The Ulakhan fault surface rupture and the seismicity of the Okhotsk–North America plate boundary

. New ﬁeld work, combined with analysis of high-resolution aerial photographs, digital elevation models, and satellite imagery, has identiﬁed an active fault that is traceable for ∼ 90 km across the Seymchan Basin and is part of the Ulakhan fault system, which is believed to form the Okhotsk–North America plate boundary. Age dating of alluvial fan sediments in a channel system that is disturbed by fault activity suggests the current scarp is a result of a series of large earthquakes ( ≥ M w 7 . 5) that have occurred since 11 . 6 ± 2 . 7 ka. A possible channel feature offset by 62 ± 4 m associated with these sediments yields a slip rate of 5 . 3 ± 1 . 3 mm yr − 1 , in broad agreement with rates suggested from global plate tectonics. Our results clearly identify the Ulakhan fault as the Okhotsk–North America plate boundary and show that tectonic strain release is strongly concentrated on the boundaries of Okhotsk. In light of our results, the likelihood of recurrence of M w 7 . 5 earthquakes is high, suggesting a previously underestimated seismic hazard across the region.

The basins along the scarp showed signs of intermittent flooding, and transport of some alluvial material, including occasional deposits of pebbles up to 5 cm diameter. Field reconnaissance, aerial photographs and satellite imagery suggest that present day drainage runs west northwest into lake Rovnoye, and parallel to the scarp ( figure 3, figure 4 ). It terminates into the series of narrow, elongated depressions encountered in 2011, which form a linear trend oriented 115 • , and extend for a further 2.5 km eastwards. There are approximately 10 of these depressions, with smaller vertical offsets than on the scarp, further west. Still further eastwards, the series of smaller depressions give way to 3 much larger (∼500 m x 100-200 m, and 10 m deep) basins, aligned in the same 115 • orientation, and spread over a distance of ∼ 2.5 km. These appear to be dried out, thermokarst lakes (e.g., Bouchard et al., 2016). These also coincide with the apparently more recently active part of 15 the fan surface, judged by the distinctiveness of the preserved braided channel systems in satellite imagery. Similar depressions also occur in other places on the Buyunda fan surface which are not on the trend of the scarp.
The linear features we have identified are aligned close to parallel to the predicted Okh-NAm linear velocity vectors (Sella et al., 2002;Seno et al., 1996;Apel et al., 2006). There is an especially close match to the present day, REVEL, GPS-based, global euler vector, with magnitude ∼ 6 mmyr −1 . The best fitting euler vector (Seno et al., 1996), based on spreading rates 20 and earthquake focal mechanisms, and hence considered a longer term estimate (up to 3.5 M yr), is slightly more oblique to the trend, and of lower magnitude (∼ 4 mmyr −1 ) (figure 4).
We can thus trace a linear feature across the inner fan surface, from west to east, made up of a distinct fault scarp, a series of narrow, elongated depressions, and a series of dry, thermokarst lakes over a total distance of ∼ 10 km. The linear trend terminates in the modern day Buyunda river channel. Given its linear nature, the sometimes pronounced scarp, and the fact that 25 the trend of the feature fits closely with predicted Okh-NAm motion from global euler vectors Sella et al. (2002), we suggest that this is evidence of a recently active, mostly strike-slip fault, likely to be the Okh-NAm plate boundary, and we will now present arguments for potentially very large earthquakes and ruptures along it.

