Changing food webs before and during the Last Glacial Maximum based on stable isotopes of animal bone collagen from Lower Austria

We investigated palaeofood web structures using stable isotope analyses on animal bone collagen from four Upper Palaeolithic sites dated to the Early Gravettian (Krems‐Hundssteig and Krems‐Wachtberg: 33–31k cal a bp, Langenlois: 31–29k cal a bp) and to the Early Epigravettian (Kammern‐Grubgraben: 24–20k cal a bp). In both periods, δ13C values show niche partitioning between hare, horse and mammoth on one side, and reindeer and ibex on the other, indicating different diets and habitats between both herbivore groups. The δ15N differences between carnivores and herbivores suggest a difference of one trophic level during the pre‐Last Glacial Maximum (pre‐LGM) period at the Early Gravettian sites and a tendency towards secondary carnivores during the LGM at Kammern‐Grubgraben. δ15N values of pre‐LGM mammoths are elevated in relation to other herbivores but shifted to the level of other herbivores in the LGM. A general δ15N value shift in herbivores of 3.3‰ from the pre‐LGM to the LGM is related to climatic deterioration. This may have led to the disappearance of certain ecological niches and to a shift from broader to overlapping ecological herbivore niches shortly before the LGM, as demonstrated by SIBER analyses.


Introduction
Animal bones represent a frequently well-preserved and widely abundant archive at Palaeolithic sites.Their isotopic composition is commonly used to reconstruct local palaeofood webs and past ecosystems (Bocherens et al., 2015;Brock et al., 2010;Drucker et al., 2012;DeNiro and Epstein, 1978;Hobson, 1999;Hoke et al., 2019).The reconstruction of food webs is based on the fact that isotopic fractionation occurs between animals from different trophic levels.Moreover, isotopic differences at the baseline of food webs occur across geographical regions and in different climates.Potential shifts in the isotopic composition due to trophic level or geographic origin are well reflected in animal tissues, such as bone collagen (DeNiro and Epstein, 1978;Hobson, 1999).
For bone collagen, carbon stable isotopes show a trophic enrichment of approximately 0.8-1.3‰ in carnivores in comparison to their prey (Bocherens and Drucker, 2003;Krajcarz et al., 2016), whereas nitrogen stable isotopes increase by about 3.0-5.0‰per trophic level (DeNiro and Epstein, 1981;Schoeninger and DeNiro, 1984;Bocherens and Drucker, 2003;Fox-Dobbs et al., 2007).Unlike tooth dentin with its incremental growth, bone is a renewing tissue, so its isotopic composition exhibits an averaged value over several years of nutritional behaviour (Drucker 2022).The collagen in the femoral bones of human adults, for instance, reflects a period of more than a decade (Hedges et al., 2007).
A number of stable isotope studies on bone collagen have been conducted for the frequently prevailing steppe-and tundra-like ecosystems of the last glacial period in central Europe, in particular for Marine Isotope Stage (MIS) 3 (59-29k cal a BP) and MIS 2 (29-11.7kcal a BP) (Voelker et al., 2002).Due to the predominance of large herbivorous mammals, this ecosystem is also known as 'mammoth steppe' (e.g.Zimov et al., 1995Zimov et al., , 2012;;Bocherens, 2003;Bocherens et al., 2005Bocherens et al., , 2014Bocherens et al., , 2015;;Drucker et al., 2003;Drucker and Henry-Gambier, 2005;Fox-Dobbs et al., 2008;Yeakel et al., 2013, Schwartz-Narbonne et al., 2019).This biome is characterised by a large biomass of plants and herbivores despite the partly very cold and dry conditions during the last glacial.It has been proposed that high insolation and low humidity led to increased summer snowmelt and a longer growing season (Guthrie, 1982;Zimov et al., 2012;Drucker, 2022).A rapid turnover of nutrients in loess sediments may also have favoured high vegetation productivity and, thus, provided a base for grazing and browsing by animal communities, which in turn fertilised the soils through manure and carcasses (Zimov et al., 1995;Drucker, 2022).Depending on the geographic region and the time period considered, the Eurasian mammoth steppe is characterised by a changing landscape of continuous and discontinuous permafrost (Delisle et al., 2007;Blaser et al., 2010;Vandenberghe et al., 2014).According to Řičánková et al. (2015), the Altai-Sayan, Kazakhstan, and the Eastern European Plain are the areas that best serve as modern equivalents.
