Sedimentology of a Late Quaternary lacustrine record from the south‐eastern Carpathian Basin

The Upper Pleistocene geoarchives in the south‐eastern Carpathian Basin are represented predominantly by loess–palaeosol records. In 2015, a 10 m sediment core composed of clay‐rich lacustrine sediments was recovered by vibracoring a dry lake basin located between the Vršac Mountains (Serbia) and the Banat Sands in the south‐eastern Carpathian Basin; a location relevant for placing regional archaeological results in a palaeoenvironmental context. Here, we present results from geoelectrical prospection and a lithostratigraphic interpretation of this sequence supported by a detailed granulometric study supplemented by ostracod analysis. An age model based on luminescence dating is discussed against sedimentological proxy data and its implication for palaeoenvironmental change. The cores show a stratigraphy of lighter ochre‐coloured and darker greyish sediment, related to the deposition of clay and silt trapped in an aquatic environment. Geophysical measurements show ~20 m thick lacustrine sediments. The grain‐size distributions including the variability in fine clay are indicative of a lacustrine environment. Fine particles were brought into the depositional environments by aquatic input and settled from suspension; also, direct dust input is constrained by grain‐size results. Riverine input and aeolian dust input interplayed at the locality.


Introduction
The Carpathian Basin (Middle Danube Basin) is a key area to investigate the influence of the continental, Mediterranean and Atlantic climate interaction over Europe during the Quaternary (Stevens et al., 2011;Obreht et al., 2016). Large areas of the basin are covered by floodplain deposits and aeolian sediments (aeolian sands, loess, and loess-like sediments; Haase et al., 2007;Lindner, 20120167;Lehmkuhl et al., 2018Lehmkuhl et al., , 2020. Investigating these aeolian sediments led to a better understanding of Quaternary palaeoclimate evolution (see e.g. Marković et al., 2015). The understanding of past environmental conditions relevant for loess and loess-like sediments relies on biological proxies; in this area usually malacological assemblages (e.g. Marković et al., 2007;Molnár et al., 2010;Sümegi et al., 2012aSümegi et al., , 2013a, stable isotopes (Banak et al., 2016) and n-alkanes. For an overview, see Obreht et al., (2019). It is widely accepted that Late Pleistocene palaeoenvironments in the southern Carpathian Basin were dominated by open steppe grasslands, and partly forest steppes (Zech et al., 2009;Sümegi et al., 2012b;Marković et al., 2007Marković et al., , 2018. Establishing the environmental conditions and dominating vegetation is still challenging. Also, constraining the exact timing of palaeoenvironmental changes is debated (Willis et al., 2000;Rudner and Sümegi, 2001;Willis and van Andel, 2004). In contrast, lacustrine geoarchives can be excellent recorders of e.g., pollen and other biological proxies, thus allowing for more reliable palaeoenvironmental reconstructions. These are available for long and high-resolution sequences close to the Mediterranean region and in the mountain ranges of the southern Balkans, such as Tenagi Philippon (e.g. Tzedakis et al., 2006;Pross et al., 2015), Lake Ioannina, Lake Ohrid (e.g. Just et al., 2019;Wagner et al., 2019, and references therein) and Lake Prespa (Panagiotopoulos et al., 2013). Unfortunately, only one long lacustrine record from the surrounding Carpathians exists, from Lake St. Ana in Romania (Magyari et al., 2009(Magyari et al., , 2014Tóth et al., 2018), representing mainly mountain environments and does not span the entire last glacial cycle (Feurdean et al., 2014). However, those records indicate a more complex environmental evolution of the Carpathian Mountains and adjacent areas during the last glacial cycle, with varying phases of grassland and forest domination (Sümegi et al., 2013b). Therefore, the rising challenge is to connect the palaeoenvironmental conditions as inferred from lowland areas of the Carpathian Basin to the surrounding mountain areas of the Carpathian range (see also, Staubwasser et al., 2018).
Until now, there are not many lacustrine or peat records (Sümegi et al., 2011b, and references therein) from the lowland areas in the south-eastern Carpathian Basin reaching into the last glacial cycle, and most records are overbank or oxbow lake deposits related to the large river systems in the basin. However, several dry lake basins are present in the area around Vršac (Serbia) containing sediments deposited in palaeolakes and shallow swamps (Fig. 1). Here we present and discuss data highlighting the importance of one of these lacustrine geoarchives from the edge of the Carpathian Basin. Besides the lithostratigraphic core description, we present a chronology based on luminescence and radiocarbon dating, and grain-size data.
