← Back to Sessions

Earthquake Source Depth Determination Using Single Station Waveforms and Deep Learning

Description: 

Earthquake source depth plays a crucial role in characterizing earthquakes, assessing seismic hazards, and understanding subsurface structures. However, accurately determining source depths is often challenging. Traditional travel-time-based location methods struggle to constrain depths due to imperfect station distribution and the strong trade-off between source depth and origin time. Identifying depth phases at regional distances is further hindered by strong wave scattering, which is particularly challenging for low-magnitude events.

Deep learning techniques have great potential for deriving source depth information directly from continuous data streams based on waveform features. In this work, we propose a novel depth feature extraction network (named VGGDepth), which directly maps seismic waveforms to depth information using three-component single-station waveforms. The network structure is adapted from VGG16 in computer vision and is designed to take three-component single-station waveforms as inputs and depth labels as outputs, achieving a direct mapping from waveforms to depth.

We train the network by segmenting continuous waveform streams into time windows, which serve as inputs to the model. Two scenarios are considered: (1) training and testing on the same seismic station and (2) generalizing training and testing to multiple stations within a particular region. The results show that both VGGDepth models achieve high-depth prediction accuracy. We demonstrate the effectiveness of our methodology using seismic data from the 2016-2017 Central Apennines, Italy earthquake sequence and the 2019 Ridgecrest, USA earthquake sequence.

Session: Building and Decoding High-resolution Earthquake Catalogs With Statistical and Machine-learning Tools [Poster]

Type: Poster

Date: 4/15/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Wenda

Student Presenter: No

Invited Presentation: 

Poster Number: 36

Authors