Iterative Deterministic and Bayesian Seismic Imaging of the Wasatch Fault Zone, Utah

Session: Seismic Imaging of Fault Zones [Poster]
Type: Poster
Date: 4/29/2020
Time: 08:00 AM
Room: Ballroom
Description: 

We create a high-resolution regional-scale seismic velocity model of the Wasatch fault zone in north-central Utah by incorporating newly-created catalogs of local earthquake P and S wave arrival time data (1981–2019) in conjunction with Rayleigh wave phase and ellipticity data derived from ambient seismic noise. The ambient noise and earthquake datasets exhibit complementary sensitivity to seismic velocity structure: earthquake body waves offer strong lateral constraints at seismogenic depth, frequency-dependent Rayleigh wave phase data resolve vertical gradients and Rayleigh ellipticity (or H/V) measurements are extremely sensitive to the upper few kilometers.

We iteratively apply deterministic double-difference tomography with Bayesian joint inversion of surface wave data. Specifically, we employ a machine learning algorithm (convolutional neural network) to significantly expand the number of S picks in the available catalog from the University of Utah Seismograph Stations. A least squares inversion is applied to earthquake P and S wave absolute arrival times and the derived differential travel times. We thus create models of Vp and Vs on a ~3km laterally and ~1km vertically discretized grid for the north-central region of Utah. This velocity model is then used as a starting model in a Markov Chain Monte Carlo joint surface wave inversion of Rayleigh wave H/V and phase velocity determined at each lateral grid point. Because the Monte Carlo method has the advantage of being less likely to become trapped in a local minimum, we iterate between the two methods until a final regional model is achieved. We observe new detailed patterns of seismic velocity structure in the Wasatch fault zone and north-central Utah, with implications for the ongoing tectonic evolution of the Intermountain West and corresponding seismic hazard. The resulting models (Vp and Vs) are directly applicable to ground motion simulation-based seismic hazard predictions and tectonic history interpretations.

Presenting Author: Elizabeth M. Berg

Authors