From Mars to Earth: Deep Learning Denoising for Seismic Monitoring

by Nikolaj Dahmen (ETH Zürich)

Tuesday May 12, 2026, 2:00 PM → 4:00 PM Europe/Paris

Lamarck 522 (IPGP - Campus des Grands Moulins)

A fundamental challenge in seismic data analysis is the presence of noise that corrupts waveforms and masks weak event signals, limiting the completeness and quality of seismicity catalogs and with that the analysis of seismic processes. This challenge is particularly acute for NASA's InSight mission on Mars, where a single-station setup, low magnitudes, and large epicentral distances combined with highly variable background noise lead to low signal-to-noise ratios that limit event detection and characterization. Deep learning models have emerged as a game changer in seismic data processing, and here I show how a deep learning-based denoising approach that separates event and noise energy improves upon conventional processing across a wide range of signal-to-noise conditions, substantially advancing the analysis of marsquake signals, from signal detection, phase picking, and manual and automated waveform analysis. Transferring a similar framework to earthquake data, specifically the Swiss seismic network, I demonstrate its broad applicability and efficiency in processing large data volumes, significantly improving earthquake catalog completeness at the low-magnitude end and enhancing operational monitoring settings - in both cases leading to a more detailed picture of seismicity and ultimately advancing the observation and characterization of seismic processes.

https://eaps.ethz.ch/en/people/profile.nikolaj-dahmen.html