Bayesian on-fault b-value estimates inferred from paleoseismology and the NSHM23-WUS model

Kevin R. Milner, Morgan T. Page, & Edward H. Field

Submitted August 30, 2026, SCEC Contribution #15401, 2026 SCEC Annual Meeting Poster #152

Although the Gutenberg-Richter (GR) magnitude-frequency distribution (MFD) with a b-value of approximately 1 has been robustly observed globally and for regions, the magnitude-frequency distribution of large ruptures on individual faults remains poorly constrained. The 2023 update to the USGS National Seismic Hazard Model (NSHM) for the western U.S. (NSHM23-WUS) enumerated those uncertainties by imposing GR relationships on faults spanning an assumed uniform b-value distribution from b=0 to b=1. The final solution on-fault MFDs were determined through a fault-system inversion matching the assumed GR b-value, slip rate data from deformation models, and subject to segmentation rules. For faults with paleoseismic recurrence estimates, the inversion also used those data as direct constraints and mediated any disagreements, attempting to fit each constraint equally relative to their uncertainties. The addition of the on-fault b-value constraint, absent from the 2014 model, resulted in generally shorter recurrence intervals on some faults in NSHM23-WUS than in that prior model, notably on the northern San Andreas fault. Although paleoseismic data were well-fit in aggregate relative to their uncertainties, areas of systematic misfit remain, including on the northern San Andreas and Wasatch faults.

We present a Bayesian analysis of on-fault b-values for faults in California and Utah using NSHM23-WUS’s assumed uniform b-value distribution as a prior distribution, and develop posterior distributions using a Gaussian log-likelihood based on misfit to paleoseismic data. Paleoseismic studies supply recurrence estimates at distinct points along faults rather than for entire fault sections, so we developed a connectivity-based weighting scheme to encode inferences for a given fault section from all connected paleoseismic sites (either on the same or on nearby fault sections). The weight calculation assumes the average prior b-value (b=0.5) and branch-dependent segmentation rules. The posterior b-value distributions are calculated before the fault-system inversion and could thus be directly used as inputs in future NSHMs to narrow uncertainties and better fit available paleoseismic data in the final model. Our preliminary results show better agreement with paleoseismic data using lower b-values (b~0) for the northern San Andreas fault, and higher b-values (b~1) for the San Jacinto and Wasatch faults.

Key Words
Paleoseismology, NSHM23, San Andreas, b-value, ERF

Citation
Milner, K. R., Page, M. T., & Field, E. H. (2026, 08). Bayesian on-fault b-value estimates inferred from paleoseismology and the NSHM23-WUS model. Poster Presentation at 2026 SCEC Annual Meeting.

Related Projects & Working Groups
Earthquake Forecasting and Predictability (EFP)