Evaluating the SAIPy Performance using a Local Seismic Network for Volcano-Tectonic Earthquakes Monitoring

TL;DR

This work evaluates SAIPy, an open-source deep-learning toolkit for seismic analysis, in a local volcanic network to test its generalization beyond single-station tectonic events. It extends SAIPy with a network-based framework that performs time-based P-arrival association and pattern-based clustering across nine stations, enabling improved event detection in a caldera. Results show expanded event discovery and generally reliable phase picks and P-wave polarity estimates for local volcano-tectonic activity, though magnitude estimates exhibit inter-station variability and some biases due to site effects and instrument responses. The study demonstrates SAIPy’s potential for processing continuous network data in active volcanic regions and suggests retraining with physically standardized units and more diverse volcanic data to enhance magnitude estimation across networks and sensor types.

Abstract

In this study, we evaluated the performance of SAIPy, an open-source Python package for deep learning-based seismic data analysis, by applying its single-station monitoring tools and extending its use to a seismic network based approach, using data from a local seismic network deployed in a Caldera. Although the integrated models into SAIPy for earthquake detection,magnitude estimation, seismic phase picking, and P-wave polarity classification, were originally trained on tectonic signals, we assess their performance in a more complex seismic environment that includes volcano-tectonic events, along with signal interference from distant earthquakes.We also demonstrate the advantages of integrating outputs using multiple stations to improve event detection. SAIPy was able to identify a significantly larger number of local events than those included in previously published catalogs. SAIPy demonstrated reliable phase picking and P-wave polarity estimation, particularly for local volcano-tectonic events, with some limitations observed in the magnitude estimation for complex volcanic signals. These results support the utility of SAIPy for processing continuous seismic data and suggest that future retraining using data with physically standardized units, removing instrumental response, and including data from more diverse seismic sources, could improve its generalization for magnitude estimation to complex scenarios and different seismic networks and sensor types.

Figures (8)

• Figure 1: Seismic stations used in this work, which were part of a temporary network operative during 2017, in the framework of the P24 CeMIE-Geo project bib_PRIM_AGUbib_GMZ.
• Figure 2: Output plots from the single-station monitoring by SAIPy. Example of three stations PR15, PR23, and PR45. This earthquake reported by bib_PRIM_AGU as Ml 2.5, occurred on September 9, 2017, at 06:45:20, and 2 seconds later was recorded by these stations. It is worth noting that the reported location of this earthquake is located approximately midway between these three stations, just in the Caldera area. In the row of the top, are the P and S picking, and the magnitude estimated per station. The average of the magnitudes is very close to the reported Ml. In the middle row are the probabilities of the P and S picking, and in the bottom row, we can observe the P-arrival time and the respective polarities.
• Figure 3: Output plot from the detection by multiple stations. Only vertical components of the seismogram are shown. Example using the same earthquake as in Figure \ref{['fig:Figure2']}, occurred on 9 September 2017, which was detected by the nine stations used in the testing, sorted by detection time from the top to the bottom.
• Figure 4: Close-up view of the three components of the seismograms, with the P and S phase picks by SAIPy, of the event detected in nine stations showed in \ref{['fig:Figure3']}. In particular, this is an volcano-tectonic event located in the Caldera are. We could observe a very well Phase picking with differences of P-S arrivals no longer than 3 seconds. This visualization was customized to facilitate the review of picking accuracy.
• Figure 5: One hour waveform recordings at six stations of the temporary network on September 8, 2017, starting at 05:00:00. Only vertical components of the seismogram are shown. In the left, the data are filtered to highlight low frequencies between 0.01 and 0.8 Hz, allowing the observation of regional earthquake signals, including aftershocks of the Mw 8.2 Tehuantepec earthquake that occurred at 04:49:17 (UTC). In the right, the data are filtered in the 1–45 Hz frequency range, which corresponds to the operating range of SAIPy, revealing a sequence of closely spaced local earthquakes. For comparison purposes, red lines that indicate the events detected using SAIPy's multi-station approach (with detections at two or more stations) are shown in both panels. Only two of the local events, during this time period, were reported by bib_PRIM_AGU (see Table \ref{['tab1']}).