Timing of exotic, far-travelled boulder emplacement and paleo- outburst flooding in the central Himalaya

Timing of exotic, far-travelled boulder emplacement and paleooutburst flooding in the central Himalaya Marius L. Huber1,2, Maarten Lupker1, Sean F. Gallen3, Marcus Christl4, Ananta P. Gajurel5 1Geological Institute, Department of Earth Sciences, ETH Zurich, Zurich 8092, Switzerland 2now at: Université de Lorraine, CNRS, CRPG, F-54000 Nancy, 5 3Department of Geosciences, Colorado State University, Fort Collins, Colorado 80523, USA 4Laboratory of Ion Beam Physics (LIP), Department of Physics, ETH Zurich, Zurich 8093, Switzerland 5Department of Geology, Tribhuvan University, Ghantaghar, Kathmandu, Nepal Correspondence to: Marius L. Huber (marius.huber@univ-lorraine.fr)

typical monsoonal discharges are too low to mobilize large grain sizes that cover channel beds thereby setting the pace of landscape evolution (Cook et al., 2018). The timing, frequency and magnitude of these events as well as their impact on landscape evolution as discontinuous, singular events is however still poorly understood and documented, emphasizing the need to bridge the gap between modern observations and the long-term evolution of landscapes.

70
To better understand catastrophic erosion and mass transport events as well as their potential impact on landscape evolution and related hazard, we focused on the occurrence, provenance, and mechanisms and timing of large boulder emplacement in central Himalayan river valleys. In the studied valleys, numerous large boulders of ca. 10 m in diameter and more, are found in or near the present-day channel beds. The lithology of many of these large boulders differs from those present on the adjacent hillslopes but are the same as geologic units located 10's of kilometres upstream. Their elevation, well below the last glacial 75 maximum ice extent (e.g. Owen and Benn, 2005), excludes glacial transport. The exact transport mechanisms of such exceptionally large grain sizes remain unknown and may be linked to reoccurring catastrophic events such as LOFs. In this contribution, we test this hypothesis using the boulder geometry and channel hydraulic geometry to estimate paleo-discharges required for their mobilization and use cosmogenic nuclide exposure dating to constrain their emplacement age. We discuss our findings in the context of landscape evolution and natural hazards in the Himalaya and point to possible directions for 80 future research.

Study area
The central Himalaya of Nepal accommodate ca. 18 mm/yr of the convergence between India and Eurasia (Ader et al., 2012) that translates into high tectonic uplift rates reaching 10 mm/yr (Bilham et al., 1997;Lavé and Avouac, 2001) and a dramatic topographic gradient where elevations rise from about 200 meters above sea level at the range front to >8000 m of elevation 85 across a horizontal distance of ~150 km. Great earthquakes (Mw ≥ 8.0) along the central Himalayan front happen irregularly with estimated reoccurrence intervals of less than 750 to 1000 years (e.g. Bollinger et al., 2014;Sapkota et al., 2013).
The Trishuli and Sunkoshi rivers are the targets of this investigation and are located in central Nepal, northwest and northeast 100 of Kathmandu, respectively. They span the Greater Himalayan range from the arid high-elevation Tibetan Plateau to the Himalayan range front and are tributaries of the Narayani and Kosi Rivers, respectively. Both catchments are separated by a main drainage divide in their headwaters and the moderately-sized Melamchi Khola drainage basin at lower elevation (Fig. 1).
The Indian summer monsoon, with highly seasonal distributed rainfall, affects both catchments with peak rainfall from June to September (Bookhagen and Burbank, 2010) that is marked by high river discharge and sediment transport (Andermann et 105 al., 2012).
LOF events have been reported for both river-reaches. The Trishuli has been affected by the breach of lake Longda-Cho in Tibet during August 1964 (ICIMOD, 2011), but little is known about the downstream impact of that event. The studied Sunkoshi reach has been affected by repeated LOF events in 1935, 1964, 1981(Shresta et al., 2010ICIMOD , 2011;110 Cook et al., 2018) and the last event of 2016 was closely monitored, revealing the impact of the GLOF, originating from the breach of a Tibetan proglacial lake, on the channel bed morphology (Cook et al., 2018). The upstream catchments of both river-reaches are heavily glaciated and a number of potentially hazardous lakes have been identified (Maharjan et al., 2018) so that it is likely that LOF activity extended in the past.

