Systematic Estimation of Earthquake Source Parameters for Continental Australia: Magnitude Conversions and Ground-Motion Variability

Allen (2025) presented a new continental-scale Fourier spectral attenuation model that was used to resolve source parameters for well-recorded earthquakes occurring within Australian continental crust since 1994. Using an updated dataset from Allen (2025), moment magnitudes (M) and stress drops (Δσ) are derived for 318 earthquakes from magnitude 3.1≤M≤6.6. The study explores the impact of Δσ on the relativity between local magnitude (ML) and M, and shows that earthquakes having high stress drops possess relatively higher values of ML. The between-event residuals of Fourier amplitude spectra are subsequently evaluated relative to ML–M conversion relationships and the direct use of Δσ as a ground-motion predictor variable. The study finds that knowledge of an earthquake’s source characteristics can lead to significant reductions in the standard deviation of the between-event ground-motion residuals (τ), relative to the ergodic continental-scale attenuation model. Building on these observations, this study explores the use of the k-means unsupervised machine learning clustering to develop a priori information to reduce between-event ground-motion residuals. The approach clusters earthquakes in similar geographical locations (longitude and latitude), regulated by similarities in earthquake source attributes (i.e., Δσ). Relative to the ergodic case, results of this partially nonergodic approach demonstrate that regionalized reductions in τ can be achieved through the use of these cluster-based corrections. Statistical hypothesis testing to assess the independence of the cluster-based samples of earthquake stress drops demonstrates that—in many cases—the difference between the means of the populations is statistically significant. In other cases, however, the difference is not statistically significant. This observation justifies the grouping of some regions and the development of regional ground-motion logic trees that can reduce the epistemic uncertainty of seismic-hazard models. The study confirms that an earthquake’s stress drop is a first-order variable in modeling earthquake ground motions for Australian earthquakes.