Adam Booth , Poul Christoffersen, Andrew Pretorius, Joseph Chapman, Bryn Hubbard , Emma C Smith, Sjoerd de Ridder, Andy Nowacki , Bradley P Lipovsky, Marine Denolle

The recognition of Distributed Acoustic Sensing (DAS) as a valuable tool for glaciological seismic applications is growing. However, besides the logistical challenges of installing fibre-optic cable, the volume of DAS data that can be collected in a field campaign poses computational challenges. In this paper, we show the potential of active-source DAS to image and characterise subglacial sediment, 20-30 m thick, beneath a fast-flowing Greenlandic outlet glacier, but highlight the difficulty of analysing a counterpart 3-day (9 TB) record of cryoseismicity. We describe experiments with data compression using the frequency-wavenumber (f-k) transform, that provides ~300-times improvement in the computational efficiency of the detection of cryoseismic events via a convolutional neural network. In combining active and passive-source and the machine learning framework, the potential of large DAS datasets can be unlocked for a range of future applications.