Machine-learning-based earthquake locations reveal the seismogenesis of the 2020 Mw 5.0 Qiaojia, Yunnan earthquake

Abstract

Summary A moment magnitude (Mw) 5.0 earthquake hit Qiaojia, Yunnan, China on May 18, 2020. Its hypocenter is only approximately 20 km away from the Baihetan reservoir, the second largest hydropower station in China. The Baihetan Reservoir is located at the junction of multiple fault zones on the eastern boundary of the Sichuan-Yunnan rhombic block, an area with high background seismic activity. The Baihetan Reservoir was planned to be impounded in April 2021 and the MW 5.0 earthquake occurred during its water-retaining. Thus, it is critical to investigate the seismogenesis of the Qiaojia MW 5.0 mainshock and evaluate the risk of inducing earthquakes near the Baihetan Reservoir after impoundment. In this study, we built a complete and accurate earthquake catalog to analyze seismicity in the reservoir area before and after the MW 5.0 Qiaojia earthquake. We adopted a machine learning-based seismic phase picker, PhaseNet, to automatically detect seismic picks from continuous raw seismic data. Seismic phase picks were associated and located using sequential earthquake association and location methods, including REAL, VELEST and hypoDD. We eventually obtained high-precision locations of 1640 earthquakes by the hypoDD. The distribution of earthquake locations indicates that a concealed fault nearly vertical to the surface accommodated the MW 5.0 Qiaojia mainshock. The majority of its aftershocks is located within a narrow depth range of 8–13 km, indicating that the stresses in the hypocentral area were concentrated near the hypocenter of the MW5.0 earthquake. Along with focal mechanism solutions, we suggested that the Mw 5.0 Qiaojia earthquake is more likely a tectonic earthquake. However, we cannot exclude the possibility that earthquakes could be induced after the impoundment of Baihetan Reservoir, because the identified concealed fault is located in the middle of many large fault zones and only 20 km away from the Baihetan Reservoir.

Seismicity and Magmatic System of the Changbaishan Intraplate Volcano in East Asia

“…For instance, the PhaseNet phase picker comes with a training data set containing ∼700,000 seismic waveform samples collected from northern California more than 30 years (W. Zhu & Beroza, 2019), and it has significantly improved the accuracy of S ‐wave arrivals picking compared to the STA/LTA method. In addition, the PhaseNet method has been widely applied to many earthquake location studies (e.g., Liu et al., 2020; Tan et al., 2021; R. Wang et al., 2020; L. Zhou et al., 2022) and magmatic system studies in Mayotte and Hawaii (Retailleau et al., 2022; W. Zhu et al., 2022). Nevertheless, some weak seismic phases may be missed out when using just a single method (Figures S1 and S2 in Supporting Information S1).…”

“…In addition, the PhaseNet method has been widely applied to many earthquake location studies (e.g., Liu et al, 2020;Tan et al, 2021;R. Wang et al, 2020;L. Zhou et al, 2022) and magmatic system studies in Mayotte and Hawaii (Retailleau et al, 2022;.…”

Automatic earthquake detection and phase picking in Muong Te, Lai Chau region: an application of machine learning in observational seismology in Vietnam

“…The three notable deep learning pickers are a general phase generalized seismic phase detection (GPD) (Ross et al, 2018), PhaseNet (Zhu and Beroza, 2019), and the current state-of-the-art Earthquake Transformer (EQT) (Mousavi et al, 2020). These deep-learning-based pickers have improved the quantity and quality of earthquake detection, and phase picks in various studies (Liu et al, 2020;Wang et al, 2020;Xiao et al, 2021;Zhou et al, 2021). Moreover, a new study evaluating the performance of deep learning pickers indicated that the Earthquake Transformer could be considered the most reliable picker for local earthquakes within an epicentral distance of about 350 km (Münchmeyer et al, 2021).…”

A multi-task encoder-decoder to separating earthquake and ambient 1 noise signal in seismograms

“…Many ML algorithms are built to (1) automatically perform complex prediction task; (2) create a representation that approximates numerical simulations or captures relationships; (3) reveal new patterns, structures, or relationships from data (Bergen et al, 2019). ML algorithms are powerful in many different seismological tasks, including but not limited to waveform classification and earthquake detection (Li et al, 2018;Perol et al, 2018;Kong et al, 2019;Mousavi et al, 2019bMousavi et al, , 2020Beroza et al, 2021;Johnson et al, 2021), phase picking and association (Li et al, 2018;Meier et al, 2019;Zhu et al, 2019a;Liu et al, 2020;Mousavi et al, 2020;Walter et al, 2021), source location and characterization (Perol et al, 2018;Mousavi and Beroza, 2019;Ren et al, 2020;van den Ende and Ampuero, 2020;Kuang et al, 2021;Münchmeyer et al, 2021a;Zhou et al, 2021), earthquake early warning (Li et al, 2018;Münchmeyer et al, 2021b), and many others.…”