Master of Science
Planetary Science and Exploration
The presence of strong aftershocks can increase the seismic hazard following a large earthquake and should be considered for operational earthquake forecasting and risk management. Aftershock forecasts are generated from seismicity models during the evolution of the aftershock sequence. This work compares quantitative test results of the forecasting abilities for three competing aftershock rate models - the modified Omori law, the Epidemic Type Aftershock Sequence model, and the compound Omori law - to identify the best performing model for forecasting the largest aftershock during the early aftershock sequence. Forecasts of large aftershock probabilities are generated by either the Extreme Value distribution or the Bayesian Predictive distribution for the forecasting time interval. Testing is conducted retrospectively on five sequences for a fixed forecasting time interval of seven days during the early aftershock sequence. None of the models and forecasting methods consistently outperforms the others regardless of the training time interval.
Summary for Lay Audience
Large earthquakes and their subsequent aftershocks are destructive. The presence of strong aftershocks can increase the seismic hazard and should be considered for operational earthquake forecasting and risk management. In this study, selected large earthquakes and their aftershocks are analyzed using three models, the modified Omori law, the Epidemic Type Aftershock Sequence model, and the compound Omori law. The models are applied to data available during various lengths of time as the earthquake sequence progresses. For each length of time, the behaviour of the sequence is simulated for the next seven days. From the simulation, the probability of a large aftershock can be computed as a forecast. This study mimics operational forecasting by progressively increasing the amount of available data for each forecast. The probability for large aftershocks is computed using two different methods - the Extreme Value distribution and the Bayesian Predictive distribution. Several statistical tests are applied to the forecasting methods and evaluated for performance. Models are scored based on quantitative values for their performance and the results are used to compare the performance of the respective models and forecasting methods. The goal of this work is to determine whether one model and forecasting method consistently scores better than the others when forecasting large aftershocks using data available during the early aftershock sequence.
The results of the statistical testing indicate that there is no best model nor forecasting method. The most suitable model and method may be regional or sequence dependent. This suggests that the choice of using one model over another should be carefully considered. To forecast the probability of the largest aftershock occurring during a short time period more reliably, the early aftershock behaviour of sequences requires more detailed analysis in future studies.
Dong, Elisa, "Testing Aftershock Forecasts Using Bayesian Methods" (2022). Electronic Thesis and Dissertation Repository. 8481.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License