A 2600-year-long paleoseismic record for the Himalayan Main Frontal Thrust (western Bhutan)

In spite of an increasing number of paleoseismic studies carried out over the last decade along the Himalayan Arc, the chronology of historical and prehistorical earthquakes is still poorly constrained. In this paper, we present geomorphologic and paleoseismic studies conducted over a large river-cut exposure along the Main Fontal Thrust in southwestern Bhutan. The Piping site reveals a 30 m high fault-propagation fold deforming late Holocene alluvial deposits. There, we carried out detailed paleoseismic investigations and built a chronological framework on the basis of 22 detrital charcoal samples submitted to radiocarbon dating. Our analysis reveals the occurrence of at least five large and great earthquakes between 485± 125 BCE and 1714 CE with an average recurrence interval of 550± 211 years. Coseismic slip values for most events reach at least 12 m and suggest associated magnitudes are in the range of Mw 8.5– 9. The cumulative deformation yields an average slip rate of 24.9± 10.4 mm yr−1 along the Main Frontal Thrust over the last 2600 years, in agreement with geodetic and geomorphological results obtained nearby.

In this study, in order to investigate large Himalayan earthquake series, we selected a site in southwestern Bhutan where a ∼ 30 m high natural section is exposed by erosion at the outlet of a trans-Himalayan river called the Wang Chu. After describing the Bhutan Himalaya setting, we present the geomorphological and paleoseismic investigations carried out around and along this exposure. Our results allow us to discuss the timing and the magnitude of five surfacerupturing events that occurred in Bhutan during the last 2600 years.

Active tectonics in Bhutan
From north to south, Bhutan can be divided into four distinct tectonic units (Fig. 1b): the Tethyan Sedimentary Series (TSS), the Higher Himalaya (HH), the Lesser Himalaya (LH), and the Siwaliks (Sw). All these units are bounded by major faults including the South Tibetan Detachment (STD), the Main Central Thrust (MCT), the Main Boundary Thrust (MBT), and the Main Frontal Thrust (MFT), which is the most recent expression of the thrust sequence that accommodated the deformation over geological timescales (Gansser, 1964;Le Fort, 1975;McQuarrie et al., 2008;Long et al., 2011a). At depth, these four major north-dipping thrust faults connect to the Main Himalayan Thrust (MHT), a mid-crustal decollement under which the Indian plate subducts beneath the Himalayas and Tibet. In terms of geometry, several studies suggest a ramp-flat-ramp geometry of the MHT (e.g., Zhao et al., 1993;Nelson et al., 1996;Cattin and Avouac, 2000;Nábelek et al., 2009;Coutand et al., 2014;Le Roux-Mallouf et al., 2015).
Present-day deformation is constrained by (1) a far-field convergence of 17 ± 0.5 mm yr −1 inferred from geodetic measurements along three profiles across western, central, and eastern Bhutan (Marechal et al., 2016) and (2) a single estimate of Holocene uplift rate of 8.8 ± 2.1 mm yr −1 , from the study of alluvial terraces along the front in central Bhutan (Berthet et al., 2014). A first paleoseismic study by Le  suggests that south-central Bhutan has been struck by at least two earthquakes during the last millennium, including (1) a M w 7.5-8.5 earthquake in central Bhutan that produced ∼ 1 m of coseismic uplift in 1714 CE (see also Hetényi et al., 2016) and (2) a M w > 8.5 earthquake that produced ∼ 8 m of coseismic uplift during medieval times (between 1204 CE and 1464 CE). This last event contributes to the debate about variations in coupling along strike and the possible deficit of seismic moment along the Himalayan Arc (e.g., Bilham et al., 2001;Stevens and Avouac, 2016) and the probability of occurrence of a subduction-type M w 9 earthquake in this region (Kumar et al., 2010;Mugnier et al., 2013;Srivastava et al., 2013;Stevens and Avouac, 2016;Le Roux-Mallouf et al., 2016;Wesnousky et al., 2017a).

