Insights into the rupture physics and geomechanics of microseismicity induced during hydraulic fracturing operations

E}xploiting the subsurface through high pressure injection of fluids is used for multiple geo-energy industrial activities, including enhanced geothermal systems, waste water disposal and hydraulic fracturing. However, earthquakes caused by industrial activity are of concern to the government, operators and the public. In the U.K. hydraulic fracturing activities were banned as a result of induced earthquakes in 2019. Injection activities continue in Cornwall as part of a geothermal project, which are causing felt earthquakes. Induced earthquakes offer a unique controlled environment to ask scientific questions about the rupture physics of earthquakes and to inform mitigation strategies to reduce the risks of injection induced earthquakes.

In this thesis I use induced seismicity to ask some fundamental questions about the rupture physics of tiny earthquakes (i.e., $M_w$ $\leq$ 0.6). I use a dataset of high quality microseismic events collected during hydraulic fracturing operations in the Horn River basin, Canada, and exploit the borehole-geophone setup, which is near the reservoir, to probe seismic events at high frequencies (i.e., $\textgreater$ 200 Hz). I focus on the largest seismic events which are linked to a re-activated structure that extends from the stimulated shale into the underlying crystalline basement. These events show the clearest phase arrivals and the best signal to noise ratio.

Using the data, I first analyse the nature of the geophone response to noise and signal. In chapter 2 I show that the resonances and high frequency compromising effects of geophones significantly hamper our ability to produce sub-catalogues of high frequency source parameters. Such features were not easily noticeable but are likely to be common in studies that use a borehole geophone setup when monitoring microsiesmicity. Here I document systematic resonance features and interpret them as near-receiver effects, although the exact provenance is still unclear. I also observe high frequency cut-offs, which can in turn generate spurious source parameter estimates, resulting in an apparent scaling of stress drop with $M_w$. Spectral ratios account for resonances better than using the raw geophone signals but do not eradicate resonances completely. My observations have been documented empirically and theoretically by others as an issue when probing high frequency microseismic events using borehole geophones and our results support these studies.


The results from Chapters 2 and 4 contribute to our understanding of how earthquake ruptures scale. Smaller earthquakes hosted within shallower crust are expected to have a lower stress drop budget than deeper tectonic earthquakes. I show that there is no evidence that challenges the independence of stress drop with magnitude (self-similarity). The absolute stress drops and rupture radii I calculate are consistent with those expected if tectonic earthquakes are scaled down to a microseismic size. However, the results also highlight the epistemic uncertainty in stress drop resulting from the chosen method, which in this case leads to an average stress drop that is twice as large when using spectral ratios compared to directly fitting source models, as reported from other datasets as well.

I investigate the spatio-temporal variation of stress drops within the studied dataset and test the
hypothesis that stress drop decreases from the point of injection, as observed in other datasets. In Chapters 2 and 4 I show that using two independent methods for estimating stress drop, there is no signal of an increasing stress drop with distance from the injection point (which is unexpected if differential stresses decrease near to the injection point). One plausible explanation for this empirical observation is that the injected fluids diffuse relatively quickly along the fault zone, thereby decreasing effective stresses over a larger spatial footprint in a shorter amount of time compared to other settings. My interpretation is consistent with a previous study of the same dataset which shows that additional pore pressure most likely drives the fault to failure.

A closer temporal analysis of the stress drop variations within clusters of co-located and highly cross-correlated events (Chapter 4) reveals that although the average stress drop is stable with respect to distance from injection there are large variations within these clusters within short time periods. A plausible explanation is that small scale pore pressure differences could cause significant differences in stress drop. However, many different theories used to explain empirical observations of stress drop differences in other datasets such as fault roughness, fault strength and small pore pressure differences could also explain these variations. Future research that provides a controlled lab study on how stress drop varies when fault properties are systematically changed would be a greatly beneficial reference for interpreting the signals of stress drop from datasets.

I delve further into the geomechanics of the fault structure in Chapter 3 with particular focus on the Fault Slip Potential (FSP) model, which has been used by others to identify which structures are critically primed for failure. My observations show very large uncertainties in the amount of additional pore pressure that an operator might use as a guiding upper limit when perturbing a reservoir. Here, I highlight the large uncertainties linked to the choice of the maximum principal stress direction one believes is affecting a reservoir, when deferring to data from the world stress map. Such uncertainties preclude robust calculations of fault stability estimates before any drilling has occurred. In-situ measurements of the maximum principal stress reduce the uncertainty in fault stability estimates and are preferable. However, the small scale variations of the maximum principal stress direction, even within the same reservoir, may still result in significant uncertainty of fault stability estimates for fault planes hosted in rocks which are below the reservoir, in the case of hydraulic fracturing of tight shales.

I show that stress drops can reveal interesting observations about the nature of induced seismicity during subsurface geo-energy exploitation. However, the paucity of high quality stress drop measurements, which is linked to the difficulties when accurately resolving high frequencies along a borehole geophone array, makes interpreting the empirical observations more difficult. I also question the confidence that we have when estimating how stable fault structures before, and during a subsurface geo-energy operation, which varies significantly depending on the tectonic length scales one believes is acting on the structure. Future studies will benefit from a better understanding of how sensitive stress drop signals are to the various attributes of a fault, wider azimuthal coverage during an operation and more in-situ measurements to characterise the stress state of a reservoir and the underlying basement.