← Back to Sessions

Deep Learning Based Approach to Integrate MyShake's Trigger Data with ShakeAlert for Faster and Robust EEW Alerts

Session: Earthquake Early Warning Live in California! Current Status and Challenges [Poster]

Type: Poster

Date: 4/23/2021

Presentation Time: 11:30 AM Pacific

Description: 

In this presentation, we propose a new earthquake detection method for earthquake early warning for the dense smartphone array in a region as well as the traditional seismic stations. The goal is to integrate these two sensor networks. The method is inspired by the scene detection in movies, that a machine learning algorithm is monitoring each frame in the movie to make the decision. For detection, each region can be divided into 10 by 10 km or 1 by 1 meter square cells to aggregate the triggers within the cells. A sequence of images will feed into a deep learning model to make the decision whether an earthquake is occurring, which the model can both handle temporal and spatial coherent signals to detect the earthquake. The training dataset was generated by using a simulation platform, so that we can generate cases with sensors at different locations without considering specific network configuration. The performance of the model will be discussed in this presentation. In addition, using the simulation platform, we can also evaluate how the smartphone seismic network helps the traditional seismic network, which will be also discussed in this presentation. This work is funded by USGS through 2020 EHP Award G20AP00058.

Presenting Author: Qingkai Kong

Student Presenter: No

Authors