Short-term earthquake probabilities during the L’Aquila earthquake sequence in central Italy, 2009

Abstract:

We compare the forecasting performance of several statistical models, which are used to describe the occurrence process of earthquakes, in forecasting the short-term earthquake probabilities during the occurrence of the L’Aquila earthquake sequence in central Italy, 2009. These models include the Proximity to Past Earthquakes (PPE) model and different versions of the Epidemic Type Aftershock Sequence (ETAS) model. We used the information gains corresponding to the Poisson and binomial scores to evaluate the performance of these models. It is shown that all ETAS models work better than the PPE model. However, when comparing the different types of the ETAS models, the one with the same fixed exponent coefficient α = 2.3 for both the productivity function and the scaling factor in the spatial response function, performs better in forecasting the active aftershock sequence than the other models with different exponent coefficients when the Poisson score is adopted. These latter models perform only better when a lower magnitude threshold of 2.0 and the binomial score are used. The reason is likely due to the fact that the catalog does not contain an event of magnitude similar to the L’Aquila main shock (Mw 6.3) in the training period (April 16, 2005 to March 15, 2009). In this case the a-value is under-estimated and thus also the forecasted seismicity is underestimated when the productivity function is extrapolated to high magnitudes.

These results suggest that the training catalog used for estimating the model parameters should include earthquakes of similar magnitudes as the main shock when forecasting seismicity is during an aftershock sequences.