Periodic Reporting for period 1 - DECODE (DECiphering the seismic signature Of aseismic DEformation)
Okres sprawozdawczy: 2023-10-01 do 2025-09-30
Earthquakes are a major threat to humankind, causing damage above 500 billion euros and more than 400,000 fatalities within the last 20 years. Still, the generation of large earthquakes remains poorly understood. Recent research revealed that large subduction earthquakes are often preceded by aseismic slip on the plate boundary. Also called slow slip events (SSEs), these aseismic ruptures are often accompanied by so-called low-frequency earthquakes (LFEs), an atypical earthquake that can repeat very often during slow slip and is depleted in high-frequency energy compared to regular earthquakes. However, the link between seismic and aseismic processes is not yet clear: not all large subduction earthquakes are initiated by precursory SSEs, not all SSEs lead to large earthquakes, not all SSEs are accompanied by LFEs. Key reason for this knowledge gap is the difficulty in detecting LFEs and SSEs so that their occurrence can be analyzed with respect to the seismic cycle. In this project, I develop machine learning (ML) techniques and apply them to new, high-density data to fill this detection gap, thus allowing a systematic study of how LFEs and SSEs relate to earthquakes. In particular, I study the Chile subduction zone, a region with known SSE activity but without known LFEs.
In the DECODE project, I was aiming to characterize the interaction between seismic and aseismic deformation in subduction zones. To this end, I focussed on low-frequency earthquakes (LFEs) and the Atacama segment of the Chilean subduction margin. I successfully developed and validated a deep learning algorithm for the detection of LFEs that is applicable across regions. I made the developed tool easily publicly available and applicable through the SeisBench platform. However, applying this model and complementary classical approaches, I found no evidence for the existence of LFEs in the Atacama segment. Based on this result, I inferred physical limitations on potential LFEs that could still exist in this area while not being detectable. This result provides essential constraints on the environment and generative processes of LFEs/ To study the interaction between seismic and aseismic deformation despite the lack of LFEs, I developed a detailed catalog of regular seismicity in the region. Based on the catalog, I studied the interactions between seismic swarms and slow slip events. My results highlighted the existence of small-scale shallow slow slip events in Northern Chile, contributing to the moment partitioning. This is an essential observation to assess the potential for future megathrust earthquakes and their rupture extents. The results of these projects have been published in four central papers: a paper on the LFE detection method, a paper on the absence of LFEs in the Atacama segment, a paper describing the seismicity in the Atacama segment, a paper analysing the interaction between seismic swarms and slow slip events.
My results present substantial advancements beyond the state of the art. The developed LFE detection model is the first deep learning model for the task and has proven more versatile and flexible than existing classical approaches. The null results about the absence of LFEs and tectonic tremors in the Atacama segment presents the first systematic analysis of potential activity of these events in the region. The observed interaction between slow slip events and seismic swarms in the Atacama segment presents the first observation of a shallow slow slip event in Chile and also a new mode of interaction between seismicity and slow deformation. The earthquake catalog developed for the Atacama segment is one of the highest-resolution catalogs globally for subduction zones. It is able to resolve fine scale structures and dynamics that could not previously be observed. For continued exploitation of these results by the scientific community, all results are openly available under permissive licenses. We aim to further study the outcomes and data products from this project in the future.