en/ ProQuest Central"/ Unprecedented quiescence in resource development area allows detection of long-lived latent seismicity

Recent seismicity in Alberta and north-east British Columbia has been attributed to ongoing oil and gas development in the area, due to its temporal and spatial correlation Prior to such development, the area was seismically quiescent Here, we show evidence that latent seismicity may occur in areas where previous operations have occurred, even during a shutdown in operations The global COVID-19 pandemic furnished the unique opportunity to study seismicity during a long period of anthropogenic quiescence Within the Kiskatinaw area of British Columbia, 389 events were detected from April to August 2020, which encompasses a period with very little hydraulic fracturing operations This reduction in operations was the result of a government-imposed lockdown severely restricting the movement of people as well as a downturn in the economic market causing industry stock prices to collapse Except for a reduction in the seismicity rate and a lack of temporal clustering that is often characteristic of hydraulic fracturing induced sequences, the general characteristics of the observed seismicity were similar to the preceding time period of active operations During the period of relative quiescence, event magnitudes were observed between ML -0 7 and ML 1 2, which is consistent with previous event magnitudes in the area Hypocentres occurred in a corridor orientated NW–SE, just as seismicity had done in previous years, and were located at depths associated with the target Montney formation or shallower (&lt;2 5 km) A maximum of 21 % of the detected events during lockdown may be attributable to natural seismicity, with a further 8 % potentially attributed to dynamic triggering of seismicity from teleseismic events and 6 % related to ongoing saltwater disposal and a single operational well pad However, this leaves ∼65 % of the seismicity detected during lockdown being unattributable to primary activation mechanisms This seismicity is unlikely to be the result of direct pore pressure increases (as very little direct injection of fluids was occurring at the time) and we see no patterns of temporal or spatial migration in the seismicity as would be expected from direct pore pressure increases Instead, we suggest that this latent seismicity may be generated by aseismic slip as fluids (resulting from previous hydraulic fracturing injection) become trapped within permeable formations at depth, keeping pore pressures in the area elevated and consequently allowing the generation of seismicity Alternatively, this seismicity may be the result of fault and fracture weakening in response to previous fluid injection This is the first time that this latent seismicity has been observed in this area of British Columbia and, as such, this may now represent the new normal background seismicity rate within the Kiskatinaw area