Understanding the Late Pleistocene environmental history of the Vršac Basin ( Fig. 1) is archaeologically relevant due to its proximity to a number of important human fossils and early Upper Palaeolithic sites. Across the border in Romania (c. 55 km east), the Peș tera cu Oase has yielded one of the earliest securely dated modern human fossils in Europe (c. 36-34 ka 14 C years BP, Trinkaus et al., 2003Trinkaus et al., , 2012 with Neanderthal admixture (Fu et al., 2015). In and around the Vršac Basin, rich Aurignacian sites and find spots occur, namely Crvenka-At, Românesţi, Cosava, Temeresţi, Tincova and Tabula Traiana (Mihailović, 2011;Anghelinu et al., 2012;Borić et al., 2012;Sitlivy et al., 2012Sitlivy et al., , 2014Chu et al., 2014Chu et al., , 2016Chu et al., , 2019Nett et al., 2021). Given the ongoing discussion concerning the role of climate in Late Pleistocene population turnover in the region (Müller et al., 2011;Staubwasser et al., 2018;Alex et al., 2019), the Vršac core may provide a valuable cross-check to local palaeoenvironmental reconstructions for late surviving Neanderthal populations and the earliest modern humans in Europe, as derived from loess data (Schmidt et al., 2013;Kels et al., 2014;Obreht et al., 2017). It may also contribute to a better understanding of the postulated Danube catchment as a migration corridor towards central and western Europe (Chu, 2018, and references therein).
The aim of this study is to provide a stratigraphy and sedimentological and palaeoclimatological interpretation for the Late Quaternary evolution of the Vršac Basin, and place it in the context of loess geoarchives. In addition, the implications of palaeoenvironmental change in relation to human occupation in the area are discussed.

Study area
North and north-west of Vršac, several morphological depressions exist. These were occupied by lakes and swamps before their drainage in historical times (Timár et al., 2008). Our coring was performed in the Alibunar Depression (see Fig. 1), which was described as a 'Morast' (German for mire/mud/ swamp) in the historical map of Müller (1769). It was drained during the late 18th and early 19th centuries (Müller, 1769;Timár et al., 2008). The depression is located between the loess and dune fields of the Banat Sands (Deliblatska Peščara) in the west, and the Vršac Mountains in the east. It is one of several oval landforms in the area; these have a maximal elevation of c. 75 m above mean sea level (amsl) and their maximal extent in west-south-west-east-north-east direction. They form endorheic basins and are the lowest points in the landscape, being~10-15 m below separating ridges, and 40 m (loess plateau at Alibunar) deeper than the surrounding geomorphological units. The depressions are separated by longitudinal positive landforms. The ridge south-east of our coring site consists of silt-dominated and partly gravelly, presumably fluvial, overbank deposits, turning to sandy and gravel deposits downwards (Nett et al., 2021). From a tectonic point of view, the area under investigation is part of a dextral transpressive wrenching system bordering the Carpathian Basin in the south and creating closely neighbouring subsidence and uplift structures (Horváth et al., 2015;Sušić et al., 2016). The still ongoing Neogene subsidence was in the order of 1-2 km in this area (Marović et al., 2007), and the Quaternary deposits reach thicknesses of about 450 m in the northern Serbian Banat (Pigott and Radivojević, 2010). Observed west-south-west/east-north-east striking oval forms north-west of Vršac cut into loess deposits in the west and to a minor extent in the east, suggesting that the erosion connected to the maximum water tables in these depressional landforms post-date the last phase of massive loess accretion. The genesis of the Vršac Mountains with their steep northern flanks and relatively gentle southern slopes is bound to the Neogene tectonic activity as well (e.g. Matenco and Radivojević, 2012).
The Banat Sands in the south-west of the study area represent a vast body of aeolian transported deposits with longitudinal landforms (Fig. 1A), indicating general sediment input from the south-east including the Danube floodplain (Gavrilov et al., 2018). A loess ridge reaching maximum elevation of c. 250 m amsl borders the Banat Sands to the north-east. Between this ridge and the here focused depression, the landscape is covered by loess, exhibiting a surface gently dipping towards the north-east from~150 m to~100 m to the rim of the depression (Fig. 1A). The south-eastern aeolian dust transport direction, as indicated by morphology of the Banat Sands, may play a role in larger parts of the southeastern Carpathian Basin (Zeeden et al., 2007;Barbu et al., 2009;Obreht et al., 2015), and is probably related to the influence of the south-east wind (locally known as the Coşava or Košava wind; Gavrilov et al., 2018). Outside the depressions and the tectonic push-up structures like the Vršac Mountains and the ridges separating the depressions and surrounding the Banat Sands, a loess cover in plateau facies forms an aeolian blanket. In the southern Carpathian Basin, the last glacial cycle experienced high sedimentation rates of aeolian dust of c. 5-15 cm/ka in most places (see, e.g. Buggle et al., 2009;Fitzsimmons et al., 2012;Marković et al., 2014;Fenn et al., 2020), which decreases with increasing distance from the large river systems. The depressions near Vršac show a west-south-west/east-north-east orientation, and are therefore unlikely to be related to aeolian processes, since southeasterly winds were probably persistent at least from the Last Glacial Maximum (LGM; Gavrilov et al., 2018). We interpret them to be linked to tectonic activity. This direction is also indicated by Marović et al. (2007) for faults in the area, underlining the tectonic origin of the depressions.
