Transient deformation from the 2016 Meinong earthquake in Southwestern Taiwan revealed by Kalman filter analysis of GNSS time series

Weighted least-squares is commonly used in GNSS (Global Navigation Satellite System) time series analysis due to its simplicity and efficiency. However, practical applications often encounter fitting issues due to unknown transient and time-varying signals in the data. To address this, the study adopts the Kalman filter, a dynamic least-squares estimation method for stochastic processes, to estimate transient deformations associated with the 2016 Mw 6.4 Meinong earthquake in southwest Taiwan. Because crustal stress undergoes changes and requires time to recover and adjust following a large earthquake, it would be natural to include acceleration in the state vector of Kalman filter when fitting the GNSS time series. Through synthetic data tests, we demonstrated that a simple AR(1) (first-order autoregressive) process for acceleration can significantly improve the estimation of co-seismic and post-seismic displacements. Using this model, we analyzed time series from 76 continuous GNSS stations in the study area and found that large transient motions are confined to the region bounded by active faults. These ground movements lasted for about two years after the earthquake and are primarily associated with thrust fault-fold structures involving the Gutingkeng mudstone formation at shallow depths. Our findings suggest that inelastic deformation following large earthquakes can be spatially correlated with certain ductile structures in southwestern Taiwan, emphasizing the importance of considering such factors when assessing potential earthquake hazards. This working hypothesis, however, should be further tested with observations from future seismic events.

Graphical Abstract