noise in many places, and correlated with a decrease in population mobility (e.g. Lecocq et al., 2020;Dias et al., 2020).
The noise level at a seismic station can be estimated using the probabilistic power spectral density (PPSD) of its records (McNamara and Buland, 2004). Following the methodology of Lecocq et al. (2020) we compute the PPSD from 30-minute windows with 50 percent overlap so that a single value is gained for each window, calculated using Welch's method (Welch, 1967) for the Z-component of different seismic stations. This method reduces numerical noise in the power spectra at the expense of reducing the frequency resolution because of frequency binning, but this effect is minimized with a robust smoothing 65 parametrization. The 30-minute time series are then converted to an average daily PSD, and the RMS of the time-domain displacement is extracted. Anthropogenic cultural noise typically concentrates at high frequencies (> 1-10 Hz, McNamara and Buland (2004)), but is strongly diurnal (e.g. stronger during the day than at night, and stronger during the weekdays compared to the weekends (Lecocq et al., 2020)). To avoid meteorological signals, and in particular oceanic microseisms (which typically manifest below 1 Hz), we use the frequency band of 4-14 Hz to investigate seismic noise during the pandemic. 70 Figure 1 shows the reduction of seismic noise in the frequency band 4-14 Hz in Gastown, Vancouver, BC during the global pandemic. A clear reduction in noise is observed following the closure of schools (black line) and businesses (red line). During Phase I of the pandemic (i.e. between the closure of businesses and the partial reopening of the city on 5 May 2020 (green line)), noise levels remain lower than previously recorded. Following the reopening of some businesses in May and June 2020, 75 an increase in the noise is seen, although it remains lower than pre-pandemic levels, interpreted as the increased movement of people. To verify that these variations do not occur on an annual basis, we undertook the same noise analysis for the year 2019, and found no such fluctuations during the corresponding months. In fact, ground displacement remained between 20 and 30 nm at station R25AC for the entirety of 2019.
3 Seismicity in the KSMMA 80 With increasing oil and gas operations within the KSMMA over the past decade, the number of public monitoring stations has also increased. Prior to 2020, 9 public sensors maintained by Natural Resources Canada and the Geological Survey of Canada existed within the KSMMA boundary, along with 6 co-located accelerometers poised to better capture higher levels of ground motion from larger seismic events. In early 2020, 13 additional broadband seismic stations (Trillium T120 seismometers with Taurus digitizers) and two Titan accelerometers were installed within the KSMMA (expanding the EON-ROSE (EO) network) 85 as part of a joint project between the University of Calgary, Nanometrics, Geoscience BC and a number of universities in South Korea to monitor ongoing seismicity associated with hydraulic fracturing operations (Salvage et al., 2021).
The catalogue of seismic events detected in the KSMMA is based on the newly installed array and available public stations in the area. Events were detected from the incoming continuous seismic data using an STA/LTA triggering algorithm, followed by 90 a separate template-matching algorithm utilising continuously re-trained modules that classify noise from events and remove We take the catalogue of event times and P and S phases, and determine hypocentre locations using NonLinLoc (Lomax et al., 2009(Lomax et al., , 2000, a probabilistic, global-search non-linear algorithm that generates the maximum likelihood hypocenter location based on the estimated posterior probability density function for each event. A 1D velocity model, specifically calibrated for the KSMMA from compressional and shear sonic logs, formation tops and ground truth locations of previous seismicity 100 (available directly from BCOGC). Events were then re-located using HypoDD, a double difference algorithm, whereby the residual between the observed and calculated travel-time difference (or double-difference) between two earthquakes observed on a single station are related to differences in their relative hypocenter locations and origin times (Waldhauser and Ellsworth, 2000). To calculate magnitudes we use a form of the Richter (1935) magnitude formula that has been modified to better reflect local attenuation characteristics within the KSMMA (Babaie-Mahani and Kao, 2020). In line with calculations conducted by 105 Natural Resources Canada (NRCan), we calculated M L using the maximum amplitude from the vertical component simulated on a Wood-Anderson (WA) seismometer, rather than the horizontal component, which has been used elsewhere.
Historically, seismicity within the KSMMA appears to occur within spatially distinct regions that fall within a corridor orientated NW-SE (Fig. 2). In both years, the largest magnitude event occurred in an area away from the densest occurrence 110 of seismicity. Since the largest event in 2020 did not occur in the same cluster as the largest event of 2018, it appears that the occurrence of M L 3-4+ events is not necessarily confined to a single region. Temporally, seismicity within the KSMMA occurs in distinct clusters, attributed to ongoing development activity in the area (Fig. 3). In 2018, heightened periods of seismicity were observed in April, May, July and August ( Fig. 3(a)). Similar periods of heightened seismicity were observed in 2020 in March, August and September ( Fig. 3(b)). The majority of seismicity detected within the KSMMA is M L ≤ 2, and 115 consequently goes unfelt.

