Towards improved aftershocks forecasts

The destructive nature of earthquakes and the significant losses they cause underscore the essential importance of hazard assessment for effective earthquake risk mitigation. Earthquake forecasts is an important ingredient for time-dependent seismic hazard assessments. Accurate forecasting models, as an important backbone of seismic hazard analysis, can help mitigate damage, save lives, and inform preparedness strategies. This thesis contributes to advancing earthquake forecasting, with a particular focus on aftershocks, which can be as destructive and deadly as the mainshocks, by exploring various perspectives and approaches. Understanding the underlying physical mechanisms and statistical patterns of aftershocks is crucial for enhancing the accuracy and effectiveness of practical forecasting models. The occurrence of aftershocks is mainly related to the stress changes of the mainshock, although other effects such as the distance to the ruptured fault may also play an important role. First, I analyse the 2017-2019 Kermanshah (western Iran) sequence regarding the potential to forecast the spatial distribution of aftershocks based on information about the mainshock and its largest aftershocks. For the complex Kermanshah sequence, I applied five stress scalars, including three receiver-dependent metrics: (i) Coulomb failure stress (∆CFS) on master fault orientation (MAS), (ii) ∆CFS on optimally oriented planes (OOP), and (iii) ∆CFS assuming fault variability (VM), and two receiver-independent stress scalars (iv) maximum shear (MS) and (v) von-Mises stress (VMS), along with a distance-slip probabilistic model (R). Furthermore, I accounted for the potential impact of large secondary earthquakes and, recognizing the limitations of the receiver operating characteristic (ROC) metric in imbalanced datasets, also used the more appropriate MCC-F1 metric for testing. I observed that the receiver-independent stress scalars, together with the R model, provide better forecasts than the classical CFS values that rely on the specification of receiver mechanisms. I found that the performance of the best methods improves when the source information of large aftershocks is additionally considered. These results highlight the significance of incorporating all events in a sequence to improve earthquake forecasts, which may be achieved by incorporating probability maps from stress metrics and the R model into the spatiotemporal Epidemic Type Aftershock Sequence (ETAS) model. The ETAS model is the most widely used and powerful statistical model for aftershock forecasting. While current earthquake catalogs provide high-precision depth values that contain valuable information, the ETAS is two-dimensional. Thus, I extended the spatiotemporal ETAS model to 3D by considering hypocentral distances instead of epicentral distances, which also allows incorporating available 3D spatial probability maps of large events into its spatial kernel. To explore the most appropriate parametric form of the spatial kernel, I first examined different triggering functions for the depth difference of the aftershock-mainshock pairs identified by the Nearest-Neighbor (NN) method, showing that a magnitude-dependent power-law kernel fits best the earthquakes data in Southern California. Therefore, I incorporated the corresponding kernel into the 3D-ETAS model with space-dependent background activity, where I additionally allow for depth-dependent aftershock productivity. The application to Southern California showed that the model fits the data well. I also found that the aftershock productivity strongly depends on depth, similar to the seismic moment released by the mainshocks, which may be related to depth-dependent seismic coupling. While the distribution of aftershocks around the mainshock is anisotropic, the spatial probability density function of the ETAS model is commonly assumed to be isotropic due to insufficient information. In addition, its parameter estimation can be highly biased due to catalog incompleteness after the mainshock. Thus, I extended the recently developed 2D temporal ETASI, which accounts for short-term incompleteness, to 2D and 3D spatiotemporal ETASI, considering additional spatial occurrence probabilities in the framework of ETAS and ETASI to improve aftershock forecasting. I replaced the isotropic spatial kernel with anisotropic kernels estimated by a spatial probability map of stress scalars, including MAS, VM, MS, and VMS, and the nearest distance to the ruptured fault of the mainshock (R). The fit to six prominent mainshock-aftershock sequences in California demonstrates that the ETASI model outperforms the standard ETAS model. Furthermore, positive information gains indicate that using stress calculations as additional input information can improve the parameter fit. This improvement is weaker in 3D, which is likely related to greater positional uncertainty in the depth domain. However, incorporating the probability map calculated as a function of the nearest distance to the mainshock rupture leads to the best performance in all model variants. The findings in this thesis provide valuable insights into aftershock forecasting by investigating and developing physics-based, statistical and hybrid approaches. The results contribute to a better understanding of aftershocks behavior and forecasting, which can assist in the development of more accurate probabilistic seismic hazard analysis and risk reduction strategies.