The Ml 5.7 (Mw 5.3) Lake Muir earthquake occurred at 04:56:24 UTC on 16 September 2018 in a rural area of southwest Western Australia, ∼60 km east of the town of Manjimup and a similar distance north of the town of Walpole (Fig. 2). Approximately 20 km from the epicentre, relatively modest Modified Mercalli Intensity (MMI) values of VI were reported (Allen et al., 2019). With the exception of one unoccupied dwelling in the immediate epicentral area, which was extensively damaged, only minor damage and no injuries were reported. The event was widely felt throughout the Perth Basin, including the Perth metropolitan region, over 300 km away. Focal mechanisms suggest a reverse faulting mechanism, with a minor dextral transcurrent component and moderately east and northwest dipping nodal planes (Fig. 2, Table 2). Geoscience Australia recorded a magnitude Ml 3.4 foreshock 3 d prior to the mainshock. A protracted aftershock sequence, comprising hundreds of located events, was punctuated by a Ml 4.6 event on 12 October at 16:31:30 UTC (Fig. 3, Table 2). Almost 2 months after the 16 September Mw 5.3 event, a Ml 5.4 (Mw 5.2) event occurred on the 8 November at 21:07:01 UTC. Within the error estimates of the original hypocentral determinations, this event was collocated with the 16 September event (see weblinks in Table 2). The focal mechanisms calculated for the November event indicate predominantly dextral strike-slip faulting, with steeply northwest and southwest dipping nodal planes. The percentage double-couple from the U.S. Geological Survey (USGS) W-phase moment tensor solution is 31 % (Fig. 2, Table 2). This event was felt much more strongly and widely than the slightly larger first event, with MMI of VII to VIII being recorded close to the epicentre. There are several factors that might explain the relatively greater number (and density) of felt reports from the 8 November earthquake, including the following:(1) the time of day, which was early morning, when people are more likely to be stationary and thus more likely to perceive ground shaking; (2) differences in the ground motion radiation pattern and/or stress drop of the earthquake that may have yielded greater shaking in population centres at perceptible frequencies; and/or (3) greater community awareness of the earthquake sequence and where to find sources of further information. Surface rupture relating to the September event was initially identified with InSAR (Fig. 2) and has been subsequently verified by ground survey (this paper).

The Lake Muir earthquake sequence occurred within 1700–1600 Ma rocks of the Biranup Zone of the Proterozoic Albany–Fraser Orogen (Fitzsimons and Buchan, 2005; Spaggiari et al., 2009) (Fig. 2). During the Mesoproterozoic Albany–Fraser Orogeny these rocks were thrust to the north over the 3000–2600 Ma rocks of the Northern Foreland of the Albany–Fraser Orogen (the portion of the Archaean Yilgarn Craton that was reworked during the Albany–Fraser Orogeny) along moderately south-dipping faults. In the Lake Muir region, the dominant east- to east-southeast-striking structural grain is associated with the major structures bounding these thrust sheets. East-striking structural and lithological trends relating to the Albany–Fraser Orogen are transected by northwest-trending structures related to Proterozoic to earliest Phanerozoic movement on the Darling Fault Zone to the west (Janssen et al., 2003). The Boyagin dike swarm cuts across much of the southwest, including the Yilgarn Craton and Albany–Fraser Orogen (Prider, 1948; Myers, 1990; Harris and Li, 1995), and is subparallel to this structural trend. Minor north- and northeast-trending structures may relate to the Gondwana breakup or later movement focused on the Darling Fault, parallel structures to the west thereof (e.g. the Dunsborough and Busselton faults) and associated oblique structures (Harris, 1996; Olierook et al., 2015). Eocene alluvial sediments (Werillup Formation) and marine siltstone and sandstone (Pallinup Formation) locally infill topography in the crystalline basement (Wilde and Walker, 1984). Deposits have been correlated to map a Paleogene palaeo-channel system (Smith, 2010). These sediments and ferricrete developed within them are reported to be locally faulted by up to 5–7 m vertically based upon evidence from drill holes (Chakravartula and Street, 2000; Smith, 2010). The sparsity of boreholes precludes determination of fault orientation.

The Lake Muir region lies near to the southern boundary of a broad band of relatively high seismicity crossing the southwest corner of Western Australia known as the Southwest Seismic Zone (SWSZ; Doyle, 1971), which is one of the most seismically active regions in Australia (e.g. Leonard, 2008; Allen et al., 2018). Earthquake activity in the SWSZ appears to have increased significantly since the 1940s (Leonard, 2008), and it has generated five of the nine known Australian historic surface ruptures (all associated with reverse faulting, Table 1). In addition to large shallow events and scattered smaller events, the SWSZ has produced several dozen earthquake swarms in the last 40 years, including the Burakin swarm of 2000–2002 during where ∼18000 events of mixed focal mechanism (including six events in the magnitude range Mw 4.0–4.6) were recorded in a period of only a few months (Leonard, 2002; Allen et al., 2006; Dent, 2016). While most swarm centres occur within the SWSZ, they have a broader distribution across the southwest of Western Australia (Dent, 2016), with an essentially random pattern that is similar to fault scarps relating to prehistorical events (Clark, 2010; Clark et al., 2012). The relatively uniform spatial distribution of north-trending reverse fault scarps is consistent with scarp formation under conditions imposed by the contemporary east–west oriented compressive crustal stress regime (e.g. Rajabi et al., 2017b) and suggests that seismogenic strain release is uniformly distributed over the Yilgarn Craton over the timescale recorded in the land surface (ca. 100 kyr or more) (Leonard and Clark, 2011).

