The Pamir Frontal Thrust Fault: Holocene Full‐Segment Ruptures and Implications for Complex Segment Interactions in a Continental Collision Zone

The Pamir Frontal Thrust (PFT) of the Trans‐Alai Range in Central Asia is the principal active fault of the intracontinental convergence zone between the Pamir and Tien Shan. Its northward propagation is reflected by frequent seismic activity and ongoing crustal shortening. Recent and historic earthquakes exhibit complex rupture patterns within and across seismotectonic segments bounding the Trans Alai, challenging our understanding of fault interaction and seismogenic potential. We provide paleoseismic data from five trenches along the central PFT segment (cPFT) and interpret five and possibly six paleoearthquakes that have ruptured since ∼7 ka and 16 ka, respectively. Our results indicate that at least three major earthquakes ruptured the full‐segment length and possibly crossed segment boundaries with a recurrence interval of ∼1.9 kyr and potential magnitudes of up to Mw 7.4. We did not find evidence for great (i.e., Mw ≥8) earthquakes. However, discrepancies between slip and rupture extent during apparent partial segment ruptures in the western half of the cPFT, combined with significantly higher scarp offsets, indicate a more mature fault section with potential for future fault linkage. We estimate an average rate of horizontal motion for the cPFT of 4.1 ± 1.5 mm/yr during the past ∼5 kyr, which does not fully match the GNSS‐derived present‐day shortening rate of ∼10 mm/yr. This suggests a complex distribution of strain accumulation and potential slip partitioning between the cPFT and additional faults and folds within the Pamir that may be associated with a partially locked regional décollement.

. Earthquake epicenters since 1968 in red (M > 5.5) and white (M < 5.5) from the U.S. Geological Survey seismic catalog with mixed magnitude types (USGS Earthquakes); in blue for the period 2008-2014 from a regional network, indicating for instance the aftershock sequence from the 2008 M w 6.6 Nura earthquake . Focal mechanisms are from the Global Centroid-Moment-Tensor catalog (Dziewonski et al., 1981;Ekström et al., 2012), Fan et al. (1994) and Sippl et al. (2014). GNSS horizontal surface velocities are relative to stable Eurasia and are color coded according to (c) each segment is consistent with potential moment magnitude earthquakes of up to M w 7.0 during a full-segment thrust rupture. An asymmetric distribution of the vertical separation along offset fluvial terraces within the central segment of the PFT N (cPFT) (Arrowsmith & Strecker, 1999), rupture propagation between the cPFT and neighboring western transfer zone during the 1978 M w 6.6 Zaalai earthquake and its aftershock sequence (Nikonov et al., 1983), and widely distributed seismogenic deformation along the eastern segment and within the Trans-Alai Range during the 1974 M w 7.0 Markansu earthquake (Fan et al., 1994) suggest that mechanical fault interaction between the different segments of the PFT N has occurred in the recent past (Arrowsmith & Strecker, 1999). Combined with geological evidence of complex transfer zones characterized by a combination of strike-slip and dip-slip deformation (Arrowsmith & Strecker, 1999;Nikonov et al., 1983;Pavlis & Das, 2000;Strecker et al., 2003), the different interactions between fault segments may be caused by along-strike variations in structural maturity (e.g., Manighetti et al., 2007).
Another fundamental parameter for estimating fault activity is slip rate relative to the rate of strain accumulation spanning the entire fault zone (e.g., Anderson et al., 1996;Youngs & Coppersmith, 1985). Advances in GNSS (Global Navigation Satellite System) data quality and density provide improved estimates of interseismic motion and (largely) elastic strain accumulation over decadal timescales (e.g., Abdrakhmatov et al., 1996;Brooks et al., 2011;Ischuk et al., 2013;Thatcher, 2009;Weiss et al., 2016;Zubovich et al., 2010). GNSS profiles across the northern Pamir reveal a significant decrease in horizontal surface motion from the interior of the orogen towards areas north of the PFT N over a relatively short distance of ∼60 km ( Figure 1). This velocity gradient points towards the possibility of localized deformation along the leading edge of the Pamir and the potential for ruptures typical of megathrust events, as seen for instance across the Himalayan thrust front (e.g., Berger et al., 2004;Bilham, 2019;Bilham et al., 1997;Bollinger et al., 2014;Kumar et al., 2006;Lindsey et al., 2018;Mugnier et al., 2017). At first glance, however, the decadal shortening rates of ∼10 mm/yr inferred along the PFT (Ischuk et al., 2013;Metzger et al., 2020;Zubovich et al., 2010Zubovich et al., , 2016 are at odds with the estimated Holocene dip-slip rates of ∼2-6 mm/yr along the cPFT (Arrowsmith & Strecker, 1999;Burtman & Molnar, 1993;Nikonov et al., 1983).
In this study, we focus on the ∼35-km-long cPFT and report on five paleoseismological excavations that we conducted with the aim to develop a seismogenic history of this portion of the range front. Based on detailed geomorphic analysis, geochronological results, and paleoseismological observations, we (i) synthesize a segment-wide chronological framework of seismogenic faulting and (ii) evaluate the Holocene dip-slip rate along the cPFT. Despite ample geological and geodetical evidence for large-scale convergence, we did not find evidence for great M w ≥ 8.0 earthquakes. Instead, our results reveal at least three Holocene M w ≤ 7.4 earthquakes that probably ruptured the entire length of the cPFT. Furthermore, we found indicators for potential advanced fault maturity and complex segment interaction along the PFT N . Our estimated Holocene rate of horizontal motion for the cPFT does not equate to the decadal scale (i.e., GNSS-derived) shortening rate but rather suggests more widespread neotectonic activity across the PTS. Nikonov et al. (1983) pioneered earthquake-geology studies across the northern Pamir. They also identified voluminous mass-movement deposits in the Alai piedmont, which have been analyzed in more recent studies (Reznichenko et al., 2017;Robinson et al., 2015), and inferred that these deposits were related to seismogenically triggered collapse of parts of the Trans Alai mountain front. Based on a compilation of 50 paleo-earthquake-related indicators, Nikonov (1988) estimated a possible maximum magnitude of ∼M7.6 and an average recurrence time of ∼1.4 kyr for major earthquakes along the PFT.

Earthquake Geology
In subsequent studies on the structural and tectono-geomorphic evolution of the Pamir mountain-front, Arrowsmith & Strecker (1999) and Strecker et al. (1995Strecker et al. ( , 2003 built on these observations and used Nikonov et al.'s (1983) stratigraphic framework for different Quaternary fluvial terrace sequences (Qt 1-4 ) to suggest that the Trans-Alai Range front comprises three seismotectonic segments-western, central, eastern-that are inferred to be kinematically linked by dextral transfer faults and oblique thrust faults with semiindependent activity (Figure 1). This is largely based on the distribution and deformation of the prominent Qt 3 terrace (old nomenclature fgQ 3 /fgQIII; Nikonov et al., 1983). Qt 3 is generally preserved in the hanging wall of the central PFT (cPFT) and extends for several km into high-elevation sectors of the Trans-Alai Range ( Figure 2). The fluvial terrace deposits reflect a former extensive braided-stream environment in the mountain valleys that transitioned into alluvial fans in the piedmont; the deposits constituting these terrace surfaces are in erosional or conformable depositional contact over Qm 2 moraines and join the terminal Qm 3 moraines that presumably formed during the penultimate, Early Holocene (∼10 ka) glaciation in this region (Arrowsmith & Strecker, 1999;Strecker et al., 2003). The formation of Qt 3 was followed by faulting and range uplift combined with regional incision (Arrowsmith & Strecker, 1999). Occasional aggradation or lateral erosion allowed the formation of the lower and younger terrace Qt 4 (fgQ 4 /fgQIV; Nikonov et al. (1983). Arrowsmith & Strecker (1999) report radiocarbon ages for the Qt 3 surface from the Ylaisu (∼7.2 ka) and Syrinadjar (∼6.4 ka) River terraces ( Figure 2). Combining the Qt 3 age with vertical cumulative offsets of 18 m at the Syrinadjar River site along the fault trace of the cPFT resulted in a minimum Holocene slip rate of up to 6 mm/yr (Arrowsmith & Strecker, 1999;Burtman & Molnar, 1993;Nikonov et al., 1983). In contrast, based on restorations of regional cross sections from seismic reflection data and field observations, Coutand et al. (2002) estimated long-term horizontal shortening rates in the Alai Valley to be less than <1 mm/yr spanning the Neogene (ca. < 25 Ma).

Deformation and Seismic Activity Along the PTS of the Alai Valley
Over a relatively short north-south distance of ∼60 km across the PTS (∼39°10'-39°50'N) (Figure 1), GNSS measurements relative to stable Eurasia (i.e., in an Eurasia-fixed reference frame) document a decrease in present-day NNW-SSE-oriented motion from ∼25 mm/yr in the high sectors of the Pamir to ∼10 mm/yr north of the Alai Valley (Ischuk et al., 2013;Metzger et al., 2020;Mohadjer et al., 2010;Reigber et al., 2001;Zubovich et al., 2010Zubovich et al., , 2016Yang et al., 2008). Thus, the northern front of the Pamir appears to accommodate more than one-third of the ∼30-40 mm/yr of total shortening between India and Eurasia (Argus et al., 2010;Dal Zilio et al., 2019;Molnar & Stock, 2009). In contrast to surrounding areas to the north and south, this elevated strain is reflected in focused crustal seismicity (e.g., Fan et al., 1994;Schurr et al., 2014;Sippl, Schurr, Tympel, et al., 2013;Sippl et al., 2014;Teshebaeva et al., 2014). The present-day high level of seismicity associated with the PTS is concentrated in a narrow, ∼15 to 25-km-wide belt north of the MPT, with hypocentral depths in the uppermost 25 km of the crust and along the tectonically active, south-dipping faults of the PFT (Figure 1) (Fan et al., 1994;Schneider et al., 2013;Schurr et al., 2014;Sippl, Schurr, Tympel, et al., 2013;. Deformation along the MPT and PFT is dominated by thrust faults with strikes varying between N080° in the west to N060° in the east, and local strike-slip faulting (Figures 1 and 2) (Arrowsmith & Strecker, 1999;Coutand et al., 2002;Schurr et al., 2014;Sobel et al., 2013;Strecker et al., 1995). The south-dipping MPT represents the former frontal thrust that initiated during the Late Oligocene to Early Miocene, presumably with coeval northward propagation towards the southern Tien Shan. Out-of-sequence thrusting continued along the MPT throughout the Miocene (Coutand et al., 2002), whereas the former northern faults seem to have been abandoned and were progressively covered by the Alai basin fill. Presently, the MPT appears to be inactive (Cao et al., 2013;Sobel et al., 2013), with low Quaternary shortening rates of <1 mm/yr reported from the Chinese sector of the Pamir (Sobel et al., 2011;Thompson et al., 2015). Hence, the present-day regional convergence seems to be accommodated along the PFT, which initiated during the Late Miocene to Early Pliocene (Fu et al., 2010;Sobel et al., 2013;Thompson et al., 2015). The level of current tectonic activity of the PFT S , however, also appears to be rather low (Arrowsmith & Strecker, 1999;Nikonov et al., 1983). In contrast, the PFT N , which is presumably the youngest manifestation of the northward propagation of the PFT in the southern Alai Valley, is characterized by a sharply defined fault trace with pronounced tectonic scarps reaching heights of ∼15 m.
In this region, a number of earthquakes with M w ≥ 6.0 have been recorded: the 1974 M w 7.0 Markansu, the 1978 M w 6.6 Zaalai, the 2008 M w 6.6 Nura, and the 2016 M w 6.4 Sary Tash events ( Figure 1). The earthquake focal mechanisms indicate oblique reverse, dextral-slip and thrust-dominated faulting with depths ranging from 2 to 17 km (Ekström et al., 2012;Fan et al., 1994;Sippl et al., 2014). Despite the large-scale convergence and impressive heights of the fault scarps in Quaternary deposits along the Pamir mountain front, no instrumental or historical records exist for events >M7.0 in this region.

