Transects across an apparent beach berm and the lunette were surveyed using a MALÅ ProEx ground-penetrating radar (GPR) system with a 500 MHz antenna and integrated high-resolution GPS. The GPR data were collected in transects forming a rough grid parallel and perpendicular to the trend of hypothesised beach and lunette landforms. The GPR was hand-dragged at a speed of ∼ 4 kph and fired using time firing at a rate of 10 Hz, resulting in an average along-track resolution of 0.11 and 0.07 m vertical resolution, based on a centre frequency of 500 MHz. After acquisition, radar data were processed using GPR-SLICE software (DC drift; user-defined signal gain; bandpass lo = 350 MHz, hi = 650 MHz; background removal). Profiles were topographically corrected using elevation data from the GPS system and spot-checked using known elevations. While absolute topography was not reliable, relative elevation was consistently reproducible. Individual profiles were converted to depth–distance using the published radar velocity for wet sands of 0.07 m ns-1 in the beach ridges and dry sands of 0.12 m ns-1 in the lunette (Neal, 2004). Depth–distance profiles were used to evaluate sediment thickness and observe true geometry of radar reflectors.

The sub-surface sediments were logged using a hand auger to a depth of between 0.6 and 1.2 m, depending on sub-surface conditions. Sub-samples were collected for grain size analyses. In addition, gravels from the sand and gravel barrier were treated with HCl for 12 h in order to identify weathering products such as manganese–iron pisoliths. Four samples were collected for optically stimulated luminescence (OSL) dating using steel tubes, wrapped in black plastic, and transported to the Max Planck Institute for Evolutionary Anthropology in Leipzig for analysis.

Sample preparation and measurement for OSL dating were undertaken in the luminescence dating laboratory of the Department of Human Evolution, Max Planck Institute for Evolutionary Anthropology in Leipzig. The OSL samples were prepared under subdued red light using published methods (Fitzsimmons et al., 2014). This involved sieving, applying HCl acid and hydrogen peroxide digestion to remove carbonates and organic matter respectively; and isolating pure, 180–212 µm quartz grains. The outer ∼ 10 µm alpha-irradiated rind of each grain was removed by etching in hydrofluoric acid, and the sample was then subjected to a final sieve to remove finer fragments which had broken off during etching. The quartz grains were then prepared as small aliquots (18 discs; 1 mm diameter) for preheat testing and as single grains (600 grains; six single-grain discs) for equivalent-dose (De) measurement.

De measurements were undertaken using an automated Risø TL-DA-15 equipped with blue-light-emitting diodes (for preheat and initial dose estimate testing) and a TL-DA-20 reader with a single-grain attachment containing a green laser emitting at 532 nm, for light stimulation of single aliquots and single grains respectively (Botter-Jensen et al., 2000). Irradiation was provided by calibrated 90Sr/90Y beta sources. Equivalent doses were determined on single grains using the single aliquot regenerative dose (SAR) protocol of Murray and Wintle (2000, 2003). Preheat temperatures of 260 ∘C were chosen based on the results of the preheat plateau tests (Fig. S2 in the Supplement) for the natural and regenerative doses, with a preheat temperature of 220 ∘C for the test doses (0.94 Gyr).

Individual grains were analysed for their suitability for OSL dating based on the selection criteria of Jacobs and Roberts (2007). The single-grain dose distributions of all samples are > 40 % overdispersed with complex dose populations (Table S1 in the Supplement), and therefore the finite mixture model (FMM) was used to identify dose populations (Galbraith and Green, 1990). The OSL dating results are summarised in Table 1. Equivalent-dose distributions for the four samples are shown as radial plots, with the FMM-derived dose populations highlighted, in Fig. 3.

Uranium, thorium, and potassium (40K) activities were measured in the “Felsenkeller” laboratory at VKTA Rossendorf in Dresden, Germany, using low-level gamma-ray spectrometry. Dose rates were calculated using the conversion factors of Stokes et al. (2003) with β-attenuation factors taken from Mejdahl (1979). Beta counting was based on 1 g homogenised sub-samples and used for the beta component of the dose rate. Measured water contents ranged from 5 to 10 %, and these values were used for all samples. Cosmic dose rates were calculated from Prescott and Hutton (1994).

There are no dune or beach deposits on the western side of the lake (Fig. 1). The main geomorphic feature on the eastern side of the lake is the lunette and the low basalt ridge. The lunette comprises a north–south-oriented ridge less than 2 m high adjacent to the lake, a swale behind that is occupied by a small stream, and a small sand flat area that extends up to 50 m east of the lake shore.

