Seasonal climate variations during Marine Isotope Stages 3 and 2 inferred from high‐resolution oxygen isotope ratios in horse tooth enamel from Lower Austria

We present sequential oxygen isotope records (δ18Ophosphate vs. VSMOW) of horse tooth enamel phosphate of six individuals from two adjacent Palaeolithic sites in Lower Austria. Three molars from the site Krems‐Wachtberg date to 33–31k cal a bp, and three molars from Kammern‐Grubgraben to 24–20k cal a bp. All teeth show seasonal isotope variations, which are used to reconstruct the annual oxygen isotope composition of drinking water (δ18Odw) and palaeotemperatures. Measured δ18Ophosphate values ranged from 8.6 to 13.0‰ and from 10.8 to 13.9‰ at Krems‐Wachtberg and Kammern‐Grubgraben, respectively. An inverse modelling approach was used to reconstruct summer and winter temperatures after a correction for glacial oceanic source water δ18O. Reconstructed annual δ18Odw was −16.4 ± 1.5‰ at Krems‐Wachtberg and −15.3 ± 1.4‰ at Kammern‐Grubgraben, resulting in annual temperatures of −5.7 ± 3.1 and −3.5 ± 2.9°C, respectively. Summer and winter temperatures reconstructed from individual teeth exhibit high seasonal variations with moderate summer temperatures and extremely low winter temperatures typical for a polar tundra climate. Isotopic differences between individuals are attributed to interannual climate variability or to different drinking water sources. Our reconstructed temperatures are, overall, consistent with previously reported values from European horse teeth, when taking regional differences into account.


Introduction
The last glacial climate repeatedly shifted from cooler (stadial) to warmer (interstadial) periods (Andersen et al., 2004;Rasmussen et al., 2014).This alternation from cooler to warmer climatic periods had a significant impact on the population density and distribution of potential prey species for Palaeolithic hunter-gatherers.Since Palaeolithic human land-use is closely related to the occurrence of animal food resources, it also depends strongly on the underlying climatic situation and phenological patterns (D'Errico andSánchez-Goñi, 2003, Finlayson, 2004;Maier et al., 2023).The methods used to reconstruct the palaeoclimate are generally based on proxies such as sediment analyses (Brauer, 2004), pollen spectra (Chevalier et al., 2020) or mammal assemblages (Fernández and Peláez-Campomanes, 2005).
When sampled sequentially, the isotopic variability in their tooth enamel thus reflects the seasonal variability of the isotopic composition of the ingested water.Palaeoclimate reconstructions often assume that the ingested water is isotopically equivalent to local precipitation (Pederzani and Britton, 2019).The δ 18 O value of precipitation (δ 18 O precipitation ) is strongly dependent on the temperature during the condensation of water vapour in clouds (Dansgaard, 1964;Gat, 1996;Clark and Fritz, 1997).Conversion equations are therefore required, also for the estimation of palaeotemperatures.Other isotope effects, known as altitudinal, continental or latitudinal effects, are related to the movement of air masses from the moisture source to the condensation area and are also partly dependent on temperature (Dansgaard, 1964;Gat, 1996).However, the ingested water can also come from sources other than precipitation, such as groundwater, lake and spring water, meltwater, leaf water or evaporated surface water, all of which can influence the estimation of the δ 18 O of drinking water (δ 18 O dw ).In addition, species-specific drinking behaviour also contributes to isotopic variation (Blyth, 2011;Pryor et al., 2014;Skrzypek et al., 2016;Pederzani and Britton, 2019).For example, there is a non-linear relationship between oxygen fluxes and body mass, which is why mammals with larger body size tend to reflect the δ 18 O of precipitation better than small mammals, as the former take up a higher proportion of water (Bryant and Froelich, 1995).
This study investigates δ 18 O variations in tooth enamel of horses from the last glacial.Since horses are obligate drinkers (Hinton, 1978;Groenendyk et al., 1988;Scheibe et al., 1998;Houpt et al., 2000), it can be assumed that their oxygen isotope composition of tooth enamel reflects the local environmental water.In addition, larger animals, such as horses, are often well represented at Upper Palaeolithic sites due to their importance for the subsistence of hunter-gatherers.Compared with migratory species, the oxygen isotope composition of horse teeth is more likely to reflect conditions near the site (Iacumin and Longinelli, 2002;Britton et al., 2019;Pederzani et al., 2021), as horses are generally more sedentary animals (Kaczensky et al., 2008).
The dating accuracy at both sites, Krems-Wachtberg and Kammern-Grubgraben, is not precise enough to attribute human occupation to either stadial or interstadial conditions.The aim of this study is therefore to identify possible climatic differences between the two sites and periods as well as to infer seasonal variations in the oxygen isotope composition of horse tooth enamel phosphate (δ 18 O PO4 ) and to elucidate the palaeoseasonality, palaeotemperature and palaeoenvironmental conditions in this particular region.
The absolute palaeotemperatures reconstructed for the late MIS 3 and MIS 2 help to better understand the climatic development and population dynamics in the study area.In particular, against the background of a massive population decline during the transition from the Gravettian to the Epigravettian in central Europe and the low persistence of human occupation during the LGM (Maier et al., 2021), the study contributes to a better understanding of possible adaptation strategies of hunter-gatherer communities.