Remote Sensing-Based Geologic Mapping of Fault Zones and Quaternary Geology
We used the TanDEM-X 12-m digital elevation model (DEM) with a vertical accuracy of ∼2 m (Krieger et al., 2007;Rizzoli et al., 2017) combined with the high-resolution World Imagery basemap in ArcMap TM to update the mapping for fault traces along the PFT in the southern Alai Valley and Trans-Alai Range. In addition, we combined field-based observations with mapping from previous publications (Arrowsmith & Strecker, 1999;Coutand et al., 2002;Geologic map of the Kyrgyze Republic, 1964a, 1964bNikonov et al., 1983;Strecker et al., 2003) to prepare an updated version of the geologic map along the Trans Alai mountain front emphasizing the Quaternary geology and active faults ( Figure 2). We extracted eighteen scarp-perpendicular topographic profiles from the TanDEM-X data along the cPFT for along-strike fault-scarp analysis. We followed the method of Walker et al. (2015), to determine upper and lower fan-surface slopes, and to compute a mean vertical separation estimate using linear regression combined with Monte Carlo sampling (for references see Text S1 in Supporting Information S1).

Digital Surface Models
Digital surface models (DSM) were generated from each study sites by acquiring low-altitude (between 35-120 m above ground level) aerial photographs during field campaigns using an unmanned aerial vehicle (UAV). To reduce the positional error of the resulting models, ground-control points (GCP) were placed and measured using kinematic differential GNSS (dGNSS) system. All positions were recorded in absolute World Geodetic System 1984 (WGS 84) based on the Universal Transverse Mercator (UTM) grid zone 43N coordinate system 10.1029/2021JB022405 7 of 44 (EPSG:32643). The collected raw data were then imported and translated using LEICA Geo Office. We used the photogrammetric modeling software Agisoft PhotoScan Pro (v.1.2.1 and v.1.4.3) to generate the DSMs by performing image-based modeling by feature recognition and photo alignment using Structure-from-Motion (SfM) algorithm (e.g., Bemis et al., 2014;Johnson et al., 2014). The final DSMs were produced with the PhotoScan DEM generation tool, resulting in a 4-10-cm per pixel resolution. The Geospatial Data Abstraction Library (GDAL) was used to generate shaded relief, aspect and slope maps of the DSM, which were then visualized and interpreted using GIS software.

Paleoseismic Trenching
The remote mapping and preliminary field surveys helped guide our selection of the five paleoseismic excavation sites for this study. The workflow after excavation included cleaning the trench walls with scrapers and brushes, construction of a 1 x 1 m grid, photographing the walls for reproduction of image-based seamless ortho-rectified high-resolution photomosaics (see Reitman et al., 2015;Patyniak et al., 2017), on-site interpretation of stratigraphic units, logging of earthquake-related structures, and sampling for geochronology. We also included data from an earlier, unpublished paleoseismological study (Arrowsmith et al., unpub. data, 1999) at Komansu, which included photomosaics from analog photographs, drafted and interpreted trench logs, and geochronology results. We revisited the Komansu site in 2017 and compiled additional data.

Geochronology
For radiocarbon dating, we collected single specimens of macroscopic charcoal and terrestrial gastropods. For locations where these materials were not present, we collected bulk-soil samples from dark organic-rich soils. In general, the bulk samples provide a minimum depositional age for a given unit, assuming that soil development and charcoal incorporation postdate the deposition (e.g., Bennett et al., 2018). All samples were dated with low-energy 14 C accelerator mass spectrometry at the Poznan Radiocarbon Laboratory (Poland) and the University of California Irvine Keck-CCAMS facility (Table 1). All radiocarbon ages were individually calibrated using the analysis software OxCal v.4.4 and calibration curve IntCal 20 (Reimer et al., 2020).
Homogenous predominantly silty sand units were sampled for infrared simulated luminescence dating (IRSL) by hammering aluminum tubes with a diameter of 5 cm (or 2-cm thick PVC in 1999) horizontally into cleaned, vertical trench sections; these were retrieved and light-tight closed. Heterogeneous units with coarser grain sizes were sampled at night with a low-intensity red LED headlamp into double-bagged 4 mil black conductive bags. Additional material from the surrounding area was sampled for dose-rate estimation through low-level gamma-spectrometry. Samples collected in 1999 were processed in the same year by S. Forman at the Luminescence Dating Research Laboratory at the University of Illinois, Chicago. These luminescence analyses were completed on the 4-11 µm poly-mineral fraction. All samples collected after 1999 were processed at the Institute of Earth and Environmental Science (University of Freiburg, Germany). For IRSL processing details, see Text S2 in Supporting Information S1. Similar to other studies in active mountain belts (Preusser et al., 2006), the quartz grains investigated here have a low luminescence sensitivity. Consequently, feldspars were measured using the single aliquot regenerative dose protocol (Reimann & Tsukamoto, 2012). In storage tests, we observed little if there is any signal loss (<1 g / decade), which imply the absence of fading in the samples investigated. The results are summarized in Table 2. For IRSL and post-IR IRSL (pIR), we observe in several cases a significant offset of the ages determined by the two approaches, with pIR being up to three times older than IRSL. We interpret this as reflecting partial bleaching (see Discussion) of the pIR signal (Gray et al., 2015) and rely exclusively on the IRSL ages for our interpretations.
All sample locations are noted in the trench log. To model probability-density functions (PDF) of each earthquake age, we produced age models using the resulting 14 C and IRSL dates and stratigraphic constraints (see below) following the OxCal modeling approach (Bronk Ramsey, 2008).

Earthquake Evidence and Age-Quality Ranking
To indicate our confidence regarding each interpreted surface-rupturing paleoearthquake including the age, all observations within the exposed trench walls were assigned a quality ranking, following the approach of Scharer    (2007). The quality scheme ranges from 1 to 4 (best) and is based on an evaluation of strength of indication of paleoearthquake evidence of the observed features. A summary of all event indicators and associated earthquakes is listed in Table 3. Finally, we assigned an additional ranking to the OxCal-modeled earthquake age by considering (1) the completeness of bracketing of the earthquake-event horizon, and (2) the uncertainty of the resulting modeled age of the earthquake (Table 4). These ratings are used here to systematically and qualitatively assess the results to build our final interpretation; in the future, they might be useful for additional quantitative analysis (e.g., correlation models, etc.).

Excavation Sites
Our paleoseismological investigations focused on profiling the fault-zone topography and excavating trenches at five sites along the cPFT ( Figure 2). The area of investigation covers ∼25 km of the total ∼35-km-length of the central segment. From west to east, we excavated paleoseismological trenches at Achyk-Suu (T1 and T2; opened 2017), Ylaisu (T3; opened 2018), Komansu (T4; opened 1999, revisited in 2017) and Tashkungey (T5; opened 2018). The sites are separated from one another by 1, 2, 10 and 12 km, respectively. The site names correspond to the principal rivers adjacent to the excavation sites ( Figure 2).
The trace of the cPFT defines a nearly continuous, north-facing N080° striking fault scarp with maximum slopes between 15° and 20° and heights of up to ∼15 m cutting through the fluvial and alluvial surfaces. The fault-scarp morphology includes a combination of hanging-wall collapse and pressure-ridge features implying gentle thrust fault dip-angles (<30°) associated with unconsolidated friable to clayey surface material and moderate dip-slip displacements (Philip et al., 1992). The uplifted alluvial-fan and fluvial terrace surfaces along the range front generally have gentle north-facing slopes with dips ranging from ∼0° to ∼4°, incised by northward-directed river channels, and partly covered by glacial till and landslide deposits (Figures 2 and 3). The vegetation is characterized by grass cover, with root horizons 0.5-1.0 m deep. Following the Quaternary geological interpretation and regional correlation by Strecker et al. (2003) (see above), the fluvial terraces at the trenching sites correspond to the prominent Qt 3 and Qt 4 terrace deposits (Figures 2 and 3).

Description of Stratigraphy and Structural Context Exposed in the Excavations
The unfaulted alluvial-fan stratigraphy is consistent along the cPFT ( Figure 3). The sequence starts with a massive unit visible in the lower 1-3 m of the trench exposures, consisting of laterally discontinuously bedded and subrounded to rounded poorly sorted fluvial gravel (unit 1). At the T1, T2, T4 and T5 locations, and within the upper 20 to 50 cm of unit 1, a well-developed horizon of CaCO 3 -coated gravel indicates a stage I to II soil-development (unit 1b; associated with the Qt 3 surface formation) ( Figure 4). The fluvial-gravel beds (unit 1) generally dip gently north due to the northward transport direction of the regional depositional system; however, at the fault zones the beds of unit 1 as well as the unit 1/1b transition are folded parallel to the scarp. Unit 1 is capped by unit 2, a compacted layer with moderate thickness (10-50 cm) of fine-grained sediments capping terrace Qt 3 . We infer that a change in sedimentation between unit 1 and unit 2 marks the timing of initial floodplain abandonment, regrading, and incision below the Qt 3 surface associated with melting of the Qm 3 glaciers. Thus, the contact between gravel and capping material dates or postdates terrace formation and will be used as a geomorphic marker. The topsoil layer is a thin dark and organic-rich A horizon in the upper 10 to 30 cm below the surface.
The excavations expose multiple fault zones, which are represented by 15º to 30° south-dipping narrow (5-20-cmwide) zones with oriented and broken clasts or wider (up to 1-m-wide) zones with an imbricate fabric of fault   Location according to trench log documentation in Figure 5. b Ranking quality: 1 = speculative; 2 = indistinct; 3 = slightly visible; 4 = distinct. c Summarized ranking of all indicators associated with one site earthquake.
splays and injected material between bodies of fault breccia ( Figure 5). The principal indicators of ground-rupturing events are single or stacked deformed colluvial deposits adjacent to the fault zone. The colluvial material accumulates in a lens-shaped drape over the lower portion of the scarp and is usually thickest at the toe where the slope change is greatest, comparable to the modern topsoil (TS) (e.g., T2 and T5 in Figure 5). We infer that these colluvial deposits formed as a result of coseismic or immediate postseismic collapse of the overhanging scarp tip and subsequent debris-dominated and wash-dominated fault-scarp degradation (see Discussion), which buried the fault tip (e.g., McCalpin & Carver, 2009;Meghraoui et al., 1988;Patyniak et al., 2017;Rockwell et al., 2014). In a subsequent surface rupture, the lens is cut as the rupture drives the hanging wall through and over the deposit and ground surface. The resulting remnant has a wedge shape but is not to be confused with postrupture "colluvial-wedge" formation seen for instance in normal faulting settings (e.g., McCalpin, 2009). The scarp-face colluvium consists of poorly-sorted and reworked, matrix-supported deposits of silty sand that vary in color. In some cases, these strata contain gravel or consist of poorly sorted, clast-supported gravel. The colluvial material is often slivered into the fault zone. Secondary indicators of faulting include (1) nearly vertical fissures in the gravel deposits, indicated by pronounced CaCO 3 cementation along the fissure trace; (2) fault-induced material deposition against and in front of the scarp face inferred to have been deposited during temporary ponding; (3) offset markers from distinct unit boundaries or internal bedding; (4) deformed channel lenses interbedded within the fluvial deposits, (5) patches of injected sediments along and/or within the fault zone and (6)

Paleoearthquake Chronologies From Five Excavation Sites Along the cPFT
For each excavation site (Figure 4), we provide trench logs and associated photomosaics ( Figure 5) to display the stratigraphic and structural interpretations we used to infer individual surface-rupturing events ( Figure 6). In Figures S3-S7 in Supporting Information S1, we provide high-resolution photomosaics and detailed stratigraphic description for each site. Because of the stratigraphic and age consistency between unit 1 and unit 2 along the cPFT (see 4.2) and in order to provide a more consistent basis for subsequent modeling of earthquake timing, we will use the estimated mean ages of 20.1 ± 0.6 ka and 7.3 ± 0.1 ka, respectively. After a careful reevaluation of the stratigraphic and chronologic context in our multifault trench exposures, we assume coeval activation of these fault zones during some surface rupturing events. In the following section, we summarize all interpreted earthquake evidence (Table 3) along the different fault zones to define event horizons for each earthquake per site ( Figure 6). Combined with the available age control (Tables 1 and 2), we constrain the event timing and provide an event catalogue ( Table 4) that we later use for paleoearthquake correlation along the cPFT. Additional OxCal modeling information can be found in the Supporting Information (Codes S1-S5 and Figures S8-S12 in Supporting Information S1).

