The age and potential causes of the giant Green Lake Landslide, Fiordland, New Zealand

Landslide deposits preserved in the geological record afford opportunities to better inform hillslope and seismic hazard and risk models, particularly in regions where observational records are short. In the Southern Alps of New Zealand, small coseismic landslides are frequent, but the geological record preserves several instances of more substantial (> 1 km3) but infrequent mass failures. With an estimated volume of 27 km3, the giant Green Lake Landslide represents one of the largest subaerial landslides on Earth. Previous work has suggested this deep-seated mass movement was most likely triggered by high-intensity seismic shaking, but that local structural weakness and/or glacial debuttressing may help to explain the anomalously large failure volume. Resolving the potential contribution of the latter is important given predictions concerning alpine deglaciation in the coming decades to centuries. Key to resolution are secure chronologies of landslide emplacement and past glacier change. Here we present in situ cosmogenic 10Be exposure ages from the Green Lake Landslide that suggest an emplacement age of 15.5 ± 0.7 ka. Recent work shows that glacial retreat in the region was underway by 19 ka, indicating that the Green Lake Landslide was emplaced 3–4 kyr after the onset of glacier retreat. Given the geometry of the former confining valley glacier, we expect that the deglaciation-landslide age gap is closer to the upper end of this estimate. If correct, this conclusion places greater weight on the roles of local geological structure and/or a great earthquake as factors contributing to the exceptionally large volume of this event.


Introduction
Large landslides (> 10 6 m 3 ) are rare but effective erosive agents and hazards in alpine environments, with long-term environmental impacts (Guthrie and Evans 2007;Hewitt et al. 2008). Dated pre-historic landslides can help to assess landslide causes, populate size-frequency relationships for improved hazard assessment, and quantify the longevity of environmental impacts. Landslides can also be useful for identifying and quantifying major landscape disturbances, such as climate changes (e.g. paraglacial responses) or paleo earthquakes (Ballantyne et al. 2014).
Covering an area of 45 km 2 and containing c. 27 km 3 of material, the pre-historic Green Lake Landslide deposit in the southeast Fiordland region of southern New Zealand is one of the largest subaerial landslide deposits on Earth (Fig. 1). Hancox and Perrin (2009) combined geological investigations and numerical modelling to suggest a key precondition, a preparatory factor, and a trigger behind the landslide: (i) an ancient (> 2.6 Ma) low-angle fault zone causing reduced shear strength of local topography, (ii) debuttressing caused by glacier retreat at the end of the last glacial, and (iii) strong shaking generated by a high-magnitude earthquake on the Alpine Fault, respectively. However, even in the presence of a weak structure, the large landslide volume is unusual given the relatively low local relief (~ 1200 m) of the pre-failure topography ( Fig. 1) and the high-strength crystalline rock, inviting further consideration of the conditions of failure.
The timing of glacial retreat relative to landslide emplacement has not been precise enough to allow for a robust evaluation of the importance of glacial debuttressing as a preparatory factor. As for the trigger, new work has suggested that an Alpine Fault earthquake may not be the most likely seismic source, with other, more proximal faults as candidates for strong shaking (Robinson and Davies 2017). Despite its importance in the New Zealand landscape and landslide inventory, the causes of the gigantic Green Lake Landslide, and the reason for its anomalous size, remain uncertain.
More complete understanding of the causes of pre-historic landslides is facilitated by secure chronological constraint (Pánek 2015). Hancox and Perrin (2009) reported two radiocarbon dates from silts interpreted to have accumulated in a former lake body created when the Green Lake Landslide impounded the nearby Grebe River (Fig. 2). One date, taken from the top of the silt package, yielded an age of 11.1-11.8 cal. kyr BP (recalibrated using SHCal20; Hogg et al. 2020), whilst a further date from the overlying peats is stratigraphically consistent, yielding an age of 8.7-9.1 cal. kyr BP (SHCal20; Hogg et al. 2020). These dates represent minimum-limiting ages for the Green Lake event. The maximum age limit is not quantitatively constrained, but Hancox and Perrin (2009) reason that the Green Lake Landslide occurred after the most recent glaciation of the Grebe/Monowai catchment, as the deposit appears to be unmodified by glaciers. Reasoning that the former Grebe Lake impounded by the landslide may have persisted for 1-2 millennia prior to infilling/draining, Hancox and Perrin (2009) hypothesised that the close temporal connection between deglaciation and landslide emplacement may imply a causative link. Given the predictions of future alpine glacier melt in tectonically active regions globally (Bolch et al. 2012;Zieman et al. 2016;Anderson et al. 2021), the possibility for deglacial triggering of giant deep-seated landslides warrants further examination. In this study, we use cosmogenic 10 Be surface exposure dating of the Green Lake Landslide to offer a more precise comparison between the timing of landslide emplacement with recent constraints on the timing of late Quaternary deglaciation (e.g. Moore et al. 2022).

