← Back to Sessions

Estimation of Time-Dependent Strain-Rates With Gaussian Process Regression

Session: Crustal Stress and Strain and Implications for Fault Interaction and Slip I

Type: Oral

Date: 4/22/2021

Presentation Time: 10:15 AM Pacific

Description: 

We use a Gaussian process regression (GPR) to determine spatiotemporally continuous maps of time-varying strain-rates from GNSS measurements, implemented as Python-based Geodetic Network Smoother (PyGeoNS). GPR is a Bayesian method, which we use to interpolate the GNSS data in both space and time. Within our framework, the prior is placed on the covariance structure of the interpolated field, and not on any assumed parameterization of the displacement field. The prior effectively constrains the spatial variation of strain-rates, along with the degree to which transient displacements are allowed, and hence the prior can be used to determine strain-rate fields at various spatial and temporal scales. The posterior is then a distribution of arbitrary functions that interpolate the GNSS data, which we then differentiate using a radial basis function, finite difference scheme to determine the strain-rates. The resolved strain-rates have quantified uncertainties and can include discontinuities in surface displacements across fault segments with potential creep. We have applied PyGeoNS to resolve transient strain-rates in the Pacific Northwest, along with Southern California (resolving minimal strain across the creeping segment of the San Andreas fault) and South Island, New Zealand. Our specific focus in this presentation is on strain-rates prior to, and following, the 2010 El Mayor-Cucapah, 2019 Ridgecrest, 2010 Darfield/Canterbury, and 2016 Kaikoura earthquakes.

Presenting Author: Eric A. Hetland

Student Presenter: No

Authors