地殻変動データの多変量解析 -ARIMAモデルによる予測の試み-

Abstract

An attempt was made to apply a multi-variable ARIMA model (autoregressive integrated moving average model) to observed non-stationary variations of crustal strains, ground tilts, precipitation and water discharge. The purpose here is to find correlations among these observed values, to make predictions for their future variations, and to detect some possible anomalies that might be ralated to the change of tectonic stress field and the earthquake occurrence. The model fitting is applied to the records obtained at the Kamitakara Geophysical Observatory. The applicability of this model is carefully tested by the randomness of noise involved in each of the observations, and also by examining independence of each variable after removal of dependent components. The preliminary results suggest that the observed areal strains and ground tilts are somewhat affected by long-period fluctuations of precipitation in a range between 20 and 50 days but not by water discharge in the observation vault, and that the amount of water discharge is subjected to variations of areal strains in addition to the effects of rainfall. It is also suggested by the present analysis that the large amplitude variations of strains and tilts observed during some period in 1978 are within the range of prediction errors and may not be any significant changes due to tectonic origins.