Buyunda fan surface and hydrology
Although today the Buyunda river is in a braided channel to the east of the Buyunda fan and is actively incising its earlier fan 30 deposits, the fan surface is composed of several generations of earlier braided channel systems representing earlier courses of the Buyunda river. These have complicated, discordant relationships to one another, but in general, they become less distinct and presumably older, in a westerly direction.
One of the most commonly used means to establish fault slip rates in strike-slip regimes are offset markers, such as alluvial channels, terraces or other stable landforms (e.g., Grapes and Wellman, 1988;Hubert-Ferrari et al., 2002;Hetzel et al., 2002;Rodgers and Little, 2006) where they cross a fault. It is often hypothesised that some channels may be offset by a single earthquake and simultaneously abandoned by the stream that flows into them. Under these circumstances, a channel will become a passive marker for the current and all subsequent earthquake offsets. However, there are a number of potential problems with this idea. Firstly, the channel may have already been inactive for other reasons, prior to an offsetting earthquake, 5 and associated ages of channel deposits would therefore not be synchronised with the start of offset motion. This is probably only relevant for cases where single earthquake offsets are being measured. In cases where multiple offsets have occurred, the significance of a time delay between abandonment, and the beginning of offsets will become less as more earthquakes occur.
Secondly, channels may reestablish flow between offset segments in a phenomenon known as dog-legging (Rodgers and Little, 2006), and hence, sedimentary ages within channels will be far younger than the timing of offsetting motion. Dog-legging may 10 also occur where a new channel exploits an existing fault scarp along part of its length, causing a deflection in its course which is unrelated to seismic events. The best tectonic offset markers are generally linear features such as edges of incised terraces (Hetzel et al., 2002) or straight segments of a channel. However, in many cases, associating age determinations of sedimentary features with their offsets is problematic.
The Buyunda fan inner lobe surface is crossed by many channels which are intersected by the fault scarp. In general, 15 these form abandoned, braided systems of similar character to the present day Buyunda river (figure 3). Braiding leads to continuously curved features, which makes identifying tectonic offsets more difficult. The eastern end of the fan in the area of the larger, thermokarst features, nevertheless contains a number of channels with straight segments that cross the fault, often almost perpendicular to it, and none of which appear to be offset. We suggest that this region is the youngest portion of the fan surface, and was active after the last major earthquake on this fault segment, which obliterates traces of the active scarp. This 20 system is bounded on its western edge by a channel with a strongly fault parallel orientation along the scarp, which could be interpreted as a fault offset, but actually appears to mark the edge of this particular generation of deposits.
Further westwards towards the region of the pronounced fault scarp, the braided channels are less distinct and fresh in aerial and satellite imagery (figure 4). The fault scarp builds a ridge which cuts through drainage. To the south and north of the ridge, aerial and multispectral landsat imagery shows saturated zones, and fossil drainage. Between them, the scarp builds a 25 500 m wide "dry" zone (figure 4). Ground saturation in this area is due to outflow from the adjacent Okhotnik river and fan system. The area directly north of the scarp is a topographic low on the edge of the inner lobe of the Buyunda fan. Water presumably percolates into the scarp area, fed from the Okhotnik river. The step in topography due to the scarp offsets the hydraulic gradient, leaving the scarp dry, and the areas to its north and south saturated. The saturated zones highlight a fossil channel system which formed a pre-scarp drainage. The channels with higher permeability sediments, are strongly illuminated 30 in multispectral satellite imagery (figure 4). The fossil channel system drained to the northwest.
At the westernmost end of the scarp and ridge, directly adjacent to lake Rovnoye, there are two small areas of raised topography, visible on the aerial photographs and the DEM, slightly north of the main trend of the fault scarp (figure 4). The southernmost of these forms the main fault scarp at this point as can be seen from fault normal, topographic profiles ( figure   5 Solid Earth Discuss., https://doi.org /10.5194/se-2018-103 Manuscript under review for journal Solid Earth Discussion started: 1 October 2018 c Author(s) 2018. CC BY 4.0 License. 6). The fault thus passes between the two topographic highs, before terminating and resuming a few metres south along the main scarp. The topographic features both have straight, eastern edges, trending to the north. These features are interpreted as channel banks of an earlier, north draining system, the fossil remnants of which can be seen ∼ 500 m further north. The southern side of this channel is also visible to the south of the ridge. The channel banks are cut by the trace of the fault, and offset by ∼ 60 -65 m in a fault parallel direction, which we suggest may be a measure of fault offset in this location. We suggest abandonment of the channel occurred due to the fault scarp and ridge that were formed during earthquakes blocking 5 and shutting down the existing drainage. New drainage developed parallel to the scarp and began to flow into the lake at the scarp's western end, but flows were weak due to drainage reorganisation by the scarp and ridge. Hence little erosion has taken place since scarp formation, leaving the offset markers of the channel edges well preserved to this day.