The lunette on the eastern shore is composed of poorly sorted medium sand grading upwards into fine sand with accessory silt contents of 3–15 %. Particle size results and other stratigraphic information are plotted in Fig. 1, and particle size analysis curves are provided in Fig. 1 of the Supplement. GPR transects are shown in Fig. 2a and b.

On the SE margin of LLL, there is a partly infilled outlet, immediately to the west of which there is a ca. 100 m long, 50 m wide low (< 1 m) berm. The berm is poorly to well sorted and has medium and coarse quartz-rich sand with iron–manganese, pisolithic gravel, and a silt content of 1–14 %.

The GPR proved effective at mapping stratigraphic architecture and sub-surface character to a depth shallower than 4 m in the berm (Fig. 2a and b). GPR data suggest the presence of several distinct units related to changes in lake level and the development of spit/barrier and berm formations (Shan et al., 2015; Thompson et al., 2011). The berm showed strong internal stratification and features perpendicular lines with strong sigmoidal clinoforms indicating beach progradation to the west (Thompson et al., 2011) as well as low-angle sub-parallel reflectors dipping to the east suggesting basin infill via over wash processes. This package is underlain by a convex-up package of reflections that are sub-parallel with dips to the east and west. Comparison of this feature with those identified by Shan et al. (2015) suggests the complex is underlain by a spit complex. Additional information on the character of the lower units associated with the interpreted spit are unavailable due to the existing GPR data coverage.

The internal stratigraphy of the fine-grained lunette was difficult to assess with the GPR. Evidence of extensive modern bioturbation by rabbits was observed during the radar acquisition. The shallow penetration did however show weak internal characteristics commonly associated with lunette formation (Thomas and Burrough, 2016). These included eastward-dipping high-angle reflectors that are truncated on the western-facing slope, coupled with areas of parallel to sub-parallel reflections that change to steeply dipping reflections. All reflectors are laterally discontinuous and show evidence of disturbance at all depths observed, rendering the GPR data ineffective at determining genetic processes or detailed landform characteristics.

The OSL age data are summarised in Table 1 and shown with respect to stratigraphy and catchment geomorphology in Fig. 1. The three samples collected from three different locations along the lunette suggest that the entire landform was active during the LGM, between ca. 24 and 19 ka. Our samples do not extend to the base of aeolian sedimentation, and it is likely that the lunette was formed earlier than the LGM. The secondary age populations identified by FMM are all younger than the LGM phase of deposition (Fig. 3; Table S2) and suggest phases of partial reactivation or pedogenic infiltration of material into the lunette. The younger age populations from sites LL3 and LL4 in the central part of the lunette are comparable and strongly suggest contemporaneous post-depositional infiltration of younger material or partial reactivation of the lunette in the early Holocene (ca. 9–8 ka; Table S2). Sample L-EVA 1230 (LL3) exhibits a third peak centred on 11.8 Gy BP (9.1 ka). The second major age population from the LL2 site in the southern part of the lunette dates to the mid-Holocene (5.6±0.5 ka; Table S2) and suggests spatial and temporal variability in the Holocene post-depositional pedogenesis (or reactivation) of the lunette.

The overdispersion on individual De results from the berm was too high (79.9 %; Table S1) to reliably define a depositional age, although the largest age population yields a mid-Holocene age (5.1±0.5 ka; Table 1) comparable with the reactivation of the southern part of the lunette at LL2. The minor dose populations yield ages of 11.1±1.6, 2.3±0.3 and 1.2±0.1 ka (Table S2).

The most cryptic landform in the basin is the sand and gravel berm on the SE margins of LLL. The feature was identified by Gale and Haworth (2005), who interpreted it as part of a relict older lunette feature. From visual observations alone, this is a reasonable interpretation because the low berm does look like the erosional shadow of an older ridge. Our sedimentologic and GPR structural investigations, however, discount this interpretation. Based on both GPR and field observations from pits, the feature is clearly a beach berm, with numerous small wash-over structures (see Fig. 2a).

The berm barrier feature is composed of pea-sized gravels within a finer sandy matrix. We assume the sandy matrix to be post-depositional because it is incompatible with the sedimentary structures and post-depositional infilling of openwork deposits is common. In addition, the contrast between locally sourced detrital basalt gravels and reworked quartz-rich sand and silt is striking. The matrix may have accumulated either through aeolian accession or through filtration of sands through the barrier during high lake stands when the berm would have acted as a permeable filter for the lake. Given the mostly coarse nature of the matrix (medium to coarse sand), we prefer the two-stage filtration hypothesis.