The isotopic composition of a number of faunal assemblages from the late MIS 3 and MIS 2 suggests a particular niche partitioning.This may have been caused by the great diversity of coexisting herbivores and interspecies competition (Iacumin et al., 2000(Iacumin et al., , 2006;;Drucker et al., 2003;Bocherens et al., 2015;Drucker, 2022).Based on δ 15 N and δ 13 C value ranges, the ecological niches of mammoth, horse and reindeer were most frequently investigated in different regions.Most studies on faunal composition and food web structures use material from  and Magdalenian (20-14 ka) sites (Iacumin et al., 2000(Iacumin et al., , 2006(Iacumin et al., , 2010;;Drucker et al., 2003Drucker et al., , 2018;;Stevens and Hedges, 2004;Stevens et al., 2008;Bocherens et al., 2015;Krajcarz et al., 2016;Fox-Dobbs et al., 2008) and thus focus on the periods before and after the Last Glacial Maximum (LGM).Here we use Mix et al., (2001) definition of the LGM sensu stricto, which sets it at 23-19k cal a BP.Studies of the food web structure for the LGM are largely absent.From an archaeological perspective, the transition from the Gravettian to the Epigravettian is of great interest, as the human population in central Europe decreased massively and human occupation remained at very low numbers and densities in the research area during the LGM (Maier et al., 2021).
The aim of this study is to infer possible food web changes related to climatic and environmental trends around the LGM using the carbon and nitrogen stable isotopes of animal bone collagen.For this purpose, we use the diverse faunal assemblages from several well-known Upper Palaeolithic sites in and near the city of Krems (Lower Austria), which date to 33−29k cal a BP (Early Gravettian/pre-LGM) and 24−20k cal a BP (Early Epigravettian/LGM).The geographic proximity of the sites allows meaningful conclusions to be drawn from a comparison between the different age periods.
Study sites Krems-Hundssteig 2000-2002and Krems-Wachtberg 1930 (33−31k cal a BP) The Upper Palaeolithic sites of Krems-Hundssteig and Krems-Wachtberg are located very close to one another, both in the present-day urban area of the city of Krems (Fig. 1).Topographically, the sites are situated at the transition from the narrow Danube passage of the Wachau to the so-called Tullnerfeld, an extensive alluvial plain.They are located in the lower southern to south-eastern part (Wachtberg) of a spur-shaped promontory that rises up to 398 m a.s.l. on the northern bank of the Danube.The Wachtberg area, where loess deposits are up to 20 m thick, is slightly sloping to the southeast, thus providing wind-sheltered conditions for temporarily used campsites of hunter-gatherer communities (Händel, 2017).
The Upper Palaeolithic site of Krems-Hundssteig first received attention in the late 19th century.At that time, numerous lithic artefacts and faunal remains were found in the course of large-scale quarrying of loess sediments.The lithic industry points to the Aurignacian and Gravettian.New excavations were carried out adjacent to the former quarry area between 2000 and 2002 and identified five basic archaeological horizons (AHs).The main Upper Palaeolithic horizon, AH 3, is attributed to the Gravettian and represents a sequence of layers that can be subdivided in an upper (AH 3.1 and AH 3.2), middle (AH 3.3−AH 3.4) and lower sequence (AH 3.5−AH 3.8) (Neugebauer-Maresch et al., 2008).All bone samples analysed herein originate from the Krems-Hundssteig 2000-2002 excavations and stem from AH 3.2 and 3.4, which date between 32.9 and 31.0kcal a BP (Fig. 2a).Radiocarbon dating from two lower horizons, accessed only by sounding and core sampling, exhibited age ranges of 38.9−34.5kcal a BP for AH 4 and 46.6−42.4kcal a BP for AH 5 (Neugebauer-Maresch et al., 2008;Händel 2017).From a zooarchaeological point of view, Krems-Hundssteig 2000-2002 is mainly characterised by a low density of carnivores in contrast to the neighbouring Krems-Wachtberg sites (Fladerer and Salcher-Jedrasiak, 2008).The faunal assemblage of Krems-Hundssteig considered here (AH 3.21,3.22,3.24 and 3.44) predominantly consists of mammoth, reindeer and horse with minor contributions of red deer, woolly rhinoceros, arctic fox, wolf and hare, as well as micromammals.The minimum number of individuals (MNI) for the relevant horizons was 57 (Fladerer and Salcher-Jedrasiak, 2008;Händel, 2017).