Buyunda fan age determinations and offset rate estimates
We collected sediment samples from 4 sites along the scarp for age determinations using both optically and infrared stimulated 10 luminescence of quartz and feldspars (OSL and IRSL), as well as organic material for 14 C from one further site (see appendix for details of the method, sampling and laboratory procedures). Samples came from the region directly adjacent to and north of the large scarp (figure 4). Samples 1 and 2 are on both sides of the offset channel feature we have identified, at the westernmost edge of the scarp. Samples 3, 4 and 5 are from the broad zone of channel deposits that we can identify from aerial photographs.
Samples 1-4 were analysed with both OSL and IRSL. Sample 5 was 14 C only. Sample pits encountered fine grained, sandy 15 material, with occasional evidence of graded and cross bedding. Small pebbles sometimes formed the base of cross beds.
We believe we sampled a mixture of channel or possibly overbank deposits of a fossilised, fan top, channel system. Sample collection and processing procedures are described in the appendix.
Samples were generally classified as being either well bleached where quartz (OSL) and feldspar (IRSL) ages are consistent, or partially bleached where quartz and feldspar ages differ significantly. In general, feldspar ages are only considered indicative 20 of true ages of channel deposits when they closely match quartz ages (see appendix for explanation of methodology and data tables). If feldspar ages do not match quartz ages for a particular sample, only the quartz ages are taken to be representative of the true age of the deposit. From the usable data there is a relatively narrow range of ages (∼ 8.85 ka -14.3 ka ) from samples 1-4, with a mean age of 11.6 ka (table A1, figure 7). The consistency of these values as well as the generally good quality of grains used (well bleached, low scatter) gives high confidence in the ages and suggests an early Holocene abandonment 25 of this part of the fan, and by extension probably dates the first uplift of the scarp and ridge structure that reorganised the drainage in this region. The age is much younger than has generally been assumed for Quaternary deposits in the Kolyma system in the Russian literature (Patyk-Kara and Postolenko, 2004). These values are probably the first genuinely physical age determinations carried out in this region, and certainly the first employing OSL. Combining the mean age with the associated 60 m -65 m offset gives a slip rate of ∼ 6 mmyr −1 , which agrees with the modern day, plate tectonic estimate of Okh-NAm likely that the Buyunda fan sediments have been reworked from other deposits in the Buyunda river system, and it may be that the ages reflect earlier episodes of transport and deposition in other parts of the drainage basin from which the sediments have been reworked, but there is no method available for quantifying this possibility.
14 C dating of sample 5 gives a far younger result. This is not particularly surprising, given the possible ways of introducing organic material into the subsurface long after deposition has occurred. We suggest that the consistency of the OSL results, reflecting time since channel abandonment and burial of sediment, make it likely that the 14 C age is post-depositional and 5 unrelated to the abandonment of sedimentation.

Basin wide fault and scarp features
The scarp we encountered in 2011 and 2012, can be traced across most of the Seymchan-Buyunda basin. Using remote sensing data, we can trace the fault and scarp westwards from the alluvial fan (figure 3). Aerial photo coverage also overlaps with parts of this region. The fault extends ∼ 90 km in total to the northwestern edge of the basin where it may also offset Neogene 10 continental clastic deposits. West of Lake Rovonoye, aerial photographs show several small lakes which may sit between an overlapping, en echelon portion of the fault. This is followed by a linear scarp running to the eastern edge of the incised Kolyma river valley, marked in places by shutter ridges. Several smaller rivers cross this part of the fault and have built fans across it.
Some of the fan edges suggest left-lateral offsets, but this is not consistently obvious. Linking of offset distributary channels (i.e. identifying consistent, left-lateral offsets) from one side of the fault to the other is also difficult, although this is often the 15 case on strike-slip fault systems (e.g., Hubert-Ferrari et al., 2002;Rodgers and Little, 2006).
The Kolyma river has a sharply incised eastern edge but the peneplain and terrace level above this is also incised which probably obscures the possibility of clear offsets of the sharp terrace edge (figure 2). However, the Kolyma, with the main river channel emerging from a deeply incised gorge in Jurassic and Triassic bedrock to the south and incising a delta-shaped region where it enters the Seymchan Basin, appears to be deviated in a left-lateral sense by ∼ 4 km as it enters the basin. West of the 20 Kolyma, the same fault segment continues, forming the northern boundary of Neogene outcrops. The segment then terminates in what is apparently a second large scarp but with the opposite uplift polarity to that on the Buyunda fan (figure 8). This second scarp appears to be somewhat enhanced by erosion along the fault trace, but the offset of the basin floor is clear from the fault normal, topographic profiles (figure 8). The northern side of the fault is uplifted by up to 3-5 m, just as the southern side is uplifted by a similar amount at the eastern termination of the fault segment on the Buyunda fan. As we discuss in the next 25 section, these two linked scarps and peak uplifts are easily relatable to elastic dislocation models of earthquakes on strike-slip fault segments.