Sample collection
We employed cosmogenic surface exposure dating to better constrain the age of the Green Lake Landslide as this technique affords the potential for direct age constraint of landslide emplacement. This potential is realised if rock samples meet the following criteria: (1) that the samples were emplaced in their current position without 10 Be inherited from prior exposure to cosmic rays; and (2) that the sampled surfaces have remained uneroded and fully exposed since deposition. Violation of these assumptions in individual samples can yield exposure ages that over-and underestimate the true depositional age, respectively.
We employed a judicious sampling strategy to try to satisfy these assumptions and obtain an accurate age for landslide emplacement (e.g. Ivy-Ochs and Kober 2008). First, we preferentially targeted small boulders (~ 1-3 m 3 ) from the area of the landslide termed 'Zone A' by Hancox and Perrin (2009) which is interpreted to have involved more rapid failure and fragmentation with finer debris, as opposed to the more 'intact' blocks involved in the southern part of the deposit. We consider that the large intact blocks are more likely to have been exposed to the cosmic ray flux prior to the landslide event, whereas small blocks have a higher chance of having been derived from shielded positions. This criterion is thus designed to minimise the potential for inheritance that would cause exposure ages that are older than the landslide event. Five of our six samples meet this size criterion, with only GL2803 having been collected from a megaclast (Fig. 3). The latter was collected to test our assumption that megaclasts might carry significant inheritance from prior exposure. Second, we undertook careful visual assessment of the local geomorphological situation of potential boulders to minimise the potential for collecting samples from clasts that might have moved since landslide emplacement. We specifically avoided clasts situated beneath steep slopes (e.g. at the base of  Legros (2002), with the Green Lake Landslide added here as the black square. For its volume, the Green Lake Landslide had an unusually small fall height (reflective of the relatively low local relief) and short travel distance, although the angle of reach (panel c) is consistent with the trend of the compilation.

Fig. 2
Overview of Green Lake Landslide with key chronological sites from this study and relevant previous work described in the main text (Hancox and Perrin 2009;Moore et al. 2022). Aerial photograph used with permission (VML ID: 3918; Lloyd Homer/GNS Science) Landslides 20 • (2023) megaclasts), which could have been emplaced after the landslide and would therefore yield 10 Be concentrations that underestimate the true landslide age. Third, we inspected the surfaces of potential boulders and avoided any that exhibited obvious signs of surface erosion (e.g. peeling/flaking). Post-emplacement surface erosion would also cause exposure ages to underestimate the true landslide age. Finally, we collected multiple samples (six total) in order to generate a population of samples that would help to identify potential outliers that may violate the assumptions of this method. Six samples is a fairly small population for such a purpose and this low number largely reflects the paucity of suitable material for sampling that is both visible and easily accessible on the densely vegetated surface of this remote landslide deposit. However, when adhering to a sampling strategy such as that described above, six samples is generally considered a reasonably sized dataset from which to derive a secure age for a single landform (Ivy-Ochs and Kober 2008).
We removed the top surfaces of samples with a hammer and chisel, aided by a portable rock saw (e.g. Suganuma et al. 2012).
Elevations were measured using a Trimble GeoXH differential global positioning system. Data were corrected in post-processing using the Land Information New Zealand PositioNZ geodetic network yielding sub-metre vertical precision.

