dc.description.abstract | Weighted least-squares is commonly used in GPS time series analysis due to its simplicity and efficiency. However, practical applications often face fitting issues because of unknown transient and time-varying signals within GPS data. To solve this, this research adopts the Kalman filter, a least squares estimation method for stochastic process.
Temporal correlation of colored noise poses a challenge in both Kalman filtering and least squares methods. A study by Didova et al. (2016) addresses this by applying the shaping filter proposed by Bode and Shannon (1950), using autoregressive models to create a colored noise model. They found that AR(3) models effectively simulate colored noise in GPS time series. As a result, this research uses this approach, taking 96 stations across Taiwan from the Geological Survey Institute as examples. It filters out colored and white noise from the time series for noise analysis, comparing it with the traditional Hector method by Bos et al. (2013) to examine the differences in noise analysis.
In the Kalman filter, using the integrated random walk to model trends variation tends to oversmooth signals like post-seismic deformation due to its inability to handle time-correlated velocity changes effectively. To address this, instead of treating acceleration as white noise, this research introduces an AR(1) model for acceleration in the state space model during post-seismic period, improving accuracy in estimating both post-seismic and co-seismic deformations. Furthermore, the study utilizes the Kalman filter-estimated velocity/displacement field for strain rate/strain calculations, exemplifying its application using 75 GPS stations in southwestern Taiwan before and after the 2016 Meinong earthquake, offering additional data for surface deformation analysis three years after and before Meinong earthquake. | en_US |