Geomorphology of the study area
The study site, called Piping, is located in the Lhamoizingkha area (SW Bhutan), immediately upstream of the confluence between the Wang Chu and the Ramphu Chu, a 5 km long tributary that drains a 4.5 km 2 watershed (Fig. 2a) the Subhimalayan zone-S (Siwaliks, Miocene-Pliocene), immediately north of the MFT, composed of medium-to-coarse-grained sandstone and pebble-to-cobble conglomeratic sandstone (Long et al., 2011b, and references therein) dipping 50-70 • to the north and visible over more than 300 m; the Alluvial plain, composed of young unconsolidated sediment.
The MFT separates the flat, mostly undeformed, recent to active deposits of the Alluvial plain to the south from a well-developed 4 km long flight of alluvial terraces deposited by the Wang Chu over the Manas and Siwaliks formations. These terraces are composed of well-stratified cobbles to boulders (dominant lithology is metamorphic from the Manas Formation) with a sandy matrix. Available outcrops display a relatively thin sediment cover (generally less than 6 m) deposited over clear strath surfaces cutting into the Manas and Siwaliks formations. The lower (younger) terraces (T1, T2, and T3) are located directly along the present stream at low elevations (∼ 1, ∼ 11, and ∼ 33 m above the present stream, respectively). T1 and T2 are deposited over the fault trace ( Fig. 2a) and display continuous top surfaces suggesting no significant deformation occurred since their deposition. T1 is likely immerged during the monsoon season, as attested by natural and anthropic detritus caught in the low vegetation. Intermediate terraces (T4, T5, and T6) appear as continuous ribbons perched above the present river level at ∼ 43, ∼ 80, and ∼ 90 m, respectively. Finally, higher (older) terraces T8 and T9 are strongly dissected and preserved as thick alluvial sequences (e.g., ∼ 18 m thick for T8) on top of steep buttes forming local heights at ∼ 100 and ∼ 170 m above the present river level, respectively. East of the study site, a local watershed basin called Ramphu Chu cuts into the Manas and Siwaliks formations and exits the steep piedmont at the location of the MFT where it forms a 500 m wide alluvial fan (Fig. 2a). The upstream section of the fan was deposited against the main MFT tectonic scarp and over the fault trace as visible on field photographs (Figs. 2b and 3a) and provides the main stratigraphic section studied here to unravel the recent deformation history along the MFT.

Paleoseismic exposure
An orthorectified photographic mosaic (Fig. 3a) of the site shows the 30 m high river-cut cliff and displays a 40 m wide deformation zone that separates the gray Siwaliks (unit S) to the north, topped by the south-dipping U7 terrace (Wang Chu deposits) from a horizontal 18 m thick sequence of fan deposits (U6 to U0) from the Ramphu Chu. A 50 m long by 30 m high section of the natural exposure was cleaned, partly gridded, and logged in detail (Fig. 3b and Figs. 3,5,6,7,and 8) based on stratigraphy, lithology, and grain size. Overall, 50 samples of organic matter (charcoal and plant debris) were collected, and 22 were selected for radiocarbon age determination (Table 1).

Chronostratigraphy
The stratigraphy of the northern section of the exposure (Fig. 3) is mostly constituted of massive gray sands with fine beds of white silts, pebbles, and cobbles that outline a ∼ 60 • dip to the north. This unit crops out along a ∼ 150 m long section of the river cut and exhibits a thickness of at least 90 m. It is widely observed regionally along the mountain front (Long et al., 2011a) and is attributed to the Siwaliks formation (S). Here, it is overlain with a ∼ 4 m thick clast-supported stratified cobbles-to-boulders unit (called U7 hereafter). Observed clasts are generally rounded with a significant contribution of metamorphic lithology from the Higher Himalaya formation (Long et al., 2011a). Considering stratigraphy, clast roundness, distance to the nearest outcrops of said forma-tion (∼ 25 km north of the site), and relationship to the local drainages, we interpret this unit as an alluvial terrace deposit from the trans-Himalayan Wang Chu. Unit U7 is stratigraphically above the Siwaliks (S) and lies over a clear erosion surface (strath) that cuts through the Siwaliks north-dipping stratigraphy. Its top surface is eroded north of grid point (22, 24) and preserved and overlain with a succession of finegrained units south of it (Fig. 3b); it is hereafter considered to mark the base of the Quaternary stratigraphic record at this site.