Study sites
The Upper Palaeolithic sites of Krems-Wachtberg (33-31k cal a BP), and Kammern-Grubgraben (24-20k cal a BP) are located in Lower Austria at a distance of 12 km from each other (Figure 1).Krems-Wachtberg is located at the transition of the hilly area of the Wachau to the extensive alluvial plain of the Tullnerfeld.Situated in the southern to south-eastern part of a spur-shaped promontory (398 m a.s.l.), 50 m above the present-day Danube, it is slightly inclined to the south-east and thus offered wind-sheltered conditions for campsites of huntergatherer communities (Händel, 2017).Periglacial features observed in the Krems-Wachtberg loess sequence (Terhorst et al., 2014;Händel et al., 2021b) may indicate seasonal melting of continuous permafrost.Analyses of δ 13 C from bone collagen of different herbivorous taxa from this time interval suggest comparatively diverse food sources and thus a rather heterogeneous vegetation cover (Reiss et al., 2023).The probability range of radiocarbon dates for Krems-Wachtberg includes Greenland interstadials (GI) 5.2 and 5.1, but also Greenland stadials (GS) 6, 5.2 and 5.1 (Figure 2).However, based on sediment parameters and stratigraphic observations, Händel et al. (2021b) propose that the occupation represented by AH 4.4 at Krems-Wachtberg, comprising infant burials and hearths, is connected to stadial conditions, or, more precisely GS 5.2.
Kammern-Grubgraben, on the other hand, is located at the south-eastern end of the Moravian-Bohemian Hills on a loesscovered slope between two hills, the Heiligenstein and the Geißberg (Händel et al., 2021a).Sheltered from winds to the west, north and east, it is exposed towards the south and the valleys of the Kamp and the Danube rivers.The proximity to water and the protected topographic position provided favourable conditions for hunter-gatherer communities on a seasonal basis (Maier et al., 2021;Händel et al., 2021a).The period covered by radiocarbon dates from Kammern-Grubgraben includes GI 2.2 and 2.1, but also parts of the GS 3, 2.2 and 2.1 (Figure 2; Händel et al., 2021a).
Isotope data from herbivore bone collagen at both sites were investigated for isotopic niche patterns (Reiss et al., 2023).Compared with the Krems-Wachtberg site, where a wider range of herbivore δ 13 C and δ 15 N values and a broad isotopic niche breadth reflect diversified food sources, the isotopic niches at Kammern-Grubgraben decreased considerably and resulted in closely spaced clusters.This could be related to severe climate deterioration after the occupation of Krems-Wachtberg during the long GS 3 (Figure 2).
It is tempting to attribute the occupation phases to interstadial conditions at both sites, due to a more favourable climate.However, so far there is no evidence for this assumption.Therefore, this study aims to contribute absolute temperatures on a seasonal scale to test this assumption.

Horse teeth and sampling procedure
For sequential oxygen isotope analyses, horses were preferentially selected, as they need to drink water on a regular basis (Scheibe et al., 1998;Britton et al., 2019).Therefore, they are generally considered to more likely reflect δ 18 O of local environmental water in contrast to non-obligate drinking or migratory animals (Britton et al., 2019(Britton et al., , 2023)).In addition, the high-crowned teeth in the equine hypsodont dentition have the further advantage of providing a high temporal resolution, as equine molars and premolars grow to about 6-9 cm in length and have a growth rate of about 3-4 cm/year (Hoppe et al., 2004a;de Winter et al., 2016).Extant horses (Equus caballus) reach full tooth height at the age of 1 to 1.5 years (Hoppe et al., 2004a).
All analysed tooth samples from Krems-Wachtberg originate from the 2005-2015 excavations and derive from AH 4 (WA-AH 4.11 and WA-AH 4.4).While WA-AH 4.11 represents a find layer with redeposited material and therefore allows only for general statements on occupation, WA-AH 4.4 has a wellpreserved occupation area where specific activities can be well delimited in terms of content and space (Händel, 2017).Samples from Kammern-Grubgraben stem from the main occupation phase, i.e.AH 2 (Table 1).In total, six upper cheek horse teeth were selected.As isolated third molars (M3) can be identified most accurately among equine molars, three M3s from three individual adults were analysed from Kammern-Grubgraben.The same applies for Krems-Wachtberg, with the exception that in one case there was no M3 available and so we used a second premolar (P2) instead, which can also be anatomically identified with great accuracy, in contrast to P3, P4, M1 and M2.
This choice is justified by the fact that the P2 is formed at a similar season as the M3; however about one year earlier (de Winter et al., 2016).In addition, the formation of M3 and P2 occurs post-weaning and thus any nursing effect altering the δ 18 O record is avoided (Hoppe et al., 2004b;Britton et al., 2019).The  (Einwögerer et al., 2009;Händel, 2017) and (f) Kammern-Grubgraben 'AH 2' (Gilot, 1997;Haesaerts, 1990;Haesaerts et al., 2016;Händel et al., 2021a).All radiocarbon dates were recalibrated with the IntCal20 curve (Reimer et al., 2020) using Oxcal version 4.4 (Bronk Ramsey, 2009).Bars indicate 2 sigma (2σ) ranges.Panel (g) shows calcium (Ca 2+ ) as dust accumulation in Greenland, plotted inversely and on a logarithmic scale.Blue numbers represent Greenland Stadials and red numbers the respective Greenland Interstadials (Mayewski et al., 1997;Rasmussen et al., 2014).(h) The δ 18 O values (Rasmussen et al., 2014) and (i) temperature derived from δ 15 N of entrapped air bubbles (Kindler et al., 2014) are from the NGRIP record (Andersen et al., 2004;Rasmussen et al., 2014) and reported on the Greenland Ice Core Chronology (GICC05; Andersen et al., 2006).The vertical dashed line indicates the MIS 3/2 transition.[Color figure can be viewed at wileyonlinelibrary.com] teeth grow from the crown top, making the crown top the oldest part of the tooth and the crown base the youngest part (Figure 3).
Since the tooth is formed in the sequence from the tip of the crown, the masticatory surface, to the root tip (i.e.EDJ) in the jaw, the upper areas of the occlusal surface are worn off by a constant abrasion due to the grinding of food.This gradually reduces the tooth length during the lifetime with the beginning of crown formation disappearing first.The tooth specimens varied in length and state of abrasion.Prior to sequential sampling, all surfaces were cleaned by gently removing the uppermost layer using a microdrill (Saeshin, Korea).All teeth were sampled at a roughly 2.5 mm resolution on the mesial side along the right margin (Figure 3).Due to laboratory constraints, only every second sample was analysed in most cases.For the shorter specimen KG-912, every sample was analysed to achieve a higher resolution.

