Automatic Aftershock Forecasting: A Test Using Real-Time Seismicity Data in Japan

Real-time aftershock forecasting is important to reduce seismic risks after a damaging earthquake. The main challenge is to prepare forecasts based on the data available in real time, in which many events, including large ones, are missing and large hypocenter determination errors are present due to the automatic detection process of earthquakes before operator inspection and manual compilation. Despite its practical importance, the forecast skill of aftershocks based on such real-time data is still in a developmental stage. Here, we conduct a forecast test of large inland aftershock sequences in Japan using real-time data from the High Sensitivity Seismograph Network (Hi-net) automatic hypocenter catalog (Hi-net catalog), in which earthquakes are detected and determined automatically in real time. Employing the Omori-Utsu and Gutenberg-Richter models, we find that the proposed probability forecast estimated from the Hi-net catalog outperforms the generic model with fixed parameter values for the standard aftershock activity in Japan. Therefore, the real-time aftershock data from the Hi-net catalog can be effectively used to tailor forecast models to a target aftershock sequence. We also find that the probability forecast based on the Hi-net catalog is comparable in performance to the one based on the latest version of the manually compiled hypocenter catalog of the Japan Meteorological Agency when forecasting large aftershocks with M > 3.95, despite the apparent inferiority of the automatically determined Hi-net catalog. These results demonstrate the practical usefulness of our forecasting procedure and the Hi-net automatic catalog for real-time aftershock forecasting in Japan.