T1 -Achyk-Suu Site (39.48°N 72.50°E)
The Achyk-Suu excavation site is situated at the westernmost end of the cPFT ( Figure 2). Trench T1 is part of an abandoned gravel pit that was excavated in the faulted alluvial-fan deposits close to the village of Achyk-Suu (Figures 3 and 4). The trench consisted of a 7-m-long and up to 3-m-deep east wall across the lower portion of the fault scarp in the eastern part of the pit. The scarp fit-modeling of the 600-m-long topographic profile next to the trench indicates 10.2 ± 0.2 m of vertical separation, with an upper and lower alluvial-fan surface slope of ∼3° (P3; Figure 3 and Figure S1 in Supporting Information S1). The hanging wall in the exposed trench wall reveals a ∼3 to 4-m-thick unit 1 with ∼25° north-dipping bedding of the fluvial gravel covered by a thin topsoil (TS) layer ( Figure 5). We do not observe unit 2 in this location. However, farther south along the walls of the quarry in the hanging wall, unit 2 is present where the slope of the scarp flattens out again (Figure 4 and Figure S2 in Supporting Information S1). Hence, we infer that along the scarp face unit 2 was completely eroded through time. The footwall consists mainly of a succession of fine-grained sandy deposits. The boundary between units 1 and 2 is not exposed. The lowermost unit 3 is made up of grayish, very fine sandy silt and contains a few patches of fine gravel. Unit 3 is overlain by material of a similar but more heterogeneous composition that is less compacted with a higher clast content (C1) than unit 3. The change from unit 3 to C1 is also marked by a sharp change in color, from grayish to reddish brown. C1 gradually changes color upwards to beige, of softer, homogeneous fine silty sand and fewer clasts (unit 4). The subsequent package C2 can be separated from unit 4 by a clear boundary of a thin basal layer of fine gravel, a distinct change in color from beige to reddish beige, and increased clast content. C2 is covered by gravitationally collapsed and reworked unit 1 material (C3a) that is overlain by a block of reddish silty sand (C3b). Finally, a lens of more organic paleosol material lies between unit C3b and the overlying, less organic-rich TS. For a more detailed lithological description see Figure Figure S3 in Supporting Information S1. The main fault F1 is a wide zone (∼20 cm) of chaotically oriented clasts, with a strongly undulating lower contact. The strata of unit 1 are locally disturbed by deformation along fault F2, expressed as a narrow zone of imbricated clast fabric. The upper tip of a back thrust fault Fb is exposed in the northern portion of the trench. We assume the steep dip of the bedding of unit 1 presumably results from cumulative faulting and folding along F1, a common effect of thrust faulting along shallow faults (<30°) (e.g., McCalpin & Carver, 2009; also compare with following trench logs). We interpret that C1 is scarp-derived colluvium deposited interseismically after a surface rupturing event that overthrust parts of units 1 and 2 over unit 3. The part of C1 proximal to the fault was presumably deformed and dragged towards the upper boundary of unit 4 during subsequent faulting, leading to the deposition of more mixed colluvial material during the following interseismic period (C2). However, the evidence for this interpretation is not very strong, making our ability to distinguish this event debatable (see Discussion). During the penultimate event, upthrust strata of unit 1 partially collapsed (C3a) and were displaced downhill over C2. We suggest that the overlying material of C3b was originally deposited during the subsequent interseismic period at the scarp toe on top of C3a and later pushed and compacted against C3a during backthrusting along Fb, forming a convex contact between C3a and C3b.

T1 EQ Five (A5):
The evidence for the oldest rupture observed in T1 is along fault zone F2. F2 terminates at unit 1b and an unfaulted sand lens, indicating that this fault has not been active since the final phase of the fluvial gravel deposition. To bracket the timing for A5, we used the age of unit 1 (20.1 ± 0.6 ka) and the capping unit 2 (7.3 ± 0.1 ka) as maximum and minimum age constraints, respectively. Due to the large age gap between both units, A5 has a modeled age of 14.2 ± 2.6 ka.

T1 EQ Four (A4):
The second oldest event ruptured along F1 and is associated with unit C1. We were not able to obtain absolute age control for paleoearthquake A4 and can only state that it is younger than A5. However, OxCal allows us to determine relative time constraints by modeling empty uniform phases based on interpreted  Table 3 for reference and summary of stratigraphic and geomorphologic indicators for all interpreted paleoearthquakes. Radiocarbon ( 14 C) and infra-red stimulated luminescence (IRSL) ages shown in thousands of years (mean ±2σ) correspond to Table 1 and 2, respectively. See Figures S3-S7 in Supporting Information S1 for high-resolution photomosaics, trench logs and per-event displacement measurements of the trench walls complete with detailed geological unit descriptions.
stratigraphic order (see Code S1). Thus, the age for this paleoearthquake is modeled using relative ages of the stratigraphic sequence of units 3, C1 and 4. For broader bracketing, we used the age of unit 2 (7.3 ± 0.1 ka) as a maximum age (assuming that unit 3 formed subsequently) and the age of unit C2 (see below) capping unit 4 as a minimum age. Consequently, A4 has a modeled age of 4.8 ± 1.0 ka.
T1 EQ Three (A3): Event A3 is associated with unit C2. We use two ages (L3: 2.03 ± 0.20 and R2: 1.78 ± 0.04 ka) obtained from the upper part of C2 as minimum bracketing, assuming these represent the deposition of C2 in the interseismic period. Combined with the relative timing associated with A4, A3 has a modeled age of 2.4 ± 0.5 ka.  Figure 6. Summary of the structural and stratigraphic results, including geochronologic control from trenches T1 to T5 (a-e). Colors correspond to lithostratigraphic units in the trench logs ( Figure 5). All collected radiocarbon ( 14 C) and infra-red stimulated luminescence (IRSL) samples are included in the scheme; gray shaded samples are excluded from the OxCal models because of inconsistency in the stratigraphic order and/or disagreement with surrounding samples (see Section 4.3 for explanation). All modeled unit ages are mean ±2σ (Table 1, 2). See Codes S1-S5 and Figures S8-S12 in Supporting Information S1 for the corresponding OxCal scripts and associated OxCal model as PDF multiplots.

T1 EQ Two (A2):
The penultimate event is associated with C3a. The timing of A2 is tightly constrained by the age of the underlying unit C2 and the age constraints from the overlying unit C3b (L5: 1.72 ± 0.14; Rb2: 1.48 ± 0.04). This bracketing results in a model age for A2 of 1.6 ± 0.1 ka.

T1 EQ One (Ab1; MRE):
The most recent event Ab1 ruptured along F1 without breaking the surface and along the back-thrust fault Fb. The timing for the MRE is well constrained by the age of unit C3b as maximum age and radiocarbon ages from the overlying paleosol (R1: 0.21 ± 0.10; Rb1: 0.14 ± 0.08 ka) as minimum bracketing. Combined, the bracketing yields a model age for the MRE Ab1 of 0.8 ± 0.4 ka.

T2 -Achyk-Suu Site (39.48°N 72.51°E)
Trench T2 at the Achyk-Suu site was excavated about 1 km east of trench T1 (Figure 2), across a part of the fault with a double scarp morphology (Figures 3 and 4). Conditions along the single fault scarp with heights of 11 m and steep scarp slopes (i.e., P4; Figure S1 in Supporting Information S1) made an excavation logistically challenging. The scarp at our location of choice appeared unaltered and likely recorded the same faulting events as the single scarp ( Figure S2 in Supporting Information S1). The excavation resulted in a 24-m-long, 3-m-wide, and up to 4-m-deep trench. The scarp-fit modeling of the 250-m-long, topographic profile results in a total vertical separation of 5.2 ± 0.7 m, and an upper and lower fan surface slope of ∼9° and ∼0°, respectively (P5; Figure 3 and Figure S1 in Supporting Information S1). For fault zone location, see trench log Figure 5 for each trench, respectively. b Summarized ranking of stratigraphic evidence per event from Table 3. c Normalized modeled earthquake timing quality: based on bracketing and resulting timing uncertainty 1σ; 0 = poor, 1 = moderate, 2 = good, 3 = excellent. d Closed interval preceding the earthquake. e Offset = one event dip-slip offset; cd = colluvial deposit; uncertainty assumed to be 15%. f For cd, twice the mean max-thickness of colluvial deposit. The major part of the exposed stratigraphy consists of fluvial gravel (unit 1; max. thickness 3 m) overlain by unit 2 (∼50 cm). In total, the exposed trench walls revealed four separate fault zones (F1, F2, F3, F4), where fault zones F1 and F4 coincide with the surface traces ( Figure 5). In the footwall, unit 2 is covered by a layer of fine silty sand with internal lamination and varying color between gray and reddish beige (unit 3), which we interpret as pond-related sediments. The character of these sediments is compatible with modern laminated and reddish to beige silts and fine sands we observed in the footwall ( Figure S2 in Supporting Information S1). Unit 3 sediments are covered by a more compacted, slightly reworked reddish clayey fine sand (unit 4) with homogenously distributed CaCO 3 nodules in their upper part. We interpret that this unit developed during a protracted interseismic period with soil formation and assume its upper boundary to be a paleosurface. Units 3 and 4 are only present in the footwall. Unit 4 is overlain by two sheets of inferred colluvial origin (C2, C3) that can be distinguished from each other and the lower units by a clear change in color and composition. In the intermediate sectors of the hanging wall, unit 2 is also overlain by a succession of colluvial sheets (Ca, Cb, Cc). Each can be distinguished from each other by small wedges of injected clast material and a thin trace of clasts at their lower boundary emerging proximal to F4. Ca was only poorly exposed in the east wall. The surface soil (TS) constitutes a layer of mixed sediments from gravity-, debris-and/or wash-dominated degradation or fluvial transport from the adjacent foothills SSE of the trench and is covered by a thin soil A horizon (∼10 cm). For a detailed lithological description, see Figure S4 in Supporting Information S1.
F1 exhibits a complex structure with several subparallel, internal fault strands (F1.1, F1.2 and F1.3; Figure 5), which indicates upslope fault splay migration due to cumulative faulting. Within the fault zone interior, we were able to recognize parts of units 2 and 3 that were dragged upwards retaining their stratigraphic boundary. The laminated material of unit 3 between F1.1 and F1.2 is overlain by a small, clearly distinguishable wedge-like unit of beige-colored reworked fine material (C1) covered by reddish fine material, presumably from unit 4. The termination of fault strands F1.1 and F1.2 is covered by C2, we therefore conclude that C3 must be related to an event other than C2. The width of the F1 zone increases upward and is made up of more homogenous beige-colored fine material between F1.2-F1.3, which we infer to be unit 2 material that got liquefied and injected into the fault zone during repeated faulting (e.g., Weber & Cotton, 1981). Faults F2 and F3 cut through the intermediate hanging wall and were active only once, but presumably not related to the same surface-rupturing event. Although F2 can be unambiguously identified in the west wall by a pronounced trace of rotated clasts, we were not able to clearly distinguish F2 in the east wall. Fault zone F3 cuts through units 1 and 2 upthrusts parts of unit 1 above unit 2. Above the upper termination of F3, unit 2 is thicker, which strongly indicates degradation processes that reworked unit 2 material across the small scarp-free face. The undulatory nature of the paleosurface suggests that remnants of a paleo-scarp were covered by subsequent colluvial strata Cb and Cc, which formed after faulting along F4. Similar to F1, F4 exhibits internal deformation, fault splay migration and liquefied fine materials injected between fault-breccia bodies. For instance, we could observe patches of clasts that presumably belonged to the former injected material of unit 1 proximal to the fault that was deformed during subsequent faulting, indicating multiple fault activation.

