Assessment of a Regional Stochastic Point-source Model for Hybrid Empirical Applications

The Hybrid Empirical Method (HEM) is widely used to predict ground motions in regions with sparse strong-motion data by transferring empirically derived ground-motion prediction equations (GMPEs) from data-rich host regions. A fundamental requirement for the reliability of this approach is that the stochastic point-source model adopted for the host region be consistent with the empirical GMPEs developed from the same dataset. Despite its importance, this consistency is often assumed or verified only over limited magnitude-distance conditions. This study presents a systematic evaluation of the observation-calibrated stochastic equivalent point-source model of Yenier and Atkinson (2015, YA15) for California. Both the single corner-frequency (SCF) and double corner-frequency (DCF) formulations are assessed over moment magnitudes Mw 3.0-7.5, rupture distances Rrup of 1-300 km, and frequencies from 0.1 to 10 Hz. Model predictions of pseudo-spectral acceleration are compared with the median predictions of four widely used four NGA-West2 GMPEs, and consistency is quantified using normalized residuals that account for inter-model variability. Results show that the YA15 model generally reproduces the empirical predictions well for moderate magnitudes (Mw 4.5-7.0) over most distances and frequencies. However, systematic discrepancies are identified for small-magnitude events at low frequencies in the near and intermediate distance ranges, at high frequencies in the near field, and for large-magnitude events in the near-fault region. Furthermore, the more complex DCF formulation does not demonstrate systematic improvement over the SCF model at higher magnitudes and exhibits increased variability and larger residual amplitudes in certain magnitude–distance–frequency regimes. These findings delineate the effective applicability domain of observation-calibrated equivalent point-source models in HEM applications and clarify the limitations of the point-source approximation when extrapolated beyond its primary calibration range. The results provide practical guidance for selecting host-region stochastic models in hybrid empirical frameworks and indicate conditions under which more physically detailed finite-fault simulations may be required.