Sample processing and age calculation
Physical and chemical separation of beryllium was undertaken at Victoria University of Wellington. We first crushed samples to < 1 mm in a jaw crusher and then dry-sieved to isolate the 0.25-0.75 mm size fraction. After 2 overnight leaches in 10% HCl, quartz grains were isolated from feldspar by the froth-flotation method. Quartz separates were then subject to repeated overnight leaches in a 5% HF/1% HN0 3 solution for further purification. Following addition of ~ 260 μg 9 Be, samples were digested in concentrated HF. Beryllium was then isolated by ion-exchange chromatography and Be(OH) 2 was precipitated at pH9. Samples were transferred to quartz crucibles and calcined over a flame.

Original Paper
The resultant BeO powder was mixed in a 2:3 ratio with Nb and packed into stainless steel targets. 10 Be/ 9 Be ratios were determined by accelerator mass spectrometry at Lawrence Livermore National Laboratory where measurements were made relative to 07KNSTD standard ( 10 Be/ 9 Be = 2.85 × 10 −12 ) (Nishiizumi et al. 2007). We calculated 10 Be concentrations by subtraction of total 10 Be measured in a full-process blank (~ 53,000 at. 10 Be) following the data reduction protocol outlined by Balco (2006).
We calculated exposure ages using the online exposure (formerly CRONUS-Earth) calculator, version 3 (available: http:// hess. ess. washi ngton. edu/), using the 'Macaulay' production rate calibration data of Putnam et al. (2010). Topographic shielding values were derived from field measurements of horizon elevations input to the online calculator. Sample thicknesses were constrained prior to crushing as the average of multiple manual measurements by callipers.
We make no correction for shielding by snow as the low altitude site is unlikely to experience deep or prolonged snow cover now or in the past. Some of the samples exhibited ~ 1-2-cm moss or lichen cover; however, the potential attenuation of 10 Be production due to such a thin layer of low-density matter is negligible; therefore, we have not attempted a correction. Shielding of the cosmic ray flux by broader forest (e.g. canopy, stems) is estimated to be up to c.5% (Plug et al. 2007). Our samples are derived from the edges of forested areas, with the tree growth in the adjacent grass-dominated areas thought to be limited by local microclimatic factors (e.g. frost hollows -N.Perrin, pers. comm). Due to the long-term, low-density biomass surrounding our samples, we refrain from introducing a biomass correction and note that the likely magnitude of such a correction would not alter our conclusions.
We present exposure ages calculated using the 'Lm' production scaling model defined by Balco et al. (2008) for consistency with the recent proximal glacial chronological dataset of Moore et al. (2022). The alternative scaling model of Lifton et al. (2014), commonly known as 'LSD', yields exposure ages that are c. 1% younger at our site, so our conclusions are unaffected by choice of scaling model.