Phosphate extraction
Mammal tooth enamel mainly consists of hydroxyapatite, a crystalline calcium phosphate.Prior to the analysis of oxygen stable isotopes from phosphate, the PO 4 3-group must be isolated and precipitated again as Ag 3 PO 4 (O'Neil et al.,1994;Wenzel et al., 2000;Joachimski et al., 2009).
Phosphate extraction followed the method of Joachimski et al. (2009).As the first step, 2 mg of horse tooth enamel apatite was weighed into small polyethylene beakers and dissolved by adding 66 μl 2 M HNO 3 .The solution was then neutralised by adding 66 μl 2 M KOH.In the next step, Ca 2+ was precipitated as CaF 2 by adding 66 μl of 2 M HF.Solutions were then centrifuged at 3000 rpm for 15 min.Subsequently, the supernatant was transferred to previously weighed small (2 ml) polyethylene beakers.
Phosphate precipitated as tri-silver phosphate by adding 1000 μl silver amine solution (0.34 g AgNO 3 , 0.28 g NH 4 NO 3 , 1 ml NH 4 OH in 30 ml distilled water).Subsequently, the beakers were placed in a drying oven at 60°C for 8 h to precipitate Ag 3 PO 4 crystals (O'Neil et al., 1994;Joachimski et al., 2009).Ag 3 PO 4 crystals were then cleaned with deionised water at least five times and again stored in a drying oven at 60°C.In the final step, the dried Ag 3 PO 4 crystals were homogenised using a small agate mortar and pestle.

Oxygen stable isotope analyses
Between 0.2 and 0.3 mg of powdered Ag 3 PO 4 was weighed into silver foil capsules in triplicate and burned in a high temperature reduction furnace (TC-EA) which is coupled to a mass spectrometer (ThermoFisher Delta Plus V), where samples are isotopically analysed.At 1450°C, the silver phosphate is reduced, and CO is formed as the analyte gas (Vennemann et al., 2002, Joachimski et al., 2009), which is passed in a helium (He) stream through a gas chromatography column via a Conflo-IV interface to the mass spectrometer.
The stainless-steel capillary connecting the TC-EA to the Conflo III interface was immersed in a dewar vessel filled with liquid nitrogen so that volatilised phosphorus compounds are trapped and do no contaminate analytical devices.
All values are given in delta (δ) values in per mille (‰) relative to the international standard Vienna Standard Mean Ocean Water (VSMOW) according to the equation: The samples and respective internal standards were generally measured in triplicate.The measurements were calibrated by performing a two-point calibration (Paul et al., 2007) using NBS 120c (21.7‰) and a commercial Ag 3 PO 4 (9.9‰).A laboratory standard (TueA) was used as a control standard and processed together with the samples.All standards were calibrated to TU1 (21.11‰) and TU2 (5.45‰; Vennemann et al., 2002).External reproducibility, monitored by replicate analyses of samples as well as the laboratory standard was less than ±0.2‰ (1 SD).The average oxygen isotope composition of the internationally used standard NBS-120c was 21.7 ± 0.18‰ (n = 93).

Statistical analyses and isotope inverse modelling
All statistical tests were carried out with R studio (version R-4.2.2) and Excel (version 2304).Shapiro-Wilk tests were applied to test Table 1.List of sampled horse teeth.Given are the sample ID, the archaeological horizon (AH), the length of the sampled surface measured from the enamel-dentine junction (EDJ) in mm, the number of samples analysed within a tooth (n) and the type of tooth analysed.

