymetric data, I show that the thinning and acceleration of Cadman Glacier was caused by incursions of warm Circumpolar Deep Water onto the continental shelf, where a deep channel allowed warm water to access Cadman Glacier’s grounding zone.
Motivated by a lack of up-to-date grounding line position measurements in many areas of the Antarctic Peninsula, I then develop a new method for measuring grounding line position, using a time-series approach applied to tidal motion in synthetic aperture radar offset tracking results. I apply this method, called tidal motion offset correlation (TMOC), to the Antarctic Peninsula Ice Sheet to derive a new grounding line position dataset for the whole Peninsula, in some locations updating grounding lines which were last measured from tidally sensitive techniques in the 1990s. With this dataset, I observe a maximum grounding line retreat from 1996 to 2019 of 16.3 ± 0.5 km.
Finally, earlier results in this thesis raise the question of whether surface meltwater runoff can reach the glacier bed and influence ice dynamics in the Antarctic Peninsula. To address this, I develop a deep learning methodology to detect subglacial meltwater plumes in the region. I demonstrate the performance of this model against unseen data and use it to process Sentinel-2 optical satellite imagery from 2015 to 2022 to map meltwater plumes. The results show that plumes are widespread around the northern Antarctic Peninsula and correlated with modelled meltwater runoff, providing strong evidence that surface meltwater does reach the glacier bed and hence contributes to the summertime acceleration of glaciers in this region.
In summary, this thesis develops our understanding of the ice dynamics of the Antarctic Peninsula Ice Sheet by making new observations of the patterns of ice dynamic variability in the region, by producing new methods and datasets to investigate these patterns and investigating the connection between ice dynamic changes and environmental forcings.