On top of unit U7, we observed an 18 m thick succession of deposits comprised of 20 to 40 cm thick massive bluishgray silt layers and clast-supported gravel layers with a sandy matrix. Major sediment packages are delimited along contin- uous near-horizontal (in the undeformed section) limits and named U6 (deepest) to U0 (shallowest). They exhibit abundant detrital charcoal lumps, most of them reaching 1 cm in diameter and displaying freshness, compactness, and angularity indicative of a priori short transport and storage times. Overall, 50 samples were collected from units U6 to U0, of which 22 were selected and submitted for radiocarbon dating (Table 1). Fine calibration was performed with OxCal 4.2 using a depositional model where samples from the same unit are defined as a phase (e.g., Lienkaemper and Bronk Ramsey, 2009) and yielded dates consistent with the observed stratigraphic order.
-Unit U6: the lowest unit lies over unit U7 over the northern section of the exposure (north of x = 22) where it is ∼ 2 m thick, while its base is presently below the water table in the southern section and could not be logged (Fig. 3). It is comprised of massive fine to very fine silts, blueish gray in color, interbedded with 30 to 40 cm thick poorly stratified lenses of matrix-supported angular gravels, containing ∼ 50 % of fine to coarse sand. The top of U6 is marked by a relatively smooth poorly expressed erosion surface. The age of the unit is constrained by seven samples with a narrow distribution of radiocarbon ages comprised between 2480 ± 30 and 2625±30 years BP (Table 1) suggesting a relatively fast deposition process. A single obvious outlier (sample PI-C46 with a radiocarbon age of 37 700 ± 800 years BP) was considered reworked and therefore discarded from our analysis. Model calibration yields a deposition date of 670 ± 165 BCE.
-Unit U5: within the southern undeformed section of the exposure section, this unit displays a thickness of ∼ 1.5 m (south of x = 59 m in Fig. 3b). It exhibits a sim- ilar grain size distribution to that of U6 but with distinct gravel and sand lenses: the bottom section is marked by well-defined fine gravel lenses, while the top section is evidenced by a ∼ 1 m thick coarse sand and gravel lens. The top of unit U5 is defined by a weakly expressed erosional surface that probably reflects a short depositional hiatus rather than established rill processes. Unit U5 yielded six samples, four of which have ages between 2180 ± 30 and 2285 ± 30 years BP, again indicative of a relatively fast deposition process. The two remaining samples collected at the base of the unit (PI-C11 and PI-C12) are significantly older than other samples from U5 and even U6 (2905 ± 30 and 2860 ± 30 years BP, respectively). We suspect they have been reworked from the lower section of U6 or from an even older unit, and we choose therefore to discard them from our analysis. Model calibration yields a deposition date of 290 ± 120 BCE.
-Unit U4: this unit is 3 to 4 m thick in the southern section of the exposure (south of x = 55 m in Fig. 3b) and thins out to the north where is forms an onlap against U5 then U6 at x = 38 m. U4 is almost entirely composed of matrix-supported gravels with a few silt lenses and terminates with a continuous ∼ 15 cm thick sand layer. This unit did not yield any adequate sample for radiocarbon dating, probably on account of the higher-energy regime at the time of its formation.
-Unit U3: this unit displays a very constant thickness of ∼ 1.5 m over the whole exposure (between x = 24 and x = 98). It is comprised of massive silts with 20 to 30 cm thick lenses of coarse sand and fine gravel. U3 yielded three samples with radiocarbon ages of 1730 ± 30, 1960 ± 30, and 2560 ± 30 years BP. Since the latter sample is contemporaneous with U6, it is considered reworked and removed from any subsequent analysis. Model calibration yields a deposition date of 240 ± 100 CE.
-Unit U2: this unit also exhibits a constant thickness of ∼ 1.5 m over the exposure. It is, however, comprised of matrix-supported gravels with a few sand lenses, which suggests a slightly higher-energy fluvial regime. It yielded three samples with radiocarbon ages 1520 ± 30, 1770 ± 30, and 2405 ± 30 years BP. Similarly, since the latter is contemporaneous with U6, it is considered reworked and removed from subsequent analysis. Model calibration yields a deposition date of 440 ± 70 CE.