Results
Exposure ages from the six samples are widely spread, ranging from 6.7 ± 0.6 to 18.0 ± 0.4 ka (Table 1; Fig. 2). Three of the exposure ages (GL2803, GL2806, GL2807) are statistically indistinguishable given their analytical uncertainties and together yield a mean age of 15.5 ± 0.7 ka. Sample GL2801 is slightly older than this cluster at 18.0 ± 0.4 ka; meanwhile, the remaining two samples (GL2802 and GL2806) yield low 10 Be concentrations and young ages (8.5 ± 1.7 ka and 6.7 ± 0.6 ka). The relatively large uncertainty of GL2802 results from the low quartz yield (~ 0.8 g), which limited the measurement time.
The inconsistent exposure ages indicate that samples selected may violate the two key assumptions of the cosmogenic surface exposure dating method (e.g. see "Methods"). We consider the most likely emplacement age for the Green Lake Landslide to be 15.5 ± 0.7 ka, which is drawn from the mean of the three consistent samples that comprise half of the total age population. These samples yield consistent ages despite being distributed across three separate locations on the landslide surface (Fig. 2), which lessens the potential for them to have each come from a similar pre-slide location. If correct, then this interpretation implies that sample GL2801 carries some minor inheritance from exposure prior to the landslide event, which is not unexpected given the limited transport distance and propensity for landslides to retain surface materials (Akçar et al. 2012). It is possible that this inheritance provides some information about the timing of local deglaciation; however, other scenarios (e.g. more complex exposure or slow accumulation at depth via muon production) cannot be ruled out. If this interpretation is correct, the two youngest samples (GL2802 and GL0108) must have been affected by post-depositional removal of 10 Be (e.g. erosion) or suppression of production (e.g. by previous burial/shielding from the cosmic ray flux). We would consider the latter more likely, as significant (~ 1 m) erosion would be necessary to explain the 7-9 kyr discrepancy of these ages from the inferred depositional age of 15.5 ± 0.7 ka. Erosion of this magnitude is not supported by field observations (e.g. flakes/fragments/scarps).
We consider two alternative interpretations of the timing of the Green Lake Landslide emplacement from this age dataset. First is that the true age is closer to 18.0 ± 0.4 ka, as represented by GL2801, implying that all other samples have suffered post-depositional erosion or burial to produce exposure ages younger than the true depositional age. We dislike this interpretation largely because it rests on a single data point and it seems unlikely that half of the dataset were affected to the same degree by post-depositional effects despite a wide geographic spread. The second alternative interpretation is that the two youngest ages (7-9 ka) are the true emplacement age and that the four older ages have inherited 10 Be from exposure prior to the landslide event. This scenario is unlikely as it is out-of-sequence with the previous radiocarbon constraints that indicate the landslide occurred before 11.1-11.8 cal. kyr BP (Hancox and Perrin 2009;Fig. 4). To summarise, the simplest explanation of our age dataset is that the Green Lake Landslide was emplaced at 15.5 ± 0.7 ka.

Discussion
The temporal relationship between deglaciation and landslide emplacement Previous work suggested the Green Lake Landslide may have closely followed local deglaciation (Hancox and Perrin 2009); however, the precise timing of both the landslide and local glacier retreat was poorly constrained. Our emplacement age of 15.5 ± 0.7 ka for the Green Lake Landslide is 2-4 kyr older than previously assumed by Hancox and Perrin (2009) from radiocarbon dating landslide-dammed lake sediments. Our older age indicates the 14 ka age for deglaciation reported by Hancox and Perrin (2009) is too young, by at least 1.5 ka. This age is likely based on uncalibrated radiocarbon dates from organic material overlying tills and moraines (e.g. Suggate and Moar 1970) that were used to construct early quantitative glacial chronologies, but equates to c. 17 ka cal. BP when applying the most recent calibration data (e.g. Hogg et al. 2020). More recently, direct dating of moraines by cosmogenic surface exposure dating has affirmed that the final moraines deposited during the Last Glacial Maximum in the Southern Alps were abandoned at 17-19 ka (e.g. Barrows et al. 2013;Putnam et al. 2013;Rother et al. 2015;Barrell et al. 2019;Tielidze et al. 2022). In Fiordland, the density of glacial chronological information remains low relative to the more accessible glacial sequences farther north. Regional glacier modelling is consistent with the inference that Monowai Catchment was glaciated during the Last Glacial Maximum (Golledge et al. 2012), but no direct age control exists for its deglaciation. Moore et al. (2022) recently presented a cosmogenic 10 Be moraine chronology from Rocky Top Cirque, situated on the eastern side of the southern Green Lake Landslide head scarp. Moraine ages from this site show that a small cirque glacier was undergoing gradual retreat from 19 ka in response to minor (< 100 m) rise in the local snowline and that the catchment was fully deglaciated shortly after 17.2 ± 0.2 ka (Fig. 5).
If the snowline changes reconstructed by Moore et al. (2022) are regionally representative, then we expect that the former Monowai Glacier was also experiencing negative mass balance from at least 19 ka. However, differences in the geometry of the former Monowai and Rocky Top glaciers mean that the length response may have differed between these ice masses, despite common climatic forcing. Rocky Top Glacier was short and relatively steep, covering 200-300 m elevation over its 1 km length (~ 0.2-0.3 rise/run), in comparison to the former Monowai Glacier that descended ~ 1.3 km across its ~ 45 km length (0.02-0.03 rise/run). In a warming climate, large portions of low-angle glaciers can readily switch from areas of net accumulation to net ablation from only minor snowline rises. Thus, for low-angle glaciers, larger length reductions are required to accommodate mass imbalances than for steeper glaciers (Eaves . Furthermore, inception of proto-lake Monowai may have further exacerbated glacier mass loss and retreat due to calving. Considering these local chronological constraints on the deglacial snowline (Moore et al. 2022), and glaciological theory, we expect that the former Monowai Glacier was likely experiencing negative mass balance and substantial retreat from at least 19 ka, which is at least 3-4 kyr prior to emplacement of the Green Lake Landslide.

