The Egiin Davaa prehistoric rupture, central Mongolia: a large magnitude normal faulting earthquake on a reactivated fault with little cumulative slip located in a slowly deforming intraplate setting

This paper considers the potential for large magnitude earthquakes within continental interiors on faults that may have very low slip rates and little in the way of cumulative displacement. We focus on the Egiin Davaa prehistoric normal faulting earthquake from central Mongolia (e.g. Khil’ko et al. 1985; Baljinnyam et al. 1993), which formed a semi-continuous rupture 80 km in length and for which we estimate a magnitude of 7.3–7.7. Despite its size, the rupture occurred in the Cenozoic along an ancient fault with <300 m of cumulative throw. A study of the Egiin Davaa rupture is hence important for understanding the potential for large destructive earthquakes on faults with little or no expression and for investigating the processes of fault initiation and reactivation.

Well-documented examples of large normal faulting earthquakes include the 1556 Huaxian earthquake of central China, which may have involved a throw of c. 8 m along a c. 100 km-long rupture (e.g. Yuan et al. 1991), and the 1739 Yinchuan earthquake, which generated a scarp with a maximum throw of c. 4.5 m over a length of c. 100 km (e.g. Zhang et al. 1986; Deng & Liao 1996). More recently, the 1959 Hebgen Lake (M_w 7.3) and 1983 Borah Peak (M_w 6.9) earthquakes of the western USA involved peak slips of 6 and 2.7 m, respectively (e.g. Stover & Coffman 1993). Most recently, the M_w 7.1 Yutian earthquake in Tibet generated surface offsets of 3–4 m over a rupture length of 45 km (Elliott et al. 2010). However, in contrast with the Egiin Davaa rupture, all of these examples occurred on faults with large cumulative displacements, the activity of which in the late Quaternary can be inferred with relative ease from the geomorphology.

The study of large earthquakes within slowly deforming continental interiors is challenging because modern examples are rare and the interval between successive earthquakes can be very long. Reconstructions of earthquakes from historical and prehistoric records are hence required to understand parameters such as the slip-vector azimuth and estimated magnitude. Moreover, by reconstructing ancient earthquakes and by determining their slip vectors, we may be able to glean insights into the kinematics of faulting (i.e. the way in which the population of active faults accommodates the overall tectonic motions) in these slowly deforming regions.

Using records of prehistoric earthquakes to investigate the kinematics of faulting is critical for investigations of the active tectonics of Mongolia (Fig. 1), where the faults are typically very long (100 km or greater), earthquake magnitudes can be large (M_w>7) and the slip rates are typically slow, leading to very long recurrence intervals (e.g. Frankel et al. 2010; Rizza et al. 2011). In addition, the historical record of earthquakes in Mongolia is fragmentary and few were documented before the twentieth century. In compensation for the general lack of instrumentally recorded earthquakes across large parts of Mongolia, we are aided in the study of prehistoric earthquakes by the preservation of their surface ruptures in the landscape for long periods of time (e.g. Khil’ko et al. 1985; Baljinnyam et al. 1993; Walker et al. 2008). Few of the prehistoric ruptures have been dated directly, an exception being the Jid fault palaeorupture in western Mongolia, the age of which is bracketed at 970–850 years (Walker et al. 2006).

This study presents the results of 80 km of fault mapping, topographic profiling and the interpretations of two palaeoseismic sites along the Egiin Davaa normal fault within the Hangay Mountains of central Mongolia (Fig. 1). Evidence for Quaternary faulting is widespread within the Hangay Mountains, but there is no record of large earthquakes in the instrumental record (Fig. 1; Cunningham 2001; Walker et al. 2008). The Egiin Davaa fault is one of several normal faults within the Hangay Mountains (e.g. Baljinnyam et al. 1993; Cunningham 2001; Walker et al. 2007, 2008). It is expressed as a scarp in late Quaternary surfaces with a typical height of 4.5 m. The sharp geomorphology and continuity of displacement across and along the fault has led some researchers to the interpretion that it formed in a single earthquake rupture potentially related to an historical event recorded in the sixteenth century (e.g. Khil’ko et al. 1985; Baljinnyam et al. 1993).

In the following sections, we assess whether the Egiin Davaa fault scarp was likely to have been caused by a single earthquake and we provide estimates of its likely magnitude, age and slip direction. We also assess the variation of displacement along its length and evidence for segmentation. To achieve these aims, we combined evidence from the geomorphology (as observed in the field and from high-resolution KOMPSAT-2 satellite imagery), the offset on late Miocene basalt flows, palaeoseismic trenching and the analysis of topographic profiles measured orthogonal to the scarp.

The active tectonics of Mongolia are related to the ongoing India–Eurasia continental collision (Baljinnyam et al. 1993). Approximately 7 mm a⁻¹ of NNE-directed shortening occurs across the Mongolian Altay Mountains in the western part of the country (Calais et al. 2003; Fig. 1). This shortening is accommodated on a series of NNW–SSE right-lateral strike-slip faults that are thought to rotate about a vertical axis (Bayasgalan et al. 2005). However, in central Mongolia, global positioning system (GPS) measurements indicate distributed east–west left-lateral shearing (Calais et al. 2003). The GPS measurements are compatible with the east–west left-lateral faulting observed across central Mongolia and there is hence no requirement for vertical axis rotation in this region (Bayasgalan et al. 2005).

The Hangay Dome sits in central Mongolia between two prominent zones of east–west striking left-lateral faults: the Gobi–Altai to the south and the Bulnay fault to the north (Fig. 1). The Bulnay fault zone was responsible for two earthquakes of M_w 8+ in 1905 (e.g. Baljinnyam et al. 1993; Schlupp & Cisternas 2007; Rizza et al. 2015) and c. 250 km of the Bogd fault in the Gobi–Altay ruptured in 1957 (e.g. Florensov & Solonenko 1965; Baljinnyam et al. 1993; Kurushin et al. 1997; Rizza et al. 2011). In contrast with its surroundings, the interior of the Hangay Dome has experienced no significant instrumentally recorded earthquakes (Fig. 1) and hence the distribution and style of active faulting is relatively poorly resolved compared with its surroundings.