-Unit U1: this unit is ∼ 3 m thick over the exposure. It displays a stratigraphic content very similar to that of unit U2 and lies over a weak erosional surface forming the top of U2. For logistics and safety reasons, unit U1 could not be sampled for age determination.
-Unit U0: this is the ultimate deposit of this section. It displays a variable thickness of ∼ 20 cm to up to 4.5 m with a strongly eroded top surface within the deformed zone, north of x = 52 m (Fig. 3b). The top of U0 marks the abandonment of the section before it was intensely and almost entirely incised by a local gully (x = 52-70 m). Although this unit was directly accessed at the location of the uppermost log (box marked " Fig. 8" in Fig. 3), we could not retrieved adequate material for age determination.
Within this succession, clast lithology and roundness are constant, thus suggesting a common nearby source for units U6 to U0 distinct from that of U7. Gravels are very angular and made of quartzite and phyllite from the Manas Formation, sands are fine-grained and well-classed, and silts are massive and blueish gray in color, where not oxidized. Although grain size distribution varies across units from gravel-dominant (with sand lenses) to silt-dominant (with sand and gravel lenses), this does not necessarily reflect significant variations in transport flow velocity (e.g., Miller et al., 2014). Overall, we interpret units U6 to U0 to derive from the same nearby low flow-velocity source consistent with the recent alluvial fan mapped at the outlet of the Ramphu Chu watershed basin (Fig. 2). Two additional units display specific wedge-shaped geometries: W2 between U5 and U4 and W1 deposited against U0 and immediately below the modern soil. Both units exhibit little stratigraphy, with a clear debris and wash facies for W1 and intense internal deformation typical of a slump for W2 (see details below), and are interpreted as colluvial wedges (more details in the following section). W1 is stratigraphically the youngest unit observed here. Two detrital wood samples (PI-C23 and PI-C24) yield modern ages. Since roots found in the region sometimes resemble tree-trunk bark in terms of size, density, and texture, we suspect the ligneous samples PI-C23 and PI-C24 may derive from in situ roots and may not be representative of W1's true age. These samples are discarded in our analysis.
Additionally, it is quite notable that the undeformed part of the 18 m thick Ramphu Chu section (south of x = 54 in Fig. 3b) presents a quasi-continuous (erosion surfaces are poorly expressed and stratigraphic limits are virtually flat) succession of silt, sand, and gravel deposits constrained by 15 radiocarbon samples (Table 1). To better assess the timing of deposition for the uppermost units, we assume that deposition was mostly continuous, and we build an age-versus-height relationship for all samples retained for our analysis (Fig. 4). Our approach yields an average deposition rate of 7.1 ± 0.2 mm yr −1 between 805 ± 30 BCE (U6) and 520 ± 95 CE (U2), with potential short-term variability between silt and gravel beds (e.g., Kumar et al., 2007). On that basis, and considering a similar constant sedimentation rate until the final deposition of U0, we may extrapolate the deposition rate and propose a tentative date with large uncertainties (2σ ) for the top of U1 at 940 ± 200 CE. Since U0 is strongly eroded, we did not attempt to date its top surface.

Exposure description
Large-scale deformation across the MFT at the Piping site is illustrated by fault-propagation folding affecting terrace unit U7 shown on Fig. 3. U7 crops out ∼ 34 m above the present stream (grid point (0, 34) in Fig. 3b), dips increasingly to the south, is sheared by a system of north-dipping thrust fault splays (F2 to F5 in Fig. 3b), dips reverse to the north and disappears underneath a massive 8 to 10 m thick fault gouge (unit G in Fig. 3b and following). Since U7 does not crop out south of the main fault zone, it is necessarily deeper than the present river level (at least below U6) and has hence recorded more than 34 m of uplift since its deposition.
Subsequent units U6 to U0 are mostly undeformed from the southernmost tip of the exposure to the center of the studied section (i.e., south of x = 54 m in Fig. 3b). There, they exhibit various stages of deformation, from warping with minor faulting (U0 to U3) to folding (U4) and intense faulting with duplexing (U5 and U6), indicating than the older units of the Ramphu Chu fan have cumulated more deformation. Furthermore, fault strand F5 cuts through the whole section and reaches the surface with a near-vertical dip and affects U2 to U0 with an apparent normal geometry. To describe faulting and abutting relationships in detail and identify surfacerupturing events, we focus on two excerpts presented at high resolution in Figs. 5 to 8.