Elastic displacement modelling and scarp polarity
Elastic displacement theory has long been applied to analysis of co-seismic slip in earthquakes. Although there are many degrees of sophistication of these models today (e.g., Okada, 1992), the simplest case of a vertical, strike-slip fault which adequate for our purposes. The method calculates the displacement components around a Volterra dislocation in an elastic half space, by solving the equations of elasticity for boundary conditions of stress free bounding surfaces, using a Green's function method (Steketee, 1958a, b). As such, a Volterra discontinuity is a surface with constant offset or displacement across it. This is a reasonable first approximation to a fault that has undergone an earthquake displacement. The parameters in the model are Eur and NAm, with the majority of tectonic displacement occurring on the plate bounding faults; a northwestern corner of Okh consisting of a series of blocks, mostly elongated north-south, moving independently of one another by relatively even amounts over many earthquake cycles, and thus no single, clear, plate bounding fault. Although some internal deformation is occurring within Okh as shown by the 1971 intraplate earthquake, it is one order of magnitude smaller than that expected for full release of plate tectonic strain according to Hindle and Mackey (2011). GPS data from northwestern Okh is sparse and has been interpreted in a variety of ways (Steblov et al., 2003;Apel et al., 2006). Our field observations on the Ulakhan fault give 5 a first insight into the paleoseismology of the region and add important data in this context.

Paleoseismology, earthquake recurrence and seismic hazard
The fault scarp we encountered requires a recent earthquake or series of earthquakes of large magnitude and also forms part of a single, 90 km fault segment that we can trace across the Seymchan-Buyunda basin. The opposite uplift polarities at the likely tips of this segment are a morphotectonic signature probably uniquely explained by rupture or repeated ruptures on a single 10 fault segment, as suggested by elastic dislocation modelling. The magnitude of uplift implied by the Ulakhan fault scarp over ∼ 11.5 ka also matches well with that predicted for the combination of a 90km fault segment and the slip magnitude available due to plate tectonic strain accumulation in this time period (Hindle and Mackey, 2011).
Our interpretation of the field observations has several implications. It confirms a likely significant seismic hazard in the region, with a likelihood of ≥ M w 7.5 earthquakes occurring within the Seymchan Basin (the 90 km length, 10 km deep, 15 15 m slip event modelled in this paper is equivalent to an M w 7.7 earthquake), and hence affecting both populated areas and large infrastructure, in particular the Kolyma hydro-electric dam located at Ust Srednekan (figure 2). It firmly constrains the location of the plate boundary to follow the trace of the Ulakhan fault, and suggests the slip and strain partitioning due to plate tectonic motions is concentrated (> 90% = 5-6 mmyr −1 ) on the plate boundaries. This in turn implies that internal strain of the northwestern Okhotsk plate is confined to release of small amounts (probably < 0.1 -0.5 mmyr −1 ) of accumulated slip. This 20 may mean that the largest earthquakes possible in the plate are no bigger than the M w 6.4 Artyk earthquake of 1971, although ultimately this will also depend on their frequency.
The wider question of recurrence times of large earthquakes on individual fault segments can also be partly addressed by our new results. Hindle and Mackey (2011) considered two possible scenarios for strain release along the plate boundaries of Okh.
In the first, strain was only seismically released along the Okh-NAm boundary (∼ 1150 km total length). This was considered 25 a possibility due to the absence of any seismicity, or indeed any clearly defined structure for the plate boundary, along Eur-Okh, which would creep aseismically instead. In this case, the average recurrence times for large earthquakes on any segment of the Okh-NAm boundary were estimated to lie between ∼ 0.7-1.2ka. The second scenario had seismic strain release occurring along both Okh-NAm and the adjacent portion of Okh-Eur (total fault length ∼ 2500 km). In this case, recurrence times were estimated at ∼ 3.0 -4.9 ka. The lower estimate was based on average earthquake sizes M w 7.6 -7.8. The 3 ka recurrence

