Ensemble Prediction and Uncertainty Quantification of the Propagation of Weak Seismic Pulses with Minimal Site Characterization

To make a prediction for the propagation of seismic pulses, one needs to specify physical properties and subsurface ground structure of the site. This information is frequently unknown or estimated with significant uncertainty. We developed a methodology for the ensemble prediction of the propagation of weak seismic pulses for short ranges. The ranges of interest are 10-100 of meters, and the pulse bandwidth is up to 200 Hz. Instead of specifying specific values for viscoelastic site properties, the methodology operates with probability distribution functions of the inputs. This yields ensemble realizations of the pulse at specified locations, where mean, median, and maximum likelihood predictions can be made, and confidence intervals are estimated. Starting with the site's Vs30, the methodology creates an ensemble of plausible vertically stratified Vs profiles for the site. The number and thickness of the layers are modeled using inhomogeneous Poisson process, and the Vs values in the layers are modeled by Gaussian correlated process. The Poisson expectation rate and Vs correlation between adjacent layers take into account layers depth and thickness, and are specific for a site class, as defined by the Federal Emergency Management Agency (FEMA). High-fidelity three-dimension thin layer method (TLM) is used to yield an ensemble of frequency response functions. Comparison with experiments revealed that measured signals are not always within the predicted ensemble. Variance-based global sensitivity analysis has shown that the most significant parameter in the TLM for the prediction of the pulse energy is the shear quality factor, Qs. Some strategies how to account for significant uncertainty in this parameter and to improve accuracy of the ensemble predictions for a specific site are investigated and discussed.