Characterizing the Epistemic Uncertainty in the USGS 2014 National Seismic Hazard Mapping Project (NSHMP)

To account for the epistemic uncertainty in probabilistic seismic hazard and risk analysis, multiple scientifically justifiable and defensible models are usually considered with a logic tree. Closed-form solutions for the distribution of the resultant hazards, also called a logic-tree distribution (LTD), are complex and difficult or impossible to obtain. Quantification of this distribution, although more common in seismic hazard estimates for critical facilities, is generally impractical for many hazard and risk-related applications in which time and budget resources are often limited. Conventionally, the weighted mean resulting from the logic tree is reported, and decision makers are left blind to the divergent views that exist in the larger scientific community. In this article, we implement the U.S. Geological Survey (USGS) 2014 National Seismic Hazard Mapping Project (NSHMP) model using the Robust Simulation approach, in which results arising from multiple models are represented through an ensemble of solutions. To construct the ensemble, we sample the logic tree of the USGS hazard model, with the likelihood of each sampled branch proportional to the weight of that branch in the logic tree. The LTD of peak ground acceleration (PGA) and spectral acceleration at 1 s (SA1), for exceedance probabilities of 10% and 2% in 50 yrs, are examined for five seismically active regions: southern California, northern California, Pacific Northwest, the Wasatch Front fault zone, and the New Madrid seismic zone. Our mean hazard of the sampled ensemble matches the USGS-published hazard values and the sampled ensemble reveals the full LTD. The shape of the LTD can be largely explained with a lognormal probability distribution. This suggests that the lognormal standard deviation (hereafter, theta) of the LTD may be mapped and used as a simple metric to express the range of technically defensible interpretations considered in the USGS 2014 NSHMP.