The Role of Seismic and Slow Slip Events in Triggering the 2018 M7.1 Anchorage Earthquake in the Southcentral Alaska Subduction Zone

The M7.1 2018 Anchorage earthquake occurred in the bending part of the subducting North Pacific plate near the geometrical barrier formed by the underthrusting Yakutat terrane. We calculate the triggering potential related with stress redistribution from deformation sources including the M9.2 1964 earthquake coseismic slip, postseismic deformation, slip from regional M > 5 earthquakes, and the cumulative slip of previously detected slow slip events over the past 55 years. We investigate the deeper shallow depth (20–60 km) seismicity response to these perturbations using an epidemic type aftershock sequence model to describe earthquake‐to‐earthquake interactions. The statistical forecast captures the triggered seismicity during the 1983 M6+ aftershocks in Columbia Bay but performs poorly during the slow slip event period between 1992.0 and 2004.8 that presents a statistically significant rate change (β, Z > 2; M < 4.0). We find that stress effects from the 1964 postseismic source and the 12‐year‐long slow slip event (~M7.8) contribute to the 2018 Anchorage earthquake occurrence and that slow slip events modulate the deeper shallow depth seismicity patterns in the region.


Introduction
Large magnitude (M > 8.0) earthquakes in subduction zones are most commonly megathrust ruptures that are associated with multiple hazard phenomena such as severe ground shaking, tsunamis, and landslides. However, intraslab earthquakes, such as the 2018 M7.1 Anchorage (Ruppert & Witter, 2019), the 2017 M7.1 Puebla (Alberto et al., 2018), and the 1999 Oaxaca (Singh et al., 2000) in Mexico among others, are associated with extended structural damage and loss of life, highlighting their contribution to seismic hazard (Choy & Kirby, 2004). It is difficult to identify intraslab faults before they rupture, and their often undetermined recurrence intervals make them particularly difficult to account for in time-dependent seismic hazard assessments. Therefore, understanding mechanical interactions between intraslab and interslab earthquakes, and their triggering mechanisms is necessary to identify high-hazard locations and periods (e.g., Lin & Stein, 2004).
Earthquakes can be triggered by the passage of seismic waves (dynamic triggering) (e.g., Hill & Prejean, 2015, and references therein) and/or distortion of the crust caused by coseismic slip on faults (static triggering) (e.g., Harris, 1998, and references therein), with effects persisting from days to years following a large mainshock (e.g., Omori, 1894;Parsons, 2002). Until recently, only past earthquakes were considered as triggering sources, but the advance of geodetic networks reveals that aseismic deformation on the plate boundary fault can equal, if not exceed, the cumulative moment release of subduction earthquakes (Kano et al., 2019;Kobayashi & Tsuyuki, 2019;Voss et al., 2017). Dmowska and Lovison (1988) and (Lin and Stein (2004) demonstrated a general case for increased failure stress caused by interactions between interplate and normal intraslab earthquakes. Mikumo et al. (2002) analyzed the link between the M7.8 1978 to M7.5 1999 Oaxaca and the M8.1 1985 to M7.1 1997 Michoacan earthquake pairs and suggested that the intermediate-depth normal events occur as a mechanical consequence following "large thrust earthquakes with low-dip angles and shallow depths." Lay et al. (2017)  Samoa/Tonga, and 2011-2012 Japan trenches. Choy and Kirby (2004) suggested that the spatial and temporal connection between large magnitude interface-intraslab pairs in the 1997 Chile, 2001El Salvador, and 1999-2000 Kodiak sequences, supporting various coupling modes between deeper normal and shallow (h < 70 km; Choy & Kirby, 2004) thrust earthquakes. Wallace et al. (2014) suggested that the intraslab 2014 M6.4 Eketahuna earthquake in New Zealand was triggered by the 2013 Kapiti slow slip event, and Segou and Parsons (2018) supported that incorporation of aseismic phenomena, such as afterslip, was influential in triggering the deep, normal-fault M7.1 Puebla earthquake in Mexico.