T2 EQ Five (AS5):
The oldest evidence for a paleoearthquake retrieved from the stratigraphic relationships in T2 is associated with fault F2. We infer, similar to earthquake A5 (trench T1), that this fault was active during the final phase of unit 1 deposition and before the formation of unit 2. Thus, we bracket the timing for AS5 with the age of unit 1 (20.1 ± 0.6 ka) and of unit 2 (7.3 ± 0.1 ka). The bracketing leads to a modeled age for AS5 of 14.3 ± 2.9 ka.
T2 EQ Four (AS4): From the stratigraphical and chronological context, we infer that AS4 is associated with faulting along three faults (F1, F3 and F4). We assume that faulting along F1 cut through unit 2 and caused the formation of a ponded drainage and deposition of unit 3 at the toe of the lowest scarp related to initial scarp formation. Less strong faulting along F3 and faulting along F4 is associated with Ca also cut into and/or through unit 2. Following these observations, we assume that F3 and F4 were coeval, with F1 during this event. To bracket the age of this event, we used the age of unit 2 (7.3 ± 0.1 ka) to define the maximum age constraint. For the minimum age constraint, we used two radiocarbon samples from the base of unit 3 (R3e: 3.77 ± 0.06 ka; R1e: 3.85 ± 0.06). AS4 can be modeled at 4.7 ± 1.0 ka.
T2 EQ Three (AS3): Subsequent faulting along F4, cut through, and deformed the upslope part of Ca. Subsequent diffusion processes deposited colluvial package Cb. Unit C1 serves as a poor indicator for coeval fault activation along F1.2 and we infer that its upthrusting during subsequent events that ruptured along F1.1 toward its current position (see below). Following this indicator and the visible contact with unit 4, we suggest that unit 4 was deposited after AS3. Consequently, as a maximum age estimate, we use a radiocarbon age (R1e: 3.85 ± 0.06 ka) sampled from below C1 that agrees with the age of unit 3, and three radiocarbon ages (R7e: 1.78 ± 0.05 ka; R6e: 1.38 ± 0.04 ka) from unit 4 as a minimum age constraint. Combined, AS3 is modeled at 3.4 ± 0.4 ka.
T2 EQ Two (AS2): Considering our age control from C2 (R2w: 1.12 ± 0.05 ka; R1w: 1.06 ± 0.06 ka; Rb1w: 1.04 ± 0.05 ka; R8e: 1.38 ± 0.05 ka); Cc (R10e: 1.14 ± 0.04 ka; R9e: 0.8 ± 0.1 ka), we suggest that these colluvial strata are related to one surface rupturing earthquake that reactivated both F1 and F4, respectively. Sample R8e, however, is inconsistent and it may represent reworked material from unit 4 that was displaced towards the upper area of C2 during subsequent faulting along F1.3 (MRE; AS1), where the leading edge of the fault deformed the parts of C2 proximal to the fault during bending subhorizontally. Combined, four radiocarbon samples from C2 and Cc are used as minimum age constraints and the age of the underlying unit 4 (R7e and R6e) serves as a maximum age constraint. Together, AS2 is modeled at 1.2 ± 0.1 ka.

T2 EQ One (AS1, MRE):
The most recent event along F1 is linked to the uppermost massive colluvial strata of C3. C3 covers the termination of fault strand F1.2 and F1.3; we therefore conclude that both were activated simultaneously with F1.1 during AS1. Motion along them upthrust older units (unit 2 and C1) and cut through C2 along F1.3. We suggest that this event activated F4 at the same time, however, the faulting probably did not rupture the surface. Instead, faulting cut into the lower part of Cc and shoved material along the fault, which thickened the southern part of Cc. It also appears that this fault motion steepened the paleo-scarp, which allowed the deposition of slumped material at its toe (TS). We obtained two radiocarbon ages from C3 (R5w: 1.06 ± 0.06 ka; R6w: 1.14 ± 0.05 ka); however, these ages coincide with the ages from the underlying unit C2. We suggest that these ages probably rather represent an inherited age from reworked unit C2 material that was later recycled into C3. Following this interpretation, we use the age of C2 and Cc (R10e, R9e) as maximum age constraint and five radiocarbon samples (Rb2w: 0.49 ± 0.03 ka; R7w: 0.38 ± 0.05 ka; R4e: 0.11 ± 0.08 ka; Rb4e: 0.15 ± 0.09 ka; R5e: 0.19 ± 0.10 ka) representing the age of the topsoil (TS) as broader minimum age constraint. Using all of this information, we were able to model AS1 at 0.8 ± 0.2 ka.

T3 -Ylaisu Site (39.48°N 72.52°E)
During our preliminary geomorphic assessment of potential trenching locations, we discovered a trench cut into the fault scarp at the Ylaisu site ( Figure 2) that had been excavated for agricultural purposes. The trench position was perpendicular to the fault scarp, and the exposed walls were straight and undamaged when we decided to clean and to log this exposures ( Figure S2 in Supporting Information S1). The 11-m-long, 3.5-m-wide and up to 3-m-deep trench is located on a terrace of an alluvial fan east of a ∼5-m-deep river channel (Figures 3 and 4). The scarp-fit modeling of the 300-m-long topographic profile across the scarp revealed a vertical separation of 8.4 ± 0.7 m, and an upper and lower fan surface slope of ∼0° and ∼3°, respectively (P6; Figure 3 and Figure S1 in Supporting Information S1). In the surrounding foothills west of the trench, we observed fault-scarp splays of the cPFT with opposite thrust vergence that are possibly conjugate with respect to the main scarp at the trench location ( Figure 3). These structural complexities around Ylaisu suggest a possible oblique-slip component during the ruptures.
The exposed hanging wall mainly consists of the massive 2 to 3-m-thick unit 1 with an undulatory and reworked boundary against unit 2 ( Figure 5). We observed a massive channel-like body (Ch1) composed of sandy, matrix-supported gravel with patches of silty sand lenses and burrows. A second, distinctly red and fine silty-sandy deposit (Ch2) is exposed only in the west wall. Finely laminated reddish silty-sand deposits (unit 3) in front of the scarp indicate sedimentation during ponding after scarp formation, similar to the setting described in T2. We were able to excavate a limited area to deepen the northern part of the trench. This, however, only exposed a shallow part of unit 3 in the footwall (∼20 cm) that did not reveal the contact to unit 2, and the upper termination of fault zone F1. For detailed lithological description, see Figure S5 in Supporting Information S1.
The deformation in T3 is expressed at three separate fault zones and one backthrust. The main thrust-fault zone F1 in the north upthrusts unit 1 and cuts into unit 3. The gently north-dipping bedding and frontal bend of unit 1 indicates pressure-ridge formation along the ∼23°-dipping F1, reminiscent of features observed during the Al Asnam earthquake Philip et al. (1992). Above and south of F1, the backthrust Fb offsets the northern end of Ch2; in the east wall, a thin layer of unit 1 was smeared upwards as indicated by clasts that are aligned subparallel to the inferred fault strand Fb. The south dipping fault zone F2 cuts through unit 1. In the east wall, the top of F2 splays into several, nearly vertical fissures, whereas in the west wall, F2 can be traced along a single strand; both die out before the upper boundary of unit 1. In the east wall, F2 offsets a sandy layer and in the west wall, the lower part of Ch2. Fault zone F3 is nearly vertical, deforms Ch1 and Ch2 in the west wall, and terminates approximately at the same depth as F2. The deformation patterns of F2 and F3 resemble a positive flower structure, suggesting transpressive, coeval deformation with both faults rooting in one structure. However, the evidence for coeval or independent activation is ambiguous.

T3 EQ Three (YL3):
The timing of YL3 is not very well constrained. Considering the vicinity of trenches T1, T2 and T3 (distance along cPFT = ∼3 km) and comparable event indicators (i.e., fault cuts through unit 1, terminates before its upper boundary, and no scarp/colluvium), we suggest that YL3 can be correlated with A5 and AS5 from trenches T1 and T2, respectively. The only indication regarding chronology is provided by the deformation of channel Ch1 (L2: 21.3 ± 1.3 ka) by F3, and channel Ch2 (L1: 3.15 ± 0.21 ka) by F2, which provides a maximum age constraint on timing. However, L1 was omitted from the modeling because it is out of stratigraphic order and inconsistent with IRSL ages from unit 1 (L7: 24.6 ± 2.7 ka; L6: 17.9 ± 1.1 ka). Consequently, to bracket the timing, we used the age of unit 1 (20.1 ± 0.6 ka) and of unit 2 (7.3 ± 0.1 ka). Combined, the age for YL3 is modeled at 13.9 ± 2.9 ka.

T3 EQ Two (YL2):
Fault zone F1 is not well exposed, preventing a detailed reconstruction of the earthquake history. However, because we infer that YL3 did not produce a significant scarp, a younger event must have occurred to allow ponding at its front and the deposition of unit 3, similar to T2. The age constraints from four radiocarbon ages (R3w: 5.21 ± 0.09; R1e: 3.31 ± 0.04 ka; R2w: 2.35 ± 0.01 ka; R1w: 2.90 ± 0.03 ka) and two IRSL ages (L5: 4.20 ± 0.37; L4: 3.62 ± 0.39 ka) from unit 3 are favorably comparable to those from unit 3 in T2. Thus, we assume the pond formation is associated with the same event (AS4). Without any further indicators, we use the average age of unit 2 (7.3 ± 0.1 ka) as maximum age and unit 3 as minimum age constraint. Together, these provide a modeled age for YL2 of 6.1 ± 0.6 ka.

T3 EQ One (YL1; MRE):
With respect to the exposed part of deformation features, we interpret that the most recent event ruptured along F1 and coevally backthrusts along Fb. Deformed internal layering unconformably overlain by undisturbed strata within unit 3 indicates that the deformation took place during its final deposition. Following this interpretation, we use the older phase of deposition of unit 3 (lower part; see above) as a maximum age constraint and two radiocarbon ages obtained from the upper part of unit 3 (Rb1w: 0.82 ± 0.05 ka; R2e: 0.16 ± 0.09 ka) as minimum age constraints. We omit one IRSL sample (L3: 8.68 ± 0.50) due to apparent partial bleaching (major offset of pIR and IRSL; Table 2). This bracketing yields a modeled age for YL1 of 1.7 ± 0.7 ka.

T4 -Komansu Site (39.50°N 72.65°E)
The Komansu trench site is located approximately in the central part of the cPFT, on a terrace west of the ∼30-m-deep active Komansu River channel ( Figure 2). In this area, the fault generated a double-scarp morphology, with a main thrust (northern trench) and a backthrust 50 m south (southern trench), which is well developed in the western river cut (Figure 3). The scarp-fit modeling of the 600-m-long topographic profile perpendicular to both scarps revealed a vertical separation of 5.0 ± 1.1 m at the main scarp and 0.9 ± 0.5 m at the back scarp, and an upper and lower fan surface slope of ∼2°, respectively (P11; Figure 3 and Figure S1 in Supporting Information S1). The T4 trenches were excavated at a lower terrace surface, which is inferred to be correlative with the younger Qt 4 terrace (∼4.5 ka) according to the regional correlation developed by Arrowsmith & Strecker (1999) and Strecker et al. (2003) and dating presented here (Figures 3 and 4).
The lower stratigraphy is consistent throughout both trench exposures ( Figure 5). Each footwall consists of undeformed layers of horizontally and discontinuously bedded fluvial gravels with CaCO 3 coatings (unit 1b) overlain by fine silty sand with incorporated pebbles in its lower part (unit 2b). Unit 2b (Qt 4 ) is less thick than unit 2 defined elsewhere on Qt 3 and has a more diffuse basal contact with unit 1, which could imply less stable depositional conditions and potentially higher seasonal erosional events during deposition. For detailed stratigraphic description, see Figure S6 in Supporting Information S1. All ages in the northern trench result from IRSL samples collected during the 1999 campaign. In the southern trench, two IRSL samples (L1b, L2b) were collected in 2017 in order to cross-check the reliability of the older samples.
The northern trench exposes a major ∼21°-S-dipping main thrust fault F1 and inclined bedding of unit 1b in the hanging wall, as observed in the other trenches. The layering within the gravel is not well developed, but evident in layers defined by variations in the clast/matrix ratio and the sizes of clasts. F1 manifests as a 10-cm-wide zone with broken and rotated clasts ( Figure 5). Below, and to the north of the fault, we observe fault breccia presumably injected (C0) into unit 2b during rupture and two colluvial layers (C1, C2) that are preserved with gentle northward inclinations, but with an unambiguous relative age relationship. The composition of the colluvium is distinctly different from that of the fluvial gravel. The lateral pinch-outs of the colluvial layers merge with unit 2b towards the north.
The southern trench exposes a minor ∼30°-N-dipping back thrust (Fb), with two fault strands bounding the shear zone (Fb1 and Fb2). We did not find evidence for colluvial material derived from the scarp face. Instead, the south facing scarp, which builds an efficient sediment trap, has been blanketed repeatedly by fine material with north-directed transport from the surface to the south (unit 3 and TS). Unit 3 is absent in the hanging wall of Fb. Structural relation along Fb indicates two separate faulting events and fault migration. The lower fault strand (Fb1) displaces sediments of unit 1b above unit 2b and is buried by unit 3, while the northern fault strand (Fb2) cuts through units 1b and 3, and terminates approximately 20 cm below surface. The exposed stratigraphy is covered by a ∼20-cm-thick topsoil layer (TS) that becomes gradually mixed with fines from unit 3 with increasing depth.

