京都大学21世紀COEプログラム 活地球圏の変動解明 アジア・オセアニアから世界への発信
Seminar Series of Active Geosphere #22: July 27, 2005
by
Atsushi YAMAJI
Kyoto University
"Paleostress analysis: theories and applications"
Paleostress analysis is a technique of structural geology for estimating the states of tectonic stress from fault-slip data obtained from meso-scale faults that are observed on outcrops, borehole cores, etc. They are paleostresses, because the meso-scale faults are inactive when we observe them. However, the basic technique is common to that of seismology for determining present stress at depths from focal mechanisms. Over the last decade, much progress has been made in the methodology in this field. In this lecture, I will talk about our recent contribution to the field, and introduce the results of case studies on the secular variations of stress field in Japan with a temporal resolution of < 1 m.y. Those studies not only reveal tectonic history but also give a clue to the migration of fluids, including hydrocarbons, through shallow levels in the crust.
If a dataset records polyphase tectonics, the set is said to be heterogeneous. Given such data, it is our task to determine the stresses responsible for the tectonic phases from the data. Researchers usually attempt to classify faults based on their apparent ages to divide data into homogeneous subsets. Once the subsets are obtained, mathematical inversion, which was developed mainly around 1980, can determine the optimal stress tensor for each subset. However, such fault sorting is not always possible. For this reason, researchers have tackled the problem to develop computerized techniques for the separation of stresses and the automatic sorting of faults.
Recently, we have succeeded in solving this problem by adapting information science techniques to stress tensor inversion. We have demonstrated that the inversion comes down to statistical processing on the five-dimensional unit sphere: fault-slip data are represented by 180°-long arcs on the sphere, and the stresses responsible for the heterogeneous data are indicated by their intersections. We have developed two techniques for separating stresses from heterogeneous data. One is an application of the Hough transform, a basic technique of image processing. Another is an adaptation of resampling technique, the software of which has been open for the public since the last autumn. Both of them identify points on the hypersphere where many arcs intersect in the vicinity of the points. The latter technique yields a number of stress tensors, and their clusters represent significant solutions. The spherical k-means, a technique of machine learning, was recently found useful to detect the clusters on the five-dimensional parameter space.
High-resolution paleostress analysis has been enabled by those methods with the help of high-resolution stratigraphy. We have revealed that stress field in fore-arc regions in Japan is variable. For example, fore-arc stress in central Japan has changed a few times in the last two million years. Meanwhile, the present stress field in the southwestern Ryukyu arc near Taiwan was dated to be younger than ~0.125 Ma, the last interglacial maximum. The variable stress fields indicate that Japanese island arcs are very active.