Comparison of Deep Learning Techniques for the Investigation of a Seismic Sequence: An Application to the 2019, Mw 4.5 Mugello (Italy) Earthquake. Issue 12 (22nd December 2021)

• Record Type: Journal Article
• Authors: Cianetti, S.; Bruni, R.; Gaviano, S.; Keir, D.; Piccinini, D.; Saccorotti, G.; Giunchi, C.
• Is Part Of: Journal of geophysical research. Volume 126:Issue 12(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-22
• Subjects: earthquake sequence -- deep learning based seismic events detection -- seismic data analysis automation -- seismic activity in the Northern Apennines (Italy); Geomagnetism -- Periodicals; Geochemistry -- Periodicals; Geophysics -- Periodicals; Earth sciences -- Periodicals; 551.1
• DOI: 10.1029/2021JB023405
• Languages: English
• ISSNs: 2169-9313

Abstract

Abstract: The increase of available seismic data prompts the need for automatic processing procedures to fully exploit them. A good example is aftershock sequences recorded by temporary seismic networks, whose thorough analysis is challenging because of the high seismicity rate and station density. Here, we test the performance of two recent Deep Learning algorithms, the Generalized Phase Detection and Earthquake Transformer, for automatic seismic phases identification. We use data from the December 2019 Mugello basin (Northern Apennines, Italy) swarm, recorded on 13 permanent and nine temporary stations, applying these automatic procedures under different network configurations. As a benchmark, we use a catalog of 279 manually repicked earthquakes reported by the Italian National Seismic Network. Due to the ability of deep learning techniques to identify earthquakes under poor signal‐to‐noise‐ratio (SNR) conditions, we obtain: (a) a factor 3 increase in the number of locations with respect to INGV bulletin and (b) a factor 4 increase when stations from the temporary network are added. Comparison between deep learning and manually picked arrival times shows a mean difference of 0.02–0.04 s and a variance in the range 0.02–0.07 s. The improvement in magnitude completeness is ∼0.5 units. The deep learning algorithms were originally trained using data sets from different regions of the world: our results indicate that these can be successfully applied in our case, without any significant modification. Deep learning algorithms are efficient and accurate tools for data reprocessing in order to better understand the space‐time evolution of earthquake sequences. Plain Language Summary: We focus on the testing and comparison of recently published frameworks for seismic events and seismic waves detection based on deep learning. As an example we use data from an earthquake swarm that occurred in December 2019 in the Mugello basin (N of Florence, Italy). Earthquake swarms analysis is particularly challenging because of the high seismic rates, with events closely occurring in time and space. The results of the automatic analysis are compared with results reviewed by human analysts. The deep learning algorithms used in this application were originally trained on data collected from different regions of the world: Our results indicate their capability to apply the knowledge gained in different contexts to totally unseen data. However, some differences, due to the different design of the algorithms and to training data sets, arise and suggest that careful tuning and benchmarking are required to achieve stable and reliable results in seismic data analysis workflows. Key Points: We analyze the Mw 4.5 Mugello (Italy) earthquakes sequence, December 2019, using deep learning algorithms, and compare with bulletin data Deep learning performances on detections and phase picking works very well even without ad‐hoc retraining Deep learning based earthquake catalogs show differences due to different algorithm designs