The lower section documents deformation affecting units U7, U6, and U5 (Figs. 5, 6, and 7). From grid point (28, 2) (Fig. 5b), U7 is overlain with unit (G), which is composed of massive reddish to brownish clay with sheared and fractured clasts from the Siwaliks formation as well as cobbles and boulders from U7. It exhibits intense internal deformation (see close-up in Fig. 7a) typical of a fault gouge. The localized fault contact between G and U7 corresponds to F4 in Figs. 3b and 5b. To the south, U6 crops out at the base of the exposure and is affected by fault F1, which cuts through U6 and U5, and dies out ∼ 4 m southward within U5 (Fig. 5b). F1 accommodates only minor faulting as attested by a relatively small 30 cm offset affecting the base of U5 (Fig. 7b). Secondary normal-geometry splays F6 and F7 branch out from F1 and displace the base of U5 vertically by a total of ∼ 60 cm. F7 tapers out within U5, while F6 cuts it entirely and terminates against the low-dipping fault strand F2 at a right angle. Above F2, U6 displays strongly deformed   near-vertical bedding produced by dragging along F2 (Fig. 6) and forms a fault-propagation fold. Hence, F2 is a duplex fault that accommodates major deformation within the exposure. The uppermost part of unit U6 is affected by similar duplexing deformation along the F3 fault strand, though with a much smaller offset. F2 also affects U5 where duplexing produced a clear scarp overlain with the wedge-shaped unit W2. Its stratigraphy is composed of finely layered silts and gravels similar to U5 but exhibits intense deformation with sheath folds typically associated with slumping along a slope (Fig. 7c), here consistent with the frontal slope of the scarp. We interpret W2 as a scarp-derived colluvial wedge deposited during or shortly after a coseismic displacement along F2 affecting U5. The tops of W2 and U5 are in continuation and overlain by U4, which does not exhibit noticeable deformation at this location, and show that F2 was not reactivated after the deposition of U4.
The upper section (Fig. 8) documents the northernmost fault strands F4 and F5 as they reach the surface. At the bottom of the trench (Fig. 3b), F4 and F5 originate from the main gouge zone (G) where they dip ∼ 20 • N, cut through U7 with a steeper dip of ∼ 50 • N, and merge together as strand F4-F5, cut through U3 at a near-vertical angle and U2 to U0 with a ∼ 85 • S dip. This change of dip angle and direction is expressed within the shallowest units (U3 to U0) by an apparent normal-geometry fault displacement along F4-F5 (see Fig. 8b). The detailed log of the upper section shows a ∼ 3 m wide V-shaped deformation zone bounded by F4-F5 to the north and by a diffuse deformation band affecting U3 to U0 to the south (x = 38-41 m in Figs. 3b and 8). In between, units exhibit strong warping and chaotic limits suggesting soft-sediment deformation and collapse against F4-F5. Unit U1 is overlain with U0, which is itself collapsed against F4-F5. The amount of associated vertical displacement is difficult to ascertain, due to the wide collapse zone and the fact that U0 has been eroded north of F4-F5. From the base of the hanging wall section of U1 at grid point (37,18) to the base of the footwall section of U1 at grid point (38.5, 16.5), we estimate a minimum vertical offset of ∼ 1.5 m. Finally, the whole stratigraphic succession is sealed by a ∼ 1.5 m thick wedge-shaped colluvial unit (W1) deposited over U0 and against what we interpret as F4-F5 free face. The very continuous geometry of the topography across U1, W1, and U0 suggests some erosion took place after the deposition of W1, as can be expected in a monsoon-dominated climate. Hence, W1 may have originally been significantly thicker.

