Crowd-Sourced Earthquake Detections Integrated into Seismic Processing Completed

The goal of this project is to improve the USGS National Earthquake Information Center’s (NEIC) earthquake detection capabilities through direct integration of crowd-sourced earthquake detections with traditional, instrument-based seismic processing. During the past 6 years, the NEIC has run a crowd-sourced system, called Tweet Earthquake Dispatch (TED), which rapidly detects earthquakes worldwide using data solely mined from Twitter messages, known as “tweets.” The extensive spatial coverage and near instantaneous distribution of the tweets enable rapid detection of earthquakes often before seismic data are available in sparsely instrumented areas around the world. Although impressive for its speed, the tweet-based system has weaknesses, including missed events in non-populated areas, poor earthquake locations, and a 10-percent false trigger rate. To leverage the strengths and mitigate the weaknesses of both the crowd-sourced and instrument-based seismic systems, the project team used the rapid tweet-based detections as seeds for seismic processing of event location, phase data association, and magnitude determination. The rapid crowd-source detections allow the seismic systems to focus in on a region of interest, thus reducing the number of instrumental observations necessary to process an event and, in turn, accelerate the processing.



To seamlessly integrate these crowd-sourced detections with numerous existing processing systems, the detections were converted to an internationally recognized format for seismic data exchange and were distributed via existing standard mechanisms. Algorithmic improvements were made to the core tweet-based system to provide improved locations to better support the integration of the data. After successful integration of the Twitter and seismic data, the project team integrated an additional type of crowd-sourced earthquake detections that are derived from analysis of Internet traffic and produced by the European-Mediterranean Seismological Centre (EMSC). These data, referred to as “flashsourcing,” further increase the spatial coverage of the crowd-sourced detections.



Principal Investigator : Michelle Guy, Paul S Earle

Cooperator/Partner : Jessica S Turner, Remy Bossu, Robert Steed

Accomplishments

The accomplishments for this project are described in detail below.

• The project team accomplished a number of tasks to support data sharing and integration.
  • They created, and now maintain, well-formatted, real-time, crowd-sourced earthquake detection data products, in an international standard seismic data exchange format, which seamlessly integrate with existing data distribution mechanisms and multiple data consumer applications for data sharing. An example of these QuakeML-formatted tweet-based earthquake detections is provided in figure 5. The corresponding code that produces these formatted detections is available at https://my.usgs.gov/bitbucket/projects/NEIC/repos/ted2quakeml/browse (content no longer available).
  • The project team created a metadata record for this dataset, which is available at https://www.sciencebase.gov/catalog/item/580108c3e4b0824b2d18bbd3.
  • The team now integrates, in real time, these well-formatted, tweet-based earthquake detections to the NEIC seismic processing system.
  • The team also provides tweet-based earthquake detections to EMSC in real time and receives EMSC-flashsourced earthquake detections in real time.
• The project team upgraded the search index and visual analysis interface to Elasticsearch v1.7 and Kibana v4.1. All updated code is available at https://my.usgs.gov/bitbucket/projects/NEIC/repos/ted2quakeml/browse (content no longer available).
• The project team made a containerized version of the application and successfully got it running at EMSC in France.

Current analysis results for integrating crowd-sourced detections with seismic systems show TED and Flashsourcing (peaks in web traffic from EMSC) systems detected felt earthquakes before enough seismic data were available for detection in 95 percent of the cases (figure 6). Due to deriving on the order of three felt detections daily and the timing of when the real-time data integration was established, there have not been enough statistically significant events for detailed analysis of how the seismic system is performing with the crowd-sourced detections as rapid inputs of possible earthquakes. Now that the real-time data integration is fully established, analysis will continue well beyond the life cycle of this CDI funded project.



Note:  This description is from the Community for Data Integration 2016 Annual Report.