Sampling
Fieldwork took place during two field campaigns in May and October/November of 2016 along the Trishuli and Sunkoshi main trunk river channels. In the field, boulders visibly larger than the surrounding overall bedload sediment grain-sizes, were sampled for surface exposure dating and boulder provenance analysis ( Figure 2, Table 1, Supplement, S1). Samples consisted of ca. 1 to 2 kg of fresh rock from the top of the boulders. Fractured or weathered surfaces on boulders were avoided to 120 minimize potential complications associated with post-depositional erosion. Boulders that exhibited clear evidence of postdepositional movement were not sampled. For large boulders that showed "pristine" surfaces and no evidence of postdepositional movement, flat surfaces were sampled in areas that minimized chances for burial by sediment or vegetation (e.g. the top of the boulder) ( Figure 2). Rock material was removed using a chisel or a blade saw and the average sample thickness was recorded (S1 and S3). Topographic shielding was measured in the field with a laser range finder, neglecting vegetation. 125

Paleo-hydrologic discharge estimation
Paleo-discharges necessary to transport large grain sizes were computed based on boulder sizes and channel hydraulic geometry. Boulder diameter was determined from high-resolution satellite imagery (Google Earth). A succession of imagery https://doi.org/10.5194/esurf-2020-17 Preprint. Discussion started: 20 March 2020 c Author(s) 2020. CC BY 4.0 License. of the same object covering several years was used to calculate an arithmetic mean for boulder diameter to minimize error due to imagery distortion (Table 1). Rock density was estimated based on typical densities reported for the sampled lithologies. 130 Hydraulic geometry measurements included valley cross-sections and slopes, which are needed to calculate flow channel characteristics during emplacement and were obtained by digitizing 1:25 000 and 1:50 000 scale maps of the Survey Department of Nepal (S2). These estimates were made using straight river reaches directly upstream of the studied boulders.
Three different approaches were adopted to derive peak discharge values from the boulders surveyed in the Trishuli and 135 Sunkoshi River valleys: i) the empirical approach after Costa (1983), based on a literature compilation describing average flow velocity that led to transport of large boulders; ii) Clarke (1996)'s force balance approach which computes cross-sectional averaged velocities based on methods described by Costa (1983) and Bradley and Mears (1980); iii) the approach of Alexander and Cooker (2016) that allows flow velocities to be estimated from a theoretical force balance equation and taking into account an impulsive force to account for "inherently unsteady" and "non-uniform" flow. All three approaches take advantage of the 140 incipient motion principle and compute velocities for turbulent, Newtonian fluid flow when a sediment grain of given intermediate grain diameter initiates motion on the stream's bed. The Gauckler-Manning formula (Gauckler, 1867;Manning, 1891) was then used to estimate peak discharge from the peak flow velocity determined above and the valley cross-sectional geometry using the numerical optimization scheme of Rosenwinkel et al. (2017). Peak discharges and maximum flood heights were calculated from flow velocities using the average of two valley cross-sections and bed slope for each boulder considered 145 (S2). Access to Matlab script is detailed in the code availability section below.

Surface exposure dating
The samples were prepared in the laboratories of the Geological Institute in the Earth Science Department at ETH-Zürich.
Bedrock was crushed and sieved to 250-1000 µm. Quartz was isolated by magnetic separation followed by five to seven sequential leaching steps with H2SiF6 and HCl (2/3 -1/3 by volume). Meteoric beryllium (Be) was removed by three additional 150 leachings with HF to dissolve ~10% of the sample mass at each step (Lupker et al. 2012). Pure quartz was dissolved after the addition of ca. 250 µg of 9 Be carrier solution and Be was isolated using sequential ion exchange column chromatography.
Beryllium-10 ( 10 Be) concentrations were measured for each sample at the Laboratory of Ion Beam Physics of ETH Zurich using the 0.5 MV TANDY AMS facility (Christl et al., 2013). Cosmogenic exposure ages were computed from the blank corrected 10 Be/ 9 Be ratios using the online "Cosmic Ray Exposure program" (CREp) calculator (http://crep.crpg.cnrs-nancy.fr, 155 Martin et al., 2017) using a global production rate 4.08 ± 0.23 at/g/yr, the ERA40 standard atmosphere and a correction for geomagnetic dipole moment changes. See Supplement (S3) for more detailed information on sample preparation and exposure age calculation.