Prior and Post Lockdown: 2020
In March 2020, the Province of British Columbia introduced measures aimed at slowing the spread of COVID-19, including the closure of schools and childcare facilities on 17 March, and the closure of many businesses (in particular those that included daily human interaction) on 21 March. Up until this point in 2020, similar patterns of seismicity to other years were observed 120 in the KSMMA (Fig. 3). A total of 4,268 events were detected from the onset of data collection (22 January 2020) from the updated EO array (yellow triangles, Fig. 2) to 1 April. Following the initial closure of businesses on 21 March, there is evidence of ongoing hydraulic fracture operations for ∼10 days, with associated heightened seismicity ( Fig. 3(b)). It is possible this reflects operators in the area undertaking additional hydraulic fracturing jobs during this time, as government restrictions became increasingly tight, and no "end-date" to the restrictions being suggested, or it may be that these 10 days of seismicity At the beginning of April, a period of relative seismic quiescence began in the KSMMA (Fig. 3(b)). Operations were once again restarted in British Columbia in the later summer months, after ∼4 months. Seismicity since the resumption of activities is once again temporally clustered, with a total of 2,617 events being recorded since 6 August to present. The largest magnitude event of 2020 at the time of writing occurred on 11 September at 22:37 UTC with an estimated M L of 3.1, after which proximal operations were shut down in line with the traffic light protocol introduced for the KSMMA (BC Oil and Gas Commission, 2018). A total of 73 precursory events occurred over approximately 4 hours, with events locating within a small spatial extent (∼300m x 150m), probably directly related to ongoing operations in the area due to the correlation in space and time of events 135 and injection. Events within this precursory sequence had magnitudes between M L 0.2 and M L 2.6, and were all located at depths of approximately 2.05 km. Moment tensor results for this event suggest a focal mechanism dominated by strike-slip (Salvage et al., 2021).

Evidence of reduction in seismic noise
A clear reduction in the number of seismic events was observed during the lockdown period from April to August 2020 in the 140 KSMMA ( Fig. 3(b)). Over the ∼ 4 months of relative quiescence only 389 events were detected using the EO network and available public stations in the area. For comparison, 344 events were detected on our network over a single week from 8 to 15 February when operations were fully underway. On average during this period, the magnitude of events were smaller than during time periods when activity was driven by ongoing operations.

145
A reduction in seismic noise and therefore ground motion is also evident in the KSMMA following the introduction of government restrictions in March 2020 (Fig. 4). Unfortunately, the most central seismic stations in the EO array were not installed until immediately before (March) or post lockdown (May) and therefore could not be used to analyze the long term changes in seismic noise. We chose station KSM08, located in the east of KSMMA due to the long, uninterrupted seismic data recorded at this station, as well as its proximity to recent dense clusters of ongoing seismicity (Fig. 2). Heightened seismic 150 ground motion is evident at KSM08 through January to March, as operations are ongoing (Fig. 4). A significant decrease in seismic ground motion is observed following the government restriction in late March 2020, with the average displacement sitting well below the weekday and weekend daytime mean calculated prior to lockdown. As restrictions ease, we see a large increase in ground motion following the reopening of businesses in May 2020, although this once again tails off through June and July. The re-introduction of operations in August is clear from an increase in ground displacement and seismic noise, 155 which has remained elevated (although not as high a pre-lockdown levels) since.

Latent Seismicity during relative quiescence: 2020
Seismicity occurring during the period of quiescence from April to August 2020 within the KSMMA exhibit a number of characteristics indicative that it is a (latent) consequence of previous operations in the area. Figure 5 shows the temporal and spatial evolution of seismicity during this period. Firstly, perhaps unsurprisingly, seismicity does not occur in a distinct 160 temporal pattern that exhibits clustering ( Fig. 5(a)). A small number of events (∼5) occur each day throughout the 4 months.
Event magnitudes also reveal no discerning patterns with time, with all events registering M L -0.66 to ∼M L 1.2. Furthermore, the frequency index (FI) suggests no temporal patterns during the period of relative quiescence. The FI is a proxy for the spectral content of each waveform based upon the ratio of energy in low and high frequency windows (Buurman and West, 2010), calculated at a single station. We use station KSM06 (Fig. 2) due to its proximity to the majority of the ongoing 165 seismicity during this period of relative quiescence. A negative FI means the waveform is dominated by low frequency energy (in this case 1 -40 Hz); a positive FI demonstrates a majority of energy in the high frequency band (40.1 -80 Hz). In many environments (e.g. volcanic) a lower frequency content of the waveform is proposed as evidence for the direct role of fluids in the generation of the seismicity (e.g. Lahr et al., 1994;Chouet, 1996). The seismicity detected during the period of relative quiescence within the KSMMA shows no discerning temporal characteristics.