Examination of a 2012 lidar dataset (see the Supplement) revealed the presence of an alignment of valleys and ridges that mimics structural trends evident in the continental- and regional-scale magnetic data (cf. Figs. 2 and 3) (Chakravartula and Street, 2000; Milligan and Nakamura, 2015). The main ridge-line, valley and drainage trends in the study area are broadly to the east-southeast and north, with a subordinate northwest trend (black double-ended arrows in Fig. 3). Ridges are subdued, broad and undulating. Valley floors are flat-bottomed and are locally occupied by lakes and swamps, implying the presence of significant sedimentary infill (cf. Commander et al., 2001; Smith, 2010). Lunettes are developed on the east and southeast sides of most lakes, often with evidence for minor remobilization into parabolic dunes. Knickpoint retreat from a regional topographic low to the southwest occupied by Lake Muir is observed in several instances to be rejuvenating the drainage within flat-bottomed valleys, resulting in the removal of valley-fill sediments and the accentuation of structural trends evident in the alignment of adjacent linear ridges. There is no landscape evidence for sharp vertical dislocations of valley floor sediments that might relate to Late Pleistocene or Holocene surface-rupturing seismic events.

Since the 1992 Mw 7.3 Landers earthquake (Massonnet et al., 1993), remote sensing techniques have been shown to be valuable tools for observing surface deformation and damage patterns relating to earthquakes in a range of tectonic settings (e.g. Livio et al., 2017; Klinger et al., 2018; Wang et al., 2019; Gold et al., 2019; Scott et al., 2019). Interferometric Synthetic Aperture Radar (InSAR) in particular has greatly improved the detectability of surface faulting and surface deformation, especially relating to small- to moderate-sized earthquake sequences in remote or poorly instrumented areas, such as SCRs (Satyabala, 2006; Dawson et al., 2008; Amarjargal et al., 2013; Livio et al., 2017; Gardonio et al., 2018). Surface deformation relating to 11 SCR earthquakes has been imaged using InSAR (Table 1). Following the launch of Sentinel-1A and Sentinel-1B missions in 2014 and 2016, respectively, there is now a near-global coverage of SAR acquisitions every 12 d (https://sentinel.esa.int/web/sentinel/missions/sentinel-1/overview, last access: 23 June 2019). The availability of this rich dataset has enabled global systematic studies of earthquake detectability in InSAR data (e.g. Barnhart et al., 2019; Funning and Garcia, 2019) and, for the first time in an SCR setting, has allowed us to construct coseismic interferograms for two events within the same SCR seismic sequence.

For each of the two earthquake events we generated a coseismic interferogram from two Sentinel-1 interferometric wide-swath SAR images (earthquake 1: 14 and 26 September 2018; earthquake 2: 1 and 13 November 2018) using a standard interferometric SAR processing workflow implemented with the Gamma software (Wegmüller and Werner, 1997). The topographic component of the phase signal was removed using a 1 arcsec Shuttle Radar Tomography Mission (SRTM) digital elevation model (Geoscience Australia, 2011), and the orbital component was removed using precise orbital ephemerides (POE) products. Each interferogram was subsampled (multi-looked) eight times in range and two times in azimuth to reduce phase noise. An adaptive spatial filter (Goldstein and Werner, 1998) with exponent 0.5 was applied to each interferogram to further reduce phase noise prior to automatic unwrapping using a minimum-cost flow algorithm (Costantini, 1998). The automatically unwrapped interferogram for the September earthquake had significant errors due to poorer phase coherence around the surface rupture zone. A manual approach to unwrapping this interferogram was therefore applied using the branch-cut method (Goldstein et al., 1988).

The epicentral regions of the earthquakes were visited over 3 d (12–14 November 2018), less than a week after the November event. The scarp produced by the September event was mapped, and evidence for rejuvenation of the scarp during the November event was assessed. A hand trench was excavated across the central section of the September scarp to assess fault dip and displacement.

Because InSAR captures a single displacement vector in the radar line of sight (LOS) and commonly decorrelates in regions close to earthquake surface ruptures where displacement gradients exceed the Nyquist sampling rate or cause pixels to no longer overlap (e.g. Hamling et al., 2017; Metzger et al., 2017; Gold et al., 2019), it is desirable to obtain independent validation data. In order to recover part of the surface vertical deformation envelope associated with the events, aerial photographs were acquired with a DJI Phantom 4 unmanned aerial vehicle (UAV) in an approximately 500 m wide swath along a 2 km length of the September event scarp. An approximately 2 km long cross line was also flown, extending eastward from the scarp across the region of most significant surface deformation indicated in the InSAR imagery (Fig. 3). Ground control was provided by an RTK GPS, with mean horizontal and vertical measurement uncertainties of 0.026±0.025 and 0.056±0.055 m, respectively. A third mission covered the southern extent of surface deformation indicated in the InSAR imagery (Fig. 3). The results of this mission are presented in the Supplement (Fig. S1).

The image dataset was processed using a structure-from-motion (SfM) and multi-view stereo approach, implemented in the software Agisoft Photoscan Pro 1.4.3 (Agisoft LCC, 2018). The resulting dense point cloud achieved a standard deviation of the location differences between it and the control points of 0.09 m in the horizontal and 0.01 m in the vertical, which is comparable to other studies using similar ground control point (GCP) densities (e.g. Gindraux et al., 2017). Several studies of factors impacting local photogrammetry-derived digital terrain model (DTM) accuracy (e.g. Tonkin and Midgley, 2016; Gindraux et al., 2017) report a vertical accuracy decrease of ∼0.1 m for every 100 m increase in the distance to the closest GCP. In our study the maximum distance from a control point is on the order of 200–300 m. A 6 cm resolution DTM was produced from the dense point cloud.