Field survey and boulder lithology 160
We surveyed and sampled a total of 16 boulders: ten boulders in the Trishuli River valley, six in the Sunkoshi River valley  Table 1). Only one boulder sampled and surveyed was not exposed in the river channel but was embedded in a terrace deposit just upstream of Balephi Khola's confluence with the Sunkoshi main trunk ( Figure 2). The lithology of these Sunkoshi boulders consists of a variety of gneisses (Figure 2, Table 1); some show big porphyric felspar laths deformed to augen structures (NEQ/162 79, 80 and 98). Boulder sample NEQ/162 03, the largest boulder surveyed in this study, has an augen-fabric that 180 looks more homogenous and slightly less deformed than the other samples ( Figure 2, Table 1). The lithology of all boulders sampled in the Sunkoshi is different from the schists found in bedrock on the adjacent hillslopes, and they are thus, not locally derived but allochthonous.

Paleo-hydrologic discharge estimation
The average flow velocity values calculated after Costa's (1983) empirical approach are only dependent on the intermediate 185 boulder diameter and range between 7.8 and 16.7 m/s for the surveyed boulders (S2). Flow velocities calculated using a fluid density of 1500 kg/m 3 range between 6.7 and 17.3 m/s using the Clarke (1996) method and between 4.3 and 11.2 m/s using Alexander and Cooker (2016) (S2). Despite minor differences, these three approaches produce similar results ( Figure 3). Paleodischarge derived from these velocities are shown in Figure 3 (and table in S2). Values of Costa (1983), Clarke (1996) and https://doi.org/10.5194/esurf-2020-17 Preprint. Discussion started: 20 March 2020 c Author(s) 2020. CC BY 4.0 License. Alexander and Cooker (2016) derived peak discharges range from 3.63 x 10 3 to 1.97 x 10 5 m3/s for boulders in the Trishuli 190 and 1.34 x 10 3 to 1.03 x 10 5 m 3 /s for the boulders in the Sunkoshi.

Boulder exposure ages
10 Be cosmogenic exposure ages of boulders in the Trishuli valley have ages ranging between 2.81 and 5.84 ka BP ( Figure 4; Table 1) with overall 1σ uncertainties ranging from 0.35 to 0.67 ka and mode of the summed probability density functions around 4.5 kyrs BP. One of the ten boulders, sampled upstream of Betrawati (NEQ/162 59; Figure 4; Table 1; S3) shows a 195 significantly younger exposure age of 1.06 ± 0.29 ka BP, suggesting that the age might suffer from the effects of erosion or more likely burial of the sampled flat top-surface in the terrace deposit during an earlier stage in time and thus does not represent the true emplacement age.
The Sunkoshi exposure ages can be assigned to three different age groups ( Table 1; S3).

Boulder provenances and travel distances
The schist and phyllite boulder lithologies in the Trishuli valley are associated with the Paleoproterozoic lower Nuwakot group of the Lesser Himalayan sequence and more precisely from the Kuncha formation or the overlying Dangdagaon phyllites that 210 could be found locally in the valley (Stöcklin, 1980;Upreti, 1999). The Trishuli phyllite and schist boulders might, therefore, have travelled only short distances by toppling down from adjacent hillslopes. However, similar to the gneiss boulders, they are in most cases sub-angular to sub-rounded with crescentic abrasion marks suggesting substantial fluvial transport distances (S1). Diagnostic mineralogy and fabric of the other surveyed ortho-gneiss boulders in the Trishuli catchment are not present in the Lesser Himalayan sequence and, therefore, must originate from areas upstream of the MCT (Figure 1). This observation 215 requires minimum transport distances of approximately 22 km to 46 km depending on the present boulder location (Table 1).
A variety of gneiss boulder lithologies are found in the Sunkoshi/Balephi Khola catchment. Ortho-gneiss and augen-gneiss lithologies among boulders around Balephi are of higher metamorphic grade (amphibolite to granulite facies) and must originate upstream to the north across the MCT from Higher Himalayan crystalline rocks present in these areas or from gneiss 220 to be found just below the MCT in the Lesser Himalayan footwall ("Ulleri type augen-gneiss" for NEQ/162 03, see S1) (Shrestha et al., 1986;Amatya and Jnawali, 1994;Dhital, 2015). This analysis suggests minimum transport distances of ca. 11 to 17 km ( Figures 1, Table 1). While the surveyed boulders are mostly located bellow the Sunkoshi/Balephi Khola confluence and could therefore have been transported by both rivers, field observations show an abundance of boulders present in the bed of the Balephi Khola. 225 As noted above, boulders in both valleys are well below the extent of alpine glaciers in the modern or during the last glacial maximum (e.g. Owen and Benn, 2005; Figures 1 and 2). With glacial transport excluded at these low elevations, this analysis indicates that the mobilization and transport of large grain-sizes occurred in central Himalayan river valleys over long distances (>10 km) through fluvial processes. 230

