Development of a Method for Reliability Estimation of Triggered P-arrivals using Convolutional Neural Network
Abstract
Earthquake Early Warning System (EEWS) uses P-arrivals triggered by automated algorithms at a small number of seismic stations. Wrong or imprecise arrivals may cause false or large uncertainty in the warning. However, it is not easy to distinguish seismic signals from non-seismic signals (e.g., sensor glitch, lightning strikes, or anthropogenic noises, etc.).
The goal of this study is to develop a method for estimating the reliability of automatically triggered P-arrivals. Then, the reliability of picks could be utilized for a probabilistic EEWS. We used arrivals triggered by the Korean Earthquake Early Warning System at 2016 - 2017 as a training data set for deep learning. We extracted seismograms of 1 second time window centered at triggered time and manually classified into seismic and non-seismic signals. The classified data are used to fit convolutional neural network model. This model evaluates how reliable the triggered signal is. And the result will be used as weight for determining earthquake location and magnitude of EEWS.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.S11E0411K
- Keywords:
-
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICSDE: 1910 Data assimilation;
- integration and fusion;
- INFORMATICSDE: 1942 Machine learning;
- INFORMATICSDE: 7223 Earthquake interaction;
- forecasting;
- and prediction;
- SEISMOLOGY