170
Spatially, seismicity detected during the COVID lockdown period exhibits characteristics that are similar to the previously detected seismicity in the KSMMA ( Fig. 5(b)). Most events occur in a corridor orientated NW-SE, similar to the spatial distribution of seismicity prior to lockdown. Some spatial clustering is evident (e.g. in May in the south (yellow)), but given the limited number of events this is difficult to determine with certainty. Most events during the quiescence period occur at a 175 focal depths of ∼0-4 km, which is similar for events prior to lockdown within the KSMMA, if potentially slightly shallower.
Target formations for hydraulic fracturing within the KSMMA (Upper and Lower Montney) typically sit between 2000 m and 2500 m (total vertical depth), with salt water disposal (SWD) injecting at shallower depths (M. Gaucher, Pers. Comm, 2020).
This suggests that events detected during the quiescence were generated in formations similar to those that occur when active hydraulic fracturing and SWD is ongoing.

Characteristics of Observed Seismicity
Seismicity generated during this period of quiescence appears to share many characteristics with seismicity generated during hydraulic fracturing operations within the KSMMA. Although low in number, the event rate per day remains fairly constant 185 throughout the ∼4 month period of no hydraulic fracturing operations, with no apparent temporal decay ( Fig. 5(a)). This contrasts the "usual" pattern of seismicity during active hydraulic fracturing operations, which is highly temporally (and spatially) clustered around the wells operating ( Fig. 3) (e.g. Skoumal et al., 2015). Figure 3(b) also suggests no change in The magnitude of completeness (Mc) during the lockdown period is ∼0.4. The Mc for the entire catalogue to from 22 January to 1 October 2020 (n=7216) is estimated to be 0.074, suggesting that even though relatively few events were detected during this quiescence, the detection of small magnitude events is good. Given the reduction in noise during the period of quiescence (Fig.   4), this is perhaps no surprise. The estimated b-value (Gutenberg and Richter, 1944) (Igonin et al., 2018;Eaton et al., 2014). The fact that no large magnitude events were detected during the period of quiescence (no M L >1.5) is directly influencing the estimated b-value in this case. Interestingly, higher b-values have typically been attributed to seismicity generated in normal faulting regimes (Schorlemmer et al., 2005;Amini and Eberhardt, 2019). The KSMMA is strongly influenced by the Fort St. John Graben complex, an asymmetrical half graben that has also undergone significant strike-slip and rotational movement upon reactivation of the basement faults in the area (Barclay et al., 1990), which 200 may also be directly influencing the estimated b-value. Furthermore, in hydraulic fracturing environments, b-values of >2 have been associated with the stimulation of natural fractures at depth, with smaller b-values associating with large-scale tectonic faults (Wessels et al., 2011;Eaton and Maghsoudi, 2015). In our case, this would suggest that the seismicity being generated is directly related to the complex natural fracture system, rather than any large scale faults in the area.

205
Seismicity during the quiescence appears to be spatially concurrent with previous seismicity in the area (Figs. 5(b) and 2). However, there appears to be very little correlation between the spatial extent of seismicity and the most recent hydraulic fracturing activity in the area (active in March 2020 prior to lockdown). Seismicity appears in two planar elongated features, extending in a NW-SE direction, with lengths of up to 30 km (eastern segment), if assumed to be one feature. These features are not coincident with any known faults in the area (Furlong et al., 2020). Seismicity recorded during this period of quiescence 210 is generally located at a similar depth to the target formations of the Montney (∼2 km), as well as in the formations above.
This suggests hydraulically connected pathways above the injection zone, perhaps within mechanically stronger lithologies, as has been previously suggested by Eyre et al. (2019b) in the Fox Creek region of Alberta (another area undergoing intensive hydraulic fracturing operations).

