A unified method for analysis of short time-series with observational error is proposed in order to make characteristics of the time-series clear. When 32 points are given in a certain time-series, the orthogonal array L32, whose columns correspond to discrete pulses, is used in this analysis method. These pulses are so-called factors in Design of Experiments. We pick up several main basic pulses and sub-basic pulses which are considered as orthogonal basic vectors in Vector Space, and other random pulses are assumed to be error pulses. Analysis of variance (ANOVA) is used in this method, and significant pulses are detected by F test with 5% significant level . We can easily classify any time-series into several types by the result of ANOVA. As an example, mean sea level analysis at Fukaura and Oga is tried by this method, because we are interested in the earth crust movement in relation to the 1983 Central Japan Sea Earthquake. The result shows that the patterns of the real fluctuations of mean sea level are very similar in the both areas, and abnormal upheaval of crust before the earthquake was found as a statistically significant phenomenon .
抄録全体を表示