Delphine Fitzenz, Generalization of a 3-dimensional fault interaction model including tectonics, fluids, and stress transfer, 2002, PhD thesis, Swiss Federal Institute of Technology. Thesis advisor S. A. Miller.

Summary of the principal results


We developed and used a set of modular, flexible, physics-based forward fault models as a tool to investigate the dominant physical and chemical processes at work in the earthquake process. The results of this study form the foundation for developing time-forward mechanistic assessment of seismic hazard for large-scale potentially destructive fault systems.

From semi-forward stress transfer to a fluid-controlled forward model

We developed a stress transfer model using the analytical solutions giving the internal displacement field due to a finite dislocation in an elastic half-space in the framework of the stress transfer theory. Using the slip distribution inverted for the Izmit earthquake, we investigated how the main event changed the proximity to failure on the surrounding optimally oriented planes. We also compared maps of volumetric strain and Coulomb failure stress changes with the aftershock distribution and the location of the hypocenter of the Ducze earthquake which stroke four months later. The latter was in a zone of increased proximity to failure. However, no significant conclusion could be drawn about what drives aftershocks, namely because of the limited knowledge on the regional stress and on the pore pressure state, and due to the lack of reliably relocated aftershocks. The complexity of the stress transfer patterns near the fault, reflecting both the changes in fault strike and the heterogeneity of the slip on the main fault, defined regions of enhanced proximity to failure (or the opposite) on the scale of the error on the aftershock location.
The limits of this semi-forward study pointed to the need for a forward 3-dimensional fluid-controlled fault model, in which the local stress tensor is determined by both the tectonic loading and the seismicity. The pore pressure is monitored and fault interaction can be investigated. The construction and simulation results of such a model was presented in Chapter 2. Although the model leads to slip and stress distributions and seismicity statistics in general agreement with observation, the development of pore pressure compartments and the significant role of poroelastic effects in fault interaction pointed to the need for better constrained fault hydraulics.

Rheology and hydraulics of fault zones

A model fault having properties of a ductile fault core was considered and shear creep and ductile compaction were introduced in the 3-D forward model (Chapter 3). We showed the development of weak overpressured faults as a result of ductile compaction and small-scale coseismic pore pressure redistribution, whether the fluid compressibility is pressure dependent or kept constant. Because many faults exhibit creep slip, which are not purported to be weak or overpressured, we propose two alternative mechanisms to regulate the pore pressure cycle in fault zones. The first one is related to the role of the highly permeable damage zone (Tanaka et al, 2001) and the second relates to the in-plane pore pressure redistribution length-scale. Because such hypotheses have to be tested against data, we built a regional model.

Regional model

Chapter 4 shows first results from a regional model of the transpressional San Andreas Fault near the Big Bend (California). To avoid cell-size effects in the pore pressure distribution, we computed and introduced a 2-dimensional finite difference algorithm that calculates the diffusion of fluid pressure during interseismic periods. We present the choice of the boundary conditions (tectonics) and the evolved stress maps and tectonic regimes around the model fault system due to the pore pressure and stress state and to seismic and aseismic slip on the two specified faults. Small-scale seismicity-induced stress perturbations develop near the model faults. Maps of $R$ show complex distribution of the relative magnitude of the principal stresses. The hypothesis of homogeneous stress used in focal mechanism inversions is therefore to be handled with care in tectonically active regions exhibiting a bend in the plate boundary. Maps of Coulomb failure stress, maximum shear stress, and optimal orientation for failures calculated with the evolved 3-dimensional local stress tensor show the advantage of the forward approach compared to the stress triggering approach.

Methodological aspects

The earthquake process involves processes spanning over 21 orders of magnitude in time and 15 orders of magnitude in space. It is studied by laboratory experiments (e.g., short nucleation process experiments or long deformation experiments, from microseconds to days and from micrometers to meters), field measurements and geophysics (e.g., structural geology, fault mapping, plate boundary, mantle convection, earthquakes, from milliseconds to million years and from micrometers to thousand kilometers), and space-borne observations (e.g., coseismic surface deformation, subsidence, erosion, long-term surface strain rates, from seconds to years and from millimeters to thousands of kilometers). Forward modeling earthquake generation on interacting faults including large-scale tectonics and a detailed handling of fault zone hydraulics is therefore a difficult task.

The quasi-static approach

We considered the problem of earthquake mechanics from a tectonic scale while keeping small-scale processes intact. We made significant approximations to simplify the physics, while ensuring that these approximations had substantial supporting evidence. The advantage of the quasi-static approach shown here is utilizing analytic solutions for the internal displacement field due to slip on a model subfault. This guarantees accuracy of the solutions along with other significant advantages. There is no need for a 3-dimensional mesh covering the model space; the exact displacement (and subsequent stress) field on a subfault is calculated directly from the dislocation that causes the perturbation and does not require the convergence of a numerical method (e.g., finite elements) on a series of nodes linking the perturbation to the observation point. This saves considerable computation time and also gives more freedom in the choice of the model geometry and the range of spatial scales considered. Thus, the tectonic loading is applied on dislocation planes much larger than the model faults (e.g. 800 km long, to avoid lateral boundary effects) whereas the stresses are monitored on 2 km long subfaults and creep slip is calculated using the tens of centimeter wide fault zone width. Note that in principle, the subfault size can be reduced (or alternatively, the resolution can be increased), but this results in the usual compromise between resolution and computation time.

A modular, flexible tool

The computer code is made of a series of modules articulated within a main fault interaction loop and synchronized namely through the time step calculation. After the definition of the initial conditions, a module calculates the time step required to reach failure on exactly one subfault. Once slip on a subfault is initiated, the subsequent rupture is controlled by the physics of a propagating elastic dislocation. Other modules include in-plane fluid diffusion, slip calculation to get the assumed stress drop, pore pressure redistribution, strain redistribution and stress calculation. Other processes that have to be incorporated depend mainly on the drained vs undrained assumptions. For instance, pore pressure redistribution occurs only once the system is in stress equilibrium, and porosity is updated for frictional dilatancy before pore pressure redistribution. This occurs because although  
fluids do not fill the new porosity instantaneously, we consider that pore pressure aids the propagation of the cracks responsible for the increase in porosity. This structure is very convenient both to test different approaches (e.g., different pore pressure redistribution methods, different boundary conditions in terms of pore pressure, see differences between Chapter 2 and Chapter 3) and to easily add modules.

See example of flowchart

Perspectives

The results of this thesis form a basis for many future directions in modeling the earthquake process, with the long-term goal of building physics-based, real-time self-learning models that describe the physics of major active faults. Such models can be tailored to particular tectonic regimes and fault systems that pose a major seismic risk (e.g., the North Anatolian Fault near Istanbul, the San Andreas fault at Parkfield, the Nojima Fault near Kobe), and of intraplate deformation in particularly sensitive densely populated areas. A number of future developments and studies are necessary before achieving this goal, and some of these are discussed below.

See the thesis text for more!
If you have ideas or comments, please contact me at fitzenz@usgs.gov.