Site
Sample  for normal distribution of data.Wilcoxon-Mann-Whitney tests were applied to test whether two groups come from the same true mean (Dytham, 2012).For all statistical analyses, a significance level of 0.05 was set.Delayed enamel growth and tooth mineralisation lead to damped and time-lagged seasonal isotopic oscillations, i.e. to a decrease in amplitudes and to a shift in the peak positions in the δ 18 O enamel values, respectively (Passey and Cerling 2002;Passey et al., 2005;Pederzani et al., 2021aPederzani et al., , 2021b)).To account for this phenomenon and, in the next step, to accurately reconstruct the summer and winter temperatures for each tooth, we used the inverse model described by Passey et al. (2005).For inverse modelling, we implemented the R code adapted from the Matlab code of Passey et al. (2005), available from the supplementary online repository of Britton et al. (2023).The species-specific parameters for horses were set to 22% for enamel content, 6 mm for enamel attachment length, and 28 mm for enamel maturation length (Passey and Cerling, 2002;Blumenthal et al., 2014).
Reconstruction of δ 18 O dw and temperatures δ 18 O dw originating from regional precipitation correlates with air temperature (T), which in turn relates to the period of bioapatite growth (Pryor et al., 2014).For the reconstruction of the local ambient water, we used the species-specific conversion equation for horses documented in Delgado Huertas et al. (1995), implementing the spreadsheet (Supplementary data sheet Z1) provided by Pryor et al. (2014)  In the second step, a correction of δ 18 O dw is required for the last glacial due to 18 O enrichment of the oceanic source water, which is the ultimate origin of all precipitation.For late MIS 3 an 18 O enrichment of 0.5‰ (Lea et al., 2002) is reported, whereas during the LGM, a sea level decline of 120 ± 10 m (Figure 2e; Lambeck et al., 2014) resulted in a mean increase of the global δ 18 O value of ocean water value by 1.2 ± 0.1‰ (Lea et al., 2002) or 0.8‰ (Schrag et al., 1996(Schrag et al., , 2002)).We therefore subtracted 0.5‰ (Lea et al., 2002) for the late MIS 3 (Krems-Wachtberg) and 1.2‰ (Lea et al., 2002) for MIS 2 (Kammern-Grubgraben) from the δ 18 O dw derived from Equation (2) to account for persistence of the source value throughout the terrestrial water cycle (Lécuyer et al., 2021).In the following data evaluation and discussion, we only refer to a correction of 0.5‰ for the late MIS 3 and of 1.2‰ for the LGM as we consider the correction factors reported by Lea et al. (2002) more reliable than the values reconstructed from sediment pore water in Schrag et al. (2002).However, we list the values corrected by 0.8‰ as reported in Schrag et al. (2002) in a supplementary table (Table S1) for completeness.
Long-term means of δ 18 O precipitation and temperature from 84 European sites (GNIP database; IAEA/WMO, 2023) were used to estimate mean annual temperatures.Mean temperatures of the warmest (July/August) and coldest (December/January) periods were considered representative for summer and winter temperatures, respectively (Table S2).According to the spreadsheet (Supplementary Data Sheet Z2) provided by Pryor et al. (2014) (5) At this point, it needs to be mentioned that with each conversion step, specific uncertainties increase and propagate to the final palaeotemperature estimates (Pryor et al., 2014;Lécuyer et al., 2021).
MATs were estimated from the non-modelled δ 18 O PO4 values (Table S1, S3), as the peak and trough values of the inversely modelled δ 18 O PO4 values can exhibit considerable uncertainties compared with the raw δ 18 O PO4 values (Pederzani et al., 2021a).A comparison of the different averaging methods shows no significant differences in the mean δ 18 O PO4 values and thus to the MAT calculations as described in Pederzani et al. (2021a).To ensure comparability, mean δ 18 O PO4 values were thus calculated as the mean of the summer peak and winter trough values analogously to Pederzani et al., (2021aPederzani et al., ( , 2021b) ) (Table S1, S3).To account for the time-lagged isotopic signal and the attenuation of seasonal isotopic variability (Passey and Cerling 2002;Passey et al., 2005;Pederzani et al., 2021a), summer and winter temperatures were calculated using the maximum and minimum values of the modelled data, respectively.For the inverse modelling of each tooth, an individual damping factor (ε²) was chosen and adjusted through an iterative fitting process as described in Passey et al. (2005).
All samples exhibit at least one annual cycle, i.e. at least one summer peak and one winter trough value.Of all samples, specimen WA-161267 is most striking as the seasonal variations are the least pronounced (Figure 4a).
The higher seasonal variation suggests that WA-163790 and WA-14501 originate from different years than sample WA-161267 (Figure 6a) with individual animals potentially being born years apart from each other.Alternatively, the individuals lived in different regions or used different water sources.At Kammern-Grubgraben, summer temperature estimates are slightly higher than at Krems-Wachtberg, ranging from 8.0 ± 3.9°C for KG-113 and 8.8 ± 3.9°C for KG-912 to 11.4 ± 3.7°C for KG-857 (Figure 6b).

