Stress-Driven Earthquake Cycle Model of the Active Taiwan Collision Zone

We have developed a 2-D earthquake cycle model for compressional deformation in Taiwan in which active faults slip in response to the stresses in the lithosphere. The earth is modeled as an elastic lithosphere overlying a Maxwell viscoelastic asthenosphere. Slip on major earthquake-producing faults is modeled with sudden periodic slip at a specified recurrence interval. Coseismic rupture is modeled as a full stress-drop. Creeping faults slide at zero resistive shear stress throughout the earthquake cycle. The purpose of our modeling is to infer the geometry of active faults and partitioning of slip-rates among the faults. We seek a model that is consistent with surface deformation measured by GPS, deformation patterns at depth inferred from earthquake focal mechanisms, and longer-term geologic data such as paleoseismology and subsurface structure revealed by reflection seismology. Our study focuses on central Taiwan where an extensive GPS network has recorded preseismic, coseismic, and postseismic deformation associated with the Mw = 7.6, 1999 Chi-Chi earthquake. Active deformation in central Taiwan is concentrated mainly in two regions: the Longitudinal Valley of eastern Taiwan, which is thought to be the suture zone between the Philippine Sea Plate and the Eurasian continental margin, and the western fold and thrust belt. Interseismic deformation recorded with GPS for seven years leading up to the Chi-Chi earthquake displays high shortening rates of about 35 mm/yr across the Longitudinal Valley and subsidence of up to 15 mm/yr. About 25 mm/yr of shortening is occurring across the western fold and thrust belt. We model this deformation with slip on an eastward dipping Longitudinal Valley Fault and slip on a ramp-decollement geometry under the western fold and thrust belt. We find that the data is best explained with a decollement that dips about 10° under the western foothills and then dives down at a steeper angle of approximately 40° under the Central Ranges. In our model the decollement creeps continuously throughout the earthquake cycle and the frontal ramp breaks periodically in large earthquakes and is locked interseismically. The Longitudinal Valley fault creeps between 0 and 11 km depth and between 20 and 30 km. The fault is locked interseismically between 11 and 20 km. The long-term slip rate on the Longitudinal Valley fault is nearly twice the long-term slip rate on the frontal ramp under the western foothills. This viscoelastic cycle model explains the horizontal and vertical interseismic GPS data and deformation patterns at depth inferred from focal mechanisms of earthquakes occurring just before and just after the Chi-Chi earthquake. The long-term uplift rates are also consistent with the general patterns inferred from dated uplifted terraces. We also compare the fault geometry obtained form the stress-driven cycle model with inversions of GPS data for fault geometry using standard kinematic dislocation models in which the slip is prescribed as a displacement discontinuity rather than solved for as in the stress-driven cycle model. A bootstrap analysis is performed using the kinematic model to identify a range of solutions that fit the data satisfactorily. We show that many of the fault geometry models, which fit the data using the kinematic model, do not fit the data with the stress-driven model. This indicates that we can better constrain the range of possible models by enforcing stress boundary conditions on the faults.