In the north-west of the Vršac Mountains, archaeological excavations yielded Aurignacian assemblages. The find layers occur at the southern base of a ridge composed of a mix of fine to coarse clastic sediments. Coarse and subangular material including small pebbles occurs at its base. Separated from the Vršac Mountains by a small depression, this ridge was likely formed as a tectonic push-up. Inside this ridge, coarser deposits (sand, gravels) of presently unknown facies occur, bearing rich archaeological assemblages that are thought to testify to the presence of the earliest modern humans in Europe some 40 ka ago (Mihailović, 2011;Chu et al., 2014Chu et al., , 2016Nett et al., 2021).
Coring was conducted west-north-west of the archaeological sites ( Fig. 1) in the large Alibunar Depression to obtain sediments for palaeoenvironmental reconstructions. The coring location was chosen after geophysical prospection by means of electrical resistivity tomography (ERT; Figs. 2 and S1-S3).

Electrical resistivity tomography and coring
ERT was conducted to detect stratigraphic differences within the near-surface Quaternary deposits and the underlying bedrock topography using a Syscal R1+ Switch 48 multielectrode device (Iris Instruments) and a Wenner-Schlumberger electrode array. Data were inverted incorporating the topography using the Res2Dinv inversion program (Geotomo Software). Three sediment cores were obtained by vibracoring. Steel augers (8 cm and 6 cm in diameter) were used in combination with an automotive drill rig (type Nordmeyer RS 0/2.3). Sediment cores were cleaned, photographed and described. Additional coring was conducted using closed auger heads (63 mm in diameter with a polypropylene liner of 50 mm diameter). The elevation and geographical position of all cores and ERT transects were measured using a differential GPS (type Topcon HiPerPRO).

Dating and age-depth model construction
Eight samples of polymineral fine grains (4-11 µm) were taken from halved sediment core VRT1A (diameter 5 cm, light-tight plastic liner), and were prepared for luminescence dating (Table 1) using conventional sample preparation techniques (Frechen et al., 1996). All measurements were carried out on an automated Risø TL/OSL DA 20 reader equipped with a calibrated 90 Sr beta source using the post-infrared infrared stimulated luminescence signal measured at 290°C (p-IR IRSL 290 , Thiel et al., 2011). Stimulation was carried out with infrared diodes (870 nm, FWHM = 40) and the signals were detected through an interference filter (410 nm). The initial 4 s of the signal minus a background of the last 20 s was used for p-IR IRSL 290 dating. Laboratory experiments included prior-IR stimulation temperature tests (Buylaert et al., 2012) using a temperature range from 50°C to 220°C, and preheat plateau and dose-recovery tests. For the dose-recovery tests, the samples were illuminated for 24 h in a Hönle SOL2 solar simulator and a laboratory dose in the range of the natural dose was given to the samples, the prior-IR stimulation temperature was kept constant at 50°C and the post-IR IRSL tracked the preheat temperature by −20°C. The apparent residual dose after 24 h bleaching in the solar simulator was measured using routine measurements (Thiel et al., 2011). Additionally, infrared stimulation measured at 50°C (IR 50 , Wallinga et al., 2000;Preusser, 2003) was carried out for samples C-L4422 to C-L4424 and C-L4029 to C-L4031. Equivalent doses were calculated with an arithmetic mean. Fading tests (Auclair et al., 2003) were carried out for all samples except C-L4032 and C-L4033 for both the IR 50 and the p-IR-IRSL 290 measurement protocols. Only the IR 50 ages were corrected for fading following the approach of Huntley and Lamothe (2001). The environmental dose rate was measured using high-resolution gamma-ray spectrometry. The dose rate was calculated with DRAC v.1.2 (Durcan et al., 2015), using conversion factors of Guérin et al. (2011), an a-value of 0.12 ± 0.02 and the measured water content. The internal beta dose-rate contribution of the polymineral fine grain samples was calculated by assuming a potassium content of 12.5 ± 0.5% (Huntley and Baril, 1997). Alpha and beta attenuation factors of Bell (1980) and Mejdahl (1979) were used. The cosmic dose rate was calculated following Prescott and Hutton (1994).
