Testing spatial aftershock forecasts accounting for large secondary events during on going earthquake sequences: A case study of the 2017-2019 Kermanshah sequence

Large earthquakes are always followed by aftershocks sequence that last for months to years. Sometimes, these aftershocks are as destructive as the mainshocks. Hence, accurate and immediate prediction of aftershocks’ spatial and temporal distribution is of great importance for planning search and rescue activities. Despite large uncertainties associated with the calculation of Coulomb failure stress changes (ΔCFS), it is the most commonly used method for predicting spatial distributions of aftershocks. Recent studies showed that classical Coulomb failure stress maps are outperformed by alternative scalar stress quantities, as well as a distance-slip probabilistic model (R) and deep neural networks (DNN). However, these test results were based on the receiver operating characteristic (ROC) metric, which is not well suited for imbalanced data sets such as aftershock distributions. Furthermore, the previous analyses also ignored the potential impact of large secondary earthquakes.

In order to examine the effects of large events in spatial forecasting of aftershocks during a sequence, we use the 2017-2019 seismic sequence in western Iran. This sequence started by Azgeleh M7.3 mainshock (12 November 2017) and followed by Tazehabad M5.9 (August 2018) and Sarpol-e Zahab M6.3 (November 2018) events. Furthermore, 15 aftershocks with magnitude > 5.0 and more than 8000 aftershocks with magnitude > 1 were recorded by Iranian seismological center (IRSC) during this sequence (12.11.2017-04.07.2019). For this complex sequence, we applied the classical Coulomb failure stress, alternative stress scalars, and R forecast models and used the more appropriate MCC-F₁ metric to test the prediction accuracy. We observe that the receiver independent stress scalars (maximum shear and von-Mises stress) perform better than the classical CFS values relying on the specification of receiver mechanisms (ΔCFS resolved on master fault, optimally oriented planes, and variable mechanism). However, detailed analysis based on the MCC-F₁ metric revealed that the performance depends on the grid size, magnitude cutoff, and test period. Increasing the magnitude cutoff and decreasing the grid size and test period reduces the performance of all methods. Finally, we found that the performance of all methods except ΔCFS resolved on master fault and optimally oriented planes improve when the source information of large aftershocks is additionally considered, with stress-based models outperforming the R model. Our results highlight the importance of accounting for secondary stress changes in improving earthquake forecasts.