Understanding the frequency and the triggering processes behind the occurrence of intraslab events is critical for seismic hazard since their often increased apparent stress and radiated energy can be the underlying cause of extended damage for populations located just above the bend of the subducting plate (Bilek & Lay, 2018;Choy & Kirby, 2004). Slow slip events are known to trigger small magnitude seismicity in swarms (Ozawa et al., 2003), low-frequency events that invade the locked megathrust zone through stress transfer (Kano et al., 2019), and small repeating events with slow slip involvement (Kato et al., 2012). In recent years, the precursory role of slow slip events in the nucleation processes of large shallow megathrust earthquakes has been revealed in locations with dense geodetic networks, such as in Costa Rica (Voss et al., 2018), Japan (Ito et al., 2013;Kato et al., 2012), Chile (Ruiz et al., 2014), and Mexico (Maury et al., 2018). In this study we focus on the contributions of past large earthquakes and long-duration slow slip events to investigate seismicity patterns and the triggering processes that culminated in the 2018 Anchorage intraslab earthquake.

Seismicity and Stress Changes
In Southcentral (Li et al., 2016) extending the previous work of these coauthors from Suito and Freymueller (2009) regarding deformation sources in the area.
We investigate the influence of the 1964 Great Alaska earthquake and the role of the subsequent slow slip events in triggering the 2018 M7.1 Anchorage event. We describe the physical interaction between earthquakes and slow slip events using the Coulomb failure stress estimate (CF) given by the equation, where τ f is the change in shear stress (fault parallel) on the receiver fault, μ ′ is the apparent coefficient of friction (0≤μ ′ ≤0.8), σ n is the change in normal stress, and B k is Skempton's coefficient, which accounts for pore fluid pressure (Rice, 1992). Stress values are estimated by slipping an elastic dislocation representation for each source model using Okada's (1992) equations.
reported a best fit solution supporting a west dipping plane with strike 186°and dip 65°with a deeper hypocenter at 55 km.
In Figure  We also investigate the role of slow slip events in triggering the deeper shallow (20-60 km) depth seismicity in general; the spatial extent of the data used is shown in Figure 1. Investigating the seismicity response to aseismic transients through statistical modeling based on epidemic type aftershock sequence (ETAS) model is a robust method that extends even into inversion techniques for determining the past occurrence of those transients in the absence of geodetic measurements (Llenos & McGuire, 2011). Our effort here is to model the earthquake-to-earthquake interactions through the ETAS model and to evaluate the stress effects from large regional earthquakes so that we can track the unique effect of the slow slip transients to the target seismicity. To isolate the effects of aftershocks from large regional earthquakes, we describe the expected seismicity resulting from earthquake-to-earthquake interaction using an ETAS model (Ogata, 1985) fit to the Omori decay using the function: where μ is the background rate, which is a time-independent, and spatially heterogeneous Poisson process. The summation term represents the triggering effects from all preceding earthquakes occurring at t i < t.
The parameter K regulates the short-term aftershock productivity by a parent event with magnitude M equal or above a minimum triggering magnitude; α establishes the efficiency of earthquakes in triggering aftershocks as a function of magnitude.
We used the NEIC earthquake catalog in the Cook Inlet region to assess whether any seismicity patterns occurred within the period of the slow slip events that could not be explained by earthquake-to-earthquake interaction mechanisms. We consider deeper shallow depth earthquakes (20-60 km) to avoid shallow seismicity on active faults in the area, provided as supplementary material. Our primary target in the ETAS model is the long-term aftershock decay following the M9.2 1964 Great Alaska earthquake. We use the pre-1990 seismicity (Ruppert & West, 2019) to model the long-term decay. Due to the sparse seismic networks in the 1960s to 1970s, we determine the p value = 1.27, and α = 3.0 by using higher magnitude (M ≥ 5.0) events from the first 6 months after the 1964 earthquake and extrapolate assuming a b value = 1. We then fix the p value, and we take c = 0.018, K = 0.21 from a past statistical model for Alaska (Wei et al., 2012).
