← Back to Sessions

Semiai Seismic Detection and Picking: An Application to Active and Passive Seismic Data for the Tomography of the Stromboli Volcano Island.

Description: 

In several seismic applications, including seismic location and tomography, the detection and selection of seismic waves is essential. The application of artificial intelligence to earthquake analysis is a relatively recent development which attempts to tackle these objectives. Today's machine learning methods greatly benefit from the availability of large earthquake datasets that can be employed throughout the training process. However, data related to active seismic experiments is still limited.
At Stromboli, a large seismic database combining active seismicity data (air-gun shots) and local earthquakes was created as a result of two active seismic experiments conducted at the end of 2006 and during 2014, as well as a greater number of earthquakes acquired across a wider temporal range. Therefore, an automated and extremely accurate phase arrival picking is required to analyze this large amount of data and execute a seismic velocity tomography to examine the internal structure of the volcano.
In this work, we developed an automatic SEMI Artificial Intelligence (SEMIAI) method for the detection and the picking of seismic events. The shortage of active seismic data is circumvented by employing a polarization and spectral analysis for the detection, while the picking relies on a deep neural network. This in order to distinguish the active seismic events from earthquakes and volcanic events.
As first step, the SEMIAI methodology has been applied to the Stromboli’s 2006 active seismic dataset. We focused on this dataset because it had a large number of active seismic sources and stations (1500 shots recorded at 42 stations), both on land and offshore (ocean bottom seismometer), as well as about 300 local earthquakes.
The tomography of the Stromboli Volcano island provided by the automatic dataset is practically the same as the one previously obtained by using manually picked phases, demonstrating the effectiveness and efficiency of the SEMIAI method.

Session: Opportunities and Challenges for Machine Learning Applications in Seismology

Type: Oral

Date: 4/19/2023

Presentation Time: 02:00 PM (local time)

Presenting Author: Sergio Gammaldi

Student Presenter: No

Invited Presentation: 

Authors