Earthquake size, fault dimensions and scarp slope
One of the key questions in earthquake seismology is the nature of any relationship between rupture length and average fault slip. Wells and Coppersmith (1994) anaylsed a large number of earthquakes from around the globe, and according to their empirical formulae, a single earthquake on a 90 km long fault segment should yield an average ∼ 1.9 -2.8 m slip. Hence, our estimate of ∼ 15m slip in a single event on the Ulakhan fault seems large in this context. However, the natural example of the 5 1855, M w 8.1 Wairarapa earthquake in New Zealand, which has a relatively well-constrained rupture length of ∼ 145 km and average slip ∼ 12 -16 m shows that much higher displacement-length ratios for strike-slip faults are possible (Rodgers and Little, 2006;Grapes, 1999). The Wairarapa fault is interesting as an analogue for the Ulakhan fault in several other aspects.
Studies have shown it has a Holocene slip rate of ∼ 6 -12 mmyr −1 , broadly comparable to that suspected for the Ulakhan, and is mostly undergoing strike-slip displacement (Wellman, 1972;Grapes, 1999). There have also been a sequence of large earthquakes (∼ 11) on this segment of the Wairarapa through the Holocene, demonstrated by differential uplift of a series of river and beach terraces adjacent to the fault (Grapes, 1999). In general, the Wairarapa fault provides a well-constrained example of the behaviour we hypothesize for the Ulakhan.
A comparison with the Wairarapa fault scarp, with a known source age, relatively well-constrained magnitude, and similar kinematics is also interesting for our study. The Wairarapa fault has mutliple offset terrace levels giving a high resolution, 15 earthquake "stratigraphy" in the landscape, and confirming the repeated Holocene ruptures on the segment. These terraces are due to a longer wavelength, landscape uplift pattern around the fault of up to 5 m per earthquake over a wide area (maximum terrace uplift today ∼ 40 m). This broader uplift pattern is the main driver of terrace formation and abandonment. The scarp from the 1855 earthquake shows a 1-2 m vertical offset and remains visible in many places today (Rodgers and Little, 2006;Grapes, 1999). Elastic modelling of the Wairarapa earthquake (Darby and Beanland, 1992) has suggested a listric fault geom-20 etry, partly to account for the broader uplift field. The Ulakhan fault may also be creating uplifted fluvial terraces in a similar way. However, it is difficult to separate tectonic from glacio-eustatic and other base-level related signals in this context. More generally, the Holocene behaviour of the Seymchan-Buyunda segment of the Ulakhan fault seems to be quite well modelled by a simple, vertical, strike-slip fault. However, further analysis is probably required in this context.