The Hangay Mountains cover an area of over 200 000 km², with peaks up to 4000 m elevation and with valley incision of up to 1 km. The NW–SE-trending drainage divide separates rivers that flow northwards to Lake Baikal (and eventually to the Arctic Ocean) from those that flow southwards to the Valley of Lakes. The range is dome-shaped and many of the summits preserve remnants of a presumed pre-Cenozoic erosion surface (a peneplain) cut into resistant Palaeozoic sediments and late Palaeozoic–early Mesozoic granitoids (Cunningham 2001). West et al. (2013) fitted a surface across the peneplain remnants, hence removing the short-wavelength topography associated with valley incision into the surface. No large discontinuity is observed within the surface, indicating that any fault movements within the Hangay area have only a secondary effect on shaping its topography.

Neither the timing nor the cause of doming in the Hangay Mountains is well constrained. An increase in clastic sedimentation in the surrounding basins at 25–30 Ma is inferred to mark the onset of doming (Yanshin 1975; Cunningham 2001). The surface uplift may be associated with dynamic support from a mantle plume (Windley & Allen 1993), although the absence of a regional gravity anomaly across the Hangay region argues against such support (Bayasgalan et al. 2005). It could also be associated with buoyancy from melting associated with lateral flow around the Precambrian crustal keel (Cunningham 2001) or by magmatic underplating of the crust (Petit et al. 1998). Tertiary and Quaternary volcanism is widespread within the Hangay region (e.g. Filippov et al. 1976; Barry et al. 2003; Ionov 2007; Yarmolyuk et al. 2008; Hunt et al. 2012; Ancuta et al. 2013).

Both strike-slip and normal faults occur within the Hangay region. Many of the active faults have slip vectors consistent with the regional east–west left-lateral shearing measured by GPS (Calais et al. 2003). Walker et al. (2007, 2008) identified a number of east–west left-lateral fault systems, all showing evidence of activity in the late Quaternary, within the western and southern parts of the Hangay Mountains (Fig. 1). The occurrence of active normal faults at high elevations is likely to be influenced by the high topography within the range (Cunningham 2001). However, there is uncertainty as to whether the slip vectors on the normal faults are consistent with the regional strain field or whether they respond purely to the local extensional stresses associated with doming (e.g. Baljinnyam et al. 1993; Cunningham 2001; Walker et al. 2007, 2008). Determining the slip vectors of the normal faults, and hence determining whether they are compatible with the regional strain field, will help to distinguish between the differing scenarios.

The Egiin Davaa fault is a prime example of the relatively long NE–SW-trending normal faults that cut through the highest elevations of the Hangay Mountains (Figs 1 & 2). Elevations in the uplifted southeastern side of the fault reach c. 3500 m. The valley of the Badraig Gol (Badraig River) runs along the immediate hanging wall. The fault trace itself is situated at elevations of c. 2000–2500 m. Although there is relief of >1 km across the fault, the major part of this relief is due to fluvial incision into the dome-shaped Hangay Mountains (e.g. West et al. 2013) and the cumulative vertical displacement across the Egiin Davaa fault is <250 m. The geomorphology along the northeastern part of the fault is dominated by glacial or periglacial landforms. The landscape gradually changes to open steppe to the SW.

A semi-continuous scarp, with a vertical displacement of 4–4.5 m along its entire length, cuts across and displaces late Quaternary alluvial and glacial deposits along the southeastern side of the Egiin Davaa pass (Natsag-Yum et al. 1971; Khil’ko et al. 1985; Baljinnyam et al. 1993; Schlupp 1996; Cunningham 2001). Along its length, the scarp shows en echelon segmentation at the 100 m scale, with typical steps between individual segments of 10–20 m, along with a few larger steps of c. 100 m width at which small pull-apart features have formed. This small-scale segmentation is likely to be superficial and probably does not reflect segmentation of the fault at depth. The only larger scale discontinuity in the fault trace is found immediately west of the Egiin Davaa pass, where there is a c. 2 km step in the fault trace (Fig. 2). The continuity and steepness of the Egiin Davaa scarp has led to it being interpreted as resulting from a single earthquake. Khil’ko et al. (1985) first described the ruptures and assigned an age of 300–500 years based only on the apparent freshness of the ruptures. Baljinnyam et al. (1993), Schlupp (1996) and Cunningham (2001) provide additional descriptions of the Egiin Davaa fault and rupture.

The rupture is very easy to see along its entire c. 80 km length, both in satellite images (e.g. Figs 3 & 4) and in the field (Figs 5,, 6,, 7,, 8). The southern part of the rupture (the part roughly between the villages of Zag and Jagalant; Fig. 2) cuts across a broad low-relief alluvial plain on the SE margin of the Badraig Gol. This region is historically significant as the site where Temudjin (the future Chinggis Khaan) defeated his rival Jamuha. Near this location, the strike of the fault changes abruptly between sections striking almost north–south and others where the strike approaches east–west.

KOMPSAT-2 satellite imagery of the SW end of the mapped scarp, along with an interpretation of the geomorphology, is shown in Figure 3. Here, the fault trends WSW–ENE and cuts the surfaces of a series of coalescing alluvial fans (highlighted in dark orange in the interpreted image; Fig. 3) originating from small catchments in hills to the south. A right-stepping arrangement of en echelon fault segments towards the western end of the image is suggestive of a left-lateral slip component (cf. Baljinnyam et al. 1993). In addition, two small basins are present between the overlapping ends of the more widely separated segments in the eastern part of the image, which may result from localized extension at extensional jogs in the fault trace. The alluvial surface has been incised on its upthrown (southern) side by stream channels (beige on the interpretation). Small active fans have formed at the outflow of the incised channels on the northern (downthrown) side of the fault.

