← Back to Sessions

Evaluating Spatial Smoothing for the 2023 USGS National Seismic Hazard Model

Description: 

A long-term model for earthquake occurrence is a critical component for probabilistic seismic hazard assessments (PSHA). One approach to constructing these models is declustering and spatially smoothing an earthquake catalog in order to forecast the location of future seismicity. Past versions of the U.S. Geological Survey National Seismic Hazard Model (NSHM) relied on spatial smoothing methods using a two-dimensional Gaussian kernel of either fixed or adaptive (variable) bandwidth (Frankel, SRL, 1995; Helmstetter et al., SRL, 2007; Moschetti, BSSA, 2015). These two methods were represented as separate logic tree branches that were weighted and combined for the final hazard model. As the NSHM is updated for 2023, we re-examine these methods and explore different testing strategies to evaluate their performance, which can inform how they will be implemented in future models.

Likelihood-based methods are commonly used for testing spatial earthquake forecasts (e.g., the Regional Earthquake Likelihood Models and Collaboratory for the Study of Earthquake Predictability experiments, Schorlemmer et al., SRL, 2007; Zechar et al., BSSA, 2010). For the NSHM adaptive smoothing, they are used to optimize the nearest-neighbor number (N-value) that determines each event’s smoothing distance (Moschetti, BSSA, 2015). For 2023, we are re-optimizing the N-values for both the western and eastern US because of changes in boundaries between and within the two regions. We explore the effect of choosing different durations of training and testing data sets for that optimization, and what might be most appropriate for a 50-year forecast, given that the data does not yet exist to test such a forecast. We also consider whether the fixed smoothing model, which often requires the use of floor rates in low-seismicity regions, can be replaced by an adaptive smoothing model with an appropriate N-value(s).

Session: USGS National Seismic Hazard Models: 2023 and Beyond [Poster]

Type: Poster

Date: 4/18/2023

Presentation Time: 08:00 AM (local time)

Presenting Author: Andrea L. Llenos

Student Presenter: No

Invited Presentation: 

Authors