The pea-sized gravels are detrital. We suggest that the most likely origin for this feature is as a spit that developed from the basalt ridge on the SW edge of the lake and that the basalt gravels were moved along the shoreline by longshore drift. The barrier ultimately cut off an area to the SW of the present lake that was part of a larger, ancestral lake feature, for which we have no age constraint due to the lack of associated sedimentary deposits. The barrier post-dates the LGM as we have been provided with a radiocarbon result (R. Haworth, personal communication, 2016) from a depth of 1.5 m close to our LL1 sample (Beta-110588: 16 200±70 yr BP; median: 19 500 cal yr BP; calibration from Stuiver and Reimer (1993) (Calib 7.1) with a Southern Hemisphere correction (SHCal13) from Hogg et al., 2013) that provides a maximum age for the barrier. The luminescence sample based on the finer matrix material at 0.5 m depth yielded a highly dispersed dose distribution with four age populations, which is not unexpected given our hypothesis that the matrix is post-depositional. The grains may represent the accretion of fines to the barrier during high stands in the lake in the early (ca. 11 ka), mid- (ca. 5.6 ka) and late Holocene (ca. 2.3, 1.2 ka).

Based on the morphology, sedimentary composition, and internal structure, the feature along the eastern shoreline of LLL is clearly a composite beach and aeolian landform. The quartz-rich sands were most likely derived from the granites on the eastern side of the catchment, which deposited into the lake and were subsequently reworked onto the shoreline. Half the basin is comprised of basalt, yet there is little evidence for basalt-derived sediments in the lunette system. By contrast, the fine sediments in the depocentre of the lake basin are primarily derived from basalt. This implies that there is an effective sorting mechanism within the basin, whereby the basalt preferentially weathers to mud while the granite generates sand. Sorting by currents would transport the fines to the depocentre, while the sands would be transported towards the lake margins. The most parsimonious candidate for this latter process is wind-blown waves.

Present-day wind roses for LLL (BOM, 2014) demonstrate that there are two primary wind directions (Fig. 4), one from the east and the other from the west to north-west. These prevailing winds have strong seasonal components. Winter winds (August) are dominated by westerlies and provide the strongest and most persistent flows (8 % calm) consistent with eastward transport and deposition of sediments onto a lunette situated on the eastern shoreline of LLL. Summer winds (February) are dominated by easterlies associated with onshore circulation on the northern limb of the sub-tropical high-pressure cell in summer (Fig. 4). These easterly winds are on average weaker (20 % calm) but do include short periods of relatively high-intensity winds, which might be expected to result in sediment transport to, and deposition onto, the western side of the lake. It is curious therefore that all depositional landforms marginal to LLL are located on the east and south-east sides of the lake, with no deposition on the western shoreline. Examination of the wind roses indicates that sand-transporting winds (above ∼ 22 km h-1: Fryberger, 1979) were more than twice as frequent (∼ 14 % vs. ∼ 5 %) in August than in February and that the highest wind speeds occurred more frequently in August. This confirms that the most effective net sand-transporting wind, associated with lunette and berm formation, was from the west/north-west. The transport is most likely to have been primarily sub-aqueous, since the relatively poor sorting in the foredune indicates only a short-distance aeolian transport pathway.

In addition to the stronger drift potential there may also be a biological effect. The rush beds occurring in the shallower parts of the lake are most fully developed during the summer. Unlike much of Australia, winters are severe on the New England Tablelands due to the relatively high elevations, and seasonal die-back of the tall spike rush is observed today. New growth emerges in spring and dies off in autumn in cooler, high-altitude sites (Rajapaskse et al., 2006). Consequently, the summer peak in vegetation cover disrupts the wind fetch over the lake precisely at the same time as the easterly winds penetrate the tablelands, thereby further reducing the ability for waves to set up during the warmer months.