Paleo-discharge estimates
The range of discharge estimates derived in this work is a first-order estimate and carries important assumptions. The sediment concentration of the flow directly influences transport capacity through flow density and flow mechanics (e.g. Pierson and Costa 1987). In hyper-concentrated flows with 40 to 70 wt. % sediment entrained, non-Newtonian, plastic fluid behaviour and laminar flow can arise due to the establishment of shear strength in the fluid material (e.g. Pierson and Costa 1987). However, 235 if the amount of sediment entrainment remains at the lower end of this "hyper-concentrated" range, flow mechanics are still adequately approximated by Newtonian, turbulent flow of a "clear" waterflood (Costa, 1984;Pierson and Costa, 1987;Pierson, 2005;Wang et al., 2009;Hungr et al., 2014) as was assumed here. Other uncertainty arises from the extraction of valley crosssectional profiles using topographic maps (see S2 for more details). Terrace flats and channel widths are only crudely represented and do not account for past channel morphologies at the time of the floods. Since detailed riverbed morphology is 240 required for the hydraulic discharge calculation, additional uncertainty arises from the resolution of the data used here and the necessity of using the modern channel geometry for these calculations. We hypothesize that these uncertainties are the main reason for the discrepancies between paleo-peak-discharge estimates for boulders from a similar age range that were presumably transported during a single event (Figure 3). First-order discharge estimates for boulder emplacement therefore broadly range from ca. 10 3 to 10 5 m 3 /s. 245 To place our results in the context of previous studies, we compare our discharge to those from the literature and historical records. Cenderelli and Wohl (2001) and Fort et al. (2010) compared seasonal high flow floods (SHFFs) with discharges of recent GLOF and LLOF events and they appeared to be at least one order of magnitude higher than monsoonal precipitation peak discharges in central Himalayan catchments (Kali Gandaki and Mount Everest region). Peak discharges reaching 10 5 250 m 3 /s have been documented or suggested for a few historical events in the Himalaya mainly associated to LLOF such as the Great Indus flood of 1841 (Mason, 1929;Shroder et al., 1991), the great outburst in April 2000 in the Tibetan Yigong Zangbo River (Shang et al., 2003;Delaney and Evans, 2015;Turzewski et al., 2019), and the large LLOFs at Dadu River and Yalong River in the years 1786 and 1967 in Sichuan province, China (Dai et al., 2005;Runqiu, 2009). These events are, however, rarely observed even though there is sedimentological evidence that large-scale LLOF events happened regularly throughout 255 the Holocene within the same catchments (e.g. Hewitt et al., 2011;Wasson et al. 2013). To our knowledge, GLOF discharge estimates of documented events in the Himalaya reach ca. 10 4 m 3 /s (e.g. Vuichard and Zimmermann, 1987;Hewitt, 1982;Xu, 1988;Yamada and Sharma, 1993;ICIMOD et al., 2011;Cook et al., 2018) with Holocene reconstructed discharge estimates that exceeded 10 5 m 3 /s such as for example the Tsangpo River gorge outburst reconstructed by Montgomery et al. (2004). Comparison of the estimated paleo-discharges, which range between ca. 10 3 to 10 5 m 3 /s, with flow duration curves of the hydrological stations, shows that flows needed to mobilize the studied boulders generally exceed the largest flows on record ( Figure 3 and 5). Discharge records may also include LOF events and, therefore, not only reflect monsoonal precipitation 270 driven discharge. This finding suggests that typical monsoonal floods are unlikely to have been responsible for boulder emplacement and that LOFs are the most likely events. The discharge magnitude of these LOFs is on par with historically recorded LOFs in the Himalayan range.