At the bottom left-hand corner of Figure 3, the Egiin Davaa scarp crosses a riser between the alluvial fan surface and the floodplain of the Badraig Gol (shown in gold on the interpretation). South of where the scarp enters the river floodplain, a single terrace is preserved at an elevation between the fan surface and the present floodplain. The surface of this terrace, which is shown in detail in the inset image in Figure 3, is covered in boulders and preserves remnants of meandering channels of a form very similar to that of the present floodplain. As the terrace is only present south of the Egiin Davaa fault, we infer that it represents uplift of the floodplain of the Badraig Gol. Because there is only a single terrace, despite it being in the floodplain of a large river, this provides confirmation that the scarp was formed in a single earthquake. Alternatively, if the scarp had formed in multiple earthquakes, then we would expect to see fragments of multiple terraces in the margin of the river floodplain. The maximum scarp height in this section of the fault is c. 4.5 m (all profile measurement sites are marked by white stars in Fig. 2 and are labelled S1–S7). Each ‘site’ consists of multiple individual scarp profiles collected within c. 100 m of each other to give a measure of the small-scale variability in scarp height caused by 100 m-scale segmentation. The profile heights for each site along the entire rupture length are plotted in Fig. 9 and individual profile analyses are presented in the Appendix.

Figure 4 shows KOMPSAT-2 imagery of the transition from the WSW–ENE-trending fault segment described in the preceding paragraph (and shown in Fig. 3) to a north–south trend. The scarp is segmented, but in the north–south section there appears to be a left-stepping en echelon arrangement of 100 m-scale segments suggestive of a component of right-lateral strike-slip. The fault in the region shown in Figure 4 cuts across a coalesced alluvial fan surface. Near the north end of the image a small stream has incised for a distance of c. 300 m upstream of the scarp, where a prominent knickpoint is observed. A field photograph of the scarp where it is incised by the stream is shown in Figure 5a. The photograph shows the location of a palaeoseismic trench, which we excavated into the northern margin of the incised stream. Figure 5b shows four topographic profiles measured orthogonal to the scarp in this approximate location. The scarp height is variable between individual profiles, as it was at all measured sites along the fault. The variation in scarp height over distances of tens of metres is likely to result from the small-scale segmentation of the fault trace, which, in turn, may relate to the presence of a strike-slip component of the slip. The maximum scarp height in this section is c. 4 m.

Further to the NE, the fault scarp continues to be clearly visible in alluvial fans along the base of hills to the south (Figs 6 & 7). The maximum scarp height in the section of fault shown in Figure 6 (site S4 on Fig. 2) is 4.2 m. The northernmost measurements of scarp height were made close to the regions shown in Figure 7. The maximum scarp height at both sites S5 and S7 is c. 4.5 m. The measured scarp height at site S6, which is not far from S7 and has a similar strike, is only c. 2 m. We suspect that the profiles collected together as site S6 were measured near the end of one of the short en echelon segments and hence the profile measurements from that site are not representative. North of latitude c. 47° 05′ N the footwall passes into a region dominated by glacial morphology, with deep U-shaped valleys in the footwall, extensive glacial moraines in the hanging wall and with well-preserved remnants of a pre-uplift erosional surface preserved on the low-relief footwall summit (Figs 3 & 8).

For most of the mapped fault trace, the c. 4 m scarp is the only clear indication of late Quaternary slip, but in a few localities we noticed evidence of possible larger late Quaternary displacements representing the cumulative effects of more than one earthquake. The most convincing of these is a continuous c. 8 m scarp along the northern section of the fault, where it is preserved in Cenozoic volcanic rocks (e.g. Fig. 8). Along this section the fault also offsets a recessional lateral moraine of the Chuluut Gol valley by 8 m (at 47° 11′ 30.72′′ N, 99° 57′ 8.1′′ E, Fig. 8c). We also found one locality at lower elevations (at 47° 01′ 59.9′′ N, 99° 39′ 25.9′′ E, not shown in any figure), where possible fragments of an older alluvial fan, at an elevation of c. 8 m above the present day river level, are preserved at the margins of an incised river catchment. Adjacent to the Chuluut Gol, the fault offsets late Miocene basalt flows, in places forming two distinct sub-parallel segments, where at least 225 m of fault throw has accumulated since the eruption of the ⁴⁰Ar/³⁹Ar dated capping flow at 6.27±0.16 Ma (Tielke et al. 2007; Ancuta et al. 2013; Fig. 8a).

Khil’ko et al. (1985) estimated an average scarp height of 3.0–3.5 m in the NE and 3.5–4.0 m in the SW. Khil’ko et al. (1985) also noted that, in the central section of the fault, the rupture splays into two or three sub-parallel traces, such that the total scarp height might be up to 4.0–4.5 m. Our measurements of the scarp height from topographic profiles broadly agreed with these estimates. All scarp profiles are included in the Appendix and the height measurements are plotted as a function of longitude on Figure 9b. We note that, although the local fault strike at our measurement sites varies by c. 80°, the maximum scarp height does not vary substantially, being c. 4–4.5 m irrespective of either the local strike or of the position along the rupture length.

In addition to the c. 80 km of rupture described in the preceding section, we identified a second rupture sited c. 20 km SE of the main Egiin Davaa fault (Fig. 2). This rupture is c. 20 km long with a strike of c. 010°. In the perspective view shown in Figure 10, the scarp is visible and runs in front of (west of) a degraded footwall block in bedrock. Several eastwards-flowing drainages have been blocked at the fault, with their old courses preserved as dry valleys in the uplifted footwall (marked by stars on Fig. 10).