Large scale tectonics 25
An interesting aspect of the northwestern corner of Okh is its tectonic situation as a narrow sliver of a small plate caught in compression between much larger, converging ones (Eur and NAm) (figure 10). Despite the convergent motion of Eur and NAm, the resultant motion along the boundaries of Okh is generally believed to be strike-slip with northwestern Okh moving towards the south, perpendicular to convergence. Due to the proximity of the Eur-NAm euler pole to the region, the rates of convergence, and hence overall rates of slip on boundaries are low (∼ 5 mm/yr −1 ). There are few, if any other places on 30 earth directly comparable to this. In terms of deformation rates, Northern China is broadly similar (Liu et al., 2011). However, Northern China is a zone of intraplate deformation. It is also the place with the longest historical record of seismicity in existence, which has allowed unique insights into the nature of slowly deforming regions, faulting and seismicity. It appears Solid Earth Discuss., https://doi.org/10.5194/se-2018-103 Manuscript under review for journal Solid Earth Discussion started: 1 October 2018 c Author(s) 2018. CC BY 4.0 License. that Northern China is composed of a system of linked faults across the plate interior, and these move to some degree in coordinated fashion. It is only possible to establish this due to the 2000 year record of earthquakes there. By contrast, our work suggests that northwestern Okhtosk is more plate-like with slip concentrated on discrete plate bounding faults, even if there may be relatively large intraplate earthquakes occurring too. It is thus interesting that plate-like behaviour can apparently persist into the realm of very low deformation rates (< 5mmyr −1 ). At the same time, the Ulakhan fault system may exhibit 5 similar long term behaviour to much faster slipping strike-slip faults. The North Anatolian Fault (NAF) slips at 23 mmyr −1 , and has a length of ∼ 1000 km (Stein et al., 1997), a length broadly similar to the Ulakhan system. The NAF is also segmented, and sequential earthquake migration over longer time periods on different segments has been observed within historical records (Stein et al., 1997), in a similar fashion to that now postulated at much slower rates along the Ulakhan fault.

10
We have documented a substantial fault scarp along the trace of the Ulakhan fault in the Seymchan Basin in Northeast Russia.
The fault scarp is indicative of a series of large earthquakes, probably affecting a ∼ 90 km long fault segment that we suggest ruptures repeatedly as a single entity, as evidenced by opposite polarities of scarps at the two ends of the segment and the fit of this to uplift patterns generated by simple, elastic dislocation models of left-lateral, strike-slip faults of the appropriate magnitude and slip.

15
Age dating of a fluvial system that seems to have become abandoned due to formation of the scarp on the western edge of the Buyunda alluvial fan suggests the sequence of earthquakes causing abandonment began ∼ 11.5 ka. This age may also be associated with a ∼ 60 -65 m, left-lateral offset of an ∼ 11.5 ka fluvial feature.
In general, our field data suggests that the Okhotsk-North America plate boundary in this region slips on average at ∼ 5-6 mmyr −1 , thereby releasing almost all the available tectonic strain due to Eur-NAm plate convergence (Hindle et al., 2006, 20 2009; Hindle and Mackey, 2011). This slip rate is also in agreement with local predictions from the Okh-NAm euler vector, although the latter is derived based on fundamentally different assumptions to our models here.
The earthquake recurrence analysis in previous work (Hindle and Mackey, 2011), and comparison with the new field data suggests infrequent earthquakes of relatively large magnitude (> M w 7.5, every ∼ 3 ka) are most likely responsible for the Holocene, tectonic geomorphology of the basin. Given that strain accumulation must be continuing to the present day, the 25 seismic hazard in the local area needs careful assessment. Perhaps the most critical question now is when exactly did the last large earthquake occur on this segment of the fault? Given sufficient resources, it may be possible to determine this by trenching across the fault scarp.
Code availability. The FORTRAN code written for the purpose of georeferencing the aerial photographs can be requested from the author (Hindle). It may also be the subject of a future publication. Data availability. A file of the x,y coordinates (longitude, latitude) of the mapped segments of the Ulakhan fault, based on the TANDEM-X DEM and aerial photo interpretation is available from the author (Hinde). A GeoTiff file of the composite scene of aerial photographs in geographic coordinates may be requested for academic use only. This data is not to be redistributed. Anyone wishing to have it must request it for themselves directly from the authors (Hindle). Commercial or other non-academic use, including upload to public mapping platforms of any kind, is forbidden.

Appendix A: OSL Dating and results
Certain minerals like quartz and feldspar, can store energy released by radioactive decay. In the case of sedimentary material, this radioactivity is derived to a major extent from isotopes of uranium (e.g. 238 U) and thorium (e.g. 232 Th) and their daughter nuclides, as well as potassium ( 40 K), both in the sediments of interest and their surroundings. This stored energy can be released by heating the minerals or by exposure to light. The energy is released as light (luminescence) which can be measured using a 10 photomultiplier or a CCD camera. The higher the level of radioactivity and the longer the duration of exposure the more energy will have accumulated, and the more light will be released. Hence, for sediments, it is assumed that provided their minerals are sufficiently exposed to sunlight during transport, prior to deposition, in order to reset the radioactive stored energy, the total amount of radioactive energy measured by luminescence (radioactive dose in Grays (Gy)) can be divided by the dose rate (rate of energy supplied to the minerals by radioactive decay within the sediments, measured in Grays per year, Gy/yr) to give the 15 luminescence age of the rock (Preusser et al., 2008). In sediments, this age should usually reflect the time since final deposition and burial.