The elevation values from a lidar dataset collected by the Western Australian Government Department of Biodiversity, Conservation and Attractions in 2012 were subtracted from the UAV DTM produced in this study, to produce a DTM of difference (DoD; Williams, 2012). The average magnitude of the uncertainties associated with the elevation values for the lidar dataset is reported as 0.063±0.074 m. The combined location uncertainty in the DoD might therefore be expected to be on the order of 0.1–0.2 m. Details of the processing steps are presented in the Supplement.

The focal mechanism computed for the September earthquake (Fig. 2) indicates almost pure reverse motion. In this instance, the majority of landscape change would be expected to be vertical and so detectable in the DoD. In contrast, the November earthquake was dominantly strike-slip (Fig. 2). The deformation envelope detectible with the deployed technologies should therefore relate almost exclusively to the September reverse faulting event, with further deformation from the November event remaining largely undetected. To quantify vertical surface displacement associated with the September earthquake, we extracted swath profiles through the DoD (and the unwrapped InSAR images) parallel and perpendicular to the rupture using the swath profiler tool for Esri ArcGIS (Pérez-Peña et al., 2017). The swath profiles sample the topography perpendicular to the profile to a set on distance either side of the profile line. A range of statistics (percentiles, quartiles and max/min/mean) might then be assessed in order to account for the noisy character of the UAV-derived DTM.

Within 5 d of the 16 September event five seismic rapid deployment kits (RDKs) and three GPS kits were deployed in the epicentral region (Fig. 2). The seismic sensors included four short-period Lennartz LE-3Dlite seismometers and one Trillium Compact seismometer. The short-period instruments were paired with Nanometrics Titan accelerometers recording at 200 Hz to capture any strong ground motions from the aftershock sequence. The RDK locations (prefixed LM on Fig. 2) were selected to maximize azimuthal coverage of the study area, taking into account the location of one permanent Australian National Seismic Network (ANSN) station near Rocky Gully (RKGY), fortuitously located 24 km from the epicentre of the 16 September earthquake. The network geometry also considered the capture of ground-motion data at a range of source–receiver distances for subsequent attenuation studies. One RDK (LM01) was deployed at the initial epicentral location of the 16 September event. Waveform data were telemetered in real-time to the National Earthquake Alerts Centre at Geoscience Australia to support real-time aftershock detection and analysis. The RDK locations therefore required adequate connection to local 3G and 4G telecommunication networks. Additional factors considered in the deployment configuration were the local site geology, availability of sunlight to power the sensors and land access.

The three GPS kits were deployed on existing survey marks within a broader network covering the SWSZ, one of which was last occupied in 2012. The survey marks used (SWSZ45, SWSZ46 and SWSZ48) were approximately 36, 65 and 25 km away from the epicentral region, respectively (Fig. 2). Processing of the acquired GPS data from the SWSZ46 site produced an anomalous vertical displacement signal indicating 12 mm of subsidence. This signal could not be reconciled with the surface deformation related to the earthquakes, which was constrained to a near-field region smaller than 10 km from the surface rupture (Fig. 2), and so it is not discussed any further here.

Initial reports from local residents following the September earthquake indicated the presence of west-facing fault scarps intersecting several farm tracks (e.g. Fig. 4a), loss of tension in an east–west running fence line (GDA94/MGA50; 479590 m E, 6192140 m N) and cracking in farm dam walls related to lateral spreading (Fig. 4b). Field investigation demonstrated that the track intersections could be linked to form an approximately 3 km long, concave-to-the-east, west-facing crescentic scarp (Fig. 3). In detail, the scarp comprises a series of left- and right-stepping en échelon segments 100–200 m long. In the central 2 km of the scarp, each segment is associated with up to 20–40 cm of vertical displacement (Fig. 4c, d). The morphology of individual scarp segments show little evidence for a strike-slip component to motion, varying between discrete thrust surface rupture with local fault-parallel folding and cracking and discrete thrust surface rupture with mole tracks (cf. Lin et al., 2004) and extensional fissuring associated with topographic bulges. Where the scarp crossed drainage lines, presumably associated with metres of sandy alluvial sediments, it becomes a broad warp, often associated with extensional fissuring parallel to the warp-crest and warp-perpendicular cracking at step overs.

A hand trench 2.3 m long and 1.2 m deep was excavated perpendicular to the scarp trace at a location where the vertical displacement was on the order of 0.3–0.4 m (Figs. 4c and 5). Here, the scarp has an average trend of 25∘, and scarp segments are right stepping (GDA94/MGA50; 479290 m E, 6191500 m N). The exposed stratigraphy consists of an approximately 0.10–0.12 m thick organic-rich grey brown silty sand overlying ∼0.4 m of pale grey medium sand that becomes pisolitic with depth (Fig. 5b, c). This layer in turn overlies an orange/red mottled grey clayey sand to the bottom of the trench. We interpret the profile to reflect an in situ weathering profile developed in Albany–Fraser Orogen bedrock. Approximately 0.18–0.20 m of the pre-event organic-rich sandy soil had been over thrust along a 20∘ east-dipping reverse fault that steepened to 30∘ towards the base of the trench (Fig. 5c). A vertical scarp-parallel tension fissure ∼0.7 m east of the trench suggests that the fault dip steepens again at shallow depth (Fig. 4c). Drag of layering in the hanging wall along the fault resulted in the formation of a prominent hanging-wall fold. The total vertical deformation at this site is shared approximately equally between discrete slip across the fault plane and folding.

