Department of Earth and Environmental Sciences Graduate Student Symposium
Alireza Niksejel
PhD Candidate
Department of Earth and Environmental Sciences
Â鶹´«Ã½
Title: Deep-Learning Earthquake Catalogue for Cascadia Subduction Zone
Abstarct: Cascadia Subduction Zone (CSZ) differs from other convergent plate boundaries in the world for several reasons, especially for its unusually low seismicity along the trench. In the past few decades, geophysicists have made notable advances in the understanding of the CSZ. However, due to complicacy of the subduction system and lack of adequate observational data, a significant question remains poorly understood regarding the level of along-strike interplate coupling. This understanding relies on a comprehensive earthquake catalogue. Thus, the quality and quantity of earthquake detection and localization are of critical significance.
Phase picking plays a critical role in earthquake catalogue building. Recently, deep-learning phase pickers have shown promise with excellent performance on land seismic data, outperforming manual picking and traditional phase pickers. Although it may be acceptable to apply them to Ocean Bottom Seismometer (OBS) data that are indispensable for studying subduction zones, they suffer from a significant performance drop due mostly to the inherent differences between land and OBS noise and signal characteristics. To tackle this issue, I develop a generalized OBS phase picker—OBSTransformer, based on automated labelling and transfer learning. First, I compile a comprehensive data set of catalogued earthquakes recorded by 423 OBSs from 11 temporary deployments worldwide. Through automated processes, I label the P and S phases of these earthquakes by analyzing the consistency of at least three arrivals from five widely used seismic phase pickers. This results in an inclusive OBS data set containing ∼36k earthquake samples. Subsequently, I use this data set and a well-trained land machine learning model—EQTransformer for transfer learning.
I utilize the newly developed OBSTransformer picker along with a seamless earthquake location package – LOC-FLOW to construct a comprehensive earthquake catalogue for the CSZ region. This catalogue is built upon four years of continuous data from different OBS deployments. Specifically, I compare the earthquake catalogues of the Gorda Plate and Blanco Transform Fault System with those from previous studies employing traditional phase pickers. The deep-learning catalogue contains numerous earthquakes previously missed in existing studies, aligning well with bathymetric and tectonic features in the region. In the subsequent step, this comprehensive catalogue will be utilized for double-difference tomography and earthquake source/stress inversion, offering new insights into the locking/coupling status along the CSZ trench.
Biography: Alireza received his M.Sc. in Earthquake Seismology from the University of Tehran, Iran, in 2018. After two years of hands-on experience at the Iranian Seismological Center, he joined Â鶹´«Ã½'s Earth and Environmental Sciences department as a member of the Dalquake group, working under the supervision of Dr. Miao Zhang. Alireza’s interest is primarily centred around the development of Deep-Learning-based methods in seismology. As a third year student, he is currently engaged in studying seismicity and coupling rates along the Cascadia Subduction Zone.
Time
Location
Milligan Room, 8th Floor Biology-Earth Sciences Wing, Life Sciences Centre, Â鶹´«Ã½
Contact
Miao Zhang
Email: miao.zhang@dal.ca
Phone: +1 902 494 2831