19-3. Application of machine learning to forecasting ground motion characteristics and their evolution

Machine learning, specifically that based on deep neural networks, has shown outstanding performance for image and speech recognition tasks, leading to major advances in self-driving vehicles, medical diagnostics, virtual reality and augmented reality applications, etc.  When trained on large enough datasets, machine learning methods may be able to capture complex physical phenomena using models that can quickly be computed or queried in real time. Machine learning techniques have recently been applied to several problems in seismology and have been particularly successful at automatic waveform processing tasks such as signal versus noise discrimination and seismic phase classification. With the availability of large, publicly available seismic data archives and open-source machine learning software libraries, there is now an unprecedented opportunity to apply deep learning techniques to develop a new approach for rapid estimation of evolving ground motions. 

We seek a candidate to broadly consider how machine learning techniques can be applied to improve our understanding of ground motion characteristics and their evolution with time during a large earthquake. The unique features of modern machine learning methods fit the needs of improved ground motion forecasting, with potential application to earthquake early warning, ShakeMap, and earthquake scenario products. The research undertaken under this opportunity is expected to be fully novel yet would be the natural expansion of recent successful applications of machine learning to seismological problems.  

Applicants should consider how to apply machine learning to develop new techniques for rapidly estimating shaking characteristics (amplitude, duration, frequency content), with greater accuracy than existing methods that estimate shaking from source information and a ground motion model. Because this is an unexplored field, the candidate could undertake many potential avenues of groundbreaking research, including using convolutional neural networks or other machine learning approaches to forecast earthquake shaking, exploring the temporal evolution of earthquake shaking forecasts (i.e., how is our expectation of future shaking updated by the current state of earthquake shaking), or quantifying uncertainties in shaking characteristics for individual earthquakes.   

Machine learning projects depend on the existence of labeled datasets for use in training. Seismology has decades of carefully curated seismological data (e.g. catalogs of event locations and magnitudes, hand-picked phase arrivals, timing and amplitude of peak ground motion) with the appropriate metadata that can be leveraged for state-of-the-art supervised machine learning applications. However, applicants may also consider other predictor data or information as input into the training methods, or explore unsupervised machine learning methods. 

An example might be to use rapid observations of near-source ground motions to directly forecast the expected shaking amplitudes, durations, frequency content, and other characteristics across a potentially impacted region without first estimating source magnitude or distance. Those forecasts would then be updated by new ground motions observations as the rupture evolves. These improved, rapid shaking estimates could be integrated into earthquake early warning alerts, or provide rapid situational awareness through real-time ShakeMaps. Such work may also improve our estimates of ground motions in areas of sparse station observations, or aid in developing realistic ground motion distributions for scenario events. Output models may capture physical characteristics that have not otherwise been explored in traditional ground-motion modeling; querying these models can shed light onto the physical controls on shaking generation.  

Interested applicants are strongly encouraged to contact the Research Advisor(s) early in the application process to discuss project ideas. 

Proposed Duty Station: Pasadena, CA or Moffett Field, CA 

Areas of PhD: Geophysics, seismology, or related fields (candidates holding a Ph.D. in other disciplines, but with extensive knowledge and skills relevant to the Research Opportunity may be considered). 

Qualifications: Applicants must meet one of the following qualifications: Research Geophysicist, Research Computer Scientist, Research Engineer (General), Research Geodesist, Research Geologist, Research Mathematician, Research Physicist, Research Statistician, Research Physical Scientist  

(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.) 

Human Resources Office Contact: Beverly Ledbetter, 916-278-9396, bledbetter@usgs.gov 

Apply Here