Numerical shake prediction for Earthquake Early Warning: data assimilation, real-time shake-mapping, and simulation of wave propagation

Abstract:

In many methods of the present Earthquake Early Warning (EEW) systems, hypocenter and magnitude are determined quickly and then strengths of ground motions are predicted. The 2011 Tohoku Earthquake (M_W9.0), however, revealed some technical issues with the conventional methods: under-prediction due to the large extent of the fault rupture, and over-prediction due to confusion of the system by multiple aftershocks occurred simultaneously. To address these issues, a new concept is proposed for EEW: applying data assimilation technique, present wavefield is estimated precisely in real time (real-time shake mapping) and then future wavefield is predicted time-evolutionally using physical process of seismic wave propagation. Information of hypocenter location and magnitude are not required, which is basically different from the conventional method.

In the proposed method, data assimilation technique is applied to estimate the current spatial distribution of wavefield, in which not only actual observation but also anticipated wavefield predicted from one time-step before are used. Real-time application of the data assimilation technique enables us to estimate wavefield in real time, which corresponds to real-time shake mapping. Once present situation is estimated precisely, we go forward to the prediction of future situation using simulation of wave propagation.

The proposed method is applied to the 2011 Tohoku Earthquake (M_W9.0) and the 2004 Mid-Niigata earthquake (Mw6.7). Future wavefield is precisely predicted, and the prediction is improved with shortening the lead time: for example, the error of 10 s prediction is smaller than that of 20 s, and that of 5 s is much smaller. By introducing this method, it becomes possible to predict ground motion precisely even for cases of the large extent of fault rupture and the multiple simultaneous earthquakes.

The proposed method is based on a simulation of physical process from the precisely estimated present condition. This method, therefore, corresponds to “numerical shake prediction”, on the analogy to “numerical weather prediction” in meteorology.