215
The generation of induced seismicity has often been successfully correlated to a number of injection parameters, including the injected volume of fluid (e.g. Yu et al., 2019;Ellsworth, 2013) and/or the pumping rate (e.g. Goebel et al., 2017). Temporally data is too sparse to draw conclusions as to whether any of these parameters directly influence the generation of induced seismicity within the KSMMA, although given that hydraulic fracturing operations during our period of interest were ceased, we know that this seismicity cannot be a direct response of this type of fluid injection. However, there is evidence that a small 220 number of seismic events identified from April to August 2020 may be associated with salt-water disposal (SWD). Within the KSMMA, only 8 SWD wells were active in 2020, compared to hundreds of hydraulic fracturing wells. Of these, only one well was active during our period of investigation (Fig 5b). We believe the seismicity occurring on 13 April 2020 (Fig. 5a, upper  panel), where over 20 events were registered on the same day (significantly above the background rate of seismicity during this quiescence), may be due to SWD. In this case, ongoing sustained SWD occurred ∼2 km away from the events occurring 225 on this day. This offset is not unusual for SWD and associated seismicity; Schultz et al. (2014) found an offset of ∼3.5 km between SWD and associated seismicity in Alberta.

Estimation of Noise
PPSD is one of the most common methods used to characterize ambient seismic noise. However, the level of smoothing, the size of the data window used in analysis and the methodology itself may all influence the PPSD calculation and distort features 230 of interest (Anthony et al., 2020). Smoothing is primarily undertaken in order to reduce the uncertainty associated with the PPSD estimates, and means that short spikes in noise (e.g. due to wind gusts or seismic activity) do not dominate the spectrum.
In our case, the reduction in ground motion is much easier to determine from the average of the PPSD rather than individual estimates (Figs. 1 and 4, green vs. grey lines). Although we use a period smoothing of 0.025 octaves, this is likely to provide adequate spectral resolution of spectral peaks, as shown by Anthony et al. (2020) and therefore impacts our results minimally. 235 We also use a window of 30 minutes (overlapping by 50%) to try to reduce spectral leakage and variance when calculating the PPSD.
Earthquakes, and other transient signals, are likely to impact the estimation of ambient noise by generating large spikes in the data. However, the removal of seismicity from datasets is generally accepted as not necessary since they are low-probability 240 occurrences within generally high-probability ambient seismic noise (McNamara and Buland, 2004). Only teleseismic earthquakes appear to have any real affect upon PPSD calculations (Anthony et al., 2020). A number of teleseismic events have been detected in the KSMMA during the period of quiescence analysis (e.g. M w 7.8 event on 20 July 2020, 99 km off the coast of Alaska), that may influence our calculation of PPSD. However, since we see no peak in the average ground motion at these times (e.g. no substantial peak in July 2020, Fig. 4), we suggest that teleseismic events are not majorly influencing our results.

245
One signal that does clearly influence our PPSD results in Fig. 4 is wind. Poor weather reported in the KSMMA, with wind gusts exceeding 80 km/hour at times were observed at the beginning of May, during an otherwise quiet period (i.e. no hydraulic fracturing operations in KSMMA, limited movement of people due to lockdown measures). Since the noise generated from wind gusts penetrates a wide frequency band, we are unable to filter it out. Using a filter between 4 and 14 Hz tries to eliminate 250 some of these transient signals mostly associated with meteorological and oceanic conditions.

Generation of Latent Seismicity
The cessation of operations within the KSMMA in the summer of 2020 allows us a unique insight into seismicity that cannot be directly correlated with injection, which is the inferred triggering mechanism for most (if not all) of the seismicity within the KSMMA. The characteristics of the seismicity generated during this period suggest no fundamental differences in terms 255 of temporal or spatial patterns or magnitudes to previous seismicity within the KSMMA that can be correlated with injection. In fact, many of the characteristics appear to be equivalent to events detected prior to lockdown. Prior to the development of the Montney play, natural seismicity within the KSMMA was almost non-existent. The Canadian National Seismic Network (CNSN) recorded 20 earthquakes (M L 2.5 -M L 4.3) from 1984 to 2008, which are assumed to be mostly natural events (Halchuk, 2009). The closest event to have occurred with a significantly larger magnitude than this occured in March 1986 260 (M w 5.4) NE of Prince George, British Columbia (Lamontagne et al., 2008). In order to investigate the likelihood that our detected seismicity is natural seismicity, we calculate the expected recurrence rates of seismicity within the KSMMA greater than M L 2.5 from historical data, which is the magnitude of completeness used for the determination of seismic hazard maps in Canada due to detection thresholds from the Canadian public seismic network. The total number of earthquakes detected by the national network from 1984 to 2008 was 20 (Halchuk, 2009), suggesting a recurrence interval of 0.83 events per year.