We visited the southern part of this scarp in July 2008, and again in July 2013, to make observations in the field (Fig. 11) and to collect several topographic profiles orthogonal to the scarp using kinematic GPS (locations S8 and S9 on Fig. 2; see Appendix). From its southern end, the scarp rapidly increases in height to a maximum of c. 2.8 m. It is possible that the scarp increases further in height north of the section investigated in the field. Morphological dating of the scarp, compared with a similar analysis for the main Egiin Davaa scarp, indicates that the two ruptures may have formed at a similar time, possibly due to stress transfer between the two structures, or potentially as sub-events of the same earthquake.

In the preceding section we used the geomorphology of the Egiin Davaa scarp to confirm that it resulted from slip in a single earthquake. An additional 20 km of rupture may also have occurred on the Bor Hyarhan fault during that event. The existing age constraints on many of the known prehistoric earthquake ruptures in Mongolia come from qualitative assessments of their level of preservation compared with other ruptures of known age (e.g. Baljinnyam et al. 1993). Using this approach, Khil’ko et al. (1985) assigned an age of 300–500 years to the ruptures. Mongolian legends tell of a large magnitude event in this general area during the life of Avday Sayn Haan in the latter half of the sixteenth century (Natsag-Yum et al. 1971) and the Egiin Davaa rupture has been interpreted to result from that event.

In the following, we provide new estimates of the age of the Egiin Davaa rupture, both from palaeoseismic trenching and from morphological dating of the fault scarp. The first line of evidence for dating the scarp formation, however, comes from an analysis of satellite imagery. Figure 12 shows a KOMPSAT-2 satellite image of a north–south section of the fault (see Fig. 2 for location). Prehistoric burial mounds (khirigsuur), which typically date to 2500–3000 years ago (Allard & Erdenebaatar 2005; Wright 2007), are present at both the top and bottom of the scarp. The two khirigsuur at the top of the scarp (labelled 1 and 2) are sited on either side of a stream incised into the scarp. The close proximity of the burial mounds to the scarp and the incision implies that they post-date scarp formation. An age of several thousand years for the Egiin Davaa earthquake is also supported by trenching and morphological dating of the scarp.

We excavated two trenches through the Egiin Davaa fault scarp, one at 46° 58′ 39.7″ N, 99° 27′ 51.9″ E and another at 47° 0′ 25.06″ N, 99° 32′ 45.20″ E, in the summer of 2006. Both were dug by hand. The trench logs are presented in Figure 13. Trench 1 was dug into the side of an incised stream channel (Fig. 5) and so the ground surface shown in the log is not representative of the general slope of the fault scarp. The general stratigraphy exposed in the walls of Trench 1 was of coarse angular alluvial deposits interrupted by three light-coloured silt horizons containing sparse sub-angular clasts, which we interpret to be palaeosol horizons. The palaeosols (labelled P1, P2 and P3 from oldest to youngest) are likely to indicate soil formation during periods of fan surface stabilization, an interpretation supported by the presence of fossil rootlets in P1 and P2. Such periods may represent climatically modulated changes in sediment supply in the fan system, or simply periods of migration of the position of the main channel, and so their development is not necessarily related to fault slip events. The palaeosol layers are c. 10 cm thick and presumably formed as relatively planar bodies, but now all show either displacement or warping of the same magnitude consistent with surface rupturing in at least one earthquake post-dating the youngest of the three layers.

Three faults were identified in the trench walls (F1, F2 and F3). We interpret F3 to be the main fault zone. It has a dip of c. 60° W in the upper metre of the trench and shallows to a dip of c. 35° W below this. Fractures F1 and F2 dip steeply to the east and are antithetic to the main fault zone. F1 displaces the lowest palaeosol horizon (P1) by c. 30 cm (down to the east). The middle palaeosol unit (P2) is also deformed by the F1 fault, but by a combination of displacement and warping above the fault tip. The combined amount of displacement and warping in P2 is of a similar magnitude to the displacement of the lowest layer (P1). The uppermost palaeosol unit (P3) is not displaced, but is warped at a point c. 1 m north of the projection of the F1 fault. P3 is not preserved on the upthrown side of F1, but the observed warping in P3 is not less than 30 cm. Although we did not identify a discrete dislocation cutting the sediments up to P3, we inferred that the dip of the F1 fracture decreases in the upper part of the stratigraphy and intersects P3 where it is deformed. We inferred that the P3 unit represents the surface unit when F1 ruptured because the living root network may have been sufficiently cohesive to prevent the development of a discrete fault at the surface.

The highest of the palaeosol horizons (P3) is deformed by both the F2 and F3 faults. It has been broken into a number of short fragments and appears to be mixed with alluvium. We interpret the jumbled remnants of palaeosol and alluvium between F2 and F3 to represent colluvium derived from the collapse of the upthrown fault block. The antithetic F2 fracture extends into the P3 horizon. The P3 palaeosol is not preserved on the southeastern, upthrown side of the main fault (F3). This observation suggests that a significant amount of material has been removed from the upthrown block and re-deposited downslope. The pronounced scarp degradation at the trench site is a local effect because the excavation was made into the side of an incised gully. The most recent soil layer (P4), which is not fractured or deformed by faulting, thickens on the downthrown side of the fault, from which we interpret it as a colluvial wedge post-dating the slip event. The thickest section of the P4 soil is bounded by the opposing faults F1 and F3, which we interpret as the result of infilling of a shallow graben at the surface.

The displacement of the upper parts of the exposed stratigraphy by the three faults (F1–F3) is compatible with movement in a single slip event that post-dates deposition of the P3 palaeosol and predates deposition of the most recent soil layer (P4). None of the palaeosols is preserved on the upthrown side of the fault as exposed in the trench, suggesting that much of the immediate footwall has been eroded. As the palaeosol units are not preserved on the upthrown side of the fault, we are unable to determine the total vertical displacement across the fault zone. We can, however, estimate the minimum throw by assuming that the P1 unit was once located above the present land surface (e.g. Fig. 13a). Therefore we know that the displacement in this earthquake was >2 m.