South of 479130 m E, 6191120 m N (GDA94/MGA50) the scarp steps 50 m to the west, the general scarp strike is 355∘, and scarp segments become left stepping. The scarp is associated with a 0.2–0.3 m vertical landscape step which extends some 400 m south from this location, before entering dense pine plantation and becoming difficult to follow due to extensive landscape disturbance (e.g. furrowing) and a thick layer of organic matter cloaking the ground surface (Fig. 4d). Within the pine forest, the scarp maintains a vertical displacement of 0.2–0.4 m, before stepping again 50 m to the west at 479190 m E, 6190050 m N (GDA94/MGA50).

The segments south of the step-over strike ∼340–350∘ and decrease in vertical displacement from ∼0.3 to ∼ 0.1 m at the most southerly intersection of the scarp with a farm track (GDA94/MGA50; 479320 m E, 6189440 m N). At this track intersection, observed after the November event, the scarp comprises dominant southeast-trending elements which are locally offset (left stepping) across east-trending lineaments. These become tensional fissures on the eastern up-thrown side of the scarp. The vertical displacement across the scarp at this location is approximately 0.1 m. Landowners report that there was a “freshening” (rejuvenation) of the scarp at this location following the November event. The presence of fine cracking details in November, given there were significant rainfall events in October, support these anecdotes. The observations are consistent with a reverse oblique (sinistral) reactivation of this segment of the scarp. South of the track intersection, the scarp is lost in bushland. However, heavy tree limb fall and the occasional toppled tree was observed along strike for a further ∼400 m to the southeast, suggesting strong localized ground shaking during the September event (Fig. 3).

The wrapped interferogram for the September event shows an extent of surface deformation ∼12 km in an east–west direction and ∼8 km in a north–south direction (Figs. 2, 6a). A linear surface deformation front relating to the rupture can be traced for approximately 5 km. The central ∼3 km corresponds to the fault scarp mapped on the ground. The unwrapped interferogram (Fig. 6b) shows a broad shallow lobe of negative line of sight (LOS) change that extends from the west to the surface scarp (equivalent to ∼2.5 fringes in the wrapped image). From the east, a broad shallow lobe of positive LOS change (∼2 fringes in the wrapped image) transitions to a tight (∼1.5 km wide) band of negative LOS change characterized by at least 10 fringes. The images lose coherence in the 200–300 m east of the scarp and in proximity to Lake Noobijup (cf. Fig. 3). Coherence is also partly lost beneath an approximately 2 km wide (N–S) easterly trending band of pine forest (see Figs. 3 and 6a for location).

The unwrapped interferogram for the September event (Fig. 6b) shows a maximum LOS displacement towards the satellite of approximately 0.20–0.24 m along the eastern side of the central ∼1.5–2.0 km of the scarp. These values are approximately half of the maximum scarp height recorded by UAV field measurement and approximately one-third the maximum magnitude of slip along the fault plane that might be calculated from the trench exposure (see also next section). The southern 1.3 km of scarp mapped on the ground occurs mainly beneath pine forest, and the unwrapping algorithm failed to reproduce measured scarp heights of 20–40 cm in this area. Similarly, the unwrapping algorithm performed poorly in the swampy vegetated land proximal to Lake Noobijup.

The hanging wall uplift diminishes to the east of the scarp and is fringed by a broad region of positive LOS change, corresponding to depression of the land surface. The deepest region of depression of the ground surface is centred on Lake Noobijup (∼13 cm LOS increase). This area also corresponds to the highest density of aftershocks following the September mainshock (Figs. 3 and 6b). The depressed region has well-defined northern and southern extents, terminating at east- and southeast-trending structures, respectively. The areal extent of the depressed region is surprisingly large for a reverse fault rupture (cf. King et al., 1988; Ellis and Densmore, 2006).

The InSAR images for the November event (Fig. 6c, d) exhibit the classic quadrupole pattern of an almost pure strike-slip rupture and are consistent with a maximum of ∼5–8 cm of right lateral slip having occurred at the surface relating to rupture of a northeast-trending, steeply northwest dipping fault. This contrasts with the focal mechanism for the event, which suggests an oblique compressive rake (Fig. 2). Although the deformation pattern shows a sharp linear discontinuity for ∼1 km to either side of the intersection of the November failure plane with the September rupture plane, simple forward modelling using a finite rectangular elastic dislocation model (Okada, 1985) suggests that a discrete surface rupture may not have been produced (i.e. the rupture tip remained buried) (Fig. S4). Indeed, structures unambiguously relating to the main rupture were not observed in the field. Potential evidence for secondary surface deformation relating to the main failure plane of this event was seen as a series of fresh-looking east-trending left-stepping dilatational cracks expressed in a boundary track at 478970 m E, 6190390 m N (GDA94/MGA50). No vertical displacement was observed to be associated with these features.

The deformation pattern relating to the November event is also seen to be influenced by the northwest-trending structure that forms the southern termination of the September event. This lineament is discernible in the November event unwrapped InSAR from the intersection of the surface projection of the rupture planes to approximately 1.8 km southeast (Fig. 6d). Minor reverse-oblique movement on this structure during the November event may have been responsible for the reports of a local freshening of the September scarp.

Part of the vertical displacement envelope resulting from the September earthquake was recovered by differencing the DTM acquired by UAV and the 2012 lidar dataset (Fig. 7; see Sect. 2.2 and the Supplement for methods). Scarp-parallel and scarp-perpendicular profiles through the resulting DTM of difference (DoD) ubiquitously show a larger displacement magnitude than the InSAR-derived LOS displacement estimates (Fig. 7b, c). However, the general form of the two deformation envelope estimates is consistent. A broad uplifted region to the east of the scarp trace in the DoD corresponds to the area of greatest uplift evident in the InSAR data (Fig. 7a). Further, the InSAR LOS phase image indicates that the eastern end of the DoD profile is located in a broad region of topographic depression relative to the far field (∼2 fringes), while the western end is raised relative to the far field (∼2.5 fringes) (cf. Fig. 7c blue line and Figs. 6a and b). This is reflected in a ∼0.3 m vertical difference between eastern and western ends of the UAV DoD profile.

