Differencing reveals hidden changes in baseline length time-series

In order to make successful earthquake predictions, detection and monitoring of baseline changes are important for investigating their origins, including precursory crustal deformations in tectonically active areas. In this study, differencing two baselines that run approximately parallel to each other and normal to the expected crustal deformations, and that share a station is proposed for analysis. Differencing reduces common systematic baseline errors, thereby enabling detection of subtle transient systematic changes in the baseline time-series that are otherwise buried in the measurement noise. Mean shift analysis, a well-known statistical technique to determine hether the mean of a stochastic process has shifted using cumulative sum charts, can then be used to locate the change points in the time-series. The application of this method to the differences of concurrently observed very long baseline interferometry (VLBI) and global positioning system (GPS) baselines in the Japanese Keystone project, where periodic and persistent baseline changes are removed, revealed transient variations in the baseline lengths several months prior to the seismic activity in the Izu Islands that started on June 26, 2000. Reproduction of the results using GPS and VLBI, two alternative baseline measurement techniques, validated the accuracy of the proposed approach for detecting previously hidden transient changes in the baseline lengths.