To provide age constraints on the palaeo-earthquake recognized in the trench we dated the two lowest palaeosols (P1 and P2), using charcoal and fossil rootlets preserved in the soils and extracted from samples collected in 2006, at c. 7.6 and 7.4 ka (calibrated radiocarbon age BP, Table 1). These two units are displaced by faulting, which must therefore date to <7.4 ka. Unfortunately, we were unable to isolate fossil organic material within the upper palaeosol units and so only have an upper bound on the age of the scarp-forming event.

Trench 2 was located on the downthrown side of the Egiin Davaa fault at 47° 00′ 25.06″ N, 99° 32′ 45.20″ E (see Fig. 2). The scarp exhibits 2.5 m of relief at the trench site. The trench was 8.5 m long and up to 2 m deep and did not intersect the fault plane. As a result of time constraints, we were not able to lengthen the trench towards the fault. The basal unit in the trench is clast-supported alluvial gravel, which we interpret as predating the last rupture event. We did not observe buried soils in this trench, which we did in Trench 1, suggesting that the alluvial gravels are not much older than the age of the last rupture. A fissure or scarp-parallel graben-fill deposit that consists of massive silty sand, possibly aeolian in origin, is unconformably draped against the alluvial gravels. The silty sand is overlain by non-imbricated cobble-dominated colluvium, probably derived from the adjacent fault scarp. The entire sequence is capped by a 30–50 cm thick cumulic silty sand with a weak modern soil developed in it. We obtained a single radiocarbon age of 4910±60 years (calibrated radiocarbon age BP) from detrital plant fragments preserved in the massive silty sand of the scarp-parallel graben-fill deposit. Although the graben-fill deposits should provide a limiting minimum age for the last earthquake on the Egiin Davaa fault, the detrital plant remains may have been reworked from levels that predate the earthquake.

The continuity of the rupture on satellite imagery and as seen in the field points to a single rupture event. The structural relationships exposed in two palaeoseismic trenches are also compatible with slip in a single event, with a maximum age of 7.4 ka. A minimum age of 3.0–2.5 ka on the scarp comes from its spatial relationship with the khirigsuur burial mounds. Below, we use the scarp morphology and its current state of surface degradation to place further limits on the age of this earthquake rupture.

Analysis of the relative elevations of fault scarps cutting poorly consolidated sediments has been widely used to estimate fault rupture ages (Mattson & Bruhn 2001; Phillips et al. 2003; Spelz et al. 2008; Hilley et al. 2010; Koehler & Wesnousky 2011). Normal fault scarps of the Basin and Range Province in the USA were noted to have gentler slopes on older scarps than on those resulting from more recent events (Wallace 1977). Quantitative analysis of scarp degradation as a diffusive process has been used as a method of morphological dating of faulting (e.g. Bucknam & Anderson 1979; Colman & Watson 1983; Andrews & Hanks 1985; Hanks & Andrews 1989; Avouac 1993) and has been reviewed by Hanks (2000).

The principle of the technique relies on the fact that a step in topography is smoothed out over time and that the transport of mass downhill is proportional to the slope and the mass diffusivity (κ). Analysis of elevation profiles can be used to estimate the product of the mass diffusivity and time since the onset of diffusive degradation of the scarp (t). Therefore the relative succession of fault scarps in a region can be established and, if the mass diffusivity is known, then an age of the scarps may also be determined. Conversely, if the age of rupture is constrained from some other dating technique, then an estimate of the mass diffusivity for the locality can be made.

We used multiple surface topographic profiles across the fault scarp at seven locations along the rupture (Fig. 2, see also Appendix) to estimate a mean diffusion age (κt) for the scarp. The elevation data were collected from fault transects using differential GPS. The fault offset and fan slopes were estimated from points along the profile away from the scarp. These values were then used to constrain the estimates for the diffusion age (κt) using the finite-slope diffusion equation (Andrews & Hanks 1985; Hanks & Andrews 1989) and the near-field scarp profile. A detailed description of our methods is given in the Appendix.

The 46 fault offsets measured at the seven sample sites ranged from 0.7 to 4.7 m, with a mean of 2.8 m. The results for the diffusion age (κt) are given in the Appendix. The mean diffusion age for the 46 profiles at the seven sites along the rupture is 27 m² with a standard error in this mean of 3 m². Taking the range of mean diffusivity ages (24–30 m²) found here and combining this with diffusivity estimates for poorly consolidated surficial deposits in the Gobi–Altay of southern Mongolia, the Dzungarian Basin and the Tarim Basin of China (Avouac 1993; Avouac & Peltzer 1993; Hanks 2000; Carretier et al. 2002) of 2.3–7.5 m² ka⁻¹, and assuming they are applicable to this region of Mongolia, would give an age range estimate for the onset of scarp degradation of 12–4 ka. Taking into account that there is also an initial period of time for a fault scarp to disintegrate to the angle of repose by non-diffusive processes, which may be of the order of centuries (Wallace 1977; Zhang et al. 1986), the minimum age of the faulting constrained by diffusion modelling is 4 ka. A minimum age of 4 ka is compatible with our interpretation that the scarp formation predates construction of khirigsuur burial mounds (Fig. 12), which are 2500–3000 years old. For the rupture to be 300–500 years old (Khil’ko et al. 1985) would require a high rapid diffusivity of 40 m² ka⁻¹ based on the mean diffusion ages found here. This high diffusivity is four times larger than any found for a range of locations by Hanks (2000) and therefore very unlikely given Mongolia’s cold and arid climate.