Timing of flood events
The exposure ages of the Trishuli boulders all cluster around ca. 5 ka BP except boulder samples NEQ/162 44 and NEQ/162 275 59, which are considered as outliers recording post-depositional erosion or shielding (Figure 4, Table 1). Boulder NEQ/162 59, which shows by far the youngest age, is still buried to a large extent and its top surface is only slightly elevated from the surrounding ground surface. There is an offset in age (up to 2 kyrs) between exposure ages of boulders currently located on the fill terrace near the town of Betrawati and boulder ages in the channel or further downstream in the wider valley reach. We interpret the younger ages of these "Trishuli upstream" boulders lying next to the riverbed as the expression of post-280 depositional shielding and burial by sediments followed by excavation through river incision in the fill terrace (Figure 2 and   4; Table 1; S1). Boulder NEQ/162 67's lithology varies in fabric from the other schist and phyllites of surveyed boulders in the Trishuli and could potentially be affected by inheritance (Table 1). Overall the exposure data and interpretation suggest that boulder transport and emplacement occurred in a single flood event in the Trishuli that happened around 4.9 ± 0.3 ka BP (arithmetic mean with 1σ error of boulder samples NEQ/162 45, 46, 47 and 66; Figure 4; Table 1). 285 A minimum of two emplacement events are noted in the exposure age data from the Sunkoshi (Figure 4; Table 1). NEQ/161 03 dates back to the late Pleistocene and has the largest maximum diameter determined among surveyed boulders in this study (Figures 2 and 4; Table 1). Boulder NEQ/162 80 yields a younger age and is partly entrained in a fill-terrace deposit ( Figure   4; Table 1; S1). We interpret both NEQ/161 03 and NEQ/162 80 as transported during the same event or series of events 290 (within dating uncertainty) that occurred at ~13 ka. This interpretation suggests that NEQ/162 80 was buried by sediments or suffered surface erosion that resulted in the observed ~2 ka age offset (Figure 4; Table 1).
Boulders NEQ/161 01 and NEQ/162 98 show strikingly similar mid-Holocene exposure ages (Figure 4; Table 1). We suppose that these boulders were already exposed or were excavated shortly after emplacement by monsoonal flows from a deposit that once buried the whole tributary junction around the town of Balephi ( Figure 4; Table 1; S1). Although the older exposure age 295 of 6.23 ± 0.92 ka BP for boulder NEQ/162 79 is associated with higher uncertainty, it can still be considered synchronous with  Table 1). Importantly, this second flooding event at ~5 ka was not as large as the preceding event at ~13 ka as it did not mobilize boulders > 10 m in diameter (Table 1).
Trishuli boulder ages (without NEQ/162 59) and two ages among the Sunkoshi boulders show a clustering around 5 ka BP across a major drainage divide in the central Himalaya (Figure 1 and 4). At least two older ages in the Sunkoshi indicate 305 flooding during the late Pleistocene ( Figure 4); similar ages were not observed in the Trishuli valley ( Figure 4). The events that emplaced these large-sized boulders, ca. 10 m in diameter or more, are rare since they represent the remnants of the last largest floods that were not re-mobilized by subsequent floods. The resolution of 10 Be exposure ages and uncertainty on the state of dated boulder surfaces does not definitely point toward a single event that affected both catchments. Boulder emplacement could potentially be the result of a series of events that occurred in a short time period of a few hundred years 310 within the range of dating uncertainty (Figure 4). However, the assumption that the boulder emplacements only record a flood of maximum magnitude (see above), the rarity of emplacement, and the absence of obvious stratification in the fill terraces ( Figure 2; S1) point toward a single mid-Holocene emplacement event in each valley.
The timing of the Trishuli and Sunkhosi events found in this study is comparable to another large event responsible for the 315 extensive fill terraces found in the central Himalayan Marsyandi river valley about 90 km to the west of the Trishuli valley (Pratt-Sitaula et al., 2004). The authors of that study found large terraces ("Middle terraces") composed of heterolithic conglomerates and boulders with lithologic evidence for transport distances of over 40 km. These fill terraces were interpreted as the result of a single massive earthquake-triggered landslide event that caused a catastrophic debris flow. Recalibrated radiocarbon ages of that infill date back to 4.6 to 5.1 ka BP (Yamanaka, 1982; ages recalculated with OxCal online calibrator, 320 Bronk-Ramsay, 2013). The ages determined for the Trishuli and Sunkhosi boulders are however older compared to the exposure age of another studied large, far-travelled, boulder capping the Pokhara formation, further West in Nepal (ca. 1680 C.E.) (Fort et al., 1987;Schwanghart et al., 2016). The emplacement of this later boulder was linked to a historical 1681 C.E. earthquake even though the magnitude and epicentre of this later event remain poorly constrained (Chaulagain et al., 2018).