Langenlois A (31−29k cal a BP)
The Upper Palaeolithic site Langenlois is located in the area of the former brickyard Kargl in the southeast of the town of Langenlois (Fig. 1).The find locations Langenlois A−C are located on a loess-covered slope exposed to the east and facing towards the river Kamp.Placed in a valley basin, the site is protected to the north by smaller hills and to the southwest by the prominent Gobelsberg elevation.During the first excavations of the site Langenlois A in the early 1960s, one archaeological horizon was identified, which is characterised by different occupational structures.Three adjacent fireplaces, an unevenly developed cultural layer, and various other features, such as workplaces for the fragmentation of bones, suggest a variety of activities.All specimens analysed here originate from the excavation site Langenlois A and date to 30.8−29.2kcal a BP (Fig. 2c;Einwögerer, 2019).The faunal assemblage is dominated by ibex, followed by horse and reindeer plus a few remains of red deer and mammoth (Einwögerer, 2000).
Kammern-Grubgraben is one of the few stratified LGM sites in central Europe and is located in the south-eastern extension of the Bohemian-Moravian Highlands (Fig. 1; Händel et al., 2021a).It is situated on a gentle loess-covered slope between two hills, the Heiligenstein in the east and the Geißberg in the west.The site is open towards the south, i.e. to the Kamp River valley and the Danube plain and confined towards the northwest by the adjacent Moravian plateau.Given the harsh environmental conditions during the LGM (Maier et al., 2021;Händel et al., 2021a), this geographic setting may have been advantageous for hunter-gatherer communities, given the wind protection and proximity to water.Sedimentological studies at Kammern-Grubgraben indicate a rather monotonous depositional environment during the LGM, dominated by aeolian deposition of silt-sized loess (Reiss et al., 2022).
Beside lithic and organic artefacts, the Kammern-Grubgraben inventory also includes a large number of faunal remains.The faunal assemblage of Kammern-Grubgraben is dominated by reindeer, followed by horse.There are minor contributions of hare, mammoth, bison, red deer, ibex, wolf, brown bear, wolverine, arctic fox, red fox, goose and ground squirrel (MNI = 234; Pfeifer et al. accepted).The zooarchaeological material stems from different archaeological layers, which were addressed differently in various excavation campaigns during the long period of research.The samples analysed here originate from both the Brandtner excavation and the excavation led by the team of Montet-White.Therefore, the labelling of the find horizons varies depending on the excavation year.The Montet-White and Brandtner excavations documented five consecutive archaeological layers (AL 1-5) and assigned these to at least five different occupation phases (Haesaerts et al., 2016;Neugebauer-Maresch et al., 2016).However, more recent investigations indicate that former ALs 2-4 probably represent a single main occupation phase, here referred to as AH 2. The earlier AH 3 (probably corresponding to former AL 5) seems to represent a much less intensive occupation, while AH 1 (former AL 1) at least partly contains relocated finds and suggests contributions from different occupational episodes including a younger, potentially Magdalenian, occupation.Radiocarbon ages of faunal remains (bones and teeth) from AH 1 and 2 (resp.AL 1-4) indicate human activity between 23.7 and 19.5k cal a BP (Fig. 2d,e; Haesaerts et al., 2016;Händel et al., 2021a).