On the role of deglaciation for landslide preparation and triggering
The concept of glacial debuttressing suggests that removal of confining glacier ice reduces the stability of oversteepened sloping valley sides to a critical state. This process has been proposed as a triggering mechanism for large pre-historic landslides in glaciated alpine regions globally (Ballantyne 2002;Soldati et al. 2004). However, McColl et al. (2010) consider the rheology of ice in the context of rock slope failure mechanics and suggest that ice is unlikely to support oversteepened rock slopes in the same manner as engineering buttresses are employed. Temperate ice exhibits a ductile response under relatively low stress; therefore, slopes in a critically stable state can begin failing before deglaciation is complete, albeit at a slower rate than in ice-free conditions. In support of this, there is growing recognition of slow, deep-seated slope failure activity in currently deglaciating alpine valleys (McColl and Davies 2013; Cody et al. 2020;Storni et al. 2020;Rechberger and Zangerl 2022). Pure glacial debuttressing-related failures may thus be considered syn-, rather than post-glacial (McColl and Davies 2013). Indeed, many post-glacial landslides in the European Alps that were previously interpreted as being triggered by deglaciation have been reappraised due to improved dating prompting alternative explanations (Ivy-Ochs et al. 2009;McColl 2012;Oswald et al. 2021). Had the initial failure of the Green Lake Landslide been established whilst the glacier was still present, it is likely that complete ice retreat (i.e. debuttressing) would have culminated in total, but possibly gradual, failure of the unstable rock slope. Given the multi-millennial lag between glacier withdrawal and emplacement of the giant Green Lake Landslide identified here, and the relatively long travel distance suggesting rapid failure, we think it is unlikely that debuttressing was the trigger of failure. Debuttressing and glacial erosion instead were probably both preparatory factors that reduced the stability of the slope, but not to a critical state sufficient to initiate failure.
Deglaciation is driven by climatic change-most notably atmospheric warming-which may also increase landslide susceptibility through enhanced weathering, degradation of permafrost, or increased precipitation, all of which can reduce rock slope stability. Our new age for the Green Lake Landslide places its occurrence at the end of a millennial-scale warming interval defined as 'post-termination amelioration' (NZce-5; Fig. 5c) in the regional pollen-based climate event stratigraphy (Barrell et al. 2013). This time is generally considered a period of rising regional temperature that coincided with a sustained southward shift in the westerlies and reduced precipitation (Whitakker et al. 2011). The landslide therefore occurred when stability conditions may have been low. However, Hancox and Perrin (2009) investigated the potential role of groundwater level changes in their slope stability analyses (e.g. from rainfall or seasonal fluctuations), and concluded that the (> 100 m) magnitude of water table increase required to be (2022) (foreground-left) adjacent to the Green Lake Landslide (background). c Relative kernel density estimates of cosmogenic ages presented in this study annotated with surrounding chronological constraints on landslide emplacement (Hancox and Perrin 2009) and proximal glaciation. Also shown are the New Zealand Climate Event (NZce) stages as defined by Barrell et al. (2013) with the colours representing relative regional temperature changes from cold (blue) to warm (orange) Landslides 20 • (2023) influential was unlikely to have occurred. Moreover, deglaciation of the valley after 19 ka would have likely resulted in a lowering of the groundwater level within the adjacent rockslopes (as conceptualised by McColl et al. (2010) for rockslopes adjacent to temperate glaciers), possibly countering the role of any potential increase in regional precipitation. Whilst little information exists for paleo permafrost conditions in New Zealand (Hales and Roering 2005), deglaciation of the Rocky Top Glacier by 17 ka suggests permafrost limits were likely higher than the majority of the Green Lake Landslide source area, which is of a similar elevation, by the time of landslide failure. Furthermore, neither frost weathering nor permafrost degradation is likely to have been sufficient to destabilise the very deep failure surface of the Green Lake Landslide; there is little evidence of cubic-kilometre scale landslides having been triggered by such processes (e.g. Huggel 2009).
Whilst there is no evidence that deglaciation or climatic processes were a direct trigger for the Green Lake Landslide, both likely primed the slope for failure, either helping to set forth a process of progressive failure considered to affect many paraglacial rock slope failures (McColl 2012), and/or moving the slope to a marginally stable state susceptible to potential triggers. Increased shear stresses from oversteepened and debuttressed slopes, weathering, and cyclic environmental processes (e.g. Grämiger et al. 2020) including seismicity ("Geological and seismic factors") may have gradually damaged the rock slope, and permitted failure of the landslide millennia after the onset of deglaciation (e.g. Spreafico et al. 2021;Ballantyne et al. 2014). Deglaciation may also have played an indirect role, for example by increasing the intensity of seismic ground motions and/or by increasing the temporal frequency of local seismicity due to glacio-isostasy. Valley-filling glacier ice can dampen the topographic amplification of seismic waves, and conversely, deglaciation can make hillslopes more prone to intense seismic shaking . Meanwhile, crustal movements due to glacial unloading have been inferred from temporal clustering of mass movement deposits during the deglaciation, particularly in regions that are otherwise tectonically inactive (e.g. Ballantyne et al. 2014). Isostatic adjustments due to deglaciation are poorly constrained in the Southern Alps. Mathews (1967) concluded postglacial rebound was modest (~ 30 m) in the central Southern Alps, where late Quaternary ice thickness was greatest. This estimate may be considered a maximum estimate for total rebound in the vicinity of Green Lake where ice loading was smaller, due to lower elevation source regions for glaciation. Furthermore, it is possible that any post-glacial rebound was more neutral at our study site as it is situated at the outer margin of the late Quaternary ice limits, and thus may have existed between the uplift and forebulge regions. Given the tectonic setting of our study at the plate margin and associated propensity for frequent, large earthquakes, we consider the simplest explanation is that any seismic trigger of the Green Lake Landslide was tectonic in origin.

