| dc.description.abstract | Plenty of GPS data were used in our seismic deformation studies. However GPS time series exhibit not only tectonic signals but also colored noises. In order to accurately estimate model parameters and their uncertainties, noises have to be evaluated using more realistic model. Some recent studies showed that continuous GPS data are best described as a combination of white noise and flicker noise. We use a maximum likelihood method to estimate amplitudes of these noises and remove them from GPS time series. After carefully corrections, secular velocities and their uncertainties are used to estimate crustal strain rates. We use a modified version of Ward (1998) for calculating strain rates in spherical geometries. Our particular method considers three different types of weights: observational errors, distance between the observation and the location of estimation, and an additional weight to account for variable station density. We also consider spatial coverage of stations and apply a smoothing operator to avoid unreasonable velocity variations without data constrained. Our approach provides more stable and interpretable strain estimates than previous studies, especially when stations are irregularly distributed or when we use larger spatial length scale of estimation.
We use a GPS-derive surface velocity field of Taiwan for the time period between 1993 and 1999 to infer interseismic slip rates on subsurface faults. We adopt a composite elastic half-space dislocation model constrained by GPS horizontal velocities projected into the direction of plate motion (306º). The model fault geometry includes a shallowly dipping décollement, in western Taiwan, and a two-segment fault representing the Longitudinal Valley Fault (LVF) in eastern Taiwan. The décollement is composed of two fault segments, one extending west under the Central Range (CR) and one extending east of the LVF, with estimated slip rates of about 35 mm/yr and 80 mm/yr, respectively. The optimal geometry of décollement is subhorizontal (2º~11º) at a depth of 8~9 km. The inferred surface location of the western end point of dislocation in the northern profile is located 15 km east of the Chelungpu fault, while in the southern section, it is located beneath the Chukou fault. Our model successfully match the horizontal velocity field and predict the location of possible future rupture, while cannot explain the vertical data and thus fail to predict the active mountain building processes in Taiwan. This failure indicates a more complex rheological model that incorporates inelastic behavior is needed.
GPS measurements of coseismic displacements from the 1999, Chi-Chi, Taiwan earthquake are modeled using both elastic half-space and layered models. The optimal slip distribution shows maximum slip is 11 m, concentrated at the northern bend of the fault and extended about 10 km in down-dip direction from the ground surface. Both models show only one or two meters of net slip at the hypocenter. We also find the large and spatially coherent residuals, which may be attributed to elastic lateral heterogeneity, topography, as well as inelastic deformation, all of which are ignored in this simplified model. Similar strategies are adopted to estimate postseismic slip distributions and fault geometries using 3-month and 15-month GPS data after mainshock. Preliminary analyses show afterslip is the main mechanism of postseismic deformation. Assuming the shallow fault dips 24~26° E, as determined by numerous studies of the mainshock, we invert for the deeper fault structure. Our results show that the fault dip shallows with depth below the hypocenter, merging into a nearly horizontal décollement at a depth of 10 km, consistent with interseismic and coseismic modeling studies. The afterslip distributions for two time periods show maximum slips of 25 cm and 46 cm in the hypocentral region at 7-12 km, respectively. Afterslip is notably absent in the region of maximum coseismic slip, consistent with the afterslip being driven by the mainshock stress change. The afterslip on the décollement in the first 3 months and over 15 months contribute 68% and 80% of the total modeled moment release, respectively. It is worth noting that the slip on the lower décollement becomes more prominent over the longer time period. The afterslip moment inferred from the 15-month GPS observation is 4.7 x 1019 N m, 44% of which occurred in the first 3 months. In contrast, the seismic moment released by the aftershocks in the same period is approximately 2.0 x 1019 N m and 85% of that moment was released before the end of 1999. This indicates that although part of the GPS-observed moment may be due to aftershocks, there is still a large amount of postseismic deformation which is aseismic. Then we adopt a more realistic layered model to estimate slip distribution. The optimal model show the depth of lower décollement is 5 km deeper than previous one, but the slip distribution and misfit are very similar.
We employed the Extended Network Inversion Filter（ENIF）to invert for the evolution of afterslip with time and space. The modeling algorithm is based on recursive liner Kalman filtering, which is used to estimate system processes at each epoch given all past and present data and forecast future observations. The prediction and update process are iterated through the entire data set to precisely describe the system. The postseismic GPS data over 15 months is used to infer the history of afterslip, it shows shallow slip decays more rapidly on the main fault than deeper décollement. The slip rate on the main fault began about 0.45 m/yr and downed to 0.1 m/yr in 200 days, while it on the décollement started form 0.55 m/yr and took about 450 days decreasing to 0.1 m/yr. The accumulated slip on lower décollement over 15 months is 1.5 times larger than it on the main fault. The coseismic slip on the décollement is not significant, while afterslip is prominent. This is consistent with the experiment result of a velocity-strengthening fault zone rheology.
To explain the possible deformation with viscous flow in the lower crust and upper mantle, we examine postseismic GPS data over 3 years. The moving directions of stations are congruent in different time periods and the magnitudes of displacements decrease with time, the rate decrease is large near the fault but small in eastern Taiwan. This might imply either a deepening of the slip or the effect of viscous flow. We adopted a viscoelastic model with 1-D layer structure, but the model predictions are discordant with GPS observations. Misfits are similar for combination of afterslip and viscoelastic models, and afterslip model alone. In either case, the discrepancy between predictions and observations still exits, especially for vertical displacements. We conclude a more realistic 3D model, included lateral heterogeneity of viscosity is really needed in the future. | en_US |