Triggers of Holocene catastrophic LOFs 325
Our results demonstrate that high magnitude peak discharge (10 3 to 10 5 m 3 /s) by lake outburst events (LOFs) are needed to explain the emplacement of large boulders in the Trishuli and Sunkoshi drainage catchments and that these events were clustered in time at around 5 kyrs in at least two distinct valleys (Figures 3, 4 and 5). To attribute these events to LLOFs or GLOFs requires evaluating our data in light of typical earthquake recurrence times and regional climate variability.
Earthquakes and their associated co-seimsic landslides provide a mechanism that could synchronously emplace large landslide 330 dams in main valleys across water divides, exposing downstream reaches to LLOF events. Climate variability, through its modulation of glacier extent and proglacial lake volumes, can also be invoked as a potential trigger of large GLOF events in multiple valleys during a short period of time as observed in our data.
The timing of the boulder emplacement in the Trishuli and Sunkoshi/Balephi Kola suggests that these high magnitude flows 335 last occurred ca. 5 kyrs ago. Older, larger boulders in the Sunkoshi/Balephi Kola suggest flows of even higher magnitude affecting trans-Himalayan valleys in the late-Pleistocene. While the dating precision available cannot strictly point towards a synchronous emplacement amongst the studied valleys (and the Marsyangdi, Pratt-Sitaula et al., 2004), major to great earthquakes (≥ Mw 7.0 and 8.0 respectively) periodically rupture large parts of the Main Himalayan Thrust resulting in surface ground shaking over distance > 100s of km along the Himalayan range (e.g. Bilham, 2019). These earthquakes can trigger a 340 large surface response with intense coseismic landsliding as was observed during the 2015 Mw 7.8 Gorkha earthquake (Roback et al., 2018). Although the 2015 earthquake did not result in significant valley blocking due to landsliding that can result in large LLOFs, the total volume of co-seismic landsliding scales with earthquake magnitude (Marc et al., 2016) and hence LLOFs might be more likely following earthquakes with magnitudes larger than Gorkha. Great earthquakes (≥ Mw 8.0) in the Himalaya susceptible to trigger large volumes of landsliding beyond what was observed during Gorkha have estimated 345 recurrence times of 750 to 1000 years (Bollinger et al., 2014;Sapkota et al., 2013). If great earthquakes had a large likelihood to trigger widespread LLOFs (as would be required to explain the emplacement of boulders and valley fills in two to three valleys), we would expect the central Himalayan valleys to be strewn with boulders of younger and possibly more diverse ages owing to the geologically frequent recurrence interval of such events. While earthquake-triggered LLOFs cannot be entirely excluded we do not favour this explanation for the emplacement of the boulders that are the focus of this investigation. 350 Climate change may be another LOF trigger through its modulation of glacier dynamics and the creation of GLOF-prone proglacial lakes. Multiple terrestrial records for the monsoon-influenced Himalaya and the Indian subcontinent indicate a wet https://doi.org/10.5194/esurf-2020-17 Preprint. Discussion started: 20 March 2020 c Author(s) 2020. CC BY 4.0 License. and strong monsoon phase during the Early Holocene Climatic Optimum (EHCO), followed by a dry phase between 5 ka to 4 ka BP. This dry phase is recorded amongst other proxy records, by a compilation of Indian monsoon records (Herzschuh, 355 2006); the transition to more arid conditions in central India that lead to a vegetation transition towards C4 grasses as recorded in the Lonar Crater Lake sediments (Sarkar et al., 2015); changes in moisture sources as indicated by a drop in δ 18 O of the Guliya ice cap on the Tibetan plateau (Thompson et al., 1997) (Figure 6). In response to the decrease in monsoonal precipitation, a number of preserved glacial landforms in the central Himalaya show a phase of glacial retreat around 5 ka BP (Abramowski et al. 2003;Finkel et al. 2003;Gayer et al. 2006;Schaefer et al. 2008;Pratt-Sitaula et al. 2011; Figure 6). This 360 retreat has not been observed in all studied valleys but glacial moraines dated to around 5 ka are reported in the Langtang valley in the upstream Trishuli catchment and in the Macha Khola, Manaslu massif ( 10 Be ages; Abramowski et al. 2003), in the Khumbu Himal (Everest region) with the Thuklha stage ( 10 Be ages; Finkel et al., 2003), as well as 3 He and 10 Be ages derived by Gayer et al. (2006) for moraines in the upper Mailun valley, Trishuli catchment suggesting a regional glacial response to aridification ( Figure 6). Although the relation between GLOF frequency and ongoing climate change remains 365 unclear (Harrison et al., 2018;Veh et al., 2019), past large phases of glacier retreat have been suggested to increase GLOF frequency (e.g. Walder and Costa, 1996;Clague and Evans, 2000;Wohl, 2013). Moraine dams form especially if fast glacial retreat follows earlier advances (Korup and Tweed, 2007). The documented phase of glacial retreat at ca. 5 ka BP, therefore, represents a possible triggering mechanism for the emplacement of the studied boulders across river drainage divides as it may have resulted in the more-wide spread occurrence of glacier lakes prone to GLOFs. Climatic forcing of GLOFs is therefore a 370 suitable explanation for the emplacement of far-travelled fills in other valleys in the central Himalaya (e.g. Pratt-Sitaula et al., 2004) and agrees with the provenance of a number of gneiss boulders originating from the glaciated Higher Himalayan crystalline. It should also be noted that abrupt climatic shifts such as the one following the EHCO occurred overall less frequently since the LGM compared to the recurrence time of great Earthquakes.