Geological and seismic factors
Although it is possible that the Green Lake Landslide did not have a specific trigger, and its timing relates indirectly to climate processes, triggering by a strong (tectonically induced) earthquake remains a likely possibility. Support for this comes from the giant size of the landslide, its location within a very seismically active landscape, close proximity to active fault structures, and the growing evidence that many of the very large and giant alpine rockslope failures in other parts of the world are likely seismically triggered (e.g. Oswald et al. 2021). However, the source of the triggering earthquake remains uncertain. Hancox and Perrin (2009) suggested a high-magnitude (> 7.5 M w ) earthquake on the southern portion of the Alpine Fault, situated 80 km to the west, as the most likely source of the MM IX-X shaking required to cause such a largevolume failure in this relatively low-relief setting. Paleoseismological evidence indicates a centennial-scale recurrence interval for Alpine Fault earthquakes over at least the last 8 ka (Cochran et al. 2017). This earthquake source is challenged by Robinson and Davies (2017) who, using regional seismic attenuation modelling, find that an Alpine Fault rupture may not be capable of generating the required ground motion for failure. They argue that an earthquake source more proximal to the Green Lake Landslide is required and suggest the nearby (< 10 km distant) Hauroko Fault as a possible candidate. Little is known about the recurrence interval of this fault. Whilst several studies have recognised evidence for activity on the offshore components of the Hauroko Fault (Norris et al. 1978;Sutherland et al. 2006;Litchfield et al. 2014), evidence for recent movement in the terrestrial portion is less clear. It should be noted that Hauroko Fault is just one candidate for a more proximal seismic trigger than the Alpine Fault, with thermochronometry data indicating significant regional uplift during the Late Cenozoic, which may have been accommodated by other proximal, but as yet unrecognised, faults (Sutherland et al. 2009a, b). Clearly, there is a need for improved resolution of active fault mapping in this region if the giant Green Lake Landslide is a product of such activity.
The lack of other landslide deposits of equivalent age (noting that many remain undated) and the weakened role for deglaciation established here place greater emphasis on the structural weakness imparted by faults as a major cause of the landslide. Hancox and Perrin (2009) suggest failure may have been facilitated by the ancient Mt Cuthbert Fault, low-angle fault zone comprising hydrothermally weakened paragneiss that strikes northwest-southeast along the landslide headscarp. We note that another mapped fault structure intersecting obliquely with Mt Cuthbert Fault may have provided a lateral release surface for the southern side of the landslide (Fig. 6). There are other landslide deposits and lineations north of the Green Lake Landslide, which are in proximity to a fault which may be an along-strike continuation of the Mt Cuthbert Fault (Fig. 6) as indicated by Hancox and Perrin (2009). Whilst these faults have previously been interpreted as inactive during the Quaternary, we consider it an open possibility that the Green Lake Landslide is the coseismic product of a rupture of one or more of these fault structures, or that these faults played a key role in locally amplifying seismic shaking. Observations from the 2016 M w 7.8 Kaikoura earthquake demonstrate the propensity for large landslides to cluster near surface rupturing faults due to elevated ground motions and rock mass damage (Bloom et al. 2022). The Kaikoura event also demonstrated how major earthquakes can involve coeval rupture of multiple faults (Cesca et al. 2017