If we make the assumption that the single detrital radiocarbon age of c. 4.9 ka extracted from the post-event sediments in Trench 2 represents the approximate age of the earthquake, we can estimate a diffusivity constant based on the mean diffusivity age (24±3 m²) of 4.9±0.5 m² ka⁻¹, a value similar to that found in the compilation by Hanks (2000) for western China and two to five times greater than those estimated from the Basin and Range Province in the USA.

The eight profiles from the Bor Hyarhan rupture at 99.70° E, 46.82° N (Fig. 2 and Appendix) are not amalgamated with the other 46 profiles in the calculation of the mean diffusion age because they occur on a structurally separate rupture. However, this southern rupture has estimated offsets in the range 1.5–3 m and diffusivity age values in the range 15–57 m², which is within the bounds found for the Egiin Davaa rupture. The mean and standard error values are 30±5 m², overlapping within error with the main rupture diffusion age of 24±3 m².

Fault plane exposures in bedrock occur at several locations along the Egiin Davaa fault. A fault plane in fractured quartzite bedrock is exposed at 47° 07′ 18″ N, 99° 44′ 27″ E, where spring water emanating from the scarp has removed the overlying soil and colluvium (Cunningham 2001; Figs 7b & 14). The exposed fault plane has a strike of c. 075° and a dip of c. 35° NNW. Striations show a slip-vector azimuth of c. 030°, indicating a large right-lateral component of slip (Cunningham 2001). This right-lateral component is at odds with the right-stepping arrangement of the rupture segments noted by Baljinnyam et al. (1993) and the possible left-lateral offsets noted where the rupture crosses incised rivers (Schlupp 1996). A slip-vector azimuth of c. 030° would also result in almost pure right-lateral strike-slip on the sections of the fault that trend close to north–south; however, we know from the topographic profiles across the fault that the scarp height is similar on both north–south- and ENE–WSW-trending sections (e.g. Fig. 9).

A potential explanation for these contradictions is that the striations relate to a geologically earlier period of activity. During our field visit we observed a series of subtle corrugations on the fault surface with an azimuth of 300° (Fig. 14). If the corrugations, rather than the striations, indicate the present direction of slip, then this would lead to a slight left-lateral slip component at the measurement site in addition to extension. A slip vector of 300° would result in similar, although not identical, amounts of extension on both fault trends, with a small left-lateral component on the sections trending ENE–WSW and a slight right-lateral component on those trending north–south. These strike-slip components are compatible with the arrangement of en echelon segments observed on east–west- and north–south-trending sections of the fault (e.g. Figs 3 & 4). Support for a slip vector of 300° comes from an observation by Baljinnyam et al. (1993), who refer to a fault plane exposure with striations indicating a slip-vector azimuth of 330–340°. They do not give a precise location of this site, but they describe the striations as being on a surface composed of Neogene basalt, which would place it near the northern end of the rupture at roughly c. 47° 12′ N, 100° 00′ E (e.g. Fig. 2). A slip-vector azimuth of c. 300° is consistent with a strain field dominated by east–west left-lateral faulting within central Mongolia (e.g. Fig. 2).

The mapped length of the Egiin Davaa rupture is c. 80 km (Fig. 2). The maximum vertical displacement is c. 4.5 m and, if the dip of 35°, which we measured from the main fault within Trench 1 (Fig. 11) and from an exposed fault plane in Figure 15, is representative, the maximum slip on the fault plane at the surface approaches 8 m. Applying the scaling relationship between M_w (moment-magnitude) and fault length yields an M_w of 7.3 (Wells & Coppersmith 1994). Using the maximum slip of c. 8 m yields a similar value of 7.25. Considering that the scarp height is similar along most of the length of the rupture, we also use 8 m as the average slip, yielding a slightly larger magnitude of 7.4.

We calculate seismic moment (M_o) using the relationship M_o=µAu (where µ is the modulus of rigidity, A is the area of rupture and u is the average slip on the fault plane). We have taken the average slip to be 8 m as the scarp height is relatively uniform along much of its length. The area of rupture is calculated from the fault length of 80 km, dip 35°, typical values in Mongolia of 20 km for the seismogenic thickness and 3×10¹⁰ Pa for the modulus of rigidity (e.g. Bayasgalan et al. 2005). We then converted this to a moment-magnitude (M_w) of 7.7 for a fault length of 80 km.

The Bor Hyarhan rupture is c. 20 km in length and has a maximum observed scarp height of 2.8 m. We have no direct constraint on the dip of the Bor Hyarhan fault, so we assume a fault dip of 45°, giving a slip on the fault plane of c. 4 m. Applying the Wells & Coppersmith (1994) scaling relationship between magnitude and maximum displacement yields a magnitude of c. 7 for the Bor Hyarhan rupture.

The geomorphology of the Egiin Davaa fault reveals evidence for only one Holocene surface-rupturing earthquake event along most of its length. There are, however, a few places where late Quaternary cumulative displacements are preserved. These sites allow some limits to be placed on the recurrence interval and average fault slip rate.

Today, the high terrain on the footwall side of the Egiin Davaa fault contains no permanent snowfield or glacier; however, during the late Pleistocene predominantly north-flowing valley glaciers crossed the trace of the fault near Egiin Davaa. Older moraines can be distinguished from younger moraines based on surface boulder frequency, boulder height and hillslope gradient. In one valley immediately north of the main Egiin Davaa divide, a recessional moraine has been displaced by slip on the fault (Fig. 2). The recessional moraine is c. 11 km down-valley from the cirques and c. 15 km up-valley from the terminal Last Glacial Maximum ice position. At this location, the moraine is offset by 8 m across the fault, probably indicating at least two rupture events since the time of moraine deposition. We obtained cosmogenic ¹⁰Be surface exposure ages from three granitic boulders from a post-Last Glacial Maximum recessional lateral moraine (at 2550 m elevation) offset by the Egiin Davaa fault (Fig. 8c; Table 2). The boulders yielded surface exposure ages of 16.2±1.8, 17.6±2.0 and 56.2±5.5 ka. The older age of 56.2 ka is not realistic given the regional glacial geomorphology and probably includes pre-deposition inheritance of ¹⁰Be. Based on the younger two ¹⁰Be ages, the moraine was probably emplaced around 16.2±1.5 ka (Fig. 16).