375
The older exposure ages derived in the Sunkoshi catchment, including the largest boulder surveyed in this study (NEQ/161 03 with 29.9 m intermediate diameter, Figure 2 and 4; Table 1), could be attributed to an LGM or post-LGM glacial retreat that lead to events preserved in the channel till today. If correct, our dataset reflects that long-term climate modulated LOFs can alter Himalayan Valleys on 10 3 to 10 4 yr timescales.

Implications for erosion and geohazards in the Himalaya 380
This study found evidence for high magnitude discharge events in the form of outburst flooding in central Himalayan river valleys. A record of exceptional flooding is preserved not only in the form of large boulders but also in the form of large alluvial fills in Himalayan valleys. As mentioned before (Wohl, 2013;Cook et al., 2018), LOFs may be responsible for channel incision and lateral erosion in the upstream reaches of mountainous rivers but may also lead to aggradation and valley fills in the lower reaches with long-lasting impacts on river morphology. LOFs may, therefore, exert a strong control on the timing 385 and locus of fluvial incision. In this study, we suggest that the largest grain-sizes that can be found in the two studied Himalayan valleys are related to large GLOF events that followed a time of regional glacial retreat after the EHCO and at the close of the Last Glacial Maximum. This interpretation argues for a climatic control on incision rates and sediment export in the Himalaya.
Furthermore, it leads to the counter-intuitive notion that the erosional engine might be most efficient during, or at the onset of, drier climatic periods as they represent periods of amplified occurrence of large, bedload mobilizing LOF events. Additionally, 390 the fact that these boulders are preserved in the modern-day channel and adjacent fill deposits suggests that channel incision halted at these locations for periods of time subsequent to extreme channel aggradation. These extreme outburst events could, therefore, be key in regulating the episodic nature of fluvial incision (e.g. Pratt-Sitaula et al., 2004;Dortch et al., 2011) and have a long-lasting effect on sediment fluxes exported downstream. Further work on the frequency and magnitude of these events is required to improve our understanding of their role in long-term channel morphology and incision. This future work 395 should also address the relative importance of these events with respect to other catastrophic events in the Himalaya as we note that GLOFs are not the only mechanism that allow for long distance boulder transport and valley aggradation. This is illustrated by the large Pokhara valley fill deposit and boulder emplacement in central Nepal which was attributed to a series of earthquakes (Schwanghart et al., 2016;Stolle et al., 2017) and had long-term impact on valley morphology and sediment fluxes (Stolle et al., 2019). 400 In the face of anthropogenic climate change, increasing land use by population growth and ongoing investments into hydropower in central Himalayan valleys, GLOF hazard and risk need to be better quantified (e.g. ICIMOD, 2011; Schwanghart et al., 2016). The recent acceleration of ice loss across the Himalayas (Maurer et al., 2019) has led to an increase in proglacial lake water volumes (Wang et al., 2015;Nie et al., 2017) but, surprisingly, has not so far lead to an observable 405 increase in GLOF frequencies (Veh et al., 2019). How these trends will evolve in the near future remains unknown. In this study, we suggest the occurrence of large GLOF events during the phases of glacier retreat after the LGM and at the mid-Holocene climatic transition. The exact timing of these GLOF events in comparison to the climate forcing remains elusive, but it may foreshadow possible future scenarios with significant threats to downstream populations and infrastructure. It should, however, be noted that the volume of water that can be stored in proglacial lakes in the upper reaches of the Himalayan 410 rivers could be significantly lower than volumes stored in proglacial lakes during glacial retreat from larger glacial extent than present. If correct, expected future GLOFs magnitudes could be lower than during the last deglaciation and the Holocene.