A1 Sample collection
Sediment was collected from pits dug in abandoned channels of the fan top drainage system of the Buyunda fan. The aim of the collection procedure was to collect sediments that were buried and cut off from light, and hence began accumulating radioactive 20 energy derived only from the surrounding sediments and internal crystalline sources. As any fresh exposure to light will release this energy, sampling must be carried out in a light proof way. Hence, we used a metal sampler with an internal, removable, light proof plastic sleeve, to take samples from the pit wall, penetrating up to 20cm. After removing the sampler from the pit wall, the ends of the plastic sleave were sealed with light proof tape immediately upon removal from the sampler, and then stored in a light proof bag. Sample pits were generally ∼1m deep (the depth at which permafrost was first encountered), and 25 samples were taken at a depth of 80cm within the pit. (fig A1)

A2 Sample preparation and measurement
Age determinations were carried out at the Klaus-Tschira-Archaeometrie Centre, in Mannheim. Sediments were sieved to separate grain sizes of 100 -200 m (coarse) and less than 100 m (fine). Organic material was destroyed using Perdrogen (30%). Acetic acid (30%) was used to remove carbonates. The coarse grain fraction was split into mineral fractions using heavy liquid separation (tungstate density 2.75 and 2.62 g/cm 3 ) to extract quartz minerals. These were etched with 48% hydrofluoric acid for 45 minutes to remove the outer 20 µm of the grain that are influenced by alpha radiation and the material below 100 m was removed by sieving. The fine grain material was further refined to 4 -11 m fraction by settling in Acetone. 5 The first step removes grains larger than 11 m and a second step excludes grain sizes smaller than 4 m (Lang et al., 1996).
Both fractions, 100 -200 m and 4 -11 m, were then deposited onto steel discs for measurement.
We used a standard Risø TL-DA-20 reader equipped with a 90 Sr/ 90 Y source for beta irradiation (strength 0.06 Gys −1 for coarse grain, 0,08 Gys −1 for fine grain) and an alpha source 241 Am for fine grain (strength 0,116 Gys −1 ). For coarse and fine grain quartz we stimulated with blue LEDs (470 ± 20 nm) and detected the luminescence signal using 7,5 mm Hoya 10 U340 filters (280 -370 nm), whereas for measuring fine grain feldspar we stimulated with infrared LEDs (870 ± 40 nm) and detected in the blue range using filter BG3 and BG39 (3 mm each, 350-420 nm).
For all coarse grain quartz samples, preheat tests have been made at 180, 200, 220, 240, 260 • C at 2 aliquots each to determine a stable preheat temperature. Each measurement cycle of the quartz samples included an infrared measurement to bleach feldspar contamination. In addition for some of the coarse grain samples dose recovery tests have been performed on 6 15 aliquots each. The measurement followed the suggestions of Lopez and Rink (2007) and Murray and Wintle (2000).
Quartz was usually extracted from the fine sand fraction between 100 -200 m. We also used polymineral fine grain samples of grain sizes 4 -11 m to measure the feldspar signal. Quartz could also be measured in this fraction. Measuring both quartz and feldspar from the same grain size fraction enables us to interpret different aspects of sedimentation. Quartz bleaches (i.e. releases luminescence energy due to exposure to light) much faster than feldspar, hence if ages of both minerals overlap within 20 errors the sediment was well bleached and the sedimentation process was rather slow meaning grains were completely reset before deposition and burial. If the ages differ significantly with the quartz age the younger, it is likely that the sediment was accumulated abruptly and minerals were not properly bleachyed (radioactive energy reset). In this case only the quartz ages are significant as they reset their radioactive dose quickly. Samples in this condition are referred to as "partially bleached" and the youngest ages from the sample are used.

A3 Results
The results are shown in tables A1 and A2.