Discussion
Palaeoclimatic conditions at Krems-Wachtberg and Kammern-Grubgraben Delayed enamel growth and tooth mineralisation lead to dampened and delayed seasonal isotopic variations (Passey and Cerling, 2002;Passey et al., 2005;Pederzani et al., 2021a;Britton et al., 2023), which is why the reconstruction of summer and winter temperatures is based on modelled data (maxima and minima of the δ 18 O modelled values).This accounts for the 'unaltered' seasonal signal, resulting in higher peaks than in the variations of the measured values (Figure 4).δ 18 O dw, as well as MAT, were reconstructed based on the measured δ 18 O PO4 (unmodelled) values and calculated based on the mean of maxima and minima (Pederzani et al., 2021a).Reconstructed mean δ 18 O dw and MAT values revealed overall low temperatures (Figure 5a, b) but strong seasonal changes during late MIS 3 at Krems-Wachtberg as well as during MIS 2 at Kammern-Grubgraben, as inferred from reconstructed summer and winter temperatures (Figure 6a, b).According to the Köppen-Geiger climate classification, a polar tundra climate prevailed during both time periods (Kottek et al., 2006;Peel et al., 2007), which is characterised by a MAT that is lower than or equal to 0°C and a maximum temperature that lies above 10°C for at least one month (Kottek et al., 2006).Modern-day climates of this type are found in polar regions, north-eastern Siberia and in high-altitude mountains, such as in the Altai-Sayan mountains, the latter often being considered to have a modern equivalent of the LGM climate in Europe (Řičánková et al., 2015;Badino et al., 2023).The overall slightly higher summer temperatures in Kammern-Grubgraben could be attributed to greater continentality due to lower sea level (Figure 2b) and an associated reduced water availability due to larger glaciated areas during the LGM.
Previous studies, however, suggested an ameliorated climate for Krems-Wachtberg compared with that of Kammern-Grubgraben (Maier et al., 2021;Reiss et al., 2023).For the time interval of Krems-Wachtberg, this is supported by a predominance of gley soils and/or cambisols in loess sequences from different, well-known sites in central Europe, such as Nussloch (Antoine et al., 2009;Moine et al., 2017), or Schwalbenberg (Vinnepand et al., 2023).δ 13 C and δ 15 N from herbivore bone collagen show broader niche widths suggesting a more diversified diet and vegetation cover (Reiss et al., 2023).The polar tundra climate is in accordance with regional pollen records that show prevailing forest tundra (MIS 3) and tundra (MIS 2) vegetation for the Alpine realm (Figure 2f; Heiri et al., 2014;Duprat-Oalid et al., 2017;Stojakowits et al., 2021;Kämpf et al., 2022).For the LGM, cold and stable conditions are evident in the NGRIP record (Figure 2), which is also supported by homogeneous loess accumulation and hardly any soil formation at Kammern-Grubgraben (Reiss et al., 2022).
However, large climatic differences between the two periods are not seen in the reconstructed palaeotemperatures in this study (Figures 5,6), even though Krems-Wachtberg exhibits overall lower temperatures and slightly higher summer temperatures were reconstructed for Kammern-Grubgraben.Regarding an occupation under stadial conditions for AH 4.4 at Krems-Wachtberg (Händel et al., 2014;Terhorst et al., 2014;Händel et al., 2021b) and interstadial conditions for AH 2 at Kammern-Grubgraben, the reconstructed seasonal temperatures are consistent.This is also evident from the NGRIP records, indicating slightly warmer conditions during the LGM (GI 2.2 and 2.1) and colder conditions during GS 5.1 and 5.2 (Figure 2).The reconstructed LGM climate is highly consistent with simulated LGM climate models for Europe reported in Strandberg et al. (2011).The regional climate model (RCA3) used in this study shows mean temperatures for the study area ranging from 6 to 12°C for the warmest month, up to −12°C for the coldest month and between −6 and 6°C for the annual mean (Strandberg et al., 2011).However, the differences when compared with the climatic conditions of the recent past are smaller: up to 9°C colder for the warmest month, between 12 and 15°C colder for the coldest month and up to 12°C colder for the annual mean (Strandberg et al., 2011).In our study, this compares to temperatures that are up to 14.0°C colder in the summer months, up to 20.5°C colder in the winter months and up to 16.5°C colder for the annual mean (Figures 5, 6).
Taking into account the 18 O enrichment of the glacial ocean during the LGM, the reconstructed mean δ 18 O dw and MAT values exhibit elevated values for Kammern-Grubgraben, which result from overall slightly elevated summer temperatures (Figure 5).As the moisture sources are bound to the ocean surface water masses, their isotopic composition may differ from that of the total ocean water and therefore show large regional variations (Luetscher et al., 2015;Lécuyer et al., 2021).Furthermore, a different transport or source area of air masses can lead to differences in isotopic composition (Dansgaard, 1964;Mayr et al., 2016).During the LGM, for instance, southerly moisture transport to the Alps may have originated from the subtropical moisture reservoir.Due to an increased sea ice cover (Figure 2e) in the North Atlantic during winter times (Strandberg et al., 2011;Luetscher et al., 2015), a southward shift of the baroclinic zone and a blocking high over the Fennoscandian ice sheet led to a deviation of the jet stream to 45°N and thus to an increased Rossby wave breaking over the North Atlantic (Kämpf et al., 2022).This phenomenon may have caused the southern moisture transport to the Alps to originate from the subtropics, inducing a meridional flow of humid and warm subtropical air masses towards the Alps, which resulted in increased precipitation, especially in autumn and spring, as indicated by the presence of liquid water in the Sieben Hengste cave system (Figure 2d; Luetscher et al., 2015).Higher summer temperatures reconstructed for the LGM site Kammern-Grubgraben might thus also be attributed to changes in air mass transport patterns.
Another explanation for the elevated values during the summer months at Kammern-Grubgraben may be related to the Fennoscandian ice sheet to the north and the adjacent Alpine glaciers to the south-west that stored much of the moisture, which may have resulted in special climatic conditions of high aridity.As such, due to enrichment by evaporation, stagnant water bodies and rivers in arid environments often exhibit seasonal isotopic patterns associated with maximum δ 18 O values in the summer months (Pederzani and Britton, 2019;Tyler et al., 2007), which could have been the case at Kammern-Grubgraben.We are aware that all the factors described above can have an influence on temperature calculations from δ 18 O PO4 of obligate drinking mammals if δ 18 O dw did not exclusively reflect local precipitation.