The luminescence ages from the lunette are coherent. All samples are dominated by grains that are LGM in age. The samples all overlap at 2σ and produce a weighted mean age of 20.4±0.8 ka, indicating that the main phase of dune activity at LLL occurred during the late LGM. Our interpretation that the dominant sediment transport mechanism was sub-aqueous therefore implies that the LGM oversaw permanent, and probably full, lake conditions at LLL. Evidence from our unpublished sedimentary archives from the depocentre of the lake supports the concept of a full lake during the late LGM (ca. 19 ka). Specifically, the lake sediments from this time interval are an unoxidised grey clay, which contains numerous sponge spicules. In addition, pollen records from these latitudes suggest the survival of rainforest at lower elevations to the east through the LGM (e.g. Moss et al., 2013), indicating persistence of moisture availability. Our argument for the persistence, and perhaps intensification, of winter westerlies throughout the LGM at LLL is also confirmed by observations made at North Stradbroke Island some 300 km to the north-north-east of our site (Petherick et al., 2009; McGowan et al., 2008). North Stradbroke Island lies at the very northern edge of the westerlies zone, and the accession of fine aeolian material into a dune lake there indicates that the winter westerlies were operative at the LGM in South East Queensland at 27.20∘ S (Petherick et al., 2009; McGowan et al., 2009), just as the westerlies operate today in this region.

A secondary peak in grain ages is observed in all three lunette samples. This peak is less well defined but in all three cases relates to the early to mid-Holocene between 9 and 6 ka. Work from the lake (Woodward et al., 2011) has already demonstrated that the early Holocene was the last phase, before the modern anthropogenically modified lake, with lake full conditions as represented by extensive Eleocharis beds. Wind waves would have been effective on the lake, and we infer partial reactivation of the lunette at this time.

We note a third grain age peak in one lunette sample (EVA1230) at ca. 3 ka. This is both the weakest individual age peak and not replicated at any other site. It is possible that this represents a dune re-activation event; bioturbation; or even Aboriginal usage of the site, which has been proposed to have intensified during the late Holocene (post 4300 years; Beck et al., 2015). At this stage this event, if real, is still poorly controlled chronologically, and we do not interpret it further.

Overall, our evidence demonstrates that, at the LGM, winter westerly winds were strong enough to form the eastern shoreline lunette in a single phase, with possible later reactivation during the early Holocene. Critically, foredune activation depends as much on high water levels in the lake allowing the wave delivery of sediment to the eastern beach as it does on sand-mobilising winds (Bowler, 1983). During the Pleistocene, elevations above 800 m in the region were subject to extensive, active development of block deposits, screes, and solifluction lobes, indicating winter cooling of at least 10.5 ∘C relative to present (Slee and Shulmeister, 2015). Reduced evaporation due to lower temperatures (e.g. Hesse et al., 2003) and transfer of flow from throughflow/baseflow to overland flow due to increased snow cover (Reinfelds et al., 2014) at this time are likely to have been sufficient to cause the change to a positive hydrological balance in the lake.

For the intervening periods, at least in the Holocene, the evidence (Woodward et al., 2014a) suggests that water levels were lower and/or even that the lake was ephemeral. It is highly unlikely that sand would be transported to the high-stand beach during low lake levels. If the entire basin floor fully dried out, pelletised clays might be expected, and yet none are observed. There are two likely reasons for this. Firstly, this high-elevation site is unlikely to become very arid even during relatively dry phases when swampy conditions probably persisted on the basin floor. Similarly, it is unlikely that salt formation is significant in this setting, and clay pelletisation may not occur. This is similar to observations from Lake George, which also lies within a cool temperate climate setting along the Great Dividing Range (Fitzsimmons and Barrows, 2010).

In summary, these records strongly suggest that for the two intervals recorded (the LGM and early Holocene) the overall circulation conditions at LLL were very similar to the present day. This region presently lies near the northern limit of westerly penetration in winter. For the intervening periods, absence of evidence is not evidence of absence; if the winter westerly winds lay at this latitude during peak warming in the early Holocene and during the LGM, it seems reasonable to suppose that this track has been persistent over the last 25 kyr.

The track of the Australian winter westerlies during the LGM has been a source of contention for some time, with both poleward and equatorward changes argued for (e.g. Harrison and Dodson, 1993; Hesse, 1994; Shulmeister et al., 2004). One possibility is that the westerly lay north of its current track during the LGM and that the timing of the westerlies at LLL shifted seasonally. A northward shift of ∼ 3∘ (350 km) in the position of the westerly wind belt during marine isotope stage (MIS) 2 was recorded in sediments from marine cores in the Tasman Sea (Hesse, 1994). Analysis of the aeolian component of lake sediments on North Stradbroke Island at 27∘ S for the period 25–22 ka indicates dust sources in the SW Murray–Darling Basin, with a secondary component from WNW of the site (Petherick et al., 2009). These findings are consistent with either no change or a possible northward shift in the westerlies but are not consistent with the poleward contraction of the westerlies in eastern Australia at the LGM.