Implication of Migration Pattern For Critical Parameters

Seismicity migration has been shown by improved pattern informatics (PI) method in retrospective studies. The retrospective studies also show that the migration patterns are dependent on the tectonic setting and parameters used in statistical calculation. For the purpose of a forecast, the critical parameters for a specified efficient forecast period have to be learned. In this study we propose that a forecast and its critical parameters, that is, the magnitude and precursory time, can be learned from a series of empirical migration patterns. We also make the forecast with half year of efficient forecast period for three densely populated seismic regions in Taiwan, Japan, and California based on the migration patterns. The seismic catalog in each region was cut in depth according to the seismogenic layer. Considering the activity of seismicity, the cut magnitudes M_c are given at 3.0, 3.5, and 4.0. The magnitude M_t of the target events are 5.0, 5.5, and 6.0, but the eventual magnitude of the event which can be forecast was determined by the performance of the migration pattern. We calculate the migration pattern with different cut magnitudes and time parameters and count the number of migration hotspots N_hs and the number of hit hotspots N_hit; the migration hotspots means the grids which seismicity migrate toward and the hit hotspots are the migration hotspots with M>M_t target events occurred in the following efficient time on them. The performance of the migration pattern depends both on the ratio of the number of hit hotspots N_hit to the number of target event N_event which occurred in the following efficient time and the ratio N_hit/N_hs. Giving the criteria of the performance that highest N_hit/N_hs with N_hit/N_event=1, we obtained the critical parameters for each region. In western Taiwan, the M>=5 events can be forecast using earthquakes with cut magnitude 4.0 and the total precursory time could be about 2.5 years. In Kanto region, the M>=5 events can be forecast using earthquakes with cut magnitude 3.0 and the total precursory time could be about 2 years. In the region of California, the M>=6 events can be forecast using earthquakes with cut magnitude 3.5 and the total precursory time could be about 4.5 years.