Potential use of different water sources
It is evident that the maxima, i.e. the summer times in the teeth of both sites are relatively constant, whereas the minima, corresponding to the winter times, differ in their amplitude (Figure 4).This could either be due to more variable winter temperatures or related to a change of the source water during winter times (Britton et al., 2023).
In general, high δ 18 O PO4 values could be a consequence of water consumption from evaporated stagnant water bodies, whereas low δ 18 O PO4 values could result from snow or glacier meltwater or water consumption from higher elevations due to the effects of water transport and altitude (Gonfiantini, 1986;Darling et al., 2003;Halder et al., 2015).Given reduced precipitation during the LGM in Lower Austria (Ludwig et al., 2019) and the proximity of both study sites to the rivers Kamp, Krems and the Danube, these rivers or local spring water are very likely sources of water.On the basis of the reconstructed winter temperatures for both periods (Figure 5), it is possible that these rivers were frozen, and the animals had to resort to other water sources.Studies of horses from central Yakutia revealed that green cryo-fodder was an important nutritional source (Petrov et al., 2020).In permafrost areas, plants are adapted to the extreme conditions insofar as intensive growth takes place mainly in the first half of summer, when plants are submerged and flooded due to snowmelt or thawing of the permafrost.After a period of regeneration, plants do not have time to go through the full growth cycle, leaving them in a green but frozen state under the snow cover in winter (Petrov et al., 2020).Feeding on these plant types must be considered as a possible source of water alongside occasional feeding on snow or local spring water especially during winter (Scheibe et al., 1998).
In central Europe, evaporation effects play only a minor role in the hydrological cycle nowadays and the river water signal mainly reflects the isotopic composition of precipitation in the catchment area (Rank et al., 2017).Hydrologically, the isotopic composition of the Danube River strongly correlates with that of the precipitation from the drainage area (Rank et al., 2017).Longterm monitoring of the modern Danube showed the lowest δ 18 O river water values in the mountain regions and the highest values in the tributaries at lower elevations (Rank et al., 1998(Rank et al., , 2017;;Fórizs, 2003).The Inn River is the most depleted (−13.1‰), and the Sio River, an outflow from lake Balaton in the Hungarian plain shows the highest δ 18 O values (−6.4‰), which could be explained by the strong lake evaporation effect (Rank et al., 2012).Rivers fed by snow or glacier meltwater are generally depleted in heavy isotopes and may lead to altered isotope values (Dutton et al., 2005;Reckerth et al., 2017).Thus, the Alpine glaciers in the Danube catchment may have caused the meltwater signal in the Danube to persist for a longer period of time.The glacial meltwater could therefore have led to lower δ 18 O values during summer, similar to modern conditions for the Rhine and Danube that show a weakening of the isotopic signals due to mixing with tributary rivers (Koeniger et al., 2022).
Considering the geographic location of the Krems-Wachtberg site, which lies on a spur formed by the incision of the Danube and Krems River into the rocks of the Bohemian Massif (Händel, 2017;Sprafke et al., 2020), this may have applied to the Krems River, which originates from the higher elevations of the Bohemian Massif.This could explain the low δ 18 O PO4 and δ 18 O dw values for horses from Krems-Wachtberg, where periodic thawing of permafrost soils and seasonal meltwater runoff may have been an important source of water for the animals (Rank et al., 2012(Rank et al., , 2017;;Meier et al., 2013;Halder et al., 2015;Jeelani et al., 2017).
As larger stagnant water bodies generally become more isotopically stable over time, their seasonal shifts generally have smaller amplitudes compared with precipitation, river or spring water (Tyler et al., 2007;Pederzani and Britton, 2019), which could be an explanation for the small seasonal variation in sample WA-161267.Drinking at downstream sites would also lead to a damping in seasonal amplitudes due to glacial melting throughout the year and a significant time lag in the isotopic signal (Kendall and McDonnell, 1998;Halder et al., 2015;Jasechko et al., 2014;Rank et al., 2017;Pederzani and Britton, 2019).Except for biologically determined thresholds, climatic factors do not usually affect animal and human behaviour directly, but are mediated by ecological cascades of cause and effect and resulting changes in phenological patterns.Together withand also driven bytemperature, the availability of bioavailable (i.e.liquid) water is a main factor for the timing and productivity of the vegetation period.The greening of the landscape, in turn, is a major incentive for movement in migratory herbivorous animals (Merkle et al., 2016).Phenological gradients in vegetation as well as migratory trajectories and the resulting large-scale distribution patterns of animal biomass create pull factors for human hunter-gatherers that differ strongly between stadial and interstadial conditions (Maier et al., 2023).Isotope data as presented in this study are crucial for a better understanding of these mediated chains of cause and effect and for refining the spatial resolution of the observations.