A limiting constraint on the overall age of the fault, and of the total cumulative throw, comes from the northern part of the fault close to the Egiin Davaa pass, where it displaces Cenozoic basalt flows that predate faulting. Around 250 m of fault throw has accumulated since the eruption of the capping flow, which has an ⁴⁰Ar/³⁹Ar age of 6.27±0.16 Ma (Tielke et al. 2007; Ancuta et al. 2013, Fig. 8a).

The geomorphology of the Egiin Davaa scarp is consistent with formation in a single surface-rupturing event with M_w 7.2–7.7. One trench excavated across the scarp yields evidence of surface rupturing, from which we are able to provide a maximum age of 7.4 ka for the fault movement. A second small trench did not expose the fault zone itself, but detrital plant remains in graben-fill deposits gave an age of 4.9 ka. Diffusional modelling of scarp profiles provides additional constraint, indicating that the scarp formed at least 4 ka ago, assuming that the diffusion coefficient is similar to that found in other arid and cold continental interiors. Finally, the distribution of khirigsuur burial mounds along the fault scarp also gives some circumstantial support to its formation before c. 3 ka. It is unlikely that the Egiin Davaa rupture formed in the historical event reported by Khil’ko et al. (1985).

The extremely well-preserved geomorphology of the Egiin Davaa rupture provides an opportunity to examine evidence for segmentation at various length scales. Segmentation at a 10–100 m scale is present along much of the rupture, which, from its size and consistent en echelon arrangement, is likely to be superficial and to result from a component of strike-slip motion. Other than this superficial segmentation, however, the rupture appears to be continuous for c. 50 km from its southwestern end, with a consistent 4 m high scarp along most of its length (e.g. Fig. 2). The southwestern segment is separated from a second segment of c. 30 km length that crosses the Egiin Davaa pass (Fig. 2).

In general, fault segment length is thought to correlate with the seismogenic thickness of the crust (e.g. Jackson & White 1989; Wallace 1989), with lengths of 20–30 km observed in places where earthquakes occur at depths of c. 15 km (e.g. in the Basin and Range Province), whereas fault segments with lengths approaching 100 km are found in regions such as parts of the East African rift, where earthquakes occur at depths of >30 km (e.g. Jackson & Blenkinsop 1997). The continuity of the Egiin Davaa rupture over a distance of at least 50 km may be indicative of relatively large seismogenic thickness, although depths determined for recent earthquakes in the Mongolian region do not typically extend lower than c. 20 km (e.g. Bayasgalan et al. 2005; Nissen et al. 2007).

We interpret 8 m of surface slip due to the shallow dip of the fault where it is exposed. This value is larger than expected from the empirical relationships between slip and length from global earthquakes, which usually have ratios of c. 5×10⁻⁵ (Scholz et al. 1986). A fault length of 80 km yields a ratio of 8×10⁻⁵. The relatively large amount of slip in large intraplate earthquakes such as that described at Egiin Davaa has implications for seismic hazard assessment within intracontinental regions, as earthquake magnitudes may be substantially larger than those estimated from global earthquake occurrences, the majority of which occur along rapidly deforming plate boundary regions.

We interpret an approximately NW-directed slip vector from fault plane corrugations and from the similar scarp heights for sections of fault trending east–west and north–south. A NW slip-vector azimuth appears consistent with the observed regional pattern of east–west left-lateral faulting (e.g. Fig. 2) and it is likely that reactivation of the Egiin Davaa fault resulted from the regional active tectonics of central Mongolia, rather than as a result of stresses introduced by the locally elevated topography (e.g. Walker et al. 2008).

Cenozoic activity on the Egiin Davaa fault is likely to have initiated no more than c. 6 myr ago, based on Ar–Ar dating of lava flows that predate fault initiation and are displaced by a maximum of 225 m across the fault. This estimate of total vertical displacement is in agreement with the findings of West et al. (2013), who showed that the long-wavelength topography of the Hangay Dome is largely uninterrupted by the mapped active faults. The valley of the Badraig Gol, which is in the immediate hanging wall of the Egiin Davaa fault, is hence likely to be caused by enhanced fluvial incision along the fault, rather than by cumulative fault movement alone. The cumulative displacement of only c. 0.3 km is much less than expected from global compilations of fault displacement and length, which predict vertical displacements of several kilometres (e.g. Schlische et al. 1996). It is hence likely that the Egiin Davaa fault is an ancient structure that has been reactivated in the late Cenozoic and which has ruptured in two large earthquakes in the last c. 16 kyr.

The 225 m displacement of the c. 6 Ma basalt flows places a minimum bound of c. 0.04 mm a⁻¹ on the rate of vertical displacement across the fault. There has been c. 8 m of vertical displacement over the last c. 16 kyr, yielding an approximate Late Pleistocene vertical rate of c. 0.5 mm a⁻¹ and an average slip rate of c. 1 mm a⁻¹. The full 225 m of vertical displacement could accumulate in a little over 500 kyr at the Late Pleistocene rate. As this rate is calculated from an 8 m scarp that we interpret to result from only two earthquakes, it is unlikely to be accurate and yet does suggest that the late Pleistocene rate is more rapid than that derived from the 6 Ma basalts.