Bone material
The find inventories include large numbers of faunal remains, both bones and teeth, of terrestrial animals, including various carnivores (red fox, arctic fox, wolf and wolverine) and numerous herbivores (woolly mammoth, reindeer, red deer, horse, bison, musk ox, woolly rhinoceros, hare, ibex and goose).Bone material was sorted according to stratigraphic aspects and taxonomically identified by K. Pasda for Kammern-Grubgraben and Langenlois.The faunal remains from Krems-Hundssteig 2000-2002 had already been taxonomically identified and documented by Fladerer and Salcher-Jedrasiak (2008) and the Krems-Wachtberg 1930 inventory by Fladerer (2001).To avoid samples with potentially poor collagen preservation, bones with a compact texture were selected preferentially for collagen extraction and subsequent stable isotope analyses.
In most cases, the same species were selected from the inventories to allow for accurate comparisons between species and sites.Whenever possible, different individuals of a species were selected from one site (Table 1).In order to increase the number of species and to obtain as many species as possible for both periods, the food webs of the Early Gravettian sites Krems-Wachtberg 1930 (KW) and Krems-Hundssteig 2000-2002 (HU) have been combined and are discussed as one coherent food web (KW + HU) in comparison to the Early Epigravettian food web of Kammern-Grubgraben (KG).This decision is justified for KW and HU based on spatial proximity (less than 250 m) and identical 14 C age range (Fig. 2a,b).In addition, we included the site Langenlois A (LK, 9 km northeast of KW + HU; Fig. 1) in this study because it contains a considerable number of ibexes, which are rare in KW + HU but also common in KG.LK appears to be around 2k years younger than KW + HU, although its age range overlaps with these sites (Fig. 2c).
In the evaluation and discussion, only bone collagen data are considered, as the values of dental collagen can be considerably enriched in 15 N compared with bone collagen, which has been confirmed for dental collagen from reindeer (Fizet et al., 1995;Britton, 2010).In addition to bone collagen analyses from KG-AH 2, four samples from KG-AH 1 were analysed, but these results are not included in the discussion, as it cannot be excluded that they originate from a Magdalenian, i.e. post-LGM occupation.Both dental collagen values and samples originating from KG-AH 1 were not further considered but are listed in Supplementary Table S1 for completeness.Fourier-transform infrared (FTIR) spectroscopy FTIR spectroscopy was used as a pre-screening method on a defined set of samples (KG: n = 7, HU: n = 4, KW: n = 4, LK: n = 3) of bone powder to detect the preservation of collagen (Cersoy et al., 2016;Lebon et al., 2016).It was further applied on the respective extracts to check for the chemical purity of bone collagen.Less than 2 mg of each powdered sample was used for FTIR spectroscopy in attenuated total reflectance mode (IR Prestige-21, Shimadzu).The device parameters were set according to the method of Lebon et al. (2016).Infrared spectra were obtained from 64 scans per run with a spectral resolution of 2 cm -1 in the wavenumber range of 4000-370 cm -1 .To account for sample heterogeneity and to determine instrumental reproducibility, each sample was measured in triplicate.Prior to every sample measurement, a new baseline (air, CO 2 ) was set.Wavenumbers in the range of 1710-1590 cm -1 and 1110-940 cm -1 represent the amide I band and the v3 phosphate ( v3 PO 4 )-band, respectively (Lambert et al., 1998;Lebon et al., 2016).

