We propose a novel framework for assessing the risk associated with seismicity induced by hydraulic fracturing, which has been a notable source of recent public concern. The framework combines statistical forecast models for injection-induced seismicity, ground motion prediction equations, and exposure models for affected areas, to quantitatively link the volume of fluid injected during operations with the potential for nuisance felt ground motions. Such (relatively small) motions are expected to be more aligned with the public tolerance threshold for induced seismicity than larger ground shaking that could cause structural damage. This proactive type of framework, which facilitates control of the injection volume ahead of time for risk mitigation, has significant advantages over reactive-type magnitude and ground-motion-based systems typically used for induced seismicity management. The framework is applied to the region surrounding the Preston New Road shale gas site in North West England. A notable finding is that the calculations are particularly sensitive to assumptions of the seismicity forecast model used, i.e. whether it limits the cumulative seismic moment released for a given volume or assumes seismicity is consistent with the Gutenberg–Richter distribution for tectonic events. Finally, we discuss how the framework can be used to inform relevant policy.

Awareness and concern regarding the impacts of seismicity induced by hydraulic fracturing have grown significantly in recent years

Several hazard and risk assessment procedures have already been proposed in the literature for various types of induced seismicity. For example,

This paper proposes a novel risk assessment framework for hydraulic-fracture-induced seismicity that directly links the volume of fluid injected
during an operation to its potential for causing nuisance ground motions, i.e. shaking that is an inconvenience to society and may raise annoyance or
distress among the public

The framework is applied to the region surrounding the Preston New Road (PNR) shale gas site in Lancashire, North West England, where recent
(2018/2019) hydraulic fracture operations resulted in 29 seismic events with local magnitude (

The proposed framework is a modified version of conventional probabilistic seismic hazard analysis

The framework is based on the assumption of a one-to-one relationship between the exceedance of tolerable ground shaking thresholds and nuisance risk,
i.e.

We apply the proposed framework to the region surrounding the Preston New Road (PNR) shale gas site in Lancashire, North West England, where hydraulic
fracturing operations in late 2018 (at PNR-1z well) and mid-2019 (at PNR-2 well) resulted in 29

We use the

Seismicity parameters used in this study, from

Ground shaking (

The considered exposure database (

We also separately consider important buildings, or critical sites in which the occupants (or equipment) may be more sensitive to the effects of
vibrations from ground shaking than those of conventional residential or commercial buildings

All buildings and important (i.e. educational and medical) buildings considered within 5

Equation (

calculate the corresponding total seismic moment, using Eq. (

choose a single random event from the magnitude distribution

use the

calculate median ground motion predictions from the GMPE at each site for the given combination of

repeat steps 2–4 until the total seismic moment of the sampled events is equal to that calculated in step 1 to within a small tolerance;

repeat steps 2–5 1000 times to generate 1000 potential catalogues corresponding to the given injected volume.

The proposed risk modelling approach is validated using data observed during the 2018 hydraulic fracturing operations at the PNR-1z well. We complete
the Monte Carlo sampling procedure for the actual volumes of fluid injected during those operations, using the UK Oil and Gas Authority's database on
PNR operations
(

Figure

Validating the risk modelling approach of this study, using data observed during hydraulic fracturing of the PNR-1z well.

We first examine the risk associated with the occurrence of specific moment magnitudes (

Magnitude-specific risk calculations. Panels

Figure

It is also seen in Fig.

Figure

We use the procedure outlined in Sect.

Injection-volume-based risk calculations. Panels

Figure

Figure

Figure

We disaggregate the results shown in Fig.

Ground motion disaggregation for

We provide the results of the disaggregation in Fig.

The application of the proposed risk framework presented in Sects.

The analysis conducted in Sect.

We assume a uniform distribution of

Uncertainty in the seismicity parameters affects steps 1, 2, and 6 of the Monte Carlo sampling procedure (Sect.

We have thus far assumed that the location of seismicity is known and that all events are co-located in space. In realistic scenarios however, there
will be some uncertainty on event locations

We assume events are produced from point sources that are uniformly distributed within 1.4

Event depths are assumed to be uniformly distributed between 1.5 and 3

Accounting for event location uncertainty requires an additional task in the Sect.

The statistical earthquake forecast model used in our analysis assumed a deterministic limit on earthquake magnitudes, based on the volume of fluid
injected (see Eq.

We now test the implications on the risk calculations of instead using the

Validating the

The

This assumption affects all steps of the Monte Carlo procedure in Sect.

The modelling approach of Sect.

To understand the effect of accounting for ground motion spatial correlation in our risk calculations, we implement the model of

Validating the assumption of spatial correlation in the ground motions of the case study: comparing predicted ground shaking with observed ground motion amplitudes. Also shown in dark grey are the mean (solid line) and first to 99th percentile predictions (shaded area) for the base (uncorrelated) case.

The

Figure

The effect of the assumptions varies across different tolerable risk thresholds and volumes of injected fluid. Assuming the

Comparing the distribution of maximum magnitudes simulated using the base case approach to rupture behaviour (i.e. magnitudes are capped in proportion to volume injected) and that of ^{3} injected volume.

Quantifying the impact of alternative modelling assumptions on the probability of exceeding various PGV levels at least once across all considered buildings in the exposure model.