The mismatch between the 6 Ma basalts and late Pleistocene slip rates indicates either that the Egiin Dawaa fault is very young (<1 Ma), or the Late Pleistocene slip is more rapid than the long-term average. Relatively young (c. 5 Ma) initiation of faulting has been found in several other deforming parts of Mongolia, including the Gobi–Altay (Vassallo et al. 2007) and at the margins of the Altay (e.g. Vassallo 2006). However, variations in slip rate and the clustering of earthquake activity on slow faults in continental interiors is seen in the palaeoseismic record of several regions (e.g. Crone et al. 1997) and it is possible that the Egiin Davaa fault has also experienced pulsed activity through the Quaternary. Switching of activity on faults may help to explain the geomorphology of other faults within the Hangay region, such as the Orhon Valley fault (Fig. 1), which have a very clear expression in the large-scale geomorphology, but which show no evidence for rupture in the late Quaternary cover.

We have shown that individual earthquake ruptures in cold, arid continental interiors may be preserved in the landscape for several thousand years. The Egiin Davaa rupture is one of the few palaeoruptures in Mongolia that has been dated, at c. 5 ka, and provides an important calibration point for quantifying the linear slope diffusion rate in Mongolia. This study corroborates evidence that many of the presently active faults in Mongolia are relatively young and may relate to a change or reorganization in the kinematics at 5 Ma or younger. There is a suggestion that activity may switch between individual faults, a process that may be widespread in slowly deforming areas, providing another level of complexity in identifying and characterizing potential sources of future events in such regions. Furthermore, the Egiin Davaa fault represents a rare example of a major earthquake on a reactivated ancient fault, on which there is little Cenozoic cumulative displacement. The fact that such a major earthquake occurred on a fault with little cumulative displacement and relatively low slip rates (probably below the resolution of all but the most long-lived GPS networks) poses a serious challenge for the identification of earthquake hazards. It is only due to the exceptional preservation of fault rupture in this area that this structure has been identified as a potential source of large earthquakes. It remains unclear how much these difficult-to-identify faults contribute to societal risk elsewhere.

The values for the vertical fault offset (2a) and fan slopes (β₁, β₂) are calculated from a linear inversion of the profile points away from the degraded scarp (typically points measured beyond 15–30 m either side of the fault scarp). On longer profiles, heights beyond where the fan slope flattens downslope or steepens upslope are discounted in the offset/slope calculations. The reported error in the heights of the individual points on the DGPS profiles is 5–6 cm. An error in the offset is estimated from weighting the points by assuming a constant error of 6 cm for each point. For some estimates of fault vertical offset using short profiles with subsequently few DGPS points away from the scarp to calculate the fan slopes and offset, the 1σ error in the calculated offset can be 30–50%. However, the bulk of offsets >2 m have errors of 3–20%.

The fault offset and fan slopes are then used in Equation (A2) to search for the minimum misfit in the L2-norm sense of the fit of the scarp for κt as a free parameter using the DGPS data points in the centre of the profile (i.e. those points in the near-field that are not used to estimate the fan slopes and fault offset). We assumed an initial scarp angle α=35° as the angle of repose, with variations of ±5° in this assumption having only a small impact on the diffusion age, κt, relative to the spread of the calculated values.

In addition, we checked for the correct identification of the fault position by searching for the minimum misfit in the scarp profile by shifting the choice of fault location by ±5 m (i.e. the point of the profile at which x=0), re-estimating the fault offset (2a) and fan slopes (β₁, β₂) and then recalculating κt for this new location. Of a total of 58 DGPS profiles examined, 12 were excluded from further analysis due to either the profiles being too short to estimate the fan slopes, small offsets that could not be estimated reliably, large erosive channels at the scarp base or profiles which departed strongly from a diffusive shape. For the southern separate rupture, eight of the 12 profiles measured were used to calculate the mean diffusivity.

The values for the vertical fault offset (2a) and fan slopes (β₁, β₂) are calculated from a linear inversion of the profile points away from the degraded scarp (typically points measured beyond 15–30 m either side of the fault scarp). On longer profiles, heights beyond where the fan slope flattens downslope or steepens upslope are discounted in the offset/slope calculations. The reported error in the heights of the individual points on the DGPS profiles is 5–6 cm. An error in the offset is estimated from weighting the points by assuming a constant error of 6 cm for each point. For some estimates of fault vertical offset using short profiles with subsequently few DGPS points away from the scarp to calculate the fan slopes and offset, the 1σ error in the calculated offset can be 30–50%. However, the bulk of offsets >2 m have errors of 3–20%.

The fault offset and fan slopes are then used in Equation (A2) to search for the minimum misfit in the L2-norm sense of the fit of the scarp for κt as a free parameter using the DGPS data points in the centre of the profile (i.e. those points in the near-field that are not used to estimate the fan slopes and fault offset). We assumed an initial scarp angle α=35° as the angle of repose, with variations of ±5° in this assumption having only a small impact on the diffusion age, κt, relative to the spread of the calculated values.

In addition, we checked for the correct identification of the fault position by searching for the minimum misfit in the scarp profile by shifting the choice of fault location by ±5 m (i.e. the point of the profile at which x=0), re-estimating the fault offset (2a) and fan slopes (β₁, β₂) and then recalculating κt for this new location. Of a total of 58 DGPS profiles examined, 12 were excluded from further analysis due to either the profiles being too short to estimate the fan slopes, small offsets that could not be estimated reliably, large erosive channels at the scarp base or profiles which departed strongly from a diffusive shape. For the southern separate rupture, eight of the 12 profiles measured were used to calculate the mean diffusivity.

The research was funded by the Royal Society of London through a University Research Fellowship awarded to RTW. KW and RC also thank the National Science Foundation (Continental Dynamics Program Award GEO-EAR-1009702) and the Keck Geology Consortium for funding and support. RTW acknowledges support through NCEO/COMET+ and the NERC-ESRC Earthquakes Without Frontiers consortium. The KOMPSAT-2 images were provided through an ESA Category-1 award to RTW. We thank Nathan Toke and an anonymous reviewer for their helpful comments and the editors for compiling this book.

Correction notice: This paper should be published Gold Open Access.