OBSTransformer: a deep-learning seismic phase picker for OBS data using automated labelling and transfer learning

Alireza Niksejel,Miao Zhang

DOI: https://doi.org/10.1093/gji/ggae049

IF: 3.352

2024-02-02

Geophysical Journal International

Abstract:SUMMARY Accurate seismic phase detection and onset picking are fundamental to seismological studies. Supervised deep-learning phase pickers have shown promise with excellent performance on land seismic data. Although it may be acceptable to apply them to Ocean Bottom Seismometer (OBS) data that are indispensable for studying ocean regions, they suffer from a significant performance drop. In this study, we develop a generalized transfer-learned OBS phase picker—OBSTransformer, based on automated labelling and transfer learning. First, we compile a comprehensive data set of catalogued earthquakes recorded by 423 OBSs from 11 temporary deployments worldwide. Through automated processes, we label the P and S phases of these earthquakes by analysing the consistency of at least three arrivals from four widely used machine learning pickers (EQTransformer, PhaseNet, Generalized Phase Detection and PickNet), as well as the Akaike Information Criterion (AIC) picker. This results in an inclusive OBS data set containing ∼36 000 earthquake samples. Subsequently, we use this data set for transfer learning and utilize a well-trained land machine learning model—EQTransformer as our base model. Moreover, we extract 25 000 OBS noise samples from the same OBS networks using the Kurtosis method, which are then used for model training alongside the labelled earthquake samples. Using three groups of test data sets at subglobal, regional and local scales, we demonstrate that OBSTransformer outperforms EQTransformer. Particularly, the P and S recall rates at large distances (>200 km) are increased by 68 and 76 per cent, respectively. Our extensive tests and comparisons demonstrate that OBSTransformer is less dependent on the detection/picking thresholds and is more robust to noise levels.

geochemistry & geophysics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Improve the accuracy and reliability of seismic phase detection in ocean - bottom seismometer (OBS) data**. Specifically, the existing deep - learning seismic phase pickers perform excellently on land seismic data, but their performance drops significantly when processing OBS data. This is mainly because OBS data are affected by ocean - specific noises, such as wind - driven waves, sounds of marine mammals, human activities and water - column reverberations, etc., which make it difficult for traditional automatic algorithms to accurately identify seismic phases.

To solve this problem, the author has developed a new OBS phase picker based on transfer learning - **OBSTransformer**. This is achieved through the following methods:

1. **Dataset construction**: Collected about 100,000 seismic waveforms recorded by 11 temporary OBS deployments around the world, and marked the arrival times of P - waves and S - waves through an automated label - generation method.
2. **Transfer learning**: Used a pre - trained land seismic phase - picker model (such as EQTransformer) as the base model, and adapted to the characteristics of OBS data through transfer learning.
3. **Noise sample extraction**: Extracted 25,000 noise samples from the same OBS network for model training and testing.

Through these methods, OBSTransformer outperforms the existing land models on test datasets of different scales. In particular, the recall rates of P - waves and S - waves at long distances (> 200 km) are increased by 68% and 76% respectively. Moreover, OBSTransformer is more robust to noise levels and less dependent on the detection / pick - up threshold.

In summary, this research aims to develop a deep - learning phase picker more suitable for OBS data through transfer learning and automated label generation, thereby improving the accuracy and efficiency of marine seismology research.