One of the key parameters for earthquake source physics is stress drop since it can be directly linked to the spectral level of ground motion. Stress drop estimates from moment corner frequency analysis have been shown to be extremely variable, and this to a much larger degree than expected from the between‐event ground motion variability. This discrepancy raises the question whether classically determined stress drop variability is too large, which would have significant consequences for seismic hazard analysis. We use a large high‐quality data set from Japan with well‐studied stress drop data to address this issue. Nonparametric and parametric reference ground motion models are derived, and the relation of between‐event residuals for Japan Meteorological Agency equivalent seismic intensity and peak ground acceleration with stress drop is analyzed for crustal earthquakes. We find a clear correlation of the between‐event residuals with stress drops estimates; however, while the island of Kyushu is characterized by substantially larger stress drops than Honshu, the between‐event residuals do not reflect this observation, leading to the appearance of two event families with different stress drop levels yet similar range of between‐event residuals. Both the within‐family and between‐family stress drop variations are larger than expected from the ground motion between‐event variability. A systematic common analysis of these parameters holds the potential to provide important constraints on the relative robustness of different groups of data in the different parameter spaces and to improve our understanding on how much of the observed source parameter variability is likely to be true source physics variability.

We investigate the nonergodic behavior of ground motions from subduction zone earthquakes and crustal earthquakes in Japan using the National Research Institute for Earth Science and Disaster Resilience strong‐motion flatfile together with various reference ergodic ground‐motion models (GMMs). For the nonergodic path effects, our nonergodic GMM has two path effects terms: a cell‐specific linear‐distance scaling, denoted as $δP2PQ$⁠, that mimics the effects of a 3D Q structure, and a source‐ and site‐specific term related to the effects of the 3D velocity structure, denoted $δP2PV$⁠. The resulting model provides spatially varying nonergodic source, site, and path terms. The GMM smoothly interpolates and extrapolates the nonergodic terms in space so that the GMM can be applied to any combination of source and site locations in the region. In regions where data are sparse, the median nonergodic terms will approach zero but with large epistemic uncertainty. Over the period range of 0–10 s, the ranges of the standard deviations of the nonergodic source, path, and site terms are 0.2–0.65, 0.25–0.40, and 0.15–0.40 natural log units, respectively. The fully nonergodic models lead to a 40%–55% decrease in the aleatory standard deviation compared to the reference ergodic GMMs. This large reduction in the aleatory standard deviation combined with the change in the median given by the nonergodic terms can exert a significant impact on the computed seismic hazard for the Japan region.

The 2009 L'Aquila, Italy, earthquake (M_w 6.3) produced an unprecedented number of seismological records for a normal faulting event, and hence represents an important case study of damaging moderate magnitude earthquakes occurring in densely inhabited areas. We have investigated the source process of the 2009 L'Aquila earthquake in both the low- and broad-band frequency ranges by analysing the teleseismic and strong motion data. A low-frequency (<0.5 Hz) source model has been determined by waveform inversion of teleseismic and near-field strong motion data, yielding slip distributions with a major asperity located about 8 km southeast of the hypocentre. The broad-band (0.2–10 Hz) source characteristics were estimated using the empirical Green's function method providing the strong motion generation area of about 42 km² located southeast of the hypocentre below the village of Onna. This coincides with the location of the asperity from the inversion analysis, indicating that the low-frequency and broad-band ground motions produced during the earthquake were generated at similar locations on the fault plane. The position of the strong motion generation area also correlates well with the distribution of observed macroseismic (MCS) intensities reported by INGV. The maximum macroseismic intensities of IX–X were reported southeast of the epicentre, with an intensity of X measured in Onna. The source properties of the 2009 L'Aquila mainshock reveal the estimated size of the strong motion generation area to be in agreement with predictions made using empirical source scaling relation for inland crustal earthquakes, implying a stress drop of approximately 10 MPa. This further indicates that the 2009 L'Aquila earthquake involved a stress drop similar to that of previously recorded earthquakes in the Central and Southern Apennines.

Physics-informed neural networks (PINNs) have great potential for flexibility and effectiveness in forward modeling and inversion of seismic waves. However, coordinate-based neural networks (NNs) commonly suffer from the "spectral bias" pathology, which greatly limits their ability to model high-frequency wave propagation in sharp and complex media. We propose a unified framework of Fourier feature physics-informed neural networks (FF-PINNs) for solving the time-domain wave equations. The proposed framework combines the stochastic gradient descent (SGD) strategy with an independently pre-trained wave velocity surrogate model to mitigate the singularity at the point source. The performance of the activation functions and gradient descent strategies are discussed through ablation experiments. In addition, we evaluate the accuracy comparison of Fourier feature mappings sampled from different families of distributions (Gaussian, Laplace, and uniform). The second-order paraxial approximation-based boundary conditions are incorporated into the loss function as a soft regularizer to eliminate spurious boundary reflections. Through the non-smooth Marmousi and Overthrust model cases, we emphasized the necessity of the absorbing boundary conditions (ABCs) constraints. The results of a series of numerical experiments demonstrate the accuracy and effectiveness of the proposed method for modeling high-frequency wave propagation in sharp and complex media.