Comparison with other Pleistocene climate reconstructions based on animal tooth enamel
Several studies report temperature reconstructions from animal enamel for the early MIS 3 and the period shortly after the LGM (Table 2).Accurate and direct comparisons of δ 18 O PO4 , δ 18 O dw and temperatures with other localities in Europe are difficult due to the lack of studies from the period studied here (Kovács et al., 2012) and the different correction factors applied (Table 2).Furthermore, the mean annual averages in different studies are also not consistently calculated as some studies used bulk samples rather than high-resolution sampling strategies, which may also hamper a direct comparison.
During the MIS 3 (51−27 ka), horse δ 18 O PO4 and δ 18 O dw values from Ajoie, Courtedoux, Switzerland range from 11.3 to 14.9‰ and from −10.0 to −14.7‰, respectively, which result in estimated mean annual palaeotemperatures that range between −1.1 and 9.5°C among different individuals (Scherler et al., 2014).However, these values might slightly decrease when accounting for 18 O enrichment of the glacial ocean for both the time before and during the LGM.δ 18 O PO4 values are thus roughly in accordance with the Krems-Wachtberg site, where the sampled horse teeth revealed a δ 18 O PO4 mean value of 11.3 ± 1.1‰.Overall, lower winter and summer values that range from 8.6 to 13.0‰ in our study result in a lower δ¹⁸O dw mean value (−16.4‰,corrected by 0.5‰) and finally in a lower MAT of −5.7 ± 3.1°C.
Additional data on δ¹⁸O dw and temperature come from two horse teeth from the MIS 3 Middle Palaeolithic site Abri du Maras (55−42 ka) in south-western France, where Britton et al. (2023) reported mean δ 18 O PO4 values of 17.8‰ and 17.1‰, respectively, resulting in a mean δ¹⁸O dw value of −7.6 ± 2.1‰ and a MAT of 12 ± 4.0°C (Table 2).This does not conflict with our values, as the French site is located at a lower latitude and the respective find layer dates to 55−42 ka, meaning that the samples are much older than those from Krems-Wachtberg.Mean annual of δ 18 O PO4 values of horse tooth enamel from the Bulgarian Bacho Kiro Cave from MIS 3 (46 to 43.3 ka) are 12.2 ± 0.8‰ and MATs range between 0.0 and −5.0°C (Pederzani et al., 2021a).
Samples from Krems-Wachtberg overall show slightly lower temperatures, indicating that AH 4.4 of Krems-Wachtberg rather falls into stadial than interstadial conditions of the late MIS 3. Regarding interannual climate variability and the different time period compared, the values are not contradictory.
For the LGM, a MAT of about 0.0°C was reported for south-western Germany, which is based on a mean annual δ 18 O PO4 value of 12.3‰ (converted from δ 18 O carbonate ) from the site of Vogelherdhöhle, resulting in a δ¹⁸O dw value of −14.5‰ (Table 6.2, Pushkina et al., 2014).
Based on δ 18 O PO4 of mammoth tooth dentine from Hungary, the oxygen isotopic composition of mean annual late Pleistocene precipitation in east-central Europe exhibited a δ 18 O dw value of −13.4‰ and a mean air temperature of 2.4°C (Kovács et al., 2012).These values are slightly higher than those reconstructed for Kammern-Grubgraben.However, Kovács et al. (2012) used a correction factor for ocean water of 0.8‰ based on Schrag et al. (2002) (Table 2).When applying the same correction of 0.8‰ to the δ 18 O dw values of Kammern-Grubgraben (Table S1), the MAT value is still about 5°C lower than Kovács et al. (2012) report for Hungary.However, further uncertainties may also arise from carbonate converted to phosphate values due to diagenetic alterations of carbonate isotopes (Zazzo et al., 2004b;Demény et al., 2019).Besides, air temperature estimates for Krems-Wachtberg (MAT: −5.7 ± 3.1°C) and Kammern-Grubgraben (MAT: −3.5 ± 2.9°C) exhibit relatively large uncertainties, leading to overlapping values.
δ 18 O PO4 values of 14.7 to 15.3‰ (converted from δ 18 O car- bonate ) from Laugerie-Haute E. in southwestern France (22.9 to 21.2k cal a BP) resulted in δ¹⁸O dw values between −11.0 and −10.3‰ leading to a temperature range from 6.0 to 7.0°C, on average when a seawater correction of 1.2‰ was applied (Table 2; Lécuyer et al., 2021).Converted δ 18 O PO4 from the site of Hornos de la Peña in northern Spain (23.4-21k cal a BP) yielded values of 15.4-15.8‰resulting in temperature estimates between 7.1 and 8.4°C corrected by 18 O enrichment of the ocean water (Table 2; Lécuyer et al., 2021).In comparison, Kammern-Grubgraben shows lower values (Figures 5, 6) with a mean δ¹⁸O dw value of −15.3‰, resulting in lower MATs, explained by the continental effect, leading to a general 18 O depletion of the isotopic precipitation signal with increasing distance from the oceanic moisture source area (Dansgaard, 1964;Gat, 1996;Winnick et al., 2014).The low reconstructed δ 18 O dw values, reflecting the progressive isotopic depletion of air masses, have also been reported for northern Europe using mammoth tooth enamel from Estonia, Latvia, Lithuania, Denmark and Poland for the period 42-24 ka (Arppe and Karhu, 2010).During this period, westerly moisture sources predominated in the southern Baltic area, similar to present-day circulation patterns.However, at least for the LGM and the central European region, this is still a matter of debate as a southerly moisture supply is considered to have been prevalent at least during summer seasons (Luetscher et al., 2015;Kämpf et al., 2022).However, regional climate models, applying circulation weather types also showed predominantly southerly and easterly flow patterns,  2), which are consistent with the values for Kammern-Grubgraben (δ 18 O dw : −15.3‰;MAT: −3.5 ± 2.9°C) in our study.Considering individual data points, the coldest temperature estimates for (late) MIS 3 are reported for the Baltic regions of Latvia and Lithuania.These temperatures are up to 13°C below today's regional values (Arppe and Karhu, 2010), which is largely consistent with the values from Krems-Wachtberg, where temperatures were up to 16°C colder than today.
For the Late Glacial, published δ¹⁸O enamel data and reconstructed δ¹⁸O dw and temperature estimates from different European regions clearly imply a climatic amelioration in eastcentral Europe (17.5-11.6kcal a BP; Ábelová, 2008;Kovács et al., 2012) and the British Isles (15.2-14.6kcal a BP; Reade et al., 2020;Table 2).In both regions this overlaps with the Bølling-Allerød interstadial (14.7-12.85kcal a BP), the first rapid warming episode of the last deglaciation in the northern hemisphere (Naughton et al., 2023).
A large part of the inconsistencies between the studies compiled in this chapter can be explained by different methods that were used.For instance, some studies used a bulk sample of the full crown height (Richards et al., 2017), others sampled a defined height of the crown (Britton et al., 2019) or use the average of the seasonal maxima and minima (Pederzani et al., 2021a;this study).For better comparability in future studies, a consistent approach for the calculation of average δ¹⁸O dw , and thus MAT, is recommended.Also, an ocean source water correction was not applied at all or different correction values were used in some studies.For glacial water and temperature reconstructions, we recommend a δ 18 O correction for oceanic source water, as previously suggested (e.g.Lécuyer et al., 2021).The use of the reconstructed ocean water values from Lea et al. (2002) is advantageous as they provide a continuous record of ocean δ 18 O for the entire glacial period.

