UC Santa Cruz

Pore fluid pressure lowers effective normal stress and weakens granular material (Terzaghi, 1936). Because of the “effective stress law”, fault failure can be induced through injection of fluids, as seen at the Rocky Mountain arsenal in the 1960s (Healy et al., 1968). A more definitive field test in Rangely, Colorado, demonstrated that when fluid pressures were raised and lowered in an oil reservoir over 4 years, seismicity rates were raised and lowered as well. Earthquakes occurred when fluid pressure in the reservoir was above ~25 MPa and the study postulated that one day, scientists might be able to control earthquakes along the San Andreas fault (Raleigh et al., 1976). During the 2010s, induced seismicity in Oklahoma rose precipitously due to massive wastewater injection activities. This wastewater is a byproduct of hydraulic fracturing of tight shale reservoirs, which is also known as super fracking (Turcotte et al., 2014). From 2014-2017, Oklahoma had more M3+ earthquakes than California, a state that is both tectonically active and is 2.3 times larger than Oklahoma. The largest mainshock events in Oklahoma history include the 2011 M5.7 Prague, the 2015 M5.1 Fairview, and the 2016 M5.8 Pawnee earthquakes. The Prague sequence also includes an M5 foreshock and M5 aftershock, which are large by induced seismicity standards.While super fracking poses a hazard to humans by polluting drinking water and creating seismicity, it has allowed earthquake scientists to ask and answer fundamental questions about earthquakes. Instead of having to constrain the causes of natural phenomena, the cause of these events is anthropogenic. The following are a few key questions in earthquake physics that can be addressed through induced seismicity studies: 1) Is there a critical pore fluid pressure for fault failure? 2) When a fault fails, what controls the mode of failure, or in other words, why does it fail in a slow (creep) or fast (earthquakes) event? 3) Are measurements of frictional strength in small-scale laboratory experiments relevant to the seismic scale, and if so, how can we use them to understand seismicity? The 2011 Prague, Oklahoma earthquakes are an extremely well-recorded sequence within well-known stratigraphy that allows us to address these questions. Earthquakes within this sequence occur in the Arbuckle group, a permeable carbonate unit where most wastewater is injected, and within the basement granitic rock. These geological units will be important in all three chapters. In addition, this earthquake sequence involved three faults, the Mw5 foreshock fault, the Mw5.7 mainshock fault, and the Mw5 aftershock fault that allows me to probe the role of fault geometry on seismicity. In Chapter 1, I use repeating earthquakes to ask what controls the mode of failure during an aftershock sequence? Repeating earthquakes, i.e. earthquakes that re-rupture the same area over and over again, are proxies for aseismic slip. A group of repeating earthquakes that ruptures the same fault is termed a family of repeating earthquakes. I find that families of repeating earthquakes occur in the Arbuckle Group and the basement granite, and form four main clusters. Three of these clusters occur at fault intersections. Here, I explore reasons why fault intersections might preferentially host aseismic slip. The traditional model of afterslip (slow slip that occurs along a fault after an earthquake) is that it occurs in velocity-strengthening areas that cannot fail seismically due to stress from an adjacent earthquake. Using this model to explain our findings would require extreme heterogeneity of the material properties of nearby slipping patches, and it does not allow for overlapping aseismic and seismic slip patches like we see here. Alternatively, extreme stress heterogeneity at these fault intersections may allow for afterslip in typically unstable areas, which is possible in a non-steady-state rate-state framework. This stress heterogeneity could also allow for earthquakes in typically stable areas when taking into account dynamic weakening. This chapter demonstrates that areas of stress heterogeneity can promote a variety of slip behaviors and that slow slip at fault intersections can promote failure on non-critically stressed faults, such as the fault that hosted the Mw5 aftershock. In Chapter 2, I measure the frictional properties of Oklahoma lithologies at various pore fluid pressures in order to determine lithological and pore pressure controls on Oklahoma induced seismicity. I perform slide-hold-slide tests to determine the frictional restrengthening behavior and velocity steps to determine the stability of the material at different pore fluid pressures. I find that the frictional healing rate of the Arbuckle dolomite is dependent on pore fluid pressure. At the highest pore fluid pressure measured, the Arbuckle dolomite weakens with hold time. I interpret such weakening to be the result of carbonate dissolution at grain contacts within the gouge layer. These experiments were run with pore fluid water undersaturated with respect to carbonate ions and therefore, during a hold, the asperities have enough time at high pressure to either dissolve and shrink a contact or, even if the contacts grow during the hold, the chemistry of the interstitial fluid could weaken the bonds at the contacts increasingly over time. Why dissolution is more effective at high pore fluid pressures than at lower pore fluid pressures