Earthquake detection and hypocenter relocation in Central Pyrenees: the case of l'Alt Urgell-Andorra seismic sequence (2021-2022)

Abstract

On 2022-02-01, the mainshock of a seismic sequence that started in 2021 struck between l'Alt Urgell (Catalonia) and Andorra la Vella (Andorra) in the Central Pyrenees area. This magnitude (Mw) 4.0 earthquake, which alerted the population of the vicinities, was followed by low-magnitude aftershocks. In this work, we aim to obtain a new catalog through machine-learning procedures. We used the easyQuake python package, which allowed us to build new catalogs after detecting earthquake arrival times within seismograms recorded by several stations in the study area. Using different pickers, we obtain those machine-learning catalogs and compare them against data from regional agencies. We observed that, depending on the deep-learning model, the new catalog matched a high percentage of the original dataset recorded by the agencies. Besides that, we noted that some earthquakes passed undetected by the routine processing of these agencies. Then, using the new arrival times, we relocated the hypocenters using a 1D velocity model of the area following a non-linear location inversion approach. Our results show low uncertainties, which suggest that the arrival times detected by the machine-learning software are accurate enough to obtain constrained hypocenters. We conclude that this procedure could be advantageous for agencies and organizations that run a regional network in addition to the standard routine procedure carried out by expert scientists and technicians.

Full text

Earthquake detection and hypocenter relocation in Central Pyrenees: the
case of l'Alt Urgell-Andorra seismic sequence (2021-2022)
José. L. Sánchez-Roldán*1, A. Echeverria2, P. Herrero-Barbero3, José A. Álvarez-Gómez1, J. Walter4, José J. Martínez-Díaz1
1Complutense University of Madrid, Deparment of Geodynamics- Stratigraphy and Paleontology, Madrid, Spain.
3Spanish National Research Council, Geosciences Barcelona GEO3BCN - CSIC, Barcelona, Spain.
2Andorra Recerca+Innovació, Sant Julià de Lòria, Andorra
4University of Oklahoma, Oklahoma Geological Survey, Norman OK, USA
1. l'Alt Urgell-Andorra seismic sequence (2021-2022)
Seismicity in the Central Pyrenees area is characterized by low-magnitude
events. However, there is evidence of larger events in the historical and
instrumental period, like in Ribagorça (1373 - IEMS > VIII) or Lles de Cerdanya
(1970 - M4.8).
On 2022-02-01, the mainshock of a seismic sequence that started in 2021 struck
between l'Alt Urgell (Catalonia) and Andorra la Vella (Andorra). The sequence
continued with low-magnitude aftershocks that alerted the population.
Regional map of the study
area with the most relevant
active structures. Seismicity in
the Central and Eastern
Pyrenees is shown in the
middle-right map (l'Alt Urgell-
Andorra seismic sequence in
blue).
2. Objectives
We study this seismic sequence and the surrounding seismicity by obtaining new
catalogs through machine-learning (ML) methods. After that, we will perform
a hypocenter relocation using a non-linear location method to compute better-
constrained solutions.
3. Machine-learning-based catalogs: models and workflow
To build the catalogs, we used the easyQuake software (Walter et al., 2021),
which allows phase picking and earthquake detection through different steps of
analysis:
Download
data
Detection
Association
Combination
1
2
3
4
Phasenet
(Zhu & Beroza, 2019)
EQTransformer
(Mousavi et al., 2020)
Original
Conservative
PhasePApy 1D
(Chen & Holland, 2016)
4. Preliminary ML catalog: 2022/01/15 - 2022/02/15
ML catalogs vs. official agency catalogs
*only M>0 quakes
How many quakes
did we get?
*
*Association performed using 3 and 4 stations as the
minimum threshold for declaring an event
Num. quakes ML
Num. quakes official catalog
Fq=
Which model fits the official data?
Num. matched quakes ML
Num. quakes official catalog
Q%=x 100
We perform this initial detection to test every machine-learning model catalog.
5. Central-Eastern Pyrenees ML catalog (2021/10/01 - 2022/06/01)
Two new ML catalogs were built using Phasenet and EQT_original (3 stations association). Combining both ML results yields a better Q%
6. Hypocenter relocation and uncertainty analysis
We relocated the combined ML catalog using the NonLinLoc software package
(Lomax et al., 2000) and a 1D P-wave velocity model of the Pyrenees
(Theunissen et al., 2018).
Num. quakes = 3962
Nov Jan Mar May
Num. quakes = 636
Uncertainty values extracted from the most likelihood hypocenters show fairly
reliable locations, considering that the arrival times were picked by ML models.
Nov Jan Mar May
References
·Chen, C., Holland, A.A., 2016. PhasePApy: A Robust Pure Python Package for Automatic Identification of Seismic Phases.
Seismological Research Letters 87, 1384–1396.
·Chiang, Andrea, & USDOE National Nuclear Security Administration. (2020, February 11). Time Domain Moment Tensor
Inversion in Python (Version 0.1).
·Lomax, A., Virieux, J., Volant, P., Berge-Thierry, C., 2000. Probabilistic Earthquake Location in 3D and Layered Models, in:
Thurber, C.H., Rabinowitz, N. (Eds.), Advances in Seismic Event Location. Springer Netherlands, Dordrecht, pp. 101–134.
·Mousavi, S.M., Ellsworth, W.L., Zhu, W., Chuang, L.Y., Beroza, G.C., 2020. Earthquake transformer—an attentive deep-
learning model for simultaneous earthquake detection and phase picking. Nat Commun 11, 3952.
·Theunissen, T., Chevrot, S., Sylvander, M., Monteiller, V., Calvet, M., Villaseñor, A., Benahmed, S., Pauchet, H., Grimaud,
F., 2018. Absolute earthquake locations using 3-D versus 1-D velocity models below a local seismic network: example
from the Pyrenees. Geophysical Journal International 212, 1806–1828.
·Walter, J.I., Ogwari, P., Thiel, A., Ferrer, F., Woelfel, I., 2021. easyQuake: Putting Machine Learning to Work for Your
Regional Seismic Network or Local Earthquake Study. Seismological Research Letters 92, 555–563.
·Zhu, W., Beroza, G.C., 2019. PhaseNet: a deep-neural-network-based seismic arrival-time picking method. Geophysical
Journal International 216, 261–273.
Acknowledgements
This research was supported by grant Model_SHaKER (PID2021-124155NB-C31) funded by MCIN/
AEI/10.13039/501100011033 and by the “European Union NextGenerationEU/PRTR”.
The author JLSR owns a predoctoral research grant (FPI ref. PRE2018-083913) funded by MCIN/
AEI/10.13039/501100011033 and by “ESF Investing in your future”.
Figures and maps were built using Generic Mapping Tools (Wessel et al., 2019). Moment Tensor
Inversion computed using MTTime (Chiang, 2020).
The easyQuake software is available on Github (https://github.com/jakewalter/easyQuake).
ORCID ID
J. L. Sánchez-Roldán