The resulting ETAS model accurately captures the peak in triggered seismicity related with the July 1983 M6.4 and September 1983 M6.3 Columbia Bay earthquakes (Figure 3). The middle to late 1983 aftershocks are described adequately by the statistical models providing an excellent benchmark time period for validating the ETAS parameterization using frequently employed log-likelihood statistics. The modified number (N) test (Zechar et al., 2010) provides a robust evaluation for the effectiveness of the statistical earthquake-to-earthquake interaction model with the metrics within the July-September aftershock period equal to δ 1 = 0.001,0.0579,0.0997 and δ 2 = 0.9995,0.9769,0.9334 corresponding to nonrejection for August-September 1983 with an underestimation in July, which is often observed in early time periods for ETAS affected by early aftershock incompleteness (Segou et al., 2013). We therefore conclude that the ETAS model is also an accurate statistical descriptor of earthquake-to-earthquake interaction for the pre-slow slip event period (1990)(1991)(1992). In the pre-slow slip event period, the average expected monthly rate of~1-2 M ≥ 3.0 events is consistent with the post-1964 decay predicted by the ETAS model with a few spikes related to aftershocks from low magnitude events. Although the moment rate function of the Slow Slip Event 1 is not available (1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004); Figure 3 in magenta), we observe that the shape of the seismicity rate follows an acceleration (1992-1995) before a phase of deceleration (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004). Rolandone et al. (2018) discusses this "typical of slow slip events" slip pattern where "slip shows a progressive acceleration before a phase of deceleration" similar to other subduction zones while investigating slow slip occurrences in central Equador.
We performed stochastic declustering (Zhuang, 2006;Zhuang et al., 2002Zhuang et al., , 2004 on the NEIC catalog  in the Cook Inlet region to remove cascading aftershocks and verified that the resulting catalog is consistent with a Poisson process at 95% confidence level, provided in this paper as supporting information. We found a mean M ≥ 3.0 rate of 2.4/30 days before Slow Slip Event 1, which then increased to 5.3/30 days during Slow Slip Event 1, and then reduced slightly to 4.8/30 days during Slow Slip Event 2. These rate changes are significant according to the Z statistic of Habermann (1981) and the β statistic of Matthews and Reasenberg (1988) at 95% confidence level, provided as supporting information.
The observed seismicity rate change pattern and the poor fit during Slow Slip Event 1 of the ETAS model (rate increase without a mainshock) provides evidence for the role of slow slip events in triggering deeper shallow depth seismicity. Following the end of Slow Slip Event 1, we observe that the average seismicity as described by our smoothed within 6-month windows seismicity (Figure 3, in black dashed) does not return to the pre-slow slip event level, but remains higher. The relatively short duration of Slow Slip Event 2 makes it difficult to assess its influence on seismicity rates, though we observe an increase in the average monthly rate (dashed black) in middle to late 2010 that the ETAS model cannot fit.
We have also considered stress change effects on the 2018 Anchorage earthquake from 15 M ≥ 5 post-1990 events. These shocks occurred within~50 km of the Anchorage epicenter and are within, or very close to, the slow slip event locations. We have considered both nodal planes from focal mechanisms and an average friction coefficient of μ ′ = 0.4. The average maximum stress load from these events is miniscule, reaching 0.88 kPa, with maximum calculated stress change at 1.397 kPa from the 1991 M5.3, and 1.302 kPa from the 1997 M5.0 events. These stress change values fall well below the observed 0.01 MPa threshold for static triggering (Hardebeck et al., 1998;Harris & Simpson, 1992).

Slow Slip Event Stress Effects on the Megathrust
We investigate the influence of Coulomb failure stress change from past earthquake sources on the first slow slip event location. The recent mechanism proposed by Takagi et al. (2019) suggests that stress shadows of seismic asperities lead to an increased slip deficit that is more prominent updip, which causes steady creep or produces transient slow slip events but does not cause long-lasting slow slip events downdip of the seismic asperity. In the case of Alaska, we find that 0.80 and 0.92 of the total long-lasting Slow Slip Event 1 locations (alternatively stated 80%, 92%) are located within the 1964 Great Alaska earthquake stress shadow of the coseismic and postseismic slip distributions, respectively. The long-lasting Slow Slip Event 1 occurs on the subducting plane further downdip of the 1964 coseismic slip distributions, but within the upper limit of the 1964 afterslip at 30 km depth (Suito & Freymueller, 2009). However, as discussed in Rousset (2019), the variation of slip amplitudes in the Cook Inlet region remains quite poorly resolved, making conclusive results difficult.