265
It is therefore unsurprising that during the period of quiescence, no events greater than M L 2.5 were detected. Following the Gutenberg-Richter formula (Gutenberg and Richter, 1944), it stands that there should be a 100-fold increase in the event rate to estimate the number of events >M L 0.5, suggesting an event count of 83. Therefore, a maximum of 21% of events detected during relative quiescence can be attributed to natural seismicity. Therefore, over 70% of seismicity generated during this period of relative quiescence cannot be explained by this mechanism, and we suggest is likely produced as a remnant to 270 previous operations, and therefore directly related to previous states of stress. With events being generated over 4 months since the cessation of operations, the state of stress at depth must be near-critical for an extended period of time in order to generate this "latent" seismicity.
The generation of seismicity in response to hydraulic fracturing is typically attributed to either fluid migration models, 275 poroelastic phenomenon, or potentially aseismic slip (e.g. Bao and Eaton, 2016;Langenbruch and Zoback, 2016;Shapiro and Dinske, 2009;Segall and Lu, 2015;Eaton, 2018;Goebel and Brodsky, 2018;Eyre et al., 2019a). In the fluid migration model, pore fluid pressures are significantly increased upon fluid injection reducing the effective normal stress within a fault zone, which is sufficient to trigger seismicity (e.g. Peña Castro et al., 2020;Bao and Eaton, 2016). Given the temporal and spatial correlation between seismicity and hydraulic fracturing operations within the KSMMA, this appears to be a likely cause of 280 seismicity. Under this model, the seismicity rate is usually observed to be proportional to the pore pressure, and is assumed to track the injection rate (Langenbruch and Zoback, 2016). Consequently, a slow and steady decrease in the rate of seismicity over time would be expected to occur, as fluid pressure leaks into the surrounding formations (Eyre et al., 2020), before seismicity returns to the background (i.e. natural) rate. Since seismicity during the period of quiescence is long-lived, shows no decay and cannot be attributed to increased fluid injection, another process must be involved in its generation. Furthermore, if pore fluid 285 pressure and relaxation as a direct consequence of immediate injected fluid was the trigger of the seismicity during this period of quiescence, we would expect the seismicity to spatially migrate directly outwards from the most recently injected wells. We see no evidence of this ( Fig. 5(b)), suggesting direct pore fluid migration cannot be held responsible for the triggering of this sequence. Seismicity triggered by pore pressure diffusion can also be estimated by determining the propagating pore pressure fluid front (r t ) related to the hydraulic diffusivity in a homogeneous isotropic saturated poroelastic medium (Shapiro and Dinske, 2009;Parotidis et al., 2003) by:

295
where D is the hydraulic diffusivity and t is time since injection. If the triggering front (r t ) closely follows the maximum distance of seismicity through time, then pore pressure diffusion is thought to play a central role in the triggering of this seismicity (e.g. Shapiro and Dinske, 2009;Parotidis et al., 2003). Diffusivity (D) is generally assumed to range in the Earth's crust between 0.1 m 2 /s and 10 m 2 /s (Scholz, 2019), although in areas affected by hydraulic fracturing is thought to generally be in the range of 0.1 m 2 /s to 2 m 2 /s (e.g. Goebel et al., 2017;Shapiro and Dinske, 2009;Parotidis et al., 2003). Yu et al. (2019) 300 suggested similar diffusivity values determined from seismicity related to hydraulic fracturing in the Montney formation to the NW of KSMMA, although others have speculated that much smaller diffusion values would be expected in shale formations (Eyre et al., 2020;Guglielmi et al., 2015). Higher values of diffusivity in hydraulic fracturing scenarios are anticipated due to faults and fractures at depth acting as fluid corridors (Caine et al., 1996), compared to in-tact shales. However, the seismicity generated in the KSMMA during the period of quiescence shows no coherence with a triggering front from the most recently 305 active injection wells (Fig. 7), suggesting that pore pressure diffusion is not the dominant mechanism responsible for triggering these earthquakes.
Other models proposed for the generation of seismicity in response to hydraulic fracturing suggest that both pore pressure and poroelastic effects are feasible mechanisms (e.g. Segall and Lu, 2015;Goebel and Brodsky, 2018). In these instances, the 310 increased pore pressure due to injection is thought to load the surrounding rock matrix, altering the stress field, often at great distances from the original injection site, if the region is well hydraulically connected. Again, however, this model suggests that seismicity is generated as a response to injecting fluid into the Earth, which was not occurring at the time of our latent seismicity. Given that the stress field would likely diminish following the cessation of fluid injection, we would also expect a decay in seismicity with time. We do not observe this. Alternatively, the trapping of fluids within a fault zone with only minor 315 fluid migration along the fault, could result in slow changes to the effective stress due to changes in pore pressure (Sibson, 1992). In this method, seismicity should migrate spatially outwards from this fault zone as the effective stress migrates. We also see no evidence of this spatial migration (Fig. 5(b)).
Recently, Eyre et al. (2019a) have suggested that aseismic slip may play an important role in the generation of seismicity, 320 whereby distal unstable regions of a fault are loaded by aseismic slip that initiated due to an increase in pore pressure within a stable zone, leading to the generation of seismicity. Once slip is initiated, far-field intraplate stresses may repeatedly reload unstable regions of the fault, leading to relatively steady seismicity rates. They suggest the driving stresses of such behaviour 10 https://doi.org/10.5194/se-2020-203 Preprint. Discussion started: 8 December 2020 c Author(s) 2020. CC BY 4.0 License.
are most likely to be elevated pore pressures (as a result of ongoing hydraulic fracturing in the area) becoming trapped within fault zones due to low permeabilities within many formations. Given that in the absence of the cessation of operations the 325 detection of latent seismicity is extremely difficult, there are few examples of long-lived seismicity associated with hydraulic fracturing operations. One recent example comes from a long-lived seismic swarm in Alberta, where seismicity was observed over 10 months after injection ceased, and was interpreted as being driven primarily by aseismic slip (Eyre et al., 2020). We favour this interpretation of aseismic slip playing an important role in the initiation of seismicity since ongoing hydraulic fracturing operations are not required to generate ongoing seismicity; instead, the previous trapping of fluids within fault zones 330 may be enough to sustain the generation of seismicity.
It is widely reported that earthquakes can be generated by transient stress changes related to the passage of seismic waves (i.e. "dynamic triggering", (e.g. Wang et al., 2015;Van der Elst et al., 2013;Hill and Prejean, 2007)). In some cases, this dynamic triggering can also be delayed by days or weeks following a teleseism, potentially related to the re-distribution of 335 pore fluid from the passing seismic waves (Brodsky and Prejean, 2005) or through initial aseismic slip on faults triggering seismicity (Shelly et al., 2011). During the period of quiescence (28 March to 6 August 2020), 43 earthquakes of >M6 were reported by the United States Geological Survey (2020), that may have the potential to cause dynamic triggering. We follow the methodology set out by Wang et al. (2015), whereby we first select only the teleseismic events that generated an estimated peak ground velocity of greater than 0.2 cm/s at any station within the KSMMA, as defined by Lay and Wallace (1995), whereby: and: where A 20 is the peak waveform amplitude when filtered at 20s; M is the magnitude; δ is the epicenter-station distance (in degrees); and T is the surface wave period (assumed to be 20 s). This method identified 40 events from the original list of 345 teleseismic events. We then calculated the β statistic (Matthews and Reasenberg, 1988) by: which is a quantitative measure of the level of dynamic triggering, representing the standard deviation in the background seismicity rate after a remote event. N 1 and N 2 are the number of earthquakes detected before (t 1 ) and after (t 2 ) the remote event, respectively. Here, we take t 1 and t 2 to be 12 hours. E(N 2 ) = N 1 xt 2 /t 1 is the expected number of earthquakes after 350 the main shock based on the background seismicity rate. If no earthquakes occur in t 1 (i.e. before the main shock), N 1 is set to 0.25 based on the equivalent range of the probability density function (Matthews and Reasenberg, 1988; Hill and Prejean, We identify 7 remote earthquakes that generate a β value ≥2 (Fig. 8), including the largest magnitude event to have occurred to date in 2020 that occurred 99 km SSE of Perryville, Alaska on 22 July at 06:12 UTC, with M w 7.8 (United States Geological Survey, 2020), although the increase in event count in the KSMMA following this remote event is difficult to determine without statistical analysis. In some cases however, such as following the M w 6.1 event on 31 May 2020, 43 km W of Lampa, Peru, a significant increase in the number of events detected in KSMMA is clear. Our analysis therefore suggests that a maximum of 360 8% of the seismicity detected during this period of relative quiescence may be attributed to dynamic triggering, in particular the events on 31 May 2020, however <70% of the detected seismicity cannot be attributed to primary activation mechanisms such as this, and therefore in our opinion are the result of "latent" ongoing processes.