Importantly, the UAV DoD captures deformation close to the scarp trace where the InSAR images lose coherence. For example, the InSAR data did not image a ∼0.1 m deep foot-wall depression that occurs within ∼100 m to the west of the scarp trace. Along the ∼2 km length of scarp that was covered by the UAV DoD, the average height of the scarp, as measured along a 100 m wide swath profile (Fig. 7b), is 0.46±0.11 m. The landscape dislocation is shown to be sharpest in the ∼10–30 m east of the scarp trace and consistent in magnitude with that measured by RTK GPS at the location of the trench (cf. Figs. 5 and 7c). The topography then continues rising at a lower rate to a maximum vertical displacement within ∼200 m east of the scarp trace, before falling relatively smoothly over a kilometre (Fig. 7c). The fact that uplift associated with the fault tip is distributed over the ∼200 m east of the scarp trace suggests that the scarp-parallel DoD swath profile, spot measurements of scarp height (green dots in Fig. 7b) and the two-dimensional UAV profiles (red dots in Fig. 7b) all underestimate total vertical deformation to some degree. A similar result was derived for the 2018 Petermann Ranges earthquake surface rupture (Gold et al., 2019).

No surface expression or ground cracking was observed on the ground along the southeast-trending feature imaged in the InSAR-derived LOS displacement data at the southern end of the September rupture (cf. Fig. 6b). However, a DoD constructed over the area (see Fig. 7a for mission outline and Fig. S1 for results) revealed the presence of a very subtle (≤5 cm high) linear topographic feature, extending over ∼ 560 m, which was coincident with the surface displacement implied by the InSAR data (Fig. 7b).

In the period from 16 September to 23 November, 884 earthquakes were recorded by stations from the permanent ANSN network and the five rapid-deployment aftershock kits temporarily installed in the epicentral region. These events include the largest aftershock (12 October, Ml 4.6) and the November Mw 5.2 event. Initial locations for these events were computed using the SeisComP3 seismological software (Weber et al., 2007) and the LocSAT location algorithm (Bratt and Bache, 1988). P- and S-wave arrival times were manually picked and reviewed by Geoscience Australia seismic analysts. The mean location uncertainties in this dataset were found to be 3.39 km in the horizontal direction and 2.33 km in the up direction. To better constrain the location and pattern of aftershocks, events from this dataset were input for relocation using the HypoDD double-difference relative location algorithm (Waldhauser and Ellsworth, 2000), implemented in the software hypoDDpy (Krischer, 2015) (see the Supplement for detail). The HypoDD algorithm minimizes errors in hypocentral locations that are commonly attributed to uncertainties in Earth structure along the event-station ray path. The final relocation included a subset of 470 events from this catalogue (Figs. 3 and S2). The mean location uncertainties in the relocated dataset were found to be 63, 116 and 228 m in the east, north and up directions, respectively.

Both the largest aftershock (Ml 4.6) and the November Mw 5.2 event did not meet criteria for relocation in HypoDD using nearby seismic stations. However, both events were clearly recorded with good azimuthal coverage at regional and teleseismic distances. Based on waveforms from 19 stations in Australia, Antarctica, Kazakhstan, Mongolia and Thailand a separate relative location of these two events was calculated, together with the September Mw 5.3 mainshock, by manually aligning the distant waveforms and performing a joint hypocentre determination (Pujol, 2000; Fisk, 2002; see the Supplement for detail). The September and November earthquakes were then anchored to the Ml 4.6 aftershock location, which was well constrained in an absolute sense, as it occurred within ∼0.6 km of the LM01 seismograph station. In this way a better estimation of the absolute location of the point of initiation for both the September and November earthquakes was produced (Table 2). For calculations, the depths for all three events were fixed to 2 km. This is justified in that the calculations were performed using P phases from regional and teleseismic stations, so the epicentral solutions are fairly insensitive to depth. The depth estimates presented in Table 2 relate to the initial locations calculated using the permanent ANSN network, which was supplemented by the five rapid deployment aftershock kits for the latter two events.

The relocated position of the September Mw 5.3 epicentre is seen to coincide with the region of most prominent surface uplift imaged in the unwrapped InSAR image (Fig. 6b) and the greatest scarp height (Fig. 7a). Aftershock epicentres occurring in the interval between the September and November mainshocks are located predominantly east of the line of the surface rupture (Fig. 6a, b). The associated hypocentres occur between ∼1.5 and 4 km depth and below a plane extending from the trailing edge of the surface rupture dipping at 50∘ to the east (Fig. 8). Considerations of the preservation of seismic moment (i.e. M0 = shear modulus × fault area × average displacement; Hanks and Kanamori, 1979) suggest that the September rupture is unlikely to have extended beyond ∼1 km depth, implying that few if any of the aftershocks nucleated on the rupture plane. This is particularly the case for the largest aftershock (Ml 4.6) – the uncertainty ellipse for this event occurs almost entirely below the depth of other aftershocks. A linear cluster of hypocentres is located directly below the southeastern end of the rupture, parallel to the northwest–southeast oriented southern terminal structure at the surface (Fig. 6b). Similarly, a loose cluster of hypocentres, including the Ml 4.6 aftershock, underlies the east–west-trending valley bounding the northern end of the reverse rupture imaged in the InSAR data (Figs. 3 and 6a). Steep dips for terminal structures at the northern and southern ends of the reverse rupture are consistent with their orientation at a low angle to the easterly maximum horizontal compression direction (SHmax) of the extant crustal stress field (cf. Rajabi et al., 2017b).

