Revealing Activate Fault Structures in the Slow-Deforming Region of Iberia by Applying Deep Learning Techniques to Dense Seismic Recordings
Description:
The seismic behavior of slowly deforming regions (≲1 mm/yr) remains a challenge due to the small tectonic loading rates, complex fault systems and episodic and migrating seismic activity that complicate the imaging of subsurface faults and hazard assessment. We combine deep learning techniques and dense seismic recordings to study Iberia, a slowly deforming region in Southwest Europe (except near the plate boundary between Nubia and Eurasia), and to gain new insights into the seismicity behavior in these regions.
From 2007 to 2014 several dense temporary seismic networks were deployed in the target region that hold untapped data. We use a deep-learning earthquake detector and phase picker, the Earthquake Transformer (EQT, Mousavi et al., 2020), to analyze the Iberian datasets. Using a small dataset of 28,622 waveforms from the region, we fine-tune the EQT model to our study region and to accommodate larger epicentral distances. When applied to this small data subset, our newly trained model detects 98% of the seismic phases in the catalog up to epicentral distances of 250 km, compared to 77% and 75% detection of P and S phases, respectively, by the original EQT model. Applying our new model to 7 years of continuous data from 555 seismic stations, we detect 28 times the number of seismic phases in the current ISC catalog. After associating the detected phases, we compile a new earthquake catalog with 69,281 earthquakes, more than doubling those in the current catalog. Most of the new detections are located near the plate boundary, but several clusters are also identified in the interior of the study region, where only sparse seismicity was detected before. Some of these events located near regions of active mining activities. Our results demonstrate that deep learning models can be adapted to identify active fault structures in slow deforming regions and with small initial information. The new catalog provides opportunities to better study the deeper seismicity in Southwest Iberia and the anthropogenic seismicity in the region.
Session: Tectonics and Seismicity of Stable Continental Interiors
Type: Oral
Date: 4/19/2023
Presentation Time: 11:15 AM (local time)
Presenting Author: Miguel Neves
Student Presenter: Yes
Invited Presentation:
Authors
Miguel Neves Presenting Author Corresponding Author mjneves@gatech.edu Georgia Institute of Technology |
Zhigang Peng zpeng@gatech.edu Georgia Institute of Technology |
Susana Custódio sicustodio@fc.ul.pt University of Lisbon, Instituto Dom Luis |
Chengping Chai chaic@ornl.gov Oak Ridge National Laboratory |
Monica Maceira maceiram@ornl.gov Oak Ridge National Laboratory |
|
|
|
|
Revealing Activate Fault Structures in the Slow-Deforming Region of Iberia by Applying Deep Learning Techniques to Dense Seismic Recordings
Category
Tectonics and Seismicity of Stable Continental Interiors