Verification test of the mean event size model for moderate earthquakes in the Kanto region, central Japan

In the Kanto region, there have been reports of decreasing b-values prior to earthquakes (M ≥ 5.5). The change of b-values is defined as the difference between the long- and short-term average earthquake magnitudes. A hazard function for moderately large earthquakes has also been proposed. This model was based on earthquakes that occurred between 1982 and 1999, and its effectiveness is measured retrospectively. In order to evaluate the effectiveness of the model through the use of independent data, the verification test is based on earthquakes observed after January 2000. Through the end of 2004, the only event where there was a decrease in mean event size was observed on June 3, 2000. This decrease resulted in a log-likelihood for the proposed model 1.3 units larger than that of the Poisson model, supporting the validity of the proposed model. Without accumulating further examples, we attempted to improve the verification test by expanding the study volume and by lowering the target cutoff magnitude in order to overcome the small sample size. When two other targets from the expanded volume were added, the difference in the log-likelihood (ΔlnL) increased to 3.6. In this case, the information rate per event was about 1.2, larger than that of the model period. Lowering the cutoff magnitude increased ΔlnL to 2.5. These extended tests led to higher confidence in the model with the larger ΔlnL value than did the primary test. From the viewpoint of the per-event information rate, each case involving targets with a magnitude 5.5 and larger resulted in better performance than in the model period.