The central ∼5 km of the September rupture, where the most significant surface displacement was observed, was modelled for Coulomb stress changes (e.g. Lin and Stein, 2004; Toda et al., 2005) (Fig. 8; see the Supplement for detail of method and parametrization). A dip of 50∘ to the east was assigned to the rupture for modelling purposes, consistent with the USGS focal mechanism for the event (Table 2), and with field data and modelling of the compressional reactivation of reverse faults (Sibson et al., 2012). The value is significantly different to the 30∘ dipping rupture plane exposed in the trench. We reconcile the two values in that the near surface may not be a good indicator of the fault dip at depth (cf. King et al., 2019, Fig. 4). This is supported by the prominent tension fissure east of the scarp at the trench location, which indicates a steeper rupture dip at depth at least locally (Fig. 4c). As the focal mechanisms of the aftershocks are not known, results are presented with stresses resolved for both reverse and strike-slip (Fig. 8a and b, respectively). In both cases, increases in Coulomb stress are modelled to have occurred down dip of the trailing edge of the rupture plane and to a lesser degree in the footwall block at depth. The majority of aftershock hypocentre locations show excellent correlation with the volume of increased Coulomb stress downdip of the rupture plane. Hypocentres in the footwall block only occur in regions of increased Coulomb stress if it is resolved for optimal strike-slip faults (Fig. 8b). A broad region of shallow stress relief above the trailing edge of the rupture corresponds with a landscape depression imaged in the InSAR (cf. Fig. 6b).

As mentioned in Sect. 3.2, simple forward modelling using a finite rectangular elastic dislocation model (Okada, 1985) (Fig. S4) suggests that a discrete surface rupture may not have been produced during the November strike-slip event. Based upon the InSAR surface deformation envelope (cf. Fig. 6c), a rupture ∼4 km long and extending from 1.0 to 2.5 km depth is preferred (Figs. S3 and S4). If Coulomb stress changes resulting from the September rupture are resolved for optimally oriented strike-slip receiver faults, the plane of the November rupture is seen to be positively stressed over more than half of its area (Fig. S4b). The case might then be made that the November event was triggered by static stress changes associated with the September event (cf. Mohammadi et al., 2019). However, additional factors may be involved as the stress increase is only on the order of 0.1–0.2 MPa in the region of nucleation of the November earthquake.

The November Mw 5.2 mainshock is located within a diffuse elongated cluster of dependant aftershock hypocentres at ∼3–4 km depth, which trends parallel to the northeast–southwest orientation of the surface deformation front imaged in the InSAR data, offset to the southeast by 0.25–1.5 km (Figs. 6c, d). Hypocentres predominantly occur at a greater depth than the preferred November event rupture plane (Fig. S4b).

The comparison of field observations, InSAR imagery and aftershock earthquake catalogue has permitted exploration of the surface and subsurface deformation field related to the Lake Muir earthquake sequence in unprecedented resolution. The September rupture, as mapped using traditional palaeoseismological means, is revealed to be part of a more extensive deformed region, involving both uplift and depression of broad areas proximal to the surface rupture. Furthermore, the spatial and temporal relationship between September and November Mw>5.0 events reveals a dependency with implications for how other earthquake swarms could be interpreted.

Whereas there is an emerging body of literature concerning human-induced/triggered earthquake sequences in SCRs (e.g. Keranen et al., 2013; Skoumal et al., 2015; Peterie et al., 2018), relatively little has been published on natural earthquake sequences (e.g. Chiu et al., 1984; Špičák, 2000; Allen et al., 2006). The latter appear to be an important component of intraplate seismicity, especially in non-extended SCR crust (e.g. Lopes et al., 2010; Clark et al., 2014a; Dent, 2016). Evidence for the spatial coincidence of similar-magnitude moderate- to large-sized earthquakes as part of these sequences, often with mixed focal mechanisms such as the Lake Muir sequence, is rare but not unprecedented. For example, the 1988 Mw 6.76 Tennant Creek earthquake sequence (Fig. 1; Table 1) began with three surface rupturing earthquakes of similar reverse mechanism within a 12 h period (Choy and Bowman, 1990; Bowman, 1992; Crone et al., 1997). Static stress triggering can account for the progression of events within the sequence (Mohammadi et al., 2019), with the rupture planes for the three largest events occurring well within a fault length of each other (cf. Caskey and Wesnousky, 1997). The sequence continues to this day, producing occasional events sufficiently large to deform the Earth's surface (Table 1). Of seven focal mechanisms constructed for the sequence, five are reverse faulting events with a similar P-axis orientation, and two are normal faulting events with tension axes coinciding with the compression axes of the reverse events (Clark and Leonard, 2003). The normal faulting events may relate to viscoelastic relaxation or adjustment (cf. DeVries et al., 2016). Similar normal faulting events were recorded in the aftershock sequence relating to the 1993 Killari (Latur) earthquake (Gupta et al., 1998).

