Machine Learning-based High-resolution Earthquake Catalog for the Middle America Trench Using Ocean Bottom and Land Seismometers
Description:
Subduction zones generate Earth’s largest earthquakes along with many low-magnitude events that remain undetected. However, the ability to monitor microseismicity in offshore settings can delineate plate-boundary fault networks and inform coupling and slip patterns. Machine learning (ML) can improve detection of seismic signals over manual picking for sparse offshore sensors, deployed in noisy marine environment. Here, we develop a high-resolution earthquake catalog for the Middle America Trench using ML applied to legacy amphibious CRSEIZE data (1999–2001), including a 2-month Osa aftershock sequence and a 6-month Nicoya deployment following the Mw 6.9, 20 August 1999. We evaluate developed ML catalogs against manually picked catalogs.
Within SeisBench, we run pre-trained phase-picking models optimized for ocean-bottom seismometer (OBS) (PickBlue-PhaseNet and OBSTransformer) and test PhaseNet with several training datasets for land recordings for both the Osa and Nicoya experiments. Picks are aggregated with a conservative cross-domain consensus rule to retain phases detected by at least two distinct OBS–land model combinations (Aziz Zanjani & DeShon 2026). We then associate events with GAMMA and PyOcto. Preliminary results for Osa shows the ML workflow detected 2,202 earthquakes, with only 835 out of 1,557 earthquakes in the manual catalog recovered. Catalogs for both Osa and Nicoya will presented. These more complete offshore seismic catalogs provide better constraints on the onset and cessation of interplate seismicity, the updip and downdip limits of seismogenic zone, and boundaries linked to a serpentinized forearc mantle wedge. Improved microearthquake distributions should also stabilize statistics such as b-value to constrain temporal and depth distributions of stress.
Session: Linking Subduction Zone Processes and Cascading Hazards in Alaska, Cascadia, Chile and Beyond [Poster]
Type: Poster
Date: 4/16/2026
Presentation Time: 08:00 AM (local time)
Presenting Author: Heather R. DeShon
Student Presenter: Yes
Invited Presentation:
Poster Number: 121
Authors
Omid Asgarvand oasgarvand@mail.smu.edu Southern Methodist University |
Heather DeShon Presenting Author Corresponding Author hdeshon@mail.smu.edu Southern Methodist University |
Asiye Aziz Zanjani aazizzanjani@mail.smu.edu New Mexico Institute of Mining and Technology |
|
|
|
|
|
|
Machine Learning-based High-resolution Earthquake Catalog for the Middle America Trench Using Ocean Bottom and Land Seismometers
Category
Linking Subduction Zone Processes and Cascading Hazards in Alaska, Cascadia, Chile and Beyond