is not resolvable from the suite of experiments run here, but I have a few hypotheses. First, more localized shear zones that occur in the high effective stress experiment have less pore space and therefore less available unsaturated water near contacts, and so dissolution is not as ubiquitous. Second, at increased contact pressure, the thickness of the interstitial layer is smaller. Finally, the strain rate at lower effective stresses is slower during the hold and therefore allows for more weakening below a critical rate for dissolution. Further work will be needed to determine which, if any, of these conditions might be responsible for frictional weakening. I find that the healing rate of the basement granitic rock is not dependent on pore fluid pressure, however I note that at the highest pore fluid pressure analyzed, stress relaxation uniquely does not follow the same stress-displacement relationship that all other experiments exhibit. In addition, I find that pore fluid pressure in both lithologies stabilizes the gouge for sliding velocities between 1-10 μm/s. Implications of this work are that pore fluid pressure promotes stable failure, and that carbonates can frictionally weaken when injected with pore fluid that is unsaturated in carbonate ions. This may drive slow slip in carbonate regions and stress faults in regions capable of nucleating earthquakes. This helps to illuminate why injection into carbonates in places such as Oklahoma, Canada, and Texas allows for widespread seismicity, while areas such as North Dakota, where high volume injection occurs into permeable sandstone, has far less seismicity. In Chapter 3, I use repeating earthquakes in order to probe whether laboratory-measured frictional behavior can be scaled to the seismic cycle. Laboratory studies predict three stages of fault healing: rapid post-seismic healing, a delay in healing due to afterslip, and a gradual increase in healing. In chapter 2, I measured stage 3 healing, i.e. the gradual increase in healing with log (time). Using the repeating earthquakes from chapter 1, and focusing only on families with greater than 3 events, I analyze their moment-recurrence time behavior. Moment is dependent on both the amount of slip and the spatial area of the earthquake. Ideally, I would first analyze their stress drop, however these are small earthquakes whose waveforms are likely attenuated. I find that there are three types of moment-recurrence time behavior in the Prague sequence that correspond to the three spatially distinct regions hosting repeaters that I saw in Chapter 1. The first group shows constant moment regardless of recurrence time and occurs at the mainshock-foreshock fault intersection in the Arbuckle unit. The second group is moment-predictable with recurrence time and occurs in the basement granite off of the mainshock-foreshock fault intersection. The third group has scattered moment-recurrence time behavior and occurs at the mainshock-foreshock fault intersection in the basement granitic unit. I interpret the constant moment group in the Arbuckle unit as evidence of a lack of healing that is caused by dissolution effects within the fault as seen in my experiments in Chapter 2. For repeating earthquakes to occur, rapid healing must have occurred following each repeating earthquake, prior to the onset of dissolution-induced weakening. I show that the moment-predictable group is consistent with laboratory rates if attenuation is assumed to be masking the event’s true source duration. Finally, I interpret the scattered moment group as indicative of either chaotic loading conditions by the surrounding aseismic slip area or chaotic pore fluid pressure conditions. Unlike previous studies of repeating earthquakes, I show that healing behavior may vary depending on lithology and pressure conditions, and that lithology-dependent laboratory healing behaviors are relevant even at seismic scales. In total, this dissertation has answered a number of questions in earthquake physics. 1) Is there a critical pore fluid pressure for fault failure? We find that fault failure thresholds can be weakened by time-dependent chemical reactions, which could allow for failure on faults at a much lower pore fluid pressure than otherwise. In other words, the effective stress law may not be the controlling factor for carbonate faults. 2) When a fault fails, what controls the mode of failure, or in other words, why does it fail in a slow (creep) or fast (earthquakes) event? We find that the dominant cause of aseismic slip during the Prague sequence is stress heterogeneity. Our laboratory analysis suggests that aseismic slip should occur at high pore fluid pressures in the Arbuckle and in the granite. 3) Are measurements of frictional strength in small-scale laboratory experiments relevant to the seismic scale, and if so, how can we use them to understand seismicity? We find that laboratory healing rates are consistent with moment-recurrence time behavior of repeating earthquakes in Chapter 3, therefore it might be possible to directly scale slide-hold-slide tests from the laboratory to the seismic scale. However, to do so, one must also consider rapid post-seismic healing and a delay in healing during afterslip. Furthermore, I show that more work is needed to include the role of fluid chemistry in our understanding of fault healing. These are some of the findings and implications for this work and more are outlined in the following chapters.