Conclusions
To our knowledge, we provide the first high-resolution phosphate oxygen isotope records of equid enamel for the Early Gravettian and Early Epigravettian in Lower Austria.We used these records to reconstruct δ 18 O dw by using a speciesspecific conversion equation.Subsequently, palaeotemperatures were estimated using transfer functions based on 84 modern European sites, for which temperature and δ 18 O of precipitation are available.We recommend correcting the reconstructed δ 18 O of the water to account for the differences between the modern and glacial isotopic composition of seawater (Lea et al., 2002) caused by the larger continental ice volumes during the last glacial.
Based on the seasonal variations, it can be assumed that at both sites, the most likely source of drinking water came from precipitation or nearby rivers.The MATs of Krems-Wachtberg and Kammern-Grubgraben were estimated to be 16.3 ± 3.1°C and 14.1 ± 2.9°C colder than today, respectively.Our observed LGM values are lower than those from western European LGM sites but can be considered consistent due to the continental effect leading to decreasing isotope ratios.The new findings suggest that occupation took place during stadial conditions at both sites, which is unexpected given previous food web comparisons (Reiss et al., 2023).
While there are temporal occupations reported for all seasons for the region and time period around the site of Krems-Wachtberg (Händel, 2017), the occupation phase for AH 4.4 at Krems-Wachtberg (Händel et al., 2021b) is originally determined as a winter settlement (Fladerer et al., 2014).But since this seasonal determination is based on a fetal/ newborn reindeer bone, the season is more likely a spring occupation, as a bone in the size of a ready-to-birth reindeer indicates a dating around the birthing season (May/June).However, the main layer (AH 2) at Kammern-Grubgraben is reported as representing winter occupation, mainly based on representation of reindeer, horse and ibex deciduous teeth (Pasda, unpublished work), which is noteworthy considering the estimated temperatures and associated climatic conditions.However, further research is needed to determine the exact season and age range of occupations and the adaptation mechanisms of Palaeolithic hunter-gatherer communities at Kammern-Grubgraben and the Upper Palaeolithic site cluster at Krems.winter temperature reconstructions the respective maximum and minimum values of the modelled data were used.
Table S2.Temperatures and δ 18 O precipitation values of 84 European sites (GNIP, IAEA/WMO, 2023).Given are long-term averages of mean annual temperatures (MAT) and the respective summer (July/August) and winter (December/January) temperatures.
Table S3.Stable isotope results.Given are the measured δ 18 O (δ 18 O unmodelled ) values with the respective standard deviation ( ± SD), and the δ 18 O modelled values (δ 18 O modelled ) with the upper (upper.CI) and lower boundary (lower.CI) of the confidence interval set at 95%.Total length refers to the model total length.

Figure 1 .
Figure 1.(a) Various European sites from MIS 3 and 2 for which stable oxygen isotope data from animal enamel and derived palaeotemperatures are available (further information in Table 2).Only sites for which geographical coordinates are reported are shown.The red box indicates the study area.(b) Location of the study sites Krems-Wachtberg and Kammern-Grubgraben.Basemap of Europe in 1 km resolution from the European Environment Agency (http://eea.europa.eu/data-and-maps/figures/elevation-map-of-europe).[Color figure can be viewed at wileyonlinelibrary.com]

Figure 3 .
Figure 3. Upper M3 (WA-163790) from the Krems-Wachtberg site.Given are the positions of the 22 samples taken from the tooth enamel.The enamel-dentine junction (EDJ) is indicated by the dashed line.[Color figure can be viewed at wileyonlinelibrary.com]

Figure 4 .
Figure 4. Measured oxygen stable isotopes of phosphate from horse teeth enamel from (a-c, open circles) Krems-Wachtberg and (d-f, open diamonds) Kammern-Grubgraben.Given are the values measured in triplicate and the respective standard deviations as whiskers.Mean inverse modelled isotopic variations are shown as grey lines with a 95% confidence interval (grey shaded areas).ε² is the respective damping factor used for inverse modelling.

Table 2 .
Arppe and Karhu (2010) different regions in Europe from MIS 3 and 2 ordered by age.Numerical codes correspond to the applied correction (1) for no ocean water correction, (2) for an ocean water correction based onSchrag et al. (2002)and (3) for an ocean water correction based onLea et al. (2002).A question mark is added if details were not published.forcentralandeasternEurope.During the LGM, precipitation in central and eastern Europe was weaker than in western Europe, which is explained by strong evaporation over the North Atlantic and changes in the frequency of circulation weather types(Ludwig et al., 2016).Arppe and Karhu (2010)reported average δ 18 O dw values from northern Europe ranging from −15.7 to −8.4‰, resulting in mean annual air temperatures of −5.6-6.3°C(Table especially Abbreviations and acronyms.δ 18 O dw , oxygen isotope composition of drinking water; δ 18 O PO4 , oxygen isotope composition of phosphate; AH, archaeological horizon; EDJ, enamel dentine junction; GI, Greenland interstadials; GS, Greenland stadials; HF, hydrofluoric acid; KG, Kammern-Grubgraben; LGM, Last Glacial Maximum; M, Molar; MAT, mean annual temperature; MIS, Marine Isotope Stage; M3, third molar; NGRIP, North Greenland Ice Core Project; P2, second premolar; TC-EA, High temperature conversion elemental analyser; TueA, Tübingen Apatite; T summer , mean summer temperature; T winter , mean winter temperature; VSMOW, Vienna Standard Mean Ocean Water; WA, Krems-Wachtberg 2005−2015.