地震波自動処理システムにおける読み取りおよび評価

抄録

In automatic processing of seismic waves, it is important not only to make efforts to read the arrival times of P and S phases with better accuracy for each observation station but to check the validity of results of reading in order to reject the misread items automatically. With this point of view, a practical method for reading and checking the seismic data in automatic processing system was designed, making a comparison with data from the manual processing system of the Regional Center for Earthquake Prediction, Kyoto University.
An outline of the method is summarized as follows:
(1) Basic principle for reading the phase times.
Reading the arrival times is equivalent to dividing a section into two parts in time series. For the criterion of dividing, AIC (Akaike Information Criterion; defined as AIC=-2{maximum log. likelihood—number of free parameters of model}) was adopted.
(2) Reading P times
a) Use of low cut filter for elimination of microseisms. b) Use of normal distribution model and also autoregressive model for further improvements.
(3) Checking P times
a) Quantification of the degree of certainty of reading values. b) Check by means of apparent velocity between observation stations. c) Check by means of hypocenter calculation.
(4) Reading S times
a) Finding out a time window for S times when a hypocenter cannot be determined. b) Applying a bivariate normal distribution model to horizontal components if available.
(5) Checking S times
a) Use of P time-(S-P) time graph (the Wadachi diagram).
(6) Hypocenter calculation
a) Use of robust estimation.
This method was tested for the seismic wave data observed by the network at the Regional Center, from 5th to 8th and from 17th to 19th December, 1984. As regards the differences between the respective P times read out manually and automatically, the relative frequency within ±0.05sec. was amounted to about 85 percent for earthquakes in the neighborhood of the network. Next, we checked the validity of S phase determination, fitting a straight line to the relation between P time—origin time and S-P time. The gradient of the line from automatic determination was estimated as 0.69 and their standard deviation is 0.32sec., while, 0.67 and 0.21sec. for manual reading. These differences, however, were so slight as to be almost negligible. By using an ordinary minicomputer, it took only several seconds to process a channel, so that, an earthquake was treated exhaustively during a few minutes.
According to the results of these tests, the method for processing the seismic wave data may be proved to be useful for the practical purpose.