Incorporating AI in Routine Seismic Network Operations in Southern California

Date: 4/26/2019

Time: 06:00 PM

Room: Grand Ballroom

Caltech and USGS operate the Southern California Seismic Network (SCSN) to provide timely disaster mitigation in the form of earthquake early warning (EEW), event notification, ShakeMap, and other data products to more than 20 million people living astride the Pacific and North America plate boundary. The EEW project (ShakeAlert) analyzes SCSN data to issue alerts to pilot users such as major utilities, hospitals, schools, and government agencies, some of them also host sensors and provide user feedback. A seismic moment tensor to identify the causative fault and evaluate tsunami hazards is also determined. In case of unusual activity, seismologists provide near real-time situational awareness to warn of increased hazards levels.

For over two years, the SCSN has been importing its entire seismic data streams of ~14Mbits/sec into the Amazon Web Services (AWS) cloud. Currently, we have the AQMS software deployed on several EC2 instances, and we routinely report origins, magnitudes, moment tensor and ShakeMaps from the cloud directly into PDL, which in turn are posted on the USGS/NEIC web pages within two minutes. The SCSN products are also made available on the SCEDC and SCSN web sites hosted by AWS, which ensures information availability during peak demand times. We are also archiving the current waveform data (since 2017) in the AWS S3 Glacier part of the cloud.

To facilitate our next step in using cloud computing, we have developed machine learning (ML) algorithms to pick phases such as P and S phases, and to determine the polarity of P-waves. We have also developed an ML phase associator (PhaseLink) to reliably detect earthquakes occurring concurrently, like during an aftershock sequence. To deploy these ML algorithms for dependable and scalable routine processing, we plan to use cloud native principles such as microservices and serverless architecture; and AWS products like Kinesis, Lambda, Batch and Analytics for seismic data processing.

Presenting Author: Egill Hauksson

Authors