Original Paper
a nearby fault, such as the Hauroko or Monowai faults which intersect the southern end of the Mt Cuthbert Fault. Clear paleoseismic evidence of recent (i.e. ~ 15 ka) rupture of these faults may be difficult to recognise given the presence of forest cover, high erosion rates, and obscuring by slope sediments, so they may have been previously misidentified as inactive. This hypothesis requires further investigation, but it aligns with the views that fault structures influence landslide distributions and hazard (Brideau et al. 2009), and large coseismic landslides often occur within zones of fault rupture (Bloom et al. 2022).
The open possibility for large earthquakes to occur in southeast Fiordland in relatively close proximity to inhabited centres of southern South Island warrants improved resolution of regional paleoseismological records (Robinson and Davies 2017). Resolving this could include the dating of other landslides in the region to identify spatial clusters of landslides, including ones that coincide with the time of the Green Lake Landslide, to help to pinpoint likely earthquake sources and their intensity. Currently, fewer than 2% of the > 150 mapped landslides that exceed 1 km 2 in area within Fiordland and the surrounding parts of the Southland and Otago have been dated (Rosser et al. 2017). Of the few that have, none has ages consistent with the Green Lake Landslide (Sweeney et al. 2013;Hancox et al. 2013;McColl and Draebing 2019), all occurring in the Holocene.

Conclusion
We sought to improve age control for the giant Green Lake Landslide in southern Fiordland through in situ cosmogenic 10 Be surface exposure dating of rock clasts emplaced at the surface of this deposit. Our results indicate an emplacement age of 15.5 ± 0.7 ka for the Green Lake Landslide. This age is 2-4 kyr older than the previous minimum-limiting ages from landslidedammed lake sediments. Comparing our age data with recent chronological constraint of nearby glacier and snowline changes at the end of the Last Glacial Maximum, we suggest that deglaciation is unlikely to have triggered the Green Lake Landslide, but may have been one of several factors (particularly weak fault structures) that contributed to its failure. This conclusion places greater weight on the alternative trigger, chiefly a locally sourced high-magnitude earthquake. There is much still to be learned about the failure conditions and causes of the extraordinary Green Lake Landslide; this new chronological constraint offers a foundation for future investigations and regional paleoseismological studies. Landslides 20 • (2023) Sutherland for suggestions concerning the paleoseismological record in southern Fiordland.

Funding
Open Access funding enabled and organized by CAUL and its Member Institutions. This work was funded by a British Geomorphology Society Early Career Researcher Grant awarded to S. Eaves. S. McColl was funded by MBIE SSIF Hazards Programme at GNS Science (contract C05X1702). This work was performed in part under the auspices of the US Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. This is LLNL-JRNL-837730.

Declarations
Competing interest The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.