In the Precambrian eastern Canadian Shield, the 1989 Mw 6.5 Ungava earthquake sequence involved a reverse-mechanism surface rupture followed after 0.9 s by a larger strike-slip rupture approximately 5.2 km away (Adams et al., 1991; Bent, 1994). The focal mechanisms for the events share a common P-axis orientation (Bent, 1994), similar to the Lake Muir mainshocks. Pre-existing bedrock faults were assumed to have hosted the ruptures, but the spatial relationship between rupture planes was not determined, ostensibly because a surface rupture relating to the larger strike-slip event was not observed. One potential explanation proposed by Bent (1994) was that the focus of the strike-slip event was deeper, and so the event did not result in surface rupture.

A comparable sequence is documented from central Australia. The reverse mechanism surface-rupturing 2012 Mw 5.4 Ernabella earthquake (Clark et al., 2014b; Fig. 1) was followed 15 months later by the strike-slip 2013 Mw 5.4 Mulga Park earthquake, which caused extensive surface cracking but no observed surface rupture (Clark and McPherson, 2013). The surface expressions relating to these events indicate that they were likely proximal (≤10 km separation) rather than collocated (Clark and McPherson, 2013). The first event occurred in the hanging wall of the crustal-scale Woodroffe Thrust fault (cf. Camacho et al., 1995; Camacho and McDougall, 2000; Lin et al., 2005), and the second event occurred in the foot wall, potentially at a depth of ∼10 km (Kevin McCue, personal communication, 2013). Again, both events share a common P-axis orientation (Clark and McPherson, 2013). In common with the Lake Muir sequence, the moment tensor for the strike-slip event had a low-percentage double couple, indicating rupture complexity.

The above examples comprise a class of seismicity in non-extended Precambrian SCR crust characterized by multiple moderate to large mainshocks sharing common compression axes. The similarity in P-axis trends of the focal mechanisms implies that the stress drop associated with each event was small compared to the extant differential stress (σ1–σ3; e.g. Hardebeck and Okada, 2018), despite the sometimes large magnitude of the mainshocks. The time interval between the largest events in a sequence varies greatly, as does the separation between rupture planes. Mohammadi et al. (2019) explored these relationships in the context of the Tennant Creek earthquake sequence and concluded that the time between two dependant events is proportional to the magnitude of the static Coulomb stress increase on the receiver fault plane (not necessarily at the hypocentral location), which in turn relates to the trigger-rupture magnitude and distance from the receiver fault. Multiple triggering mechanisms are required to explain time lags spanning several orders of magnitude (cf. Townend and Zoback, 2000; Gomberg and Johnson, 2005; Johnson and Jia, 2005; Meier et al., 2014).

The September 2000 to June 2002 Burakin earthquake swarm, located ∼20 km north of the 1979 Cadoux surface rupture (Fig. 1), is perhaps an example of a different class of seismicity to the sequences discussed above. Focal mechanisms were generated for the largest six of the approximately 18 000-event swarm (Leonard, 2002), spanning a magnitude range from Mw 4.1 to 4.6 (Leonard et al., 2002; Allen et al., 2006). These indicate a mixture of thrust and strike-slip ruptures and one pure normal faulting rupture, with diverse compression axis orientations and all within a ∼5 km radius. Preliminary joint hypocentre relocation suggests that these events and several thousand smaller events may have originated from as few as three to four source areas (Mark Leonard, personal communication, 2018). Aeromagnetic data (Milligan and Nakamura, 2015) indicate that the Burakin swarm area is characterized by a high lineament density, with major trends to the east, northeast and north. Static and dynamic stress interactions or block motion following moderate events may have stressed intersections between these trends, triggering further events with an eclectic mix of mechanisms with diverse P-axis orientations (cf. Talwani, 1988, 2014; Dentith and Featherstone, 2003). This behaviour, in contrast to the sequences that include larger events with focal mechanisms showing consistent P axes (e.g. Lake Muir), may occur preferentially in regions of extremely heterogeneous geological structure, where high stress concentration around numerous cracks and faults within the structured volume results in failure on many local fractures at low stress (Mogi, 1963), thus reducing the probability of failure on a single large fracture.

The Lake Muir sequence resulted in the production of a surface rupture and more subtle landscape change (i.e. depression of Lake Noobijup) that might be expected to be persistent in the landscape for thousands of years (cf. Quigley et al., 2010; Clark, 2010; Leonard and Clark, 2011). Questions of whether Lake Noobijup may have a seismogenic origin aside, there is no landscape evidence for Late Pleistocene to Holocene surface rupture on the structures that failed during the 2018 sequence (Sect. 1.3). Similarly, no evidence was found for more than the 2018 displacement across in situ weathered materials in the hand trench (Fig. 5). Further, no sharp landscape dislocations are evident that might relate to recent faulting of Paleogene palaeo-channel sediments (Sect. 1.2). If moderate to large earthquakes are assumed to recur on the faults that comprise the 2018 Lake Muir rupture, the long-term average uplift rate (i.e. the vertical component of the slip rate) is limited by the very low regional bedrock erosion rates of 1–5 m Myr-1 (e.g. Chappell, 2006).

The September Mw 5.3 Lake Muir earthquake was the ninth event documented to have produced surface rupture in the Australian SCR in historical times (Fig. 1, Table 1). These ruptures are located exclusively in the Precambrian SCR crust of central and western Australia (Fig. 1), and none could have been identified and mapped using topographic signature prior to the historical event (see references in Table 1). For example, Crone et al. (1997) excavated trenches across the 1986 Marryat Creek and 1988 Tennant Creek ruptures and found that while each rupture in part exploited pre-existing bedrock faults, there was no unequivocal geomorphic, stratigraphic or structural evidence to suggest a penultimate event in the preceding 50–100 kyr or more. A similar conclusion was made on the basis of trenching investigations of the 1968 Meckering surface rupture (see Clark and Edwards, 2018, and references therein).