Accounting for spatial correlation in the ground motions leads to lower probabilities of exceeding most tolerable risk thresholds, across all
injection volumes examined. This finding is consistent with previous studies of the effect of spatial dependence on risk

Seismicity parameter uncertainty and location uncertainty lead to smaller probabilities of exceeding smaller PGV values and larger probabilities of exceeding higher PGV values. This implies that the probability distribution of potential ground shaking values widens for both assumptions, which is consistent with expectations for the introduction of uncertainty. The effect of seismicity parameter uncertainty is particularly notable for smaller PGV values. For example, it substantially underpredicts pile-driving exceedance probabilities relative to the base case where the parameters are known after the fact, which may pose a problem if this threshold is implemented in policies for managing the induced seismicity. Location uncertainty is the least impactful of all assumptions examined. It is most noticeable for larger PGV values and can lead to slight increases in the probabilities of exceeding the highest thresholds observed for a given injection volume.

The calculations of this section help to increase our understanding of the type of knowledge required to conduct an accurate risk assessment with the proposed framework. We have found that rupture behaviour and event locations are respectively the most and least important pieces of information to constrain for the calculations (at least in the case study of interest).

Based on the recommendations of a hydraulic fracturing review by the Royal Society

However, such magnitude-based systems have limited connection to the actual risks associated with the induced seismicity; it is instead the intensity
of the ground motions

Alternatively, the proposed framework in Eq. (

Quantifying the impact of increasing PGV predictions by 15 % on the probability of exceeding various PGV levels at least once across all considered buildings in the exposure model, for the following injected volumes: 500, 1000, 5000, 10 000, 15 000, 20 000, 30 000, 40 000, and 50 000

A significant advantage of this approach over conventional magnitude- or ground-motion-based TLSs is that it is proactive rather than reactive, since
the injection volume can be controlled ahead of time to avoid a red light ever occurring. The findings of Sect.

Values of PGV predicted by a given GMPE may not be strictly compatible with the suggested nuisance vibration thresholds of the proposed framework. This is because GMPEs generally predict either horizontal or vertical amplitude, whereas the thresholds refer to the maximum amplitude across all three orthogonal components.

The

Figure

This study has assumed that the rate of earthquakes during hydraulic fracturing is related to the volume of injected fluid, which has two main
limitations: (1) there is no explicit temporal description of seismicity; i.e. the time period of event occurrence is not considered, and (2) the
relationship only applies during the fluid injection phase, such that additional models are needed to describe post-injection seismicity, e.g. the
decay law proposed by

This study has presented a novel framework for assessing some of the consequences of hydraulic-fracture-induced seismicity. The framework explicitly links the volume of fluid injected during operations to the risk of nuisance ground shaking, by combining statistical forecast models for injection-related seismicity, ground motion prediction equations for hydraulic fracturing, exposure models for affected regions, and suggested nuisance risk thresholds adopted from previous studies on human discomfort to vibrations.

We have demonstrated and validated the proposed modelling approach, using the UK Preston New Road (PNR) shale gas site and its surrounding area as a
test bed. In particular, we showed how the framework can be used to determine event magnitudes and injection volumes for which prescribed nuisance
risk thresholds may be exceeded at buildings near the site. For the specific case study examined, in which seismic events were deterministically
located close to the surface projection of the PNR well stimulated in late 2018, we found that ground motions equivalent in amplitude to that at which
pile driving becomes perceptible may be exceeded in the location of at least one building for event magnitudes exceeding the current UK
induced seismicity traffic light system red light event (i.e.

We have also examined the sensitivity of the calculations to various modelling assumptions, to better understand the type of information required for conducting accurate risk assessments with the proposed framework. Implementing two different models for rupture behaviour (that both aligned reasonably well with observed source data) led to significantly varied risk estimates in particular. This work therefore highlights the importance of better understanding the physics to quantify likelihoods of different types of volume-related rupture, i.e. volume-capped moment release vs. seismicity that follows a tectonic Gutenberg–Richter distribution. Use of an appropriate GMPE is also essential for obtaining accurate risk estimates. On the other hand, we found that constraining event locations would not have a significant effect on the calculations (at least for the test bed considered in this study).

Finally, we discussed ways in which the proposed modelling approach could contribute to developing risk-informed policies for the management of induced seismicity related to UK shale gas development. For example, we suggested that the framework could be used to design an injection-volume-based traffic light system for induced seismicity based on real-time updating of the model parameters, which would enable injection volumes to be controlled ahead of time to mitigate the probabilities of causing ground motions with nuisance risk potential. This proactive system could replace the reactive magnitude-based traffic light system currently used in the UK, in which the thresholds do not explicitly account for the associated risks. We expect the findings of this study to be helpful as a decision support tool for stakeholders involved in the regulation of shale gas development in the UK.

No new data were created as part of this study. Fluid injection volumes at the PNR-1z well and some related seismicity data were respectively obtained from the “Pumping Data” and “Microseismic” sections of the UK Oil and Gas Authority's database on PNR operations (

Both authors conceived and designed the research. GC drafted the written content of the manuscript, which both authors reviewed. GC performed the calculations. Both authors developed the figures.

The authors declare that they have no conflict of interest.

The authors are grateful to the UK Oil and Gas Authority and the British Geological Survey for making necessary data available. The authors thank James Verdon and Tom Kettlety (University of Bristol) for helpful feedback on seismicity source modelling. They thank Brian Baptie (BGS) and Antony Butcher (University of Bristol) for fruitful discussions on the ground motion data. Finally, the authors acknowledge the constructive comments of the two anonymous reviewers that improved the quality of this manuscript.

This work has been funded by the Natural Environment Research Council (NERC; grant no. NE/R017956/1) “Evaluation, Quantification and Identification of Pathways and Targets for the assessment of Shale Gas RISK (EQUIPT4RISK)” and the Bristol University Microseismic Projects (BUMPS).

This paper was edited by Filippos Vallianatos and reviewed by two anonymous referees.