Multivariate time-series model to estimate the arrival times of S-waves

Abstract Some computationally efficient procedures, which had been developed for the estimation of the onset time of seismic waves, are examined for their ability in determining the onset time of the S-wave in an online system. The three methods we compare are the univariate locally stationary autoregressive model (FUNIMAR or LSAR), multivariate locally stationary autoregressive model (MLSAR), and the summing AIC procedure. The summing AIC's procedure is introduced by adding AIC values at each time point for the three univariate autoregressive models. Although this simple summing can provide the same estimation as that by MLSAR at substantial computational savings, its implicit assumption of independence of each component sheds some doubt upon its results. The procedure of multivariate autoregressive model (2-V MLSAR) for horizontal components is most useful for the precise estimation of the arrival time of the S-wave. FUNIMAR is sufficient to determine the arrival time of the P-wave, but not appropriate to determine the arrival time of the S-wave.