We conclude that Slow Slip Event 1 promoted the occurrence of Slow Slip Event 2 in the period 2010-2012 because 62% of the Slow Slip Event 2 slip distribution was under positive stress change with a maximum of 0.23 MPa. We show calculated static stress changes from the cumulative slip distributions of Slow Slip Event 1 and Slow Slip Event 2 on the 3-D representation of the SLAB2.0 (Hayes, 2018) subduction interface ( Figure S3). Following Slow Slip Event 1, we calculate an extended stress increased corridor beneath Anchorage at depths ranging between 20 and 60 km. The effects of Slow Slip Event 2 are spatially limited due to its smaller moment release with stress increase primarily NE of the 2018 Anchorage earthquake epicenter ( Figure S3). We find that the positive (loading) effects of the slow slip events on the megathrust (Figure 4; in dashed orange, magenta lines) are more spatially extended when compared with the 2018 Anchorage stress changes. The slow slip events promoted slip on the megathrust at locations ranging from 20-60 km depth (Slow Slip Event 1; magenta line) and over 200 km along strike (Slow Slip Event 2; orange line) crossing between the locked and conditionally stable region, where both slow slip and seismic slip can occur. Considering the fact that the reoccurrence pattern of slow slip events is not well defined in Southcentral Alaska, it is difficult to anticipate whether the 2018 Anchorage event will shorten their recurrence interval. Dense monitoring from geodetic networks is required near Anchorage in order to determine with enhanced spatial resolution the onset of the next slow slip event.

Conclusions
We find that the 2018 M7.1 Anchorage normal intraslab earthquake is consistent with potentially being triggered by a combination of postseismic slip after the 1964 Great Alaska earthquake as well as by one to two slow slip episodes on the megathrust. The primary potential triggering sources were the slow-slip events that occurred between 1995 and 2004 that released a moment equivalent to a M7.8 earthquake. We demonstrate that the increase of low magnitude (M ≥ 3.0) deeper shallow depth seismicity (20-60 km) in the Cook Inlet region during the period between 1995 and 2004 is attributed to the slow slip event. This is evident because the rate increase within that period cannot be fit to standard ETAS models, the declustered catalog is consistent with a Poisson process, and the rate changes are statistically significant according to two methods.
Our analysis regarding the interplay between large earthquakes and slow slip events in Southcentral Alaska comes after the normal intraslab M7.1 2017 Puebla earthquake in Mexico that was likely triggered by extended delayed afterslip (M7.8) (Graham et al., 2014) along the subhorizontal subduction plane following the M7.5 2012 Oaxaca event (Segou & Parsons, 2018). A modulation of seismicity from slow slip events is identified in the Cascades region at the downdip edge of the megathrust (h = 40-60 km) (Vidale et al., 2011), but most commonly these responses are associated with shallow slow slip events for example in the Boso Pensinsula (Ozawa et al., 2003) and Hikurangi subduction zones (Delahaye et al., 2009). The above suggests that for short-term seismic hazard, incorporation of slow slip events or other aseismic phenomena is likely to play an important role in identifying high hazard locations for normal and thrust intraslab events that occur near the unstable, or conditionally stable, section of the subduction zone. Dense geodetic networks in the offshore and onshore region in Cook Inlet are required to improve the spatial and temporal resolution of the aseismic phenomena to allowing them to be incorporated into time-dependent seismic hazard assessment. In Southcentral Alaska the long-lasting slow slip events occur within the stress shadow of the 1964 earthquake and postseismic source. Although these slow slip events are found farther downdip of the 1964 coseismic slip distributions, they are within the upper limit of the 1964 afterslip slip region. The onset of slow slip events in the early 1990s may be related to 1964 postseismic effects, but assessing whether or not this is true would require geodetic data prior to the early 1990s, which does not exist.
Following the 2018 Anchorage earthquake, megathrust locations farther east and updip of the epicenter that are close to the shallow limit of the pre-earthquake slow slip event locations are positively loaded (0.01-0.4 MPa). At the same time the spatial distribution of past slow slip event related stress effects on the megathrust are far more extended. It remains to be seen the next few years whether the estimated stress effects of the 2018 Anchorage earthquake will modulate any future slow slip event locations farther east and updip of the hypocenter and/or will control the long-term deeper shallow depth seismicity rates in the region.