Conclusions
Seismicity generated in the KSMMA has always been attributed to oil and gas recovery in the area, primarily due to its hydraulic fracturing operations, events showed no temporal clustering, but instead were generated in a fairly constant manner over the ∼4 months of quiescence. No spatial correlation between the most recently active wells in the area and seismicity could be determined, however the fact that seismicity occurs at the depths of previous injection (i.e. the target formations) suggests that the area must be hydraulically linked.

375
Since there is no temporal or spatial evidence that these events are a direct consequence of the most recent hydraulic fracturing in the area (i.e. an aftershock sequence driven by pore pressure diffusion or poroelastic relaxation), and since the area is typically naturally quiet seismically (a maximum of 21% of the detected events), we conclude that most of these events are an indirect response of the increased pore pressures at depth which is causing aseismic slip on already pressurized fault zones, 380 as a result of previous fluid injection in the area. A number of events may be the result of dynamic triggering, from remote events with M w >6 (up to ∼8%), however this process cannot account for the majority of the seismicity observed (>70%).
We suggest that the prior fluid injection in the area has altered the state of stress, and caused fluids to become trapped in fault and fracture zones at depth. This allows seismicity to be primarily generated by aseismic slip loading unstable regions of these pressurized zones at depth. Once slip has initiated, far-field stresses may repeatedly reload these unstable zones, leading to the relatively stable seismicity rate that is observed.
Code and data availability. Continuous seismic data, station and associated metadata for the EO network is available through Incorporated Research Institutions for Seismology (IRIS) (http://ds.iris.edu/ds/nodes/dmc/) using Network Code EO, following a 91-day embargo period.
The velocity model used for location analysis is available directly from the British Columbia Oil and Gas Commission. Seismic noise analysis (e.g . Figs 1 and 4) were envisaged by Thomas Lecocq; the code can be found here: https://github.com/ThomasLecocq/SeismoRMS.
Research Council of Canada for providing further funding for this project. Further support was provided through the Microseismic Industry Consortium. Nanometrics is gratefully acknowledged for their contribution to this project, including the installation and maintenance of stations, and near real-time analysis of seismicity including the generation of the catalogue of seismic events. We would like to thank those at the Incorporated Research Institutions for Seismology (IRIS) for hosting the data and facilitating collaboration. We would especially like to thank Jay Hogan at Nanometrics who facilitated the successful upload of data to IRIS. We would like to thank Thomas H. A. Swinscoe