Palaeoseismological investigations of several faults in the same Precambrian SCR tectonic setting provide evidence for limited recurrence of large earthquakes, with up to four events documented on an individual fault within the last ca. 100 kyr (Crone et al., 2003; Clark et al., 2008; Estrada, 2009). These scarps – Roopena, Hyden, Lort River and Dumbleyung (see Fig. 1 for locations) – all overlie simple through-going faults imaged in aeromagnetic data (Milligan and Nakamura, 2015). The two to five Quaternary events documented on the Hyden (Clark et al., 2008) and Lort River (Estrada, 2009) scarps are all that are evident across Late Neogene duricrust. While shallow trenches across the 2–5 m high Roopena scarp exposed Precambrian bedrock on both sides of the fault (Crone et al., 2003), nearby scarps are associated with an extended Neogene to recent history of movement (Miles, 1952; McCormack, 2006; Weatherman, 2006). For example, the Randell and Poynton Faults, immediately to the south of the Roopena scarp, are associated with 30–70 m of Pliocene and younger vertical displacement (McCormack, 2006). Scarps developed in the ca. 15 Ma surface of the Nullarbor Plain (Fig. 1), which overlies Neoproterozoic mobile belt basement, are associated with up to 15–30 m of vertical surface displacement (Hillis et al., 2008; Clark et al., 2012), implying the recurrence of a dozen or so neotectonic events per fault at most. In general, scarps developed within Archean and Palaeoproterozoic crust tend to be more modest in height, less well connected (i.e. spatially isolated) and more complex in plan than scarps in Mesoproterozoic and Neoproterozoic crust (Clark et al., 2012). A pattern is emerging in Australian Precambrian non-extended SCR where “one-off” ruptures, as evidenced by the historic surface-breaking earthquakes, are filling the spaces between mapped multi-event neotectonic scarps (Fig. 1).

Elsewhere in Precambrian non-extended SCRs the record is consistent with the Australian examples but more fragmentary. Seeber et al. (1993, 1996) excavated two trenches across the 1993 Killari earthquake rupture in peninsular India. These authors found no evidence for a pre-exiting fault zone but observe that the rupture followed foliation in Cretaceous basalt country rocks, which were estimated to be several hundred metres thick. A third trench revealed a section of the rupture that exploited a pre-existing fault zone in the basalts, which coincided with a lineament evident in satellite imagery (Rajendran et al., 1996). However, no evidence was found in the topography at all three trench sites or along the lineament for a Late Pleistocene to Holocene prior rupture. The Ungava rupture in the Canadian Precambrian shield was found to be controlled by compositional layering and foliation within crystalline rocks containing concordant ductile high-strain zones (Adams et al., 1992), similar to the 2016 Petermann Ranges rupture in Australia (King et al., 2018). Although the landscape record was short, being in a recently de-glaciated region, the authors found no evidence to suggest that the Ungava rupture occurred on a fault that has hosted prior reactivation in the Phanerozoic.

The preponderance of one-off and limited-recurrence faults suggests caution in applying a traditional elastic strain accumulation model to non-extended Precambrian SCR crust (cf. Braun et al., 2009; Clark, 2010). Indeed, over the last few decades, permanent and campaign GPS studies have failed to detect a tectonic deformation signal from which a strain budget could be calculated across all classes of SCR crust; this includes in Australia (e.g. Tregonning, 2003), the central and eastern United States (Calais et al., 2005), eastern Canada (Mazzotti and Adams, 2005) and central Africa (Gardonio et al., 2018). Similar studies have used these observations, amongst others (e.g. Calais et al., 2005), to propose that one-off events and clusters of large events deplete long-lived pools of “fossil” lithospheric stress (Liu and Stein, 2016; Calais et al., 2016) and/or there is an orders of magnitude difference in the timescales of elastic strain accumulation and seismogenic strain release (e.g. Clark et al., 2015; Craig et al., 2016). Hence, the underpinning assumption that a “long-term slip rate” is a meaningful concept in all classes of intraplate crust, as per the prevailing plate margin paradigm, remains to be fully tested. Preliminary indications are that the concept may be useful for extended SCR crust worldwide (Sandiford, 2003b; Quigley et al., 2006; Cox et al., 2013; Copley et al., 2014; Rastogi et al., 2014) and for the non-extended Phanerozoic SCR crust of eastern Australia (Fig. 1), where faults with up to a few hundreds of metres of neotectonic slip occur (cf. Sandiford, 2003a; Clark et al., 2015, 2017). However, faults studied thus far in the Phanerozoic SCR crust of the eastern United States (Crone and Luza, 1990; Crone and Machette, 1995; Zellman and Ostenaa, 2016; Hornsby et al., 2019) and Mongolia (Amarjargal et al., 2013) seem more akin to the limited recurrence faults in Precambrian non-extended SCR, cautioning against a blanket application.

Conceptually, for the purposes of probabilistic seismic hazard assessment, a fault source is a seismogenic fault that has produced earthquakes in the past and can be expected to continue doing so (Musson, 2012). Our results therefore urge caution when building “active” faults into seismic hazard assessments in SCRs. Furthermore, in the absence of meaningful recurrence of large events, building relations for fault displacement hazard using rupture traces from Precambrian Australia is also fraught (cf. Boncio et al., 2018; Livio et al., 2019). Another implication for seismic hazard assessment involving the “Lake Muir” class of seismicity discussed herein, perhaps most significantly demonstrated by the 1989 Ungava rupture (Adams et al., 1991; Bent, 1994), is that the palaeoseismic and neotectonics records of large earthquakes may be missing significant seismic moment in that not all mainshocks within a sequence produce recognizable surface deformation.