Research Opportunity Description
The U.S. Geological Survey (USGS) National Earthquake Information Center (NEIC) detects, locates, and characterizes tens of thousands of earthquakes globally each year. This near real-time catalog that the NEIC produces spans the global range of earthquake sizes, tectonic environments, and seismic-station coverage. This near-real-time catalog provides the information required to further assess the potential impact of an earthquake rapidly for emergency response, and the NEIC’s subsequent long-term validated catalog acts as the primary data source for hazard mapping and seismotectonic research. Although much of NEIC’s event processing is automatic, NEIC only publishes events publicly after they have been reviewed and validated by a human analyst. This ensures the accuracy of NEIC’s initial solutions and removes the possibility of publishing false and grossly inaccurate solutions, which could undermine the public’s trust in the NEIC as an authoritative source of global earthquake information.
Although human interaction ensures the quality of published solutions, it inherently reduces the speed at which solutions can be published. After an earthquake is felt, the public immediately looks for sources of authoritative information. The human review process increases the time between when a person feels an earthquake and when the NEIC acknowledges and describes the event. The primary focus of this Mendenhall Research Opportunity is to leverage machine learning and other emergent tools to improve NEIC’s confidence in its automatic solutions, allowing for more rapid publication and, therefore, allowing NEIC to inform the public of an earthquake more rapidly.
Recent technological advancements, particularly machine learning, have begun to replace or supplement core algorithms in the earthquake monitoring pipeline. Machine learning algorithms have been shown to generalize well, have unprecedented accuracy, and are fast. Machine learning tools have been used for various earthquake characterization problems, such as improving seismic phase arrival time estimates, earthquake detection, seismic phase association, discrimination of source types, and event validation. By replacing or supplementing the NEIC’s earthquake processing pipeline with novel machine-learning tools, it may be possible to improve confidence in NEIC’s automatic solutions, allowing the NEIC to publish those solutions with less human interactions. However, most machine learning research applications in the seismology community have focused on local and regional datasets, which have a smaller variety of source types, higher quality signals, and better-constrained network configurations, thus making these applications poorly suited for the global case. Designing machine learning tools that perform on global data is particularly difficult because of the variability in global earthquake observations and station density.
Automatically published earthquake solutions sometimes rely on simple metrics, such as the number of phases recorded and associated azimuthal gaps, providing rudimentary checks sufficient for dense local networks, but they need to be improved in more complex cases encountered in regional and global earthquake monitoring. A better approach would be to develop algorithms that can reliably and accurately estimate source parameter uncertainty, particularly location and magnitude. Estimating uncertainty has been particularly difficult with machine learning models where the output is often not well calibrated to genuine uncertainty, but new approaches exist that promise to give accurate error estimates.
This Mendenhall Research Opportunity focuses on developing the scientific framework for rapid and accurate characterization of earthquakes (i.e., event detection, location, magnitude, source characteristics, seismotectonic setting) targeted explicitly at producing automatic global earthquake solutions. This research can focus on improving, supplementing, or replacing traditional approaches (e.g., STA/LTA pickers, association, magnitude estimation) or target novel procedures that will enhance the event processing framework, such as event validation. Of particular importance is developing frameworks and tools that give NEIC accurate estimates of uncertainty and provide the NEIC with confidence in automatics, particularly for events of societal importance (e.g., large, widely felt earthquakes).
We expect that proposals may be largely exploratory but will be strengthened via the demonstration of approaches that can be integrated within NEIC’s automatic solutions. While we do not expect end-to-end event processing development, we seek proposals that will demonstrate a clear vision for how machine-learning tools can move NEIC towards publishing automatic solutions. Candidates are encouraged to explore novel and state-of-the-art methods to aid in global earthquake detection or estimating earthquake properties and/or develop tools that help validate NEIC’s current automatic solutions. With the development of these technologies, research into the potential pitfalls of these algorithms and rigorous comparison with standard monitoring techniques will be required.
Interested applicants are strongly encouraged to contact the Research Advisor(s) early in the application process to discuss project ideas
Proposed Duty Station(s)
Golden, Colorado
Areas of PhD
Geophysics, seismology, computer science, or related fields (candidates holding a Ph.D. in other disciplines but with knowledge and skills relevant to the Research Opportunity may be considered).
Qualifications
Applicants must meet one of the following qualifications: Research Geophysicist, Computer Scientist, or Research Statistician.
(This type of research is performed by those who have backgrounds for the occupations stated above. However, other titles may be applicable depending on the applicant's background, education, and research proposal. The final classification of the position will be made by the Human Resources specialist.)
