Intelligent Earth Seminar: Sacha Lapins (University of Bristol)

 

Distributed Acoustic Sensing (DAS) converts ordinary fibre optic cables into extensive arrays of seismic sensors by measuring tiny strain changes along the fibre using backscattered light from laser pulses. This technology offers unprecedented spatial resolution at sub-metre scales over distances of tens of kilometres, enabling continuous monitoring and high-resolution subsurface imaging in harsh, remote or built-up environments. However, DAS data are often contaminated by strong random noise processes that make identifying or exploiting the underlying signals of interest challenging. In this talk, I present a weakly-supervised machine learning method (DAS-N2N) for suppressing strong random noise in DAS data. By exploiting the availability of multiple, redundant fibre cores within a single cable jacket, this approach uses only raw noisy data for training, mapping random noise to a chosen summary statistic, such as the distribution mean, median or mode, whilst retaining the true underlying signal. I demonstrate how this approach provides improved performance and efficiency over conventional filtering techniques using data from cables deployed on the Rutford Ice Stream in Antarctica and the Pacific Ocean floor. Additionally, I discuss how this technique can be extended to other sensing technologies, and how AI can further the utility of DAS for characterising, monitoring and managing natural and urban environments more broadly.