The resulting IR 50 and pIRIR 290 ages were used to construct an age-depth model. As we have no initial information on the depositional system and its homogeneity, we deliberately use the age-depth modelling approach of Zeeden et al. (2018), which does not make assumptions on sedimentation rates. This approach initially models the relative contributions of random and systematic uncertainty of an age dataset. This is done by using inversions as an indication of random uncertainty, and the lack of inversions as an indication of systematic uncertainty; it has been demonstrated to lead to realistic results (Zeeden et al., 2018). In a second step, only the random uncertainty is used for an age-depth model. The systematic uncertainty is added again afterwards. In our opinion, this is the most robust method available for building reliable age models based on luminescence dates. AMS 14 C dating of sediment bulk organic matter from two samples (depth: 2.2 m, 6.6 m) was carried out by BETA analytics. Samples were sieved, and the organic matter content of the < 180 µm fraction was exacted from the sediment and used for dating. This organic material did not include identifiable macroscopic remains. Calibration used the INTCAL13 database (Reimer et al., 2013); results are displayed in Table 2.

Ostracod analysis
Sediment material was screened for ostracod valves at eight discrete intervals. About 5 ml sediment was extracted from 1 cm segments and gently wet-sieved on a 150 µm screen. The residue was freeze-dried and ostracod valves were hand-picked under a stereo dissecting microscope. Identification keys and taxonomic Table 1. Dose-rate data, equivalent dose values (9.8 mm aliquots) and pIRIR 290 ages. m b. s. = meters below surface; PM = polymineral fine grains; W% = measured water content (water mass over dry sediment mass); U = Uranium; Th = Thorium; K = Potassium; age calculation is based on cosmic dose calculation after Prescott and Hutton (1994), conversion factors of Guérin et al. (2011) and the measured water content; De = equivalent dose. The internal beta dose-rate contribution of the feldspar samples was calculated by assuming a potassium content of 12.5 ± 0.5% (Huntley and Baril, 1997 papers were used to identify the ostracod species and infer palaeoenvironmental conditions (Griffiths, 1995;Meisch, 2000;Mesquita-Joanes et al., 2012).

Electrical resistivity tomography and sediment coring
In general, inversion models of all ERT measurements show a two-partite structure of the subsurface with an upper unit of low resistivity values (between 6 Ωm and 16 Ωm) and a lower unit of increasing resistivities (~50 Ωm in VRT ERT 3, 77 Ωm in VRT 1 and 148 Ωm in VRT ERT 3; Figs. 2 and S1-S3). Only in VRT ERT 2, higher resistivity values around 30 Ωm occur close to the surface between 30 m and 65 m and between 120 m and 160 m of the measured transect in channel-like structures. The transition from lower to higher resistivities is observed in all ERT transects at a depth of 17-18 m below surface (b.s.), approximately 56-57 m amsl. In order to retrieve a preferably undisturbed sedimentary sequence, coring was conducted at 115 m of transect VRT ERT 1. Three cores were drilled in direct proximity to several decimetres at the drill site. The first core (VRT1) was cored as reconnaissance core with open augers to have control on the recovered sediments; it was drilled to a depth of 10.0 m. The second (VRT1A) and third (VRT1B) cores were drilled closed to a depth of 10.0 m (VRT1A) and 5.5 m (VRT1B; Figs. S4-S6), respectively. These were taken to obtain an overlapping composite record. Figure 3 shows the sedimentary composition of vibracore VRT 1. Below the plough horizon, the upper part of the core (0.39-3.69 m, Fig. 3) is dominated by clay-rich silt and silt-rich clay of rusty greyish to greenish colours. While the upper 2.00 m show low to moderate carbonate content, sediment from 2.00 m to 4.31 m is free of calcium carbonate. The sediment between 3.69 m and 4.71 m is characterised by darker greyish colours and increasing amounts of carbonates and mollusc detritus towards it base. A distinct shift in the sedimentary composition is visible at 4.71 m, where strongly decreasing clay content and a colour change towards light rusty grey was documented. This shift is not resolved by ERT measurements. The lower transition of these silt-dominated calcareous sediments occurs at 6.23 m. Below this level, the clay content increases and the carbonate content decreases again. The lower part of the core is again dominated by clay-rich silty sediments of rusty greyish to greenish colours. Only between 7.61 m and 8.33 m, slightly decreasing clay content accompanied by a change towards lighter greyish to brownish colours were documented. The overall fine-grained composition of the drill core is well reflected by VRT ERT 1 as the core is limited to the upper part of the low resistivity unit (between 6 Ωm and 16 Ωm, corresponding to clay and silty clay). Due to high drilling resistance, the transition to the underlying unit with increasing resistivity values, most likely corresponding to coarser-grained sediments, between 17 m b.s. and 18 m b.s. could not be reached.
Dating of sediment organic matter of two 14 C samples resulted in calibrated age ranges of 9450-9525 a cal BP at 2.2 m depth and of 22 385-22 530 a cal BP at 6.6 m depth (Table 2; Fig. 4).
