Skip to main content
U.S. flag

An official website of the United States government

21-36. Improving ground-motion characterization for the USGS earthquake hazard products: Nonergodic, simulation, machine learning, and intensity-based methods


Closing Date: November 1, 2022

This Research Opportunity will be filled depending on the availability of funds. All application materials must be submitted through USAJobs by 11:59 pm, US Eastern Standard Time, on the closing date.

Please communicate with individual Research Advisor(s) on the right to discuss project ideas and answer specific questions about the Research Opportunity.

How to Apply


The U.S. Geological Survey (USGS) produces several public-facing products meant to quantify and communicate earthquake hazard, mitigate earthquake risk, and improve post-event situational awareness. These include the National Seismic Hazard Model (NSHM), the ShakeAlert earthquake early warning (EEW) system, and ShakeMap. These products all rely on forecasts from ground-motion models (GMMs), and some rely on ground-motion-to-intensity-conversion equations (GMICEs) to convert peak ground shaking into Modified Mercalli Intensity (MMI) for communicating shaking levels to the public. These products could be improved through innovations in the ground-motion modeling components. 

The NSHM forms the basis for the seismic provisions of U.S. building codes, as well as other engineering-design requirements, risk assessments, and planning decisions. The NSHM currently uses ergodic ground-motion models (GMMs), in the sense that they are empirically derived from data across many regions for a given seismotectonic category, such as shallow crustal, stable continental, or subduction zones. This approach can lead to large uncertainty, as median ground motions have been found to vary on a significantly smaller spatial scale than is considered in these types of models. In ShakeAlert and ShakeMap a GMICE is convolved with estimates of peak shaking from ergodic GMMs to estimate Modified Mercalli Intensity (MMI). Existing GMICEs have large associated uncertainties, and this is magnified when combined with the uncertainties from ergodic GMMs. Alternatively, intensity prediction equations (IPEs) can be used to directly estimate MMI as a function of earthquake source, path, and site parameters without relying on shaking intensities as input. However, existing IPEs were developed as global ergodic models, using simple statistical methods, and limited input parameters, leading to large variability in the observations, and thus large uncertainty in the resulting forecasts.  

The accuracy, precision, and effectiveness of the USGS hazard products could be improved through advancing the GMMs, GMICEs, or IPEs used to generate shaking estimates. However, the ability of existing modeling approaches to provide substantial innovations is diminishing. Significant improvements can likely only be achieved through fundamentally new or advanced methods in ground-motion modeling, including the use of a larger Did You Feel It? intensity dataset to directly estimate MMI based on novel input parameters, regionalized or nonergodic approaches to modeling ground motion, the use of more advanced statistical techniques, including machine-learning based techniques, the incorporation of 3-dimensional ground-motion simulations in model development, or the development of aleatory variability or epistemic uncertainty models to go along with median predictions.  

We seek Mendenhall Postdoctoral Scholars to investigate and develop models for ground motion with the goal of improving shaking forecasts in USGS hazard products including the National Seismic Hazard Model, the ShakeAlert EEW system, and ShakeMap. We seek candidates with interest in one or more of four possible topic areas (1) modeling intensity-based ground motions; (2) the use of 3D simulations to improve ground-motion estimates or PSHA; (3) nonergodic approaches to ground-motion modeling; and (4) the use of machine learning in ground-motion modeling. Example research directions that may be addressed by the Fellow are provided for each topic area below, but other research efforts to improve ground-motion characterizations in earthquake hazard products are also encouraged. 

  1. Modeling intensity-based ground motions:  

To address some of the issues in current intensity-based ground motion models, a large database of USGS Did You Feel It? (DYFI) responses collated into MMI values, along with the accompanying metadata, is currently in development by the Research Advisors, which could be used in any of the following projects. 

  • Developing a new IPE that can be used to estimate median MMI values as a function of a suite of predictor variables related to earthquake source, path, and site. 

  • Developing a new GMICE that can be used to estimate median MMI given other ground motion metrics such as peak ground motions or response spectra alone or in combination with source and path parameters. 

  • Developing an IPE or GMICE specifically suited to the challenges of a single USGS earthquake hazard product, such as ShakeAlert, which offers limited earthquake source information (e.g., no depth constraint). 

  • New methods for mitigating the effect of sampling bias in MMI data, such as considering biases in poorly sampled regions, underreporting at low levels of MMI, and the discrete jump going from MMI II to I.  

  • Consideration of novel predictive parameters in the modeling of MMI related to earthquake source, wave propagation, geology, the built environment, or other factors. 

  • Analysis of the most effective intensity measure(s) such as different frequencies of ground motion, for use in predicting MMI, with an emphasis on previously unconsidered metrics such as Arias intensity, cumulative absolute velocity, or significant duration. 

  1. 3D ground-motion simulations:  

Earthquake ground-motion simulations can constrain the ground-shaking from scenario earthquakes and are inherently non-ergodic; however, simulated values are highly dependent on the details of the earthquake rupture model and require various validation approaches. Research topics may also include the development of simulation-based probabilistic seismic hazard models (i.e., urban seismic hazard models).  

  • How can 3D simulations be used in probabilistic seismic hazard assessments?  

  • What features of ground-motion simulations are sufficiently characterized for application in probabilistic seismic hazard assessments?  

  • How can simulated ground motions and their variability be validated? 

  1. Development and implementation of non-ergodic ground-motion models in PSHA:  

  • How can ground-motion recordings or ground-motion measurements from simulations be used to develop non-ergodic ground-motion models?  

  • How can non-ergodic ground-motion models be implemented in the National Seismic Hazard model and what are the implications for the hazard assessments? 

  • Use recordings from small- to moderate-magnitude earthquakes and from ground-motion simulations to further constrain components of empirically based ground-motion models.  

  • Regionalize ground-motion models to account for important geologic structures, such as the sedimentary basins that underlie Los Angeles, CA; San Francisco Bay Area, CA; Seattle, WA; and Salt Lake City, UT, and to implement modified GMMs into the probabilistic seismic hazard assessments of the NSHM. 

  1. Machine learning approaches for non-ergodic GMM development:  

The recent emergence of efficient and accurate machine learning approaches, in combination with large simulation datasets, motivates the development of ground-motion models that utilize neural network architectures. This approach could be expanded upon, in a nonergodic sense, to regional ground-motion studies to better understand path, site, and source effects upon ground motion trends. 

  • How can machine learning approaches be applied to recorded and/or simulated ground-motion datasets to guide development of non-ergodic GMMs? 

  •  Can machine learning be employed to help constrain traditional regression approaches to derive improved non-ergodic models 

Applicants are expected to have strong technical skills in earthquake seismology, engineering seismology associated with the proposed work, and a desire to participate in model building. However, experience working directly with USGS earthquake hazard products is not necessary.  

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

Proposed Duty Station(s): Moffett Field, California; Pasadena, California; Golden, Colorado. 

Areas of PhD: Geophysics, seismology, engineering seismology, geology, data science, computer science, statistics, mathematics, 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 Engineer, Research Civil Engineer, Research Geologist, Research Statistician, Research Computer Scientist, Research Mathematician, Research Physicist.  

(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:  Paj Shua Cha, 650-439-2455,

Apply Here