Precursory Activity Before Larger Events in Greece Revealed by Aggregated Seismicity Data

We investigate the seismicity rate behaviour in and around Greece during 2009, seeking significant changes in rate preceding larger events. For individual larger events it is difficult to clearly distinguish precursory rate changes from other, possibly unrelated, variations in seismicity. However, when we aggregate seismicity data occurring within a radius of 10 km and in a 50-day window prior to earthquakes with, e.g. magnitude ≥3.5, the resulting aggregated time series show a clearly increasing trend starting 2–3 weeks prior to the “mainshock” time. We apply statistical tests to investigate if the observed behaviour may be simply consistent with random (poissonian) variations, or, as some earlier studies suggest, with clustering in the sense that high activity rates at some time may imply increased rates later, and thus (randomly) greater probability of larger coming events than for periods of lower seismicity. In this case, rate increases have little useful predictive power. Using data from the entire catalogue, the aggregated rate changes before larger events are clearly and strongly statistically significant and cannot be explained by such clustering. To test this we choose events at random from the catalogue as potential “mainshocks”. The events preceding the randomly chosen earthquakes show less pronounced rate increases compared to the observed rate changes prior to larger events. Similar behaviour is observed in data sub-sets. However, statistical confidence decreases for geographical subsets containing few “mainshocks” as it does when data are weighted such that “mainshocks” with many preceding events are strongly downweighted relative to those with fewer. The analyses suggest that genuine changes in aggregated rate do occur prior to larger events and that this behaviour is not due to a small number of mainshocks with many preceding events dominating the analysis. It does not automatically follow that it will be possible to routinely observe precursory changes prior to individual larger events, but there is a possibility that this may be feasible, e.g. with better data from more sensitive networks.