Collagen extraction
Collagen extraction followed a slightly modified version of the method of Bocherens et al. (1991).About 0.5-2.0g was carefully sawn from each animal bone using a rotary tool (Proxxon-Micromot, Germany) with a circular diamond blade.Wherever necessary, samples were cleaned with sandpaper and a common toothbrush before soaking them in deionised water for 30 min.Samples that were hardened or glued for preservation were additionally soaked in acetone to remove varnish from the bones.Subsequently, samples were ultrasonically cleaned in deionised water for 10 s and dried in a drying oven at 40°C overnight.Samples were then ground using a mortar and pestle.0.25-1.0g of each powdered sample (0.3-0.7 mm grain size) was weighed in centrifuge tubes and soaked in 1 M HCl to dissolve minerals, while samples were shaken on a rotator for 20 min.Subsequently, the samples were repeatedly washed with distilled water and centrifuged (Heraeus Multifuge 3L-R, Thermo Electron Corporation, USA) at 3000 rpm for 5 min, until a pH of 6 was reached.For dissolution of humic acids, the pellets were soused with 0.125 M NaOH and left under the fume hood for 20 h.Thereafter, all samples were washed again with distilled water and centrifuged until pH was neutral.In the next step, the pellets were soaked in 0.01 M HCl (pH 2) and incubated for 10-17 h in a water bath (Julabo, SW23, Germany) at 95°C to solubilise the gelatine.
Finally, the dissolved gelatine was filtrated through MF-Millipore membrane filters (5.0 μm) using vacuum flasks and filter funnels, frozen and lyophilised.

Stable isotope analysis
Samples were analysed in duplicate or triplicate at the stable isotope laboratories at Forschungszentrum Jülich (FZJ) and Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), respectively.Some 240-260 μg of collagen was wrapped in tin capsules and combusted in an elemental analyser (Flash2000, Thermo Fisher, USA, FZJ; NC 2500, Carlo Erba, Italy, FAU) and measured online with a coupled isotope ratio mass spectrometer (DeltaV plus, Thermo Fisher, USA, FZJ; DeltaPlus, Thermo-Finnigan, Germany, FAU).Carbon and nitrogen measurements were performed within one analysis run on the same sample.
Isotope results are reported as delta (δ) values per mille (‰) according to the equation: (1)  R S is the isotope ratio ( 13 C/ 12 C, 15 N/ 14 N) of the sample and R St of the respective standard.δ values are normalised to the VPDB (Vienna Pee Dee Belemnite) scale for carbon and the AIR (atmospheric nitrogen) scale for nitrogen (Coplen 2011).
Calibration of laboratory standards and scale-normalisation of δ 15 N raw values is based upon the international reference standards IAEA-N-2 (δ 15 N = 20.3‰),IAEA-N-1 (δ 15 N = 0.4‰) and USGS25 (δ 15 N = −30.4‰).The average standard deviation of replicate measurements of δ 15 N standards measured in the respective runs was 0.28‰ for IAEA-N-1 (n = 8), 0.15‰ for IAEA-N-2 (n = 8) and 0.14‰ for lab standard peptone (n = 8).Samples were also analysed for their weight percentage carbon (%) and nitrogen (%) using elemental standards for calibration.The measurement error of the element concentrations was <5%.Elemental content and atomic C/N ratios were used to check for chemical purity.

Quality criteria for collagen preservation
Impurities or diagenesis may lead to a shift in the isotope values of prehistoric bone collagen.The C/N ratio is known to be a quality criterion for collagen and was used to validate collagen purity (Guiry and Szpak, 2021).Reported C/N ratios for pure collagen vary between 2.9 and 4.0 (DeNiro, 1985;Grupe et al., 2003;Coltrain et al., 2004;Bösl et al., 2006;Hoke et al., 2019), but samples with C/N >3.6 should be rejected as the collagen purity for those cannot be guaranteed (Ambrose, 1990;van Klinken, 1999).

Statistics
Statistical calculations were carried out with R studio (version R-4.2.2).Shapiro-Wilk tests were used to check for normal distribution of data.Possible outliers were determined using Grubb's tests.Wilcoxon-Mann-Whitney tests were carried out to test the null hypothesis of whether two groups are statistically indifferent (Dytham, 2011).For all analyses performed, a significance level of 0.05 was chosen.