Summary and conclusion
We provide field observations of large boulders, ca. 10 m in diameter or more, that show lithologic evidence for travel distances of over 10s of kilometres in two central Himalayan valleys, the Sunkoshi and Trishuli. These boulders are well below the LGM 415 ice extent and therefore require exceptional flows to explain the long travel distances. Using the boulder's estimated masses and an assumption on flow density and regime, we estimated that discharges of 10 3 to 10 5 m 3 /s are required to mobilize these grain sizes. Such discharges are, however, higher than typical monsoonal floods as constrained by hydrological data from nearby locations and therefore suggest that boulder emplacement occurred through high magnitude and catastrophic lake outburst floods. 420 Cosmogenic nuclide exposure dating ( 10 Be) shows emplacement ages of between 4.7 -5.2 ka BP in the Trishuli valley and between 4.6 -5.4 ka BP and 12.2 -14.2 ka BP in the Sunkoshi valley. Our data suggest that the younger events are correlated across water divides and possibly also with other deposits in the central Himalaya (Marsyandi valley, Pratt-Sitaula et al., 2004).
The trigger for these LOF events remains difficult to constrain owing to the limits in dating precision that do not allow to 425 clearly identify one single event across drainage divide or multiple events occurring within a short amount of time (ca. 1 kyr).
Landslide lake outburst floods following a large seismic event and extensive co-seismic landsliding remains a possible explanation for boulder transport and emplacement; however, in this case, we argue against this interpretation because the typical recurrence time for great earthquakes in the Himalaya is on the order of 1 kyrs, relative to the evidence of mid-Holocene and late Pleistocene LOFs in this study. In the case of coseismic LLOF events, we would expect boulder deposits to have 430 younger and/or more widespread ages. Alternatively, glacier lake outburst floods are also able to generate the required discharges. Terrestrial climatic records show that the Indian continent and the Himalaya experience significant aridification trend between 4 to 5 ka BP that follows a period of intense monsoon during the early Holocene climatic optimum. Glacier reconstructions show evidence for regional glacier retreat during that period as a response to a weaker monsoon. We, therefore, suggest that the observed boulders were most likely mobilized by GLOF events occurring during phases of glacial retreat. 435 These findings and interpretations have important implications regarding the dynamics of channel incision and erosion rates in the Himalaya. The role of these exceptional events as geomorphic agents of landscape evolution has recently been emphasized. We supplement these findings by showing that the studied central Himalayan river channels preserve traces of these large events that occurred 4 to 14 kyrs ago, suggesting a long-lasting impact on sediment dynamics and channel evolution. 440 Furthermore, even though there is currently no evidence for an increase in GLOF frequencies as a response to anthropogenic climate change in the Himalayas, the type of evidence put forward in this work suggests that major phases of glacier retreat may be associated with GLOFs that may, if not be more frequent, at least be of large magnitude.

Acknowledgments
This work was supported by the ETH research grant 15 15-2 and the Alumni-fonds associated with the Geological Institute, 445 Department of Earth Sciences, ETH Zurich. Special thanks to Katie Schide, Shyam Khatiwada, Lena Märki and Akash Acharya for help in the field. Negar Haghipour is thanked for help with the cosmogenic nuclide sample preparation.     18O (‰) Thompson et al. (1997) -32 -25 Sarkar et al. (2015) Age [