T4 EQ Three (KO3):
The oldest event ruptured solely along F1 in the northern trench and is indicated by C0 that was presumably injected during faulting and ongoing deposition of unit 2b. IRSL ages sampled stratigraphically above C0 from unit 2b yield 10.12 ± 0.72 ka (L7) and 3.2 ± 0.22 ka (L3) (northern trench), and 3.69 ± 0.16 ka (9(L1b) and 2.57 ± 0.09 ka (R1b) (southern trench). Assuming that unit 2b represents the cover that constitutes the surface of Qt 4 (4.5 ± 0.2 ka), and because sample L7 is older than unit 2 (7.3 ± 0.1 ka) that corresponds to the older terrace surface Qt 3 , we suggest that L7 is an outlier and omit the age from our model. Consequently, we used the ages of Qt 4 and L1b, and L3 and R1b as the maximum and minimum limiting age for KO2, respectively. These ages yield an age for KO3 of 3.2 ± 0.3 ka.

T4 EQ Two (KO2):
The penultimate event is associated with unit C1 in the northern trench, which covers unit 2b and F1 cuts updip past C0, injecting fault breccia in the base of C1. The age control of C1 (L5: 3.2 ± 0.26 ka, L4: 1.51 ± 0.10 ka) and unit 3 (L3b: 1.87 ± 0.14 ka, R2b: 1.85 ± 0.07 ka) from both trenches suggests that during this event, faults F1 and Fb1 were activated. The similarity in ages (∼3.2 ka) between L3 (unit 2b) and L5 (C1) implies that L5 may have been recycled during faulting without sufficient bleaching of the material; thus, this sample will be omitted. In contrast, L4 (C1) more likely represents a true depositional age for C1, assuming it got pushed northward and disturbed during the subsequent earthquake (KO1) given its location with respect to L5 and proximity to the inferred portion of the ground surface prior to the earthquake. Following this argument, L3 and L1b serve as maximum limiting ages and L4, L3b and R2b as minimum bracketing ages, which yield a modeled age for KO2 of 2.4 ± 0.2 ka.

T4 EQ One (KO1, MRE):
The most recent evidence for earthquake faulting exposed in the northern trench is the steeper F1 strand that cuts off and deforms C1, while its upper end flattened out to a horizontal position. As a result, the hanging wall collapsed partly and the strata were displaced downhill over the preexisting toe of the scarp (top of C1). The overlying colluvial package C2 formed presumably during subsequent interseismic scarp diffusion processes. Similarly, for KO2, the age control in both trenches indicates that the MRE rupture activated both faults F1 and Fb2 simultaneously. To bracket KO1, we used the ages of C1 (L4), and unit 3 (L3b and R2b) as maximum age constraint and the IRSL age from C2 (L2: 1.43 ± 0.10 ka) and TS (L2b: 1.12 ± 0.09 ka) in the southern trench as a minimum age constraint. The age for KO1 is modeled at 1.5 ± 0.1 ka.

T5 -Tashkungey Site (39.52°N 72.77°E)
The Tashkungey paleoseismic site is located at the eastern end of the cPFT, on alluvial-fan deposits of Qt 3 , west of the active ∼20-m-deep Tashkungey River channel (Figure 2). We chose to excavate across a portion of the fault scarp which has a double-scarp morphology (Figures 3 and 4). The scarp-fit modeling of the 350-m-long, scarp-perpendicular topographic profile reveals 6.4 ± 0.5 m of total vertical separation, and an upper and lower fan-surface slope of ∼2°, respectively (P16; Figure 3 and Figure S1 in Supporting Information S1). The excavated trench was ∼19 m long, ∼2.5 m wide, and up to ∼3 m deep. The excavation exposed up to 3 m of fluvial gravel terrace deposits of the bottom (unit 1) capped by a thin (10-20 cm) layer of presumably reworked and partly eroded remnants of unit 2 (fine silty sand) in both trench walls, and is stratigraphically and chronologically consistent with the other trenches (∼21 ka). The boundary between unit 1 and unit 2 is not sharp. Instead, we observe interspersed subangular clasts incorporated into the silty sand matrix at the basal contact of unit 2. This indicates that the upper contact of unit 1b was reworked during the deposition of unit 2. The main stratigraphic sequence is faulted along two separate fault zones (F1, F2), which coincide with the surface position of the scarps. Near to and north of each fault zone, unit 2 is overlain by a succession of scarp-derived colluvial deposits C1-C3, and Ca-Cc, respectively. The organic-rich silty to sandy modern soil A horizon reaches a thickness of up to 30 cm at the scarp toes and is only poorly developed in an area of unfaulted intermediate hanging-wall deposits. For detailed stratigraphic description, see Figure S7 in Supporting Information S1.
Both fault zones show a clear 10 to 20-cm-wide zone of rotated clasts subparallel to the fault dip. Fault zone F1-related deformed colluvial packages comprise loess-like deposits similar to unit 2. Unit 2 and colluvial units C1 and C2 are only distinguishable by fault breccia injected into the base of the colluvium at F1. C3 is more mixed, shows a clear change in color from beige to reddish beige, and includes more organic material. Colluvium related to fault zone F2 is massive and structurally and stratigraphically complex. Ca consists of well-sorted silty sand. Its base is derived from a remnant of injected fault breccia (similar as at F1) that was dragged and faulted during subsequent faulting along F2. A major faulting event deformed a large wedge-shaped block with a maximum thickness of up to ∼80 cm and composed of poorly sorted gravel with chaotic orientation of pedogenic carbonate coatings and reworked matrix-supported pebbles at its tip (Cb). This unit is overlain by a lens of organic-rich paleosol, developed in a matrix-supported, poorly sorted gravel (PS). The youngest colluvium (Cc) constitutes strata of poorly sorted gravel that fines upward and merges laterally into more organic-rich fine material toward its northern tip. We infer that internal structural disorder in Cc is a result of possible bioturbation or cryoturbation. Considering the stratigraphic and chronologic relationship across the trench, we infer that C1, C2 and C3 formed coevally with Ca, Cb, and Cc after simultaneous fault activation during rupturing events.

T5 EQ Three (TS3):
The oldest earthquake that caused rupture along F1 and F2 led to subsequent deposition of C1 and Ca. We collected five radiocarbon samples from both units, respectively. Unfortunately, the resulting ages (R3e: modern; R2e: 0.17 ± 0.09 ka; R9e: 1.55 ± 0.02 ka; R1w: 0.13 ± 0.08 ka; R4e: 1.01 ± 0.06 ka) are neither in chronologic order nor do they fit the stratigraphic order, and were therefore omitted in the modeling (see Discussion for more detailed explanation). Two IRSL ages (L1e: 4.79 ± 0.35 ka; L3e: 4.45 ± 0.28 ka) from unit C1 provide a more realistic age, assuming that the underlying unit 2 formed at ∼6-7 ka. We were not able to obtain absolute age control for unit 2 in the trench; however, the terrace-surface age obtained from a pit (PT3) within the hanging wall east of the river (see Figure 3 for location) is consistent with the ages of unit 2 along the cPFT. Thus, we are confident in using the age for unit 2 (7.3 ± 0.1 ka) as a maximum age and use our IRSL ages (L1e and L3e) as a minimum age constraint. Combined, this information yields an age for TS3 of 5.4 ± 0.8 ka.

T5 EQ Two (TS2):
The penultimate earthquake ruptured along F1 and F2 deforms C1 and Ca, respectively, and was followed by inferred hanging-wall collapse at F2 depositing Cb and slow subsequent deposition of the northern tip of Cb and C2. We obtained three IRSL ages (L2e: 2.25 ± 0.14 ka; L6e: 3.06 ± 0.14 ka; L5e: 3.78 ± 0.44 ka) and one radiocarbon age (Rb2w: 2.73 ± 0.04 ka) from both colluvial packages. The age constraints provide a good stratigraphic and chronologic order, and serve as minimum age constraints. Combined with the age of unit C1 (L1e, L3e) as a maximum age constraint, TS2 is modeled at 3.7 ± 0.4 ka.

T5 EQ One (TS1; MRE):
The most recent event ruptured along F1 and F2 and is associated with units C3 and Cc, respectively. Cc covers a layer of paleosol (PS) that presumably developed after the deposition of Cb (further distinguishing this and the prior earthquake). Injection and upward-dragged Ca material south of Cb and subparallel to F2 indicates that the Cb and PS units were truncated and deformed during this younger rupture. The inferred strong deformation during the faulting event could further explain the chaotic gravel orientation in the main body of Cb. Following this interpretation, we use one IRSL (L1w: 2.03 ± 0.13 ka) and two radiocarbon ages (Rb1e: 1.93 ± 0.03 ka; R9e: 1.55 ± 0.02 ka) from PS that broadly agree with each other, as maximum age constraints. For a minimum age bracket, we used the age of one radiocarbon bulk-soil sample (Rb1w: 1.54 ± 0.05 ka) associated with Cc. Two radiocarbon samples (R8e: 0.40 ± 0.05 ka; R6e: 0.36 ± 0.14 ka) obtained from C3 were omitted due to their unrealistically young ages (see Discussion). Combined, this bracketing yields an age for the most recent event TSa at 1.5 ± 0.1 ka.

Displacement Per Event
We measured the dip-slip displacement of stratigraphic horizons that presumably were offset during a single rupture, and the upslope thicknesses of colluvial deposits on both trench walls (if unambiguously related to surface ruptures) from all trench exposures to reconstruct per-event dip-slip displacements (T1-T5; Figures S3-S7 in Supporting Information S1). However, the evolution of such indicators in thrust-and reverse-faulting environments is more complex compared to settings dominated by normal faulting, and the depositional geometry is often affected by different processes during repeated faulting, especially if ruptures occur along the same fault zone. These processes include (a) deformation of the preserved colluvial deposits due to thrusting, (b) "bulldozing" of the fault tip especially along faults with dips <20°, and (c) truncation at the upslope limits of the deposits by fault-zone migration or splaying (e.g., McCalpin & Carver, 2009). Additionally, as shown in our paleoseismic observations, we expect that several earthquakes ruptured along multiple fault strands within single trench exposures coevally, implying partitioning of the total dip-slip displacement. Reconstruction of the effects from cumulative and coeval faulting may be very ambiguous, so we use our estimates to provide approximate ranges of dip-slip magnitudes during the earthquake history preserved in the trenches. Displacements of stratigraphic horizons range from 0.4 to 1.1 m. From colluvial deposits, we estimate an approximate dip-slip range of 0.6 to 2.8 m, assuming that the maximum preserved fault-proximal colluvial thickness represents about one-half of the initial coseismic dip-slip along the fault (e.g., McCalpin & Carver, 2009;Personius & Mahan, 2005). In case of simultaneous multi-fault activation during one rupture, total per-event dip-slip displacement estimates reach up to 4.9 ± 0.7 m (Table 4).

Paleoearthquake Correlation and Recurrence Along the cPFT
Our modeled timing of single earthquakes is consistent for several events across all trenching sites. Considering the proximity of these sites, and taking into account the logic of Biasi & Weldon (2009) and the scaling relationships of Wells and Coppersmith (1994), potential minimum surface-rupture lengths based on average dip-slip displacements support the assumption that several identified paleoearthquakes likely extended from one site to another along the cPFT. The graphical paleoearthquake PDF correlation across multiple trenching sites suggests six earthquakes (E6-E1, oldest to youngest) ruptured the surface along the cPFT since ∼14 ka (Figure 7). To justify our interpretation of the correlation, we use the earthquake evidence and timing quality metric (i.e., poor, moderate, good, and excellent) based on the bracketing and resulting modeled uncertainty of the timing of the event (Table 4).
Based on this correlation, we have produced a segment-wide earthquake chronology following the approach of Personius et al. (2012) and the method established by Weldon (2009) andDuRoss et al. (2011). We evaluated and integrated multiple paleoseismic ages into a single refined age model by applying the product method (direct multiplication of the discretized PDFs), which yields better-defined earthquake ages. Applying this method to this interpretation resulted in refined earthquake ages for the segment (with ±2 σ) of E6 = 14.3 ± 2.2 ka, E4 = 3.4 ± 0.2 ka, E3 = 2.4 ± 0.2 ka, E2 = 1.5 ± 0.1 ka, and E1 = 0.8 ± 0.2 ka (Figure 7). Due to poor overlap of the site PDFs in the case of E5, we used the simple-sum-mean, which resulted in E5 = 5.3 ± 1.1 ka with a significantly higher uncertainty. The new segment-wide earthquake chronology reveals a stepwise decrease in re-currence intervals from past to present (Table 5), which contrast the irregular recurrence intervals in site-specific earthquake chronologies (Table 4).