An age-depth model builds a stratigraphic interpretation of optically stimulated luminescence (OSL) ages for the upper part of the core. Age increases almost linearly with depth (Fig. 4A). Figure 4B shows the grain-size distribution of the vibracore VRT 1. With the exception of several samples between 4.70 m and 7.70 m depth which are monomodal, the samples are poorly sorted and polymodal. However, each sample has a main mode or at least a subordinate mode within the middle-or coarse-silt fraction. For the sediments between 1.40 m and 4.20 m there is a distinct minor mode within the medium clay fraction visible. From 4.70 m to 6.10 m depth, the curve progression of the particle-size distribution has a single maximum at~20 µm; the main modes show a fining upwards from 20 µm to 26 µm. Also, the sediments between 6.

Ostracod analysis
Eight samples were analysed for ostracod valve occurrence. In general, the material was dominated by lacustrine species and

Sedimentology and stratigraphy
Vibracoring stopped at a depth of 10.0 m b.s. due to the stiffness of the sediments. The facies of the sediments underlying the lacustrine sediments remains unclear although ERT data indicate another c. 8.0 m of similar sediments. In contrast to VRT ERT 1, higher resistivity values close to the surface of VRT ERT 2 (see Fig. S2) indicate channel formation between 30 m and 60 m and are even more defined between 120 m and 160 m of the transect; these channels are most likely filled with coarser-grained sediments. In addition, the underlying deposits denominated as pre-lake sediments in Figs. 2, S2 and S3 show a surface dissected by a channel-like structure of low resistivity values, suggesting erosional processes prior to the deposition of fine-grained sediments in VRT ERT 2. In all ERT measurements, higher resistivity values of the lower unit indicate a change to coarser-grained sediments. The described shifts in grain size in correspondence with the observed stratigraphic sequence cannot be resolved by geoelectric measurements due to the decreasing sensitivity of ERT measurements with depth and overall small differences in grain size (Fischer et al., 2016;Wunderlich et al., 2018).
The cored Alibunar Basin north-west of Vršac reveals a sediment sequence forming the infill of a largely tectonically controlled local deposition centre, in which subsidence exceeds sediment input (as visible among others from the steep cliff at its western edge). Nonetheless, stratigraphic completeness and sedimentary continuity is likely but cannot be assured. The loess-like grain-size distributions (compare the density functions from aeolian loess from Krems in Sprafke et al., 2020) at 4.7 m to 6.2 m and in the uppermost c. 1.4 m of the core, suggest a prevailing aeolian sediment input into an aquatic system. Also, sediment input directly from the catchment of short creeks draining the eastern fringe of the loess plateau at Alibunar or through erosion of the loess cliffs to the west of the depression may play a role. In between (c. 1.4-4.7 m) and below (from~6.2 m downwards), the multimodal distribution including rather high amounts of clay suggests a complex sediment origin. Especially the high content of submicron material at 1.4-4.3 m depth indicates a lacustrine deposition environment, which has acted as a continuous trap for non-aeolian sediments, including a substantial part of suspended submicron particles. Submicron particles can either be deposited under prevailing still water conditions within an aquatic system (O'Melia, 1980;Torres et al., 2005;Wilkinson and Lead, 2007;Dingman, 2009), may be part of autochthone bioproduction (diatoms, calcareous algae; the presence of these has not been investigated), or can originate from soil formation (Sun et al., 2004(Sun et al., , 2006. Because we do not see any reason for a discontinuous input of aeolian fine component, we interpret a detrital origin of these particles through the distal settlement of particles from superficial runoff (e.g. Torres et al., 2005;Vandenberghe et al., 2018) from the catchment. This includes a sediment component from the Banat Sands and also from the north-east and south, where erosion landforms and discharge structures are present and connected only to the basin. In general, grain-size distributions are similar to those from loess at Semlac in western Romania , but are in most of the record supplemented by an additional finer component.
The ostracod assemblage clearly demonstrates a continuous lacustrine environment throughout the fossil record. We propose that the ostracod valves are partly transported due to the poor preservation condition or reflected by an incomplete ontogenetic record, where juvenile valves of Candona sp. are exclusively present in low abundance. High sedimentation rates and/or low oxygen availability at the sediment surface may also lead to a similar ostracod signal. A proper ontogenetic record is found in samples with dominating juvenile valves in relation to adult morphotype autochthonous populations, as in samples at 6.3 m, 5.3 m, and 4.3 m core depth. More specifically, we infer water temperature estimates from the Holarctic species, Candona neglecta, known from other reference studies (Viehberg, 2006). The species has a temperature optimum of about 4°C during its final moulting time in springtime, while the juvenile forms can tolerate water temperatures of c. 20°C and above (Meisch, 2000;Viehberg, 2006). Another dominating species is Darwinula stevensoni, a sediment-dwelling species with the ability to penetrate subaquatic seepages in lacustrine environments. The sporadic occurrence of Cyclocypris sp. hints at shallow water depth at the time of occurrence with an abundant macrophyte stand at the coring site (Meisch, 2000).