Timing of surface ruptures and associated coseismic displacements
In order to identify the various deposition, erosion, and deformation events recorded at the Piping site, we propose a schematic sequential retro-deformation combining all observations collected over the exposure ( Fig. 9; see Malik et al., 2017, for a similar approach further west). We start from a simplified log (Fig. 9a) and successively retro-deform the whole section to restore the most recent deposits to their original geometry and infer previous events where deformation remains. In parallel, we present OxCal-modeled (Bronk Ramsey, 2009) event dates constrained by 15 radiocarbon samples (see Sect. 3.1) and a chronostratigraphic model following guidelines from Lienkaemper and Bronk Ramsey (2009) (Fig. 10).
-Event 1 + Event 2: the most recent unit observed in the exposure is a ∼ 1.5 m thick colluvial wedge (W1 in Figs. 8 and 9a) deposited against a free face affecting unit U1 by faulting along F4-F5. The diffuse deformation observed within U3, U2, and U1 and the collapse of unit U0 within an open fissure are contemporaneous with an event that occurred after the deposition of U0 (Fig. 8). The removal of W1 and retro-deformation of units U0 to U3 restore the continuity of the bottom of U0 and leave large-scale folding affecting units U2 and older. Restoring these deposits to their original horizontal geometry (Fig. 9b) in agreement with the southern section of the exposure (Fig. 3b) involves (at least) bringing the highest observable point of unit U1 (erosion surface at grid point (26, 25.5) marked by the northern green star in Fig. 9b) down to the height of the U1 top observed in the undeformed section (e.g., grid point (50, 14) marked by the southern green star in Fig. 9b). This analysis yields a minimum vertical offset of 11.6 m ± 0.8 m along the 60 • ± 10 • north-dipping F4-F5 splay, which corresponds to 13.8 m ± 2.3 m of coseismic dip-slip. The amount of deformation accom-modated by faulting at the surface appears disproportionally small compared to folding at depth. This may be explained by efficient attenuation of dragging within soft sedimentary units and the emergence of a small localized surface offset. Alternately, this may suggest the occurrence of two distinct events; a recent faulting event with 1.5 m ± 0.5 m of coseismic dip-slip (called E1) and an older folding event with 12.2 m ± 2.8 m of coseismic dip-slip (called E2). For E2, the coseismic dip-slip is obtained by subtracting slip for E1 from the total slip then projecting the result onto the 60 • ± 10 •dipping rupture. Radiocarbon dating of W1 only yielded modern dates (Table 1) -likely due to contamination from actively developing soil -and does not permit us to date E1-E2 accurately. From our chronostratigraphic analysis (Fig. 10), said event(s) occurred after 895 CE and was/were associated with faulting along faults F4 and F5. Removing the now undeformed units U2 to U0 reveals that significant folding and faulting remain for units U3 and older (Fig. 9c).
-Event 3: by applying the same approach to units U4 and U3 and considering that the uppermost point of the top of unit U3 has been eroded away, we estimate the height difference between grid point (26, 25.5) and the height of the top of U3 in the undeformed section, to be 9.5 m (blue stars in Fig. 9c). This yields a minimum cumula-tive vertical offset along F3, F4, and F5 of 16 m ± 0.5 m for E3 + E2 + E1 and hence 4.4 m ± 1.3 m of vertical offset for E3 alone. Since slip propagated primarily along F3 with an average dip of 20 • ± 5 • , we estimate the coseismic dip-slip for E3 along F3 to be 14.7 m ± 7.4 m. U3 is the youngest affected unit, while U2 is the oldest unaffected unit, which indicates E3 occurred between the deposition of U3 and U2. Our radiocarbon chronology (Fig. 10) yields a date of occurrence at 300 ± 70 CE. Retro-deformation along F3, then removal of undeformed units U4 and U3, suggests residual deformation affects units U5 and older (Fig. 9d).
-Event 4: at this stage (Fig. 9e) Fig. 5b). This event is predated by the deposition of U5 and postdated by the deposition of U4, hence bracketed at 100±160 BCE (Fig. 10). This brings U5 to its original undeformed geometry forming a near-horizontal unit deposited against a pre-existing scarp formed in U6, as attested by the onlap termination visible at grid point (38,13) in Fig. 3a.