Comparisons of isotopic niches among different animal communities were analysed using the Stable Isotope Bayesian Ellipses in R (SIBER) package (Jackson et al., 2011).It must be emphasised here that isotopic niches do not necessarily correspond to ecological niches, since, for example, different trophic niches can lead to the same isotopic ranges (Jackson et al., 2011).In comparison to simple convex-hull methods, SIBER is based on multivariate ellipse metrics, with the advantage of being independent of sample size while enabling robust comparison between data sets of different sizes (Jackson et al., 2011).
We calculated convex hulls, i.e. total area and standard ellipse area (SEA, in ‰²) of herbivorous taxa represented by more than three data points.For the KW + HU and KG food webs, these include horse, hare, reindeer and woolly mammoth, for LK and KG, ibex.The SEA explains 40% of all potential specimens that fit into the respective niches, based on a Bayesian probability estimation.The SEA is sensitive to sample size, which is why we used the recommended standard ellipse area corrected for sample size (SEAc) (Jackson et al., 2011).

Preservation and purity of collagen
FTIR spectra revealed a sufficient amount of collagen in all powdered bone samples analysed (Fig. 3a).The FTIR spectra of the powdered untreated samples show a distinct peak in the amide I band range (1710-1590 cm -1 ) indicative for collagen, while the v3 PO 4 band range (1110-940 cm -1 ), indicative for bone phosphate (PO 4 ), is obviously more pronounced.
Collagen quality was confirmed by FTIR spectra of the extracted collagen of selected samples (Fig. 3b).After collagen extraction, no v3 PO 4 band could be detected anymore, indicating complete dissolution of v3 PO 4 during the collagen extraction procedure.Instead, the amide band is most pronounced, indicating that pure collagen was extracted.
Collagen yields (wt.%), varied between 0.8 and 16% and were on average 4.3% at KG and 7.0% at KW. Two samples (woolly mammoth 'KG 1110' and ibex 'KG 249') yielded collagen content <0.5%.However, other chemical characteristics, such as %N, %C and C/N ratios of these samples, are still within the acceptable range for well-preserved collagen, except for KW-MK-938 (red fox), KG-2652 (bison) and KG-1500 (hare), which all had nitrogen and carbon content that were too low (Table S1).All other samples are thus considered reliable in terms of collagen preservation and are included in the further evaluation.The high purity of all the extracted collagen samples was generally confirmed by their C/N ratios which range between 2.9 and 3.6 as expected for pure collagen (Table S1).

Isotopic results of bone collagen
All stable isotope results are shown in δ 15 N vs. δ 13 C plots separated for the Early Gravettian/pre-LGM sites KW + HU (Fig. 4a), the Early Gravettian/pre-LGM site LK (Fig. 4b) and the Early Epigravettian/LGM site KG (Fig. 4c).Shapiro-Wilk tests confirm a normal distribution of the complete dataset (δ 15 N: W = 0.97455, p < 0.05 and δ 13 C: W = 0.9042, p < 0.05).We excluded one outlier (hare 'WA-MK 923', confirmed by Grubb's test) from further evaluations and listed it only in Supplementary Table S1, as the exceptionally low δ 13 C values most likely indicate modern contamination.
In the pre-LGM food web of KW + HU, all common herbivorous species except mammoth exhibit mean δ 15 N values between 2.1 ± 1.2 (hare) and 5.3 ± 2.3‰ (woolly rhinoceros).Musk ox also exhibits a very high δ 15 N value of 7.7‰.However, they are represented by only one specimen and therefore may not be representative.As mentioned, mammoths exhibit considerably higher δ 15 N values than other herbivores, with a mean value of 8.2 ± 0.8‰ and are thus almost on the same level as the carnivorous species wolf and wolverine that exhibit δ 15 N mean values of 8.8 ± 0.7 and 8.8 ± 1.1‰, respectively (Fig. 4a).Arctic fox and red fox show slightly lower δ 15 N mean values of 7.5 ± 1.2 and 7.2 ± 0.4‰, respectively.
At the LGM site KG, δ 15 N mean values of herbivore taxa range from 1.8 ± 0.7 in hare to 3