Based on the macroscopic core description and the available grain-size data, five lithological units can be discriminated: from top to~1.20 m a silt dominated Unit 1 (#1); downwards to~4.20 m a unit rich in fine clay follows (#2); below and from~4.4 m down to c. 6.0 m, Unit 3 is similar to Unit 1 (#3). Unit 4 (#4) has the highest contribution of coarse silt and comprises the interval from~4.4 m to 8.5 m. The lowermost Unit 5 is similar to Unit 2, and forms the base of the core (from~8.5 m downwards; #5).

Challenges of age control
The chronological results of luminescence and radiocarbon dating contradict each other. While radiocarbon ages are < 23 ka for the upper 6.6 m, luminescence results increase constantly up to~120 ka at 5.6 m depth. This suggests that either the radiocarbon samples are underestimated (e.g. through contamination by younger carbon via bioturbation or by carbon exchange via roots; Gocke et al., 2017) or that the luminescence ages are overestimated (e.g. due to incomplete bleaching of the mineral grains before final deposition).
Two different luminescence dating measurement protocols were used, which generally agree with each other, although IR 50 ages are consistently younger. Since these two signals bleach at different rates (i.e. signal resetting is faster for IR 50 than for pIRIR 290 ; see also Veres et al., 2018), their agreement suggests that the investigated grains have been completely bleached prior to sediment deposition. This would also be expected from the depositional environment with a substantial part of suspended submicron particles from a distal source area and local runoff from the surrounding catchment that would have provided sufficient time for signal bleaching. The slight offset between both OSL datasets may be related to the fading correction of the IR 50 signal or a slight overestimation of the pIRIR 290 signal, which was observed, e.g., in a Serbian loess-palaeosol sequence (Avram et al., 2020). While one might argue that dating quartz minerals would be favourable in aquatic settings, since these bleach fast, it has already been demonstrated from the nearby investigation at the archaeological sites Crvenka-At (Nett et al., 2021) that dating quartz in this basin is problematic. In-lake bleaching is probably dependent on how clear the lake is/how much light penetrates to the bottom.
For 14 C dating performed on bulk sediment, it is well known that this can include younger material from root channels or from sediment movement along cracks or pipes or directly from living roots (e.g. Gocke et al., 2017). The measured 'soil organic matter' may be contaminated by younger carbon, which is known to lead to unrealistically young ages. The fact that 14 C ages show a low uncertainty which does not account for geological processes is well known (e.g. Telford et al., 2004;Trachsel and Telford, 2017).
The luminescence ages suggest low average sedimentation rates of around 4 cm/ka during the last c. 25 ka, while the 14 C age model suggests much higher sedimentation rates of 20-25 cm/ka for the uppermost 2.2 m. Comparing this to aeolian sedimentation rates in the Banat at Orlovat (Lukić et al., 2014;Marković et al., 2014;Obreht et al., 2015) and the Bačka loess plateau at Crvenka (Stevens et al., 2011), the luminescence age model suggests rather low sedimentation rates for this depocentre. Such low sedimentation during the Upper Pleistocene is challenging to explain and sharply contradicts recent results obtained from the Kisiljevo LPS located upwind 50 km to the south (Perić et al., in press). In contrast, the second 14 C date at 6.6 m depth implies rather high sedimentation rates.
When considering the 14 C ages as correct and extrapolating sedimentation rates to the lower part, the whole sequence would be formed during the last~30-40 ka. This would imply similar sediment grain sizes during the Holocene (top), the LGM (~3-7 m) and the late last glacial cycle in the lower part. While~3 m Holocene sediment fill may seem a reasonable accumulation comparable to other lacustrine geoarchives from the Carpathian Basin (e.g. Lake Fehér; Sümegi et al., 2011a), the lack of a clear sedimentological shift from the LGM to the Holocene is intriguing. In such a scenario, the relatively coarse Unit 2 would date from~18-13 ka, which does not relate to the LGM in the region (e.g. Sümegi et al., 2012a;Magyari et al., 2019) and is inconclusive or contradictory from a palaeoclimatic point of view. Grain sizes in our study are rather fine throughout, and also finer than at Semlac (Schulte et al., 2014;Zeeden et al., 2016), where loess-forming dust was suggested to be of non-local origin. Sümegi et al. (2011bSümegi et al. ( , 2013b, however, reported lacustrine sedimentation rates from the Danube floodplain and Mureş (Maros) alluvial fan lakes similar to what 14 C ages suggest for the Vršac coresbut in minerogenic deposits with signs of high organic production and containing abundant fossils. Sümegi et al. (2011b) also report genetically similar sediments of loess deposited in an aquatic environment, and report sedimentation rates of~7 cm/ ka for these sedimentswhich is in the same order as that suggested by luminescence ages. On the other hand, the Fehér Lake (Sümegi et al., 2011a), located only c. 150 km to the north-north-west, exhibits a similar lithofacies and an almost identical sedimentation rate when we rely on the obtained 14 C agesbut contains considerable amounts of autigenic carbonates and in-washed sediments (Sümegi et al., 2011a), and may therefore be expected to exhibit higher sedimentation rates than the geoarchive under investigation here. The southeastern part of the Alibunar Depression already has an area of > 100 km 2 and shows a catchment which also drains part of the Banat Sands. Lake level high stands are expected because of the presence of erosional cliffs in the loess west at Alibunar, indicating fluctuating lake levels.