-Event 5: this event is documented by the remaining scarp affecting U6 once previous events are retrodeformed and U5 is removed (Fig. 9g). Although the height of this scarp is poorly constrained, the retrodeformation analysis suggests it is at least 2 m high and was produced by slip along a shallow-dipping rupture (∼ 10 • N), similar to F2 and F3 as observed at the base of the exposure (below z = 1 m). Hence, we propose that the amount of slip involved during E5 is similar to what is inferred for E4. Furthermore, since the event took place between the deposition of units U6 and U5, it may be dated back to 485 ± 125 BCE (Fig. 10).
A striking feature of surface deformation visible in the Piping exposure is the gradual change in fault dip over time. While all fault strands converge and dip 35-40 • N below grid point (30, 2) they diverge from ∼ 10 to ∼ 50 • N (locally 90 • ) as Figure 10. Chronostratigraphic model for deposition episodes (alluvial units U0 to U6 and colluvial wedge W1) and surface-rupturing events (E5 to E1) at the Piping exposure. The model is built from abutting relationships between stratigraphy and faulting and is constrained by 18 detrital charcoal samples and one inferred age corresponding to the top of unit U1. All resulting calendar dates are rounded to the nearest multiple of 5. they propagate to the south (Figs. 3 and 9), presenting a geometry similar to tri-shear folding (Allmendinger, 1998). In detail, the oldest event (E5) occurred while the top of unit U6 constituted the ground surface (i.e., the event horizon) and is expressed along a shallow 10 • north-dipping duplex rupture. The situation is similar for E4. After deposition of units U4 and U3 adds 2.5-3 m of sediments on top of the E5 rupture, the following event (E3) emerges higher in the stratigraphic section along F3 with a steeper dip of 25-30 • . A consequent deposition episode adds at least 8.5 m of sediments (units U2, U1, and U0) over these ruptures. The most recent event(s) (E2-E1) exhibit a much steeper rupture (along strands F4 and F5) with a dip reaching ∼ 50 • within unit U7 (coarse-grained terrace deposits) and 90 • as it emerges to the present-day surface through unit U0 (fine-grained fan deposits). It is a common observation both in the field and in analog experiments that ruptures along thrust faults tend to flatten as they reach the surface under the influence of decreasing lithostatic pressure (e.g., Philip and Meghraoui, 1983;Lee et (Nakata et al., 1998, Upreti et al., 2000; (L) Panijhora (Mishra et al., 2016); (M) Chalsa (Kumar et al., 2010); (N) Sarpang (Le Roux-Mallouf et al., 2016); (O) Nameri (Kumar et al., 2010); (P) Harmutti (Kumar et al., 2010). (b) Synoptic calendar and positions of great/large earthquakes along the Himalayan front (including instrumental, historical, and paleoseismic events). Orange horizontal bars approximate minimum source lengths with or without observed surface rupture. Vertical blue bars correspond to the radiocarbon-model constraints on the timing of the different events. Vertical brown bars correspond to ∼ 2600-year-long record deduced from the present study. al., 2001). We propose that the change in deformation style from nearly horizontal (E5 and E4) to steep (E2) and then vertical (E1) displayed in the Piping trench reflects increasing vertical load on the foot of the tectonic scarp associated with the progressive buildup of the Ramphu Chu fan against it.

Summary of recurrence times, magnitudes, and slip rate
Paleoseismic investigations conducted along the MFT at the confluence between the Wang Chu and the Ramphu Chu in western Bhutan show an important cumulative deformation zone including a rich chronology of deposition phases and deformation events for the last ∼ 2600 years.
Historically, the most recent earthquake to have provoked massive destruction in the region is the 1714 CE earthquake, previously described as the 1713 CE earthquake by Ambraseys and Jackson (2003) and identified in the paleoseismic record by Berthet et al. (2014) and Le  in the Sarpang area (∼ 50 km to the east; see Fig. 11). By combining historical and paleoseismic constraints, Hétényi et al. (2016) propose that this earthquake reached M w 7.5-8.5 with a modeled rupture centered on Bhutan and largely encompassing the Piping site. A possible event E1, though insufficiently documented by unfavorable sedimentation here, would be consistent in terms of coseismic slip and chronology, and we propose that it may correspond to the 1714 CE earthquake. Similarly, event E2 is consistent with an event observed at the Sarpang site as well, dated to 1344 ± 130 CE (Fig. 11) and tentatively associated with a medieval earthquake that may have ruptured a large section of the MFT (see Le Roux-Mallouf et al., 2016, and references therein). Hence, we propose that event E2 corresponds to that second event. Events E3, E4, and E5 occurred at 300 ± 70 CE, 100 ± 160 BCE, and 485 ± 125 BCE, respectively.
Hence, according to our retro-deformation analysis and chronostratigraphic model, our results allow constraining the occurrence of five surface-rupturing events between 485 ± 125 BCE and 1714 CE with an average recurrence interval of 550 ± 211 years. When only considering events with the largest documented coseismic slip values (E2 to E5) that are the most likely to be preserved and observed in exposures, the average recurrence interval reaches 610 ± 238 years. Our results are comparable to the lower values obtained for the late Holocene by Bollinger et al. (2014) in eastern Nepal (610 to 1220 years, depending on hypotheses). Furthermore, the relatively small coseismic slip value determined for E1 (and assigned to the 1714 CE earthquake) suggests smaller, though destructive, events may occur on occasion as was the case for the 2015 Gorkha earthquake in Central Nepal (e.g., Grandin et al., 2015), although there was no surface rupture associated with it.
The retro-deformation analysis also allows estimating associated dip-slip coseismic displacements with values ranging from 1.5 m ± 0.5 m for E1 to more than 12 m for E2, E3, E4, and probably E5, a value typical of the largest events documented along the Himalayas in Nepal, Sikkim, Bhutan, and Assam and consistent with extreme magnitudes on the order of M w 8.5-9 (Le Roux-Mallouf et al., 2016, and references therein). Considering the largest events, this represents 40.4 m ± 10.8 m of slip (E2 + E3 + E4) accrued over 1629 ± 255 years (between E5 and E2) at a rate of 24.9 ± 10.4 mm yr −1 . Although the duration of our dataset may be too limited to represent the long-term behavior of the MFT, this slip rate is consistent with those derived from 8 kyr old uplifted terraces in Sarpang ( Fig. 11; Berthet et al., 2014) and from far-field GPS shortening rate measurements (Marechal et al., 2016). Together, these results suggest that the Himalayan convergence is mainly seismically accommodated along the MFT in western Bhutan as well.

Conclusion
We presented here the longest continuous record of paleoearthquakes along the Himalayan Arc from the detailed study of an 18 m thick deformed sedimentary sequence dated from 17 radiocarbon samples. Well-expressed deformation and a detailed retro-deformation analysis reveal the occurrence of five surface-rupturing earthquakes along the MFT in southwestern Bhutan during the past ∼ 2600 years. The two most recent events can be related to the 1714 CE earthquake (Hétényi et al., 2016) and a medieval event (1344 ± 130 CE) already described in south-central Bhutan (Le . More strikingly, events E3, E4, and E5 are documented here for the first time and constitute some of the oldest paleo-earthquakes characterized in the central Himalayas (Fig. 11). Together, these events give an average earthquake recurrence interval of 550±211 years (or 610±238 years for the largest) for the Main Frontal Thrust in Bhutan.
The slip rate of 24.9 ± 10.4 mm yr −1 obtained from cumulative slip is consistent with both Holocene rates obtained from uplifted terraces (Berthet et al., 2014) and the high interseismic coupling level inferred from geodetic measurements (Marechal et al., 2016), which suggests that the Himalayan convergence in western Bhutan is mainly seismically accommodated along the MFT. Moreover, this result suggests that -at least locally -the slip budget does not display significant deficit over the time period of this study (Stevens and Avouac, 2016). Finally, estimated coseismic displacements between ∼ 1.5 m and at least 12 m indicate the likely occurrence of large (between M w ∼ 7.5 and M w ∼ 8.5) and great earthquakes (M w > 8.5) at a single site. This complexity should be taken into account in probabilistic seismic hazard calculations.
Data availability. Correspondence and requests for materials should be addressed to romain.lerouxmallouf@geolithe.com.
Author contributions. RLRM, MF, JFR, and PP conducted field work. RLRM and MF prepared the paper with contributions from all co-authors.
Competing interests. The authors declare that they have no conflict of interest.