Deep learning for locating icequakes in sea ice

Durations: 5-6 months; dates to be determined.

Scientific context

The Arctic is changing rapidly, and understanding how sea ice responds to environmental forcing is becoming increasingly important. In the Marginal Ice Zone, where ocean waves interact strongly with sea ice, waves can trigger fractures and generate small seismic events known as icequakes. Locating these events and understanding how they propagate through the ice can provide valuable information on ice mechanical properties and on the processes controlling sea-ice breakup.

Seismic methods offer a unique way of monitoring these processes at very fine spatial and temporal scales. However, extracting information from seismic waveforms is challenging: different types of waves (flexural, longitudinal and transverse) need to be identified, and the location of the icequake source must then be determined. Because propagation distances in sea ice are often short, conventional earthquake-detection and location methods can perform poorly.

Deep learning offers a promising alternative. Recent advances in artificial intelligence have led to powerful tools for detecting and identifying seismic phases, even in noisy and complex waveforms. During this internship, you will work with real seismic recordings from sea ice and investigate how different approaches can be used to detect and locate icequakes.

Tasks during the internship

• implement and evaluate conventional detection methods such as STA/LTA and kurtosis-based approaches;
• test state-of-the-art deep-learning methods for seismic phase picking, such as PhaseNet and EQTransformer;
  use the detected seismic phases to locate icequake sources;
• compare these approaches with our own deep-learning method, which directly predicts icequake locations from waveform data;
• evaluate the strengths and limitations of each method in the challenging environment of sea ice.
• depending on the starting date of the internship, motivation and engagement in the internship, field experiments on sea ice in Quebec could be possible during February 2027.

The project lies at the intersection of geophysics, seismology, Arctic science and artificial intelligence, and offers the opportunity to work with real-world data while exploring modern machine-learning techniques.

Profile and skills

• Seismic wave propagation: wave dispersion, polarization, guided waves
• Python programming
• Signal processing: Fourier analysis, time-frequency analysis, cross-correlation...
• Deep Learning: basics of deep learning, being able to apply already existing CNN.

To apply, please send an e-mail with your CV and motivations.