While an age model based on the two available 14 C ages suggests a depositional history which is more similar to other much smaller lakes of oxbow character from the region, we are fully aware of issues with these. OSL ages suggest rather low sedimentation rates which are similar to dust accumulation in LPS plateau settings of the southern Carpathian Basin, but in contrast to recently obtained results from upwind LPS sites. Furthermore, low sedimentation rate scenarios do not favour an explanation that corroborates a lacustrine setting with a catchment of such a large dimension.

Sedimentological, palaeoenvironmental and archaeological implications
While the chronology is challenging, the grain-size data allow an interpretation regarding the palaeoenvironmental development, and thereby place the nearby archaeological sites (Chu et al., 2014(Chu et al., , 2016Nett et al., 2021) in a corresponding context. In addition, we discuss this geoarchive in the context of loess sedimentation.
Grain-size distributions are finest in Units 2 and 5, and considerably coarser in Unit 1 and especially Unit 3 (Fig. 4). In the Carpathian Basin, coarse aeolian grain sizes generally occur during glacials (e.g. Antoine et al., 2009;Bokhorst et al., 2011). Here we also expect an additional component of silt-sized (but not sand-sized) aquatic input, especially during high lake stands due to erosion in the catchment. While coarser grains would also be eroded, these were probably not transported to the centre of the depression, but were trapped by marginal lake vegetation. This aquatic component is expected to decrease during low lake stands because in this scenario input would not reach the centre of the depression.
Generally, environments during glacials are considered much dryer than during interglacials in the Eurasian loess belt (e.g. Heyman et al., 2013;Sirocko et al., 2016;Shao et al., 2018). In the Carpathian Basin this is evidenced by negligible soil formation during peak glacials and weakly developed steppe soils during interstadials (e.g. Marković et al., 2015Marković et al., , 2018. In contrast, warm interglacials experience strong soil formation. Consequently, high water tables in the depression are likely to be related to interglacial climates, and were less likely during interstadial periods. Also, autochthonous bioproduction is controlled by nutrient availability and temperature. In our opinion, Units 2 and 5 would relate to lake high stands, as evidenced by the clay which is probably related to distal settling. The erodibility of surface material is largely controlled by lake level and vegetation cover, which in turn depends on moisture. We propose that hinterland erosion took place predominantly during glacials when precipitation and lake level increased while vegetation cover was still restricted. This effect would be amplified by an extension of the catchment. The presence of fine sand in Units 3 and 4 is unlikely to be facilitated through riverine input in the depression centre, but is interpreted to be aeolian and may be the result of periods of enhanced storminess of the Košava wind system, as also suggested by their short duration (Fig. 4). In the relative vicinity, large dunes exist in the Banat Sands, demonstrating appropriate wind energy for sand transport through the Košava wind system (Gavrilov et al., 2018). Furthermore, long-range transport of sand particles has been demonstrated from the Atlantic (e.g. von Suchodoletz et al., 2013;van der Does et al., 2018), and is suggested to play a role here.
Following the luminescence age model (Fig. 4), Units 4 (to 5?) may be interpreted as the penultimate glacial phase and Unit 3 as corresponding to the last interglacial. Unit 2 may be related to the early Last Glacial (i.e. MIS 4-3), and Unit 1 would comprise the peak glacial and possibly the Holocene. Stratigraphy and environmental conditions reconstructed based on age models relying on the luminescence ages imply the opposite conclusions from those relying on geomorphological and sedimentological processes ruling the steppe environment of south-eastern Europe (e.g. Obreht et al., 2019, and references therein). The OSL ages imply that the lake was more extended during glacial periods (Units 5 and 2) because the fine grain sizes related to distal settling from suspensions occur in these intervals, which contradicts the idea of rather dry glacials.
Our work relates the grain-size data to relative water table and to climate states. However, a definite connection to the established climate evolution of the Carpathian Basin (e.g. Marković et al., 2015Marković et al., , 2018Obreht et al., 2019) cannot be made using either the 14 C or luminescence chronologies.
