Analysis of High Frequency Marsquake Swarms Informed by Deep Learning

NASA's InSight mission has provided an unprecedented snapshot of Mars' seismicity, despite data analysis challenges arising from low signal‐to‐noise ratios (SNR) and single‐station constraints. High frequency (HF) events—the most common type—were initially assumed to propagate through shallow crustal layers. However, several impacts that occurred late in the mission provided independent distance constraints, indicating that HF event energy must have propagated through the mantle. We analyzed the full HF data set using an extended catalog and denoised waveforms derived with deep learning (DL) techniques. Using a DL ensemble, we picked phase arrivals on denoised envelopes and estimated SNR‐dependent pick timing uncertainties based on the removed noise. We computed distances consistent with mantle paths using the latest Mars interior models, while the back azimuth remained inconclusive due to local resonance dominating the HF bandwidth. We compared and grouped HF recordings by their similarity and investigated how attenuation properties shape their envelopes. Additionally, we re‐calibrated and assigned magnitudes for the extended catalog. Overall, we (re‐)located 1,430 HF events clustered between epicentral distances of around 1,600–3,600 km, but without constraints on the back azimuth, their source region remains speculative. Based on spatiotemporal similarities, we attributed a subset of 1,357 events to a common source region and labeled them as swarm events. The analysis of envelope shape confirms and extends previous results of stratified attenuation properties. Swarm events, with magnitudes between 1.5 and 2.5, are cumulatively equivalent to a single magnitude 4 event and show a high ‐value and clear seasonal trends in seismicity.