Vertical Separation Along the cPFT
We measured a total vertical separation (VS; as defined in Yang et al., 2015) from the offset Qt3 surface along the cPFT by extracting scarp-perpendicular topographic profiles from the 12-m-resolution TanDEM-X model ( Figure S1 in Supporting Information S1). In order to assess the accuracy of the TanDEM-X model, we compared  (Table 6) along the central PFT, based on topographic profiles across the scarp extracted from TanDEM-X data (black diamonds indicate location; red diamonds indicate trenching sites). For details see Figure  S1 in Supporting Information S1 for additional details.
these profiles with profiles extracted from the 0.5-m-resolution DSMs from our trenching sites T1, T2, T3 and T5 ( Figure 3). The correlation shows that the elevations and slopes from the extracted data do not differ significantly (see profile P3, P5, P6 and P16 in Figure 3 and Figure S1 in Supporting Information S1), thus we are confident that the TanDEM-X-based profiles are suitable for reliable VS calculations. The resulting far-field cumulative VS along the cPFT ranges from ∼5-16 m, with nearly parallel hanging-wall and footwall slopes at ∼2°. The along-strike distribution shows an asymmetry with significantly higher VS (up to double) in the western half of the central segment, consistent with results presented in Arrowsmith & Strecker (1999), with a maximum VS of 16.2 ± 0.2 m at the Syrinadjar River terrace (Profile P8; Table 6, Figure 3 and Figure S1 in Supporting Information S1), and an average VS of 9.2 ± 0.3 m (Table 6).
However, as shown in other studies of reverse-fault scarps, the VS of the ground surface does not always reflect the actual fault throw (e.g., Kaneda et al., 2008), which is particularly true for older scarps that formed through multiple events (McCalpin & Carver, 2009) or where there is a profile asymmetry that likely leads to an overestimated or underestimated VS (e.g., at T2). To cross-check our VS estimates, we employ a second approach for total VS evaluation following the reverse-fault related parameters and resulting equation for parallel hanging wall and footwall ground surfaces DS = VS cos α/sin (θ+α); where DS is dip-slip displacement, α is the fan surface slope, and θ is the dip angle of the fault (Thompson et al., 2002;X. Yang et al., 2015). By rearranging the equation, we used cumulative dip-slip displacements exposed in our trenches to calculate VS. In the subsurface of trench T2 and T5, we used the contact between units 1 and 2 as a horizontal marker that can be laterally traced between the hanging wall, intermediate hanging wall, and footwall ( Figures S3 and S7 in Supporting Information S1). Together with an average fault dip of 21 ± 3° exposed in the trenches and assuming a surface slope of ∼2°, we were able to estimate a cumulative net-amount of dip-slip motion at T2 and T5 ( Figures S3 and S7 in Supporting Information S1), which results in a VS of 8.5 ± 0.2 m and 5.8 ± 0.2 m (Table 6), respectively (equivalent to profile P5 and P16; Figure 3). The calculated VS of T5 coincides with the TanDEM-X-based VS of 6.4 m. In case of T2, the calculated VS is significantly larger (8.5 m versus 5.2 m), but better matches the along-strike distribution of VS (8-10 m) (Figure 7 and Figure S1 in Supporting Information S1). In this case, we assume that possible differences in surface process dynamics (e.g., enhanced lateral sediment transport) led to the apparent profile asymmetry at T2. In summary, we suggest that the TanDEM-X-based VS estimates in our study area are representative and can be used for further calculations.

Late Quaternary Dip-Slip Rate
The dip-slip rate v is calculated from the division of accumulated offset x by the age t of a corresponding geological unit (v = x/t with Δv/v =v Δx/x + Δt/t). The cumulative dip-slip measured in the trenches is significantly lower than would be expected compared to the associated VS, highlighting the ambiguity of dip-slip reconstructions in thrust fault exposures (Figure 8). Therefore, to provide a firm basis for our dip-slip rate calculations, we used our TanDEM-X-based VS estimates to re-calculate the associated accumulated dip-slip offsets along the cPFT as- Note. Bold highlights the paleoevents that ruptured the full segment length, which are the base for the calculation of the interseismic recurrence of "full activation". With the bold we would like to make the connection clear.
a Sequence of paleoearthquakes correlatebale in their timing from all trenches along the cPFT. b Refined earthquake timing through multiplication of overlapping site PDFs. c Recurrence interval for major earthquakes that activated the entire segment during rupture. Closed recurrence (E older -E younger ) and absolute error.

Table 5 OxCal Models for Paleoseismic Sites and Refined Segment Earthquake Ages for the Central PFT
suming a thrust-dip angle of 21 ± 3°. To emphasize the variations of the VS along the cPFT, we present selected profiles from and adjacent to the trenching sites for our calculations (Table 6). However, to be more cautious, we also average the VS along the full length of the cPFT, which results in 9.6 ± 0.3 m of VS and corresponds to a cumulative dip-slip of 24.7 ± 4.1 m. The timing of paleoearthquakes varies along the segment between 14.3 ± 2.2 ka in the western (T1, T2, and T3) and 5.3 ± 1.1 ka in the central and eastern parts (T4 and T5), which hampers our ability to make a definite statement regarding fault initiation for the rate determination. Consequently, we bracket a minimum and maximum dip-slip rate using both ages, respectively. To provide accurate estimates, we used the Python tool Slip Rate Calculator v0.1.3 (Styron, 2015), which uses age and offset PDFs to calculate the slip-rate PDF via Monte Carlo methods. For both offset and age, we used a normal distribution characterized by a mean and standard deviation. The final modeling yields an average minimum and maximum dip-slip rate of 1.7 ± 0.5 and 4.7 ± 1.7 mm/yr, respectively (Table 6). Using the maximum VS from the Syrinadjar River (Profile 8, Figure S1 in Supporting Information S1) results in 41.5 ± 5.9 m of cumulative dip-slip and a minimum and maximum dip-slip rate of 2.8 ± 0.8 and 7.8 ± 2.7 mm/yr, respectively (Table 6).

Paleoearthquake Allocation and Potential Obstacles
Reconstruction of a paleoearthquake sequence relies on distinguishing between tectonic and depositional features in our trench exposures. We emphasize that the stratigraphic and geomorphic nature of the faulted alluvial fans that cover fluvial deposits in our study area preserve important characteristics related to the accumulation of colluvium. In most cases, within the alluvial fan, unit 2 constitutes the source material for the colluvial deposits that we identified in our exposures. Unit 2 is a thin (10-50 cm) layer composed of compacted fine to medium sand. The original fan slopes are <4°. If the age-thickness relationship of unit 2 is considered, the sedimentation rates in this area must be rather low. Mobilization of large amounts of debris from this unit during strong ground motion is thus not very likely in this environment. Furthermore, considering the well-preserved, present-day surface of the Qt 3 terraces, we assume that extensive episodes of erosion have decreased during the last ∼7.3 ka (see also Discussion below); furthermore, we do not consider the formation of the colluvial deposits to have been triggered by pronounced weathering or degradation related to storms or wet episodes (e.g., Rizza, Bollinger, et al., 2019). Instead, we relate colluvial strata observed in our study area (i.e., maximum thickness of 50 cm of mostly re- 16.2 ± 0.2 41.5 ± 5.9 7.8 ± 2.7 2.9 ± 0.9 Max East Profile P17 9.2 ± 0.2 23.6 ± 3.6 4.4 ± 1.6 1.6 ± 0.5 Average All Profiles c 9.6 ± 0.3 24.7 ± 4.1 4.7 ± 1.7 1.7 ± 0.5 a Estimated from the VS with a fault dip of 21 ± 3°. b Dip-slip rate calculations using 5.3 ± 1.1 ka and 14.3 ± 2.2 ka for max and min values, respectively. c Averaged over all vertical separation estimated from ground surface profiles (see Figure S1 in Supporting Information S1).

Table 6 Summary of Cumulative Vertical Separation and Dip-Slip Displacement, and Estimated Dip-Slip Rate From West to East
Along the Central PFT worked silty sand) to surface faulting, scrap formation and subsequent erosion and deposition. Consequently, we suggest that the colluvium represents weathered material generated by transport of scarp material from the new fault tip during the immediate postseismic interval of a ground-rupturing earthquake.
Reliable age models from past earthquakes require correct sampling of high-quality, datable material and rigorous interpretation of results. Typical targets for dating earthquakes in dip-slip fault exposures are colluvial-wedge deposits. In our trenches, only a limited part of the colluvial strata result from coseismic hanging-wall collapse (i.e., T1-C3a and T5-Cb), which adds potential obstacles to an unambiguous age determination, because luminescence samples might be only partially bleached and in case of radiocarbon dating, inherited ages from reworked organic material might get incorporated into the deposits. However, we interpret most of our colluvial deposits to result from processes described above and similar to those that formed the topsoil layers (TS) documented in the logs ( Figure 5). The toes of the fault scarps are draped by lenses of colluvium derived from slope wash and covered the scarp. Similar slow erosion and deposition have occurred in the past. The corresponding deposits were cut and deformed during faulting. The resulting cut-off, wedge-shaped colluvial deposits in the uppermost footwall were subsequently buried and preserved by the hanging wall of the thrust fault, reflecting the timing of surface offset.
To avoid sampling of partially bleached material for IRSL dating, it has been recommended to sample the toe of the colluvial deposits (long transport distances) and not the coarser materials immediately abutting the fault (short transport distances) (Gray et al., 2015). Only a limited portion of the toe was exposed at trenching sites T1 and T3, and at T5, the toes of different colluvial deposits merge with unit 2, thus, are not separable. Consequently, in these cases, we were forced to sample the colluvial strata proximal to the fault. We sampled the upper part of these colluvial deposits assuming a higher probability for interseismic deposition (see Discussion above). Finally,  Table 4) and age (Table 5) for each paleoearthquake. Inset shows the dip-slip rate derived from vertical separation along the central PFT. Maximum spans the time period since earthquake E5 and minimum since E6 (see Section 4.7 for explanation).
those samples that reflect partial bleaching of the pIR signal shown by a significant offset (more than two times higher) of the ages (T3: L2, L3, L4, L5, L6, L7; T5: L4e) between the pIR and IRSL (Gray et al., 2015) were retrieved from the fluvial deposits (unit 1) or deposits related to sediment-laden water columns (pond; unit 3). Such material may be associated with poor bleaching of luminescence signals due to limited sunlight exposure and short residence times in riverbeds (e.g., Ishii et al., 2022;Smedley et al., 2019). However, the ages derived from these samples are generally consistent within their stratigraphic context and along the cPFT, and thus seem reliable. Yet, we cannot fully exclude a certain overestimation of their age range because of insufficient bleaching prior to deposition.
To reduce ambiguity of our IRSL ages, we also sampled organic material for radiocarbon dating. In case of T1, the ages are consistent and we are confident of the reliability of the results. In the case of T2, T3 and T5, however, our radiocarbon samples (T2: Rb1, R11e, R13e, R4e, R5e; T3: R2e; T5: R2e to R7e, R1w) yielded ages younger than those from IRSL. Therefore, we omitted these radiocarbon ages, because they are inconsistent stratigraphically or chronologically. Except for Rb1e (T2), the dated samples contain small amounts of charcoal. We suspect that the young ages result from either (1) poor sampling, where dark root fragments were erroneously classified as charcoal, (2) contamination due to downward penetration of younger roots or the influence of younger humic acids, or (3) bioturbation processes, where seasonal freeze-thaw phases or burrowing organisms within these units transported fine material into the underlying units (e.g., Walker & Walker, 2005). The first suggestion may apply to material sampled below 50 cm depth, whereas the second and third issues apply to the material sampled at shallow depths of <20 cm.

Completeness of the Central Pamir Frontal Thrust Paleoearthquake Chronology
Our observations from five sites across different sectors of the cPFT provide a new compilation of surface-rupturing paleoearthquakes with detailed slip histories. An overlap in earthquake timing based on observations from multiple trenches indicates extended ruptures along the cPFT. Our refined earthquake chronology documents six surface ruptures (E6-E1; old to young) since the Late Pleistocene (<17 ka). Before we further address aspects of the seismogenic behavior along the cPFT, we assess the completeness of the modeled earthquake record. Our temporal evidence is generally less well constrained for events older than 3 ka. Lack of data is a common problem observed in paleoseismological studies (e.g., Nikonov, 1988;Nicol, 2016), which impacts the validity of the interpreted earthquake sequence. We compare our documented rupture evidence and associated timing with earthquake-related features (e.g., scarps, fissures, tension gashes) reported by Nikonov (1988) (Figure 9b and 9e). We caution that the chronology provided by Nikonov (1988) is partly based on lichenometry and the timing of the emplacement of mass-movement deposits is only inferred to have been seismically triggered. Consequently, there are large uncertainties in those data, especially for earthquakes older than 2 ka.
Our stratigraphic analysis did not reveal evidence for the A.D. 1978 M w 6.6 Zaalai event in any of the trenches despite previous suggestions that ruptures partly propagated into the central segment during its aftershock sequence (Arrowsmith & Strecker, 1999;Nikonov et al., 1983). Instead, the MRE (E1) was dated to 0.8 ± 0.2 ka and was only visible at the western end of the cPFT (Figure 9), with strong evidence in trench T2 (Table 3). The penultimate event (E2) occurred at 1.5 ± 0.1 ka (Figure 9). The consistency and large number of rupture indicators for E2 in all five trenches, as well as the robustly modeled earthquake timing suggest a rupture extent of ≥35km (full-segment length). Interestingly, three earthquake ages reported by Nikonov (1988) from the western and eastern transfer zones range from 1-2 ka, and thus overlap with our E2 event (Figure 9). By including these observations, we infer a rupture extent of as much as ∼65 km during E2. Event E3 (2.2 ± 0.2 ka) was modeled based on strong evidence and good timing in the intermediate Komansu site (T4, Table 3 and 4), but weak evidence and ambiguous timing constraints from the westernmost trench T1 (Table 3). Consequently, the rupture extent of E3 is uncertain. This earthquake might have ruptured only a small portion of the cPFT.
Our interpretation regarding earthquake E4 (3.6 ± 0.2 ka) results from solid evidence in T2 and T5, weak evidence in T4, and overlapping earthquake PDFs from all three trenching sites (T2, T4, T5) spanning different sectors of the cPFT. We did not find evidence in either T1 or T3, despite their vicinity to T2. The relative age constraints and weak evidence in T1 and generally ambiguous earthquake indicators in T3 for events older than 2 ka might result in an incomplete, less reliable earthquake chronology at these sites. We infer another discontinuous rupture extent in the case of E5 (5.3 ± 1.1 ka), with strong evidence from T2 and T5, weak evidence from T1 and T3, and missing evidence at the intermediate Komansu site (T4). Considering the fact that trench T4 was excavated on the Qt 4 terrace surface (∼4.5 ka; Figure 3), it is likely that any prior evidence was eroded during the incision of Qt 3 (<7.3 ka). In summary, we assume a segment-wide activation (∼35 km) during E4 and E5. Event E6 (14.6 ± 2.1 ka) presumably occurred during the formation of the Qt 3 terrace (20.1-7.3 ka). We found only weak evidence at the western terminus of the cPFT (T1-T3) based on small stratigraphic offsets detected in fluvial terrace deposits (unit 1) with poor age control. A longer rupture extent during this event, however, cannot be ruled out because of the limited exposure of unit 1 in the other trenches.
A possible impediment to retrieving earthquake evidence older than 7 ka in our study area could be caused by the effects of climatic conditions in the Pamir region during the Late Pleistocene to Early Holocene. In a paleoclimate review from Central Asia, Herzschuh (2006) reported a major transition from dry climate conditions during the Last Glacial Maximum (LGM) at ∼21 ka to wetter conditions and a peak in mean effective moisture at ∼7.5 ka (Figure 9). This period, which corresponds to the time of deposition of unit 1 and the aggradation of the Qt 3 terrace (20.1-7.3 ka), was characterized by a period of deglaciation that was followed by minor glacial readvanc-  (Table 6). (e) Hillshade image based on TanDEM-X data. Red and black lines indicate the fault trace (solid) and approximate trace (dashed) of the cPFT and PFT N+S , respectively. Black circles indicate locations of earthquake evidence from Nikonov (1988). es starting at ∼21 ka (e.g., Koppes et al., 2008;Röhringer et al., 2012). Thus, it seems likely that the increased availability of meltwater during deglaciation caused substantial changes in fluvial dynamics and associated erosion and aggradation processes along the Pamir mountain front (e.g., Fuchs et al., 2013). These climate-driven complexities not only potentially explain the depositional gap (∼13 kyr) between the formation of Qt 3 (unit 1) and its abandonment (post-dated by unit 2), but could also be a reason for the obliteration of geomorphic evidence of surface ruptures on fluvial channels during that time.
Consequently, if our trenching sites adjacent to active river channels (Ylaisu, T3; Komansu, T4; Tashkungey, T5) were indeed affected by increased glacio-fluvial dynamics that altered or removed evidence of surface ruptures, an incomplete chronology for earthquakes older than 7 ka is probable. However, the chronology of our documented surface ruptures during E1 to E5 likely constitutes a complete record of strong surface-rupturing plaeoearthquakes over the past 7 kyr and we focus on this time period in our subsequent analysis.

Partial Versus Full-Segment Ruptures
Considering our earthquake correlation along the cPFT, we suggest a 7-kyr-long history of seismogenic events that consisted of smaller, discontinuous ruptures (E1, E3) with lengths of <20 km that partially broke the western end of the central segment. These were interspersed with more extensive, continuous ruptures (E4 and E5) that likely activated the entire length of the segment (∼35 km) or even crossed its boundaries >35 km (E2) (Figure 9). This apparent higher rupture frequency in the western half of the cPFT coincides with significantly larger scarp offsets (implying a greater dip-slip rate) in the west (Figure 9).
If we compare our estimated extent of earthquake evidence (equivalent to a surface rupture length, SRL) with associated measured dip-slip displacements, we observe compatible scaling relationships for E2, E4, and E5, but discrepancies regarding the apparent partial ruptures E1 and E3 (following maximum displacement and rupture length scaling relations of Wells & Coppersmith (1994) and Manighetti et al. (2007)). In the latter cases (E1 and E3), the measured maximum dip-slips offset amount to several meters (2-3 m) and could imply SRLs of >30 km instead of the inferred <20 km (Figure 8). This discrepancy might be consistent with a common phenomenon observed in other thrust/reverse fault studies in Central Asia and elsewhere that revealed discontinuous surface-rupture traces that were shorter than expected, despite several meters of slip associated with major ≥M7 earthquakes (e.g., Ainscoe et al., 2018;Arrowsmith et al., 2017;Grützner et al., 2019;Rimando et al., 2019;Rockwell et al., 2014). In this context, E1 and E3 might constitute ruptures with discontinuous surface breaks that potentially bypassed our remaining trenching sites. Another reason for such irregular surface displacement could be potential complex interactions between faulting and folding along the PFT thrusts (e.g., Baljinnyam et al., 1993;Wang et al., 2020). Nevertheless, the significant dip-slip displacements suggest these ruptures were larger than what would be expected from the visible lateral extent of the surface rupture. This raises the question, as to whether or not the ruptures propagated into the neighboring western segment crossing geometric fault-trace discontinuities (see discussion below) during a multi-segment rupture scenario (Rubin, 1996).
To estimate the seismic potential of the cPFT, we determined the seismic moment for a full-segment rupture as well as potential ruptures cascading into adjacent segments. We calculated the scalar seismic moment M 0 = μD ave A, where µ=3·10 10 N/m 2 is the shear modulus for crustal faults (Hanks & Kanamori, 1979), D ave is the average displacement, and A is the fault-plane area (length x width). The per-event dip-slip displacements associated with our events did not exceed 5 m (Table 4); therefore, we exclude the possibility of megathrust paleoevents with M w > 8 along the cPFT during the Holocene. However, the sparse number of dip-slip estimates and the complex nature of dip-slip may result in inaccurate values for D ave . As observed in all five trenches, D ave-obs = 2.5 ± 0.5 m for E2 is the most reliable estimate and was thus used for a seismic-moment calculation for full-segment ruptures. Additionally, we used the minimum surface-rupture length (SRL) along the cPFT (lateral extent of earthquake evidence in trenches) and empirical self-consistent fault-scaling relationships for interplate dip-slip events where log(D ave-calc ) = -3.799 + 0.833•log(SRL) (class II after Scholz et al., 1986; last updated by Leonard, 2010Leonard, , 2014 to estimate a robust D ave . To define fault width, we used an average PFT fault dip of 30° (Arrowsmith & Strecker, 1999;Burtman & Molnar, 1993;Nikonov et al., 1983;Schurr et al., 2014) and assumed a seismogenic depth of 15 km based on the depth range of large instrumentally recorded earthquakes and the average-depth distribution of crustal seismicity associated with the PFT (e.g., Ekström et al., 2012;Fan et al., 1994;Schurr et al., 2014;Sippl, Schurr, Tympel, et al., 2013;. For a full-segment rupture (∼35 km; E4 and E5), we estimate a scaling-relationship derived D ave-calc = 1 m and a total seismic moment release of M 0 = 3.15·10 19 Nm, which is equivalent to a M w 7.0 earthquake (with M w = 2/3·log(M 0 ) -6). We infer a maximum extent of rupture of ∼65 km (E2) assuming segment interaction with the neighboring western and eastern transfer zones (Nikonov et al., 1983;Nikonov, 1988). In light of the relationship between the 1978 M w 6.6 Zaalai earthquake main shock (equivalent to M 0 = 0.71·10 19 Nm) and its strike-slip dominated aftershocks (Fan et al., 1994), with rupture-propagation patterns into the western transfer zone and the central segment (as reported by Nikonov et al., 1983), Arrowsmith & Strecker (1999) suggested that the ∼15-kmlong western transfer zone can be categorized as a small, semiindependent rupture segment. If we assume the same for the eastern transfer zone, a potential segment-transfer-zone coupling during E2 with a D ave-calc = 1 m for the cPFT, and 0.5 m for the western and eastern transfer zones, respectively, yields a total seismic moment of 4.44·10 19 Nm, which is equivalent to a M w 7.1 earthquake. If a surface rupture propagated further west and broke through all four segments (i.e., eastern transfer, central, western transfer, and western) with similar average dipslip displacements of ∼0.5-1 m, the event would have been associated with a minimum seismic moment release of M 0 = 7.12·10 19 Nm (M w 7.2). Assuming the western and central segments were indeed joined by a breaching hard link (e.g., Arrowsmith and Strecker, 1999) and assuming displacements based on E2 (D ave-obs = 2.5 m), we estimate a moment magnitude of M 0 = 7.76·10 19 Nm (M w 7.3) for a 35-km-long rupture (central segment), M 0 = 9.04·10 19 Nm (M w 7.3) for the central segment and adjacent transfer zones (∼65 km), and M 0 = 11.70·10 19 Nm (M w 7.4) during a propagation into the western segment (>65 km). In summary, considering our interpretation of full-segment length activation (≥35 km) during E5, E4, and E2, we estimate a mean recurrence interval of ∼1.9 kyr for earthquakes with magnitudes of M w ≥ 7, but not larger than M w 7.4 along the PFT N .
In light of the well-documented geomorphologically defined uplift pattern of the Qt 3 terrace within the central Trans-Alai Range (cPFT) versus the nonuniform distribution of fault-bounded terrace terminations at other mountain-front sectors (west and east of cPFT), Arrowsmith and Strecker (1999) suggested that the central segment of the PFT behaves independently. Measured vertical separation (VS) along the cPFT is asymmetric and increases in westward direction (Figure 9), pointing towards a mechanical interaction between neighboring faults as has been suggested by numerical analyses of such settings (e.g., Bergen & Shaw, 2010;Willemse, 1997). Reported rupture-propagation patterns during the 1978 M w 6.6 Zaalai earthquake (Nikonov et al., 1983) emphasize the complexity of ruptures in the vicinity of segment boundaries along the cPFT and the potential segment interaction during ruptures. Willemse et al. (1996) furthermore suggested that such interaction can lead to significant slip-to-length ratio variations as well as off-center location of maximum slip, which could explain the seemingly partial ruptures in the western part of the cPFT (E1 and E3). More recent studies by  further show that the largest slip during an earthquake systematically occurs on one half of the fault (asymmetric), i.e., the most mature fault section. It is not clear if these partial ruptures represent the ends of longer ruptures which have overcome segment boundaries (Manighetti et al., 2005(Manighetti et al., , 2007Wesnousky, 2008). A full-segment-length activation with possible ruptures along adjacent segments could imply an advanced fault maturity, which has important implications for future scenarios of fault segmentation in this setting. We cannot rule out that a more mature, fully linked thrust fault connects the present-day western, western transfer zone, as well as the central and eastern transfer segments at depth and that this fault might be capable of generating events with M w ≥ 7.2.
Despite these caveats, our paleo-magnitude estimates for the past 7 ka do not exceed M w 7.2 (or maximum M w 7.4, if considering E2 measurements). Li et al. (2019) proposed similar maximum moment magnitudes of M w ≤ 7.1 for future earthquakes in the eastern half of the PFT in the Chinese Tarim Basin, but suggested much larger earthquakes for the PFT in the Alai Valley, an interpretation that is not supported by our local cPFT study. The notion of an extensive interface at depth along which even larger ruptures could occur is a simple, but reasonable suggestion when considering the substantial geologic and geophysical evidence for intracontinental subduction and large-scale convergence along the PTS (e.g., Burtman and Molnar, 1993). However, if this system is characterized by earthquakes with magnitudes M w ≤ 7.4, and megathrust events do not occur, then the Pamir mountain front may only represent a discontinuous, up-dip portion of an active continental subduction zone (e.g., Burtman and Molnar, 1993) where very large earthquakes (M w > 8) would be expected. Instead, this area may indeed have reached a stage of continental collision with a decoupled subducting slab dominated by thin-skinned tectonics as suggested by other studies (Kufner et al., 2016;Sippl, Schurr, Tympel, et al., 2013;Sobel et al., 2013).

Holocene Slip Rate Versus GNSS Shortening Rates
Considering the earthquake chronology since 7 ka (E5-E1) and the associated cumulative average dip-slip displacement of 24.7 ± 4.1 m, we estimate an average Holocene (since 5.3 ka) dip-slip rate for the cPFT of 4.7 ± 1.7 mm/yr (Table 6). This new estimate exceeds previous values of 2.0 to 2.5 mm/yr, but compares favorably with the suggested possible 4 mm/yr since 6 ka determined at the western end of the cPFT (Burtman & Molnar, 1993;Nikonov et al., 1983). Burtman and Molnar (1993) also estimated an approximate N-S convergence rate of ∼3.5 mm/yr along the northern edge of the Pamir based on the relationship between the sum of seismic moment tensors from the AD 1963-1988 earthquake catalogue, an assumed representative shear modulus, and the estimated volume of the regional seismogenic layer (e.g., Kostrov, 1974). However, they suggested that this rate may be an underestimate due to the short time span represented in the earthquake data and the limited magnitude information. Alternatively, a Holocene dip-slip rate of ∼6 mm/yr based on offset terraces at the Syrinadjar River, which crosses the central segment (Arrowsmith and Strecker, 1999), agrees within error with our dip-slip-rate point estimate of 7.8 ± 2.7 mm/yr since ∼5.3 ka at the same location (Profile 8; Table 6). We infer that the VS at Syrinadjar represents the maximum VS along the cPFT (Figure 9), and that the associated slip is a local maximum along the cPFT rather than a minimum as Arrowsmith and Strecker (1999) previously suggested.
GNSS-derived horizontal surface velocities measured south of the PFT, across the Northern Pamir plateau, show that the region is moving north-northwestward with respect to stable Eurasia at a rate of ∼20-25 mm/yr (Ischuk et al., 2013;Metzger et al., 2020;Zubovich et al., 2010Zubovich et al., , 2016. Permanent GNSS stations north of the PFT, however, are moving northward at a rate of ∼10 mm/yr ( Figure 1). Consequently, the difference in horizontal surface velocities between the Northern Pamir plateau and the Alai Valley is ∼10-15 mm/yr. Thus, our revised Holocene shortening rate for the cPFT of ∼4 mm/yr accounts for less than half of the difference in the geodetically-determined, north-south convergence rate measured across the PTS. This discrepancy may be related to a variety of factors, which we discuss in greater detail below in the context of our cPFT results, but also in relation to broader tectonic implications.
Our detailed geomorphic and paleoseismic investigations provide strong evidence for fault partitioning both along and across the PTS. The distribution of cPFT scarps and associated structures suggest that the surface trace is discontinuous with local back thrusts (south of T1, T2 and T3; at T4), bends (at the Minjar River), and more complex steps and lateral fault splays (i.e., the eastern expansion of the cPFT; Figures 2 and 3). This along-strike structural diversity may result in coseismic and/or postseismic offset accommodation across secondary fault branches, as has been observed in several studies that relate fault geometry to coseismic rupture propagation (Biasi & Wesnousky, 2016, 2017, 2021. Furthermore, our analysis of high-resolution satellite imagery and TanDEM-X data reveals the presence of pronounced fault scarps ∼10 km south of the cPFT that displace Neogene conglomerates and Quaternary glacial and mass-movement deposits (PFT S , Figure 2). Nikonov (1988) also reported evidence of numerous earthquakes with ages younger than 10 ka along the PFT S . Local fault-trace diversion along the cPFT, active PTS splay faults, off-fault deformation, fault partitioning, and active blind structures not accounted for may all contribute to regional deformation and help explain the discrepancy between the geologic and geodetic rates. Addressing these possibilities will require additional paleoseismic, geomorphologic, and geodetic investigations.
A recent study of folded river terraces in the active Qilian Shan Mountains along the northeastern margin of the Tibetan Plateau showed that shortening rates based on surface fault displacements may underestimate the total deformation because unrecognized folding in the range interior, changes in the dip of thrust faults at depth that are unaccounted for, and footwall underthrusting can accommodate a significant amount of regional shortening (Wang et al., 2020). Arrowsmith & Strecker (1999) showed that the geomorphic expression of the terrace surfaces, which extend parallel to and along the PFT, do not reveal folding and secondary fractures in the terrace profiles. However, their terrace profiles are less than 7 km in length and may only cover a minimum aperture of localized deformation at the Trans-Alai Range front (i.e., the profiles were too short to capture recent deformation associated with potential folding or other PTS faults). More detailed work is required to better quantify the effects of folding in the hinterland and the geometry of faults at depth on the shortening-rate estimates including our documented rate discrepancy.
Another possible cause of the discrepancy relates to the observed geodetic surface velocity gradient moving from north to south across the PTS (Figure 1). Similar gradients are commonly observed across tectonically active mountain fronts including the Himalayas, the southern Bolivian Sub-Andes, and Taiwan (e.g., Bilham et al., 1997;Brooks et al., 2011;Lindsey et al., 2018;Hsu et al., 2003Hsu et al., , 2009Weiss et al., 2016). These gradients are thought to represent the surface manifestation of slipping-to-locked transitions along an underlying décollement and the down-dip accumulation of elastic strain, which is released aseismically and/or incrementally by earthquakes that rupture the shallow portion of the décollement and overlying splay faults both within the mountain belt and at the orogenic wedge front (e.g., Brooks et al., 2011;Weiss et al., 2016). If we apply this conceptual model to the PTS, including the PFT shortening rate deficit, the implication is that strain release is not solely absorbed by the cPFT (PFT N ), but rather that it is shared by other PTS structures located farther south, including primarily the southern PFT (PFT S ) and perhaps even the more internal MPT. Both of these faults were previously thought to be inactive (e.g., Arrowsmith & Strecker, 1999;Sobel et al., 2013). We infer that the PFT S and potentially other PTS structures like the MPT, as well as some potential blind faults north of the PTS within the Alai Valley (Robinson et al., 2015) might accommodate some of the missing 10-15 mm/yr of shortening (as already suggested above). Although beyond the scope of the current study, modeling of the geodetic data would aid in determining the geometry and kinematics of an underlying décollement (e.g., McFarland et al., 2017;Weiss et al., 2016). This is an important next step towards understanding regional strain release and seismic hazard. For example, the geodetic rates do not decrease to 0 mm/yr in the Alai basin north of the PTS. This probably reflects the northward continuation of ongoing Indio-Eurasian shortening in the Tien Shan and Fergana basin (Rizza, Abdrakhmatov, et al., 2019;Thompson et al., 2002); this may also indicate that the regional décollement is only partially locked and that slip can be transferred to blind faults within the Alai basin and/or additional high-angle splay faults farther to the north in the southern Tien Shan (e.g., Burtman & Molnar, 1993). A completely different structural configuration might also be required to explain both the field-based and GNSS observations.
Due to predominantly dextral strike-slip focal mechanisms, comparably lower seismic moment release, and no significant thrust earthquakes along the central PTS during the past decades, Schurr et al. (2014) and Sippl et al. (2014) inferred either significant aseismic creep or fault locking with a long earthquake-recurrence interval.
Our results point toward the latter scenario, considering the estimated recurrence interval of ∼1.9 kyr, combined with a clear indication of thrust dip-slip motion in our trench exposures, and no geomorphological observations along the cPFT for a dominant strike-slip component of motion. Based on these observations, we suggest that the PTS of the Northern Pamir is underlain by a regional south-dipping décollement that is partially locked. This is consistent with the model proposed by , where the upper and lower continental crust is separated by a décollement with the frontal ranges of the Pamir formed by ongoing northward propagation of deformation.

Conclusions
New results from five paleoseismic trench excavations across the central segment of the Pamir Frontal Thrust (PFT) in the Pamir-Tien Shan continental collision zone of Central Asia reveal evidence for five surface-rupturing paleoearthquakes since ∼7 ka, and possibly six events since ∼16 ka. At least three of these events represent major M w ≥ 7.2 earthquakes that likely ruptured the full-segment length (∼35 km) and possibly crossed segment boundaries (∼65 km). However, we find no evidence for ruptures associated with megathrusts that may have exceeded M w > 8. For two apparent partial segment ruptures affecting the western end of the central PFT, we observe a discrepancy between anomalously large amounts of horizontal slip of 2-3 m compared to a relatively short length (<20 km) of the surface-rupture trace. This observation is compatible with major historic earthquakes in this region that did not produce significant surface ruptures. Combined with larger scarp offsets at the western end of the central segment of the cPFT, we infer that segment interaction has occurred including complex (e.g., spillover and multisegment) ruptures, which have important implications for assessing future seismogenic fault behavior and suggest the possibility of potentially larger-magnitude earthquakes in this region. We estimate an average dip-slip rate of 4.7 ± 1.7 mm/yr (equivalent to 4.1 ± 1.5 mm/yr horizontal shortening rate) for the past ∼5 ka, and suggest that the shortening along the central PFT does not account for all of the present-day, geodetically derived shortening across the northern Pamir. Rather, strain is likely released intermittently along additional internal faults of the Pamir Thrust System and/or folds that have less pronounced geomorphic expressions than the central PFT. The fairly rapid decrease in GNSS-derived horizontal surface velocities across the mountain front suggest these structures may root within a partially locked regional décollement underlying the northern Pamir although additional modeling is required to better understand the associated fault geometry and kinematics and the potential influence on strain accumulation and release across the region.

Data Availability Statement
All TanDEM-X data were kindly provided by the German Aerospace Center (DLR) for scientific use. In this work, generated high-resolution DSM raster data and photomosaics of the trench exposures are archived and freely available from OpenTopography (https://doi.org/10.5069/G9KP80BM, G9QF8R2Z, G900009Z, G93R0R2F6) and figshare (https://doi.org/10.6084/m9.figshare.16670902.v2), respectively. All other data used in this study are provided in the main article and Supporting Information S1.