The implications of this past environmental reconstruction can provide context for the local Palaeolithic record that is otherwise known from loess and karstic deposits. Our ages are too insecure to make direct temporal connections with the modern human fossils and local Palaeolithic sites such as Crvenka-At. As a diachronic trend, Unit 2 suggests that the local lowlands were characterised by increasingly wet and diverse (representing fluvial, lacustrine, loess landscapes, dune environments) biomes during MIS 3 and we might begin to envision scenarios where these changing conditions may have impacted local Neanderthal biogeography (e.g. Mihailović, 2020). They may also have attracted intrepid modern humans that possibly had a broader dietary spectrum that may have included freshwater aquatic resources (Richards and Trinkaus, 2009;Trinkaus et al., 2009). Behavioural changes in the Banat c. 40 ka ago, are marked by abrupt changes in lithic technology, but also in the size and location of archaeological sites. Sites in the lower elevations of the Banat show little or no evidence for earlier Middle Palaeolithic technology below the early Upper Palaeolithic assemblages, suggesting that modern human settlement in the area did not map directly onto those of earlier populations. However, whether these patterns were climatically controlled, either directly or indirectly, remains an aspect of future research. Especially helpful would be archaeological finds around 40 ka in a highly resolved palaeoclimate archive, where geophysical and geochemical proxies would allow sites to be placed directly into a palaeoclimatic context. Alternatively, precise and robust dating of both archaeological sites and palaeoclimate archives is needed to link these in an unambiguous manner.

Conclusions
The investigated sediment core near Vršac is a new continuous geoarchive, which serves for a better understanding of the Late Quaternary climate and environmental dynamics in the Carpathian Basin. The inferred results complement the known environmental reconstructions that derive mainly from loesspalaeosol sequences. The lacustrine sediment represents the first non-loess geoarchive from the region that potentially exceeds the last glacial period. Its sedimentation history and microfossil record provides insight into the past environment and interplay of lacustrine and aeolian sedimentation in a former freshwater lake.
Valuable/supplementary geoelectric analyses added stratigraphic information to the depositional environment/setting of the sediment basin. The geophysical prospection estimates the thickness to a total of c. 20 m sediment. A luminescence dating chronology of the upper half was established from drill core samples, but is contrasted by (only two) 14 C ages (Fig. 4A,  Tables 1 and 2). Assumptions of sedimentation rates based on comparisons with other geoarchives from the region can aid an age interpretation. However, it challenges the concept of dry conditions during glacials and associated geomorphological and sedimentological processes.
Grain-size analysis results indicate fine sediment settling in a distal lacustrine environment from c. 1.4 m to 4.2 m and at the base of the core (below 7 m). In contrast, other intervals show grain-size distributions similar to aeolian sediments, suggesting their additional input into the lacustrine setting by winds or by fluvial erosion from the catchment of the depression. Comparison with a loess geoarchive from the region shows that the aeolian-deposited material is rather fine at this site, suggesting mostly non-local dust sources, and a transport range beyond local transport , which is not typical for most thick loess-palaeosol sequences near the main dust sources in the Carpathian Basin (e.g. Basarin et al., 2014;Obreht et al., 2015). Our results suggest that in the vicinity (several km) of aeolian high-sedimentation areas which are dominated by local dust sources (Deliblatska Peščara), areas with low aeolian sedimentation coexist. This highlights that steppe landscapes in the Carpathian Basin are diverse and not solely loess-dominated, which is relevant for understanding early human environments and food sources.

Supporting information
Additional supporting information may be found in the online version of this article at the publisher's web-site. Figure S1: Position of ERT transects VRT 1-VRT 3 and vibracores VRT 1, VRT 1A, and VRT 1B. Figure S2: Electric resistivity tomography image VRT 2 near the drill site. Figure S3: Electric resistivity tomography image VRT 3 near the drill site. Figure S4: Comparison and stratigraphic correlation of cores VRT1 (light, right core), VRT1A (leftmost core segments until 10 m), and VRT1B (middle core in the left panel). Figure S4: Core images with depth scale of VRT1A. Gray parts are disturbed. Figure S5: Core images with depth scale of VRT1B. Figure S6: Dose recovery test of samples C-L4029, C-L4030, and C-L4422-4424. The preheat temperature was held constant at 320°C using a pIR 140 IR 290 stimulation. Samples were bleached in a Hönle Sol2 solar simulator for 24 h. The applied laboratory doses were in the range of the measured equivalent doses. Figure S7: First IR stimulation temperature test of samples C-L4029 and C-L4030. The first IR stimulation temperature was increased between 50°C and 140°C, the pIRIR stimulation temperature was kept constant at 290°C. A preheat plateau was observed between 50°C and 140°C for both samples. Figure S8: Decay curve and dose response curve of sample C-L4424 using pIR 140 IR 290 stimulation. Figure S9: Decay curve and dose response curve of sample C-L4424 using IR 50 stimulation. Figure S10: Dose recovery test of samples C-L4029, C-L4030, and C-L4422-4424. The preheat temperature was held constant at 290°C using an IR 50 stimulation. Samples were bleached in a Hönle Sol2 solar simulator for 24 h. The applied laboratory doses were in the range of the measured equivalent doses.

Data availability statement
The data that support the findings of this study will be placed in the openly available PANGAEA database.