Assessing Bias and Variability in Physics-Based Ground-Motion Simulations for Moderate-Magnitude Earthquakes in Southern California

ABSTRACT We evaluate physics-based ground-motion simulations against observed motions for 30 moderate-magnitude (M 4.0–5.5) earthquakes (3604 recordings at 782 stations) in the greater Los Angeles region of southern California. Simulations use both plane-layer (1D) and laterally variable (3D) velocity structures, and performance is assessed using effective amplitude spectrum (EAS) residual analysis across the frequency band 0.1–1.0 Hz. Sensitivity of key simulation parameters, including minimum shear-wave velocity (VS), anelastic attenuation scaling, and VS30-based near-surface tapering in the 3D velocity model, is evaluated through their influence on residuals. Results show that adopting a minimum VS of 200 m/s combined with an anelastic attenuation scaling of QS=100VS reduces mean EAS bias in the upper-reliable frequency band by ∼0.4 ln units, and near-surface velocity tapering in nonbasin regions reduces bias further by ∼0.2 ln units. For events below M 5, simulations based on empirical magnitude–area scaling relations perform comparably with observations. However, for two larger events (M 5.39 and 5.12), simulations using rupture areas constrained by finite-fault studies reduce EAS residuals by ∼35% to 65% in amplitude over 0.3–0.7 Hz relative to those based on empirical relations (though both events remain underpredicted). Mixed-effects regression is used to quantify between-event (τ), site-to-site (ϕS2S), within-site (ϕSS), and total (σ) variability for the calibrated simulations and an EAS-based ground-motion model (GMM). In the reliable frequency range, τ is comparable across all approaches, the 3D simulations reduces ϕSS and σ relative to 1D site-adjusted simulations (by ∼0.05 ln units), and both simulation types yield lower ϕS2S than the GMM, compressing the geomorphic-dependent site bias evident in the GMM (by up to ±0.10–0.13 ln units) to within ±0.05 ln units for most categories. These results reinforce the importance of jointly calibrating 3D velocity models and earthquake source characterizations for physics-based seismic hazard applications.