Speaker
Description
PhaseNet and EQTransformer are two deep learning methods most commonly used for automatic earthquake detection and phase-picking. This work presents the performance of these algorithms using two weeks of data retrieved from the Botswana Seismological Network stations. Additional data used is obtained from the International Monitoring System network within a distance of 250 km from Botswana (e.g. Lobatse and Boshof). The results of PhaseNet and EQTransformer are recorded and compared with the results of the STA-LTA method. The accuracy, precision, recall, and F1 score are calculated for each method. Preliminary results show that the deep learning algorithms outperform conventional phase-picking methods in noisy data and where events overlap. Events that may be ambiguous for traditional methods are often clear outliers within the DNN model domain, as observed in this study.
Promotional text
Joseph Maritinkole is a Data Analyst and Station Operator at the Botswana Geoscience Institute. He completed his undergraduate degree in Geophysics at the University of Botswana and is now a full-time Masters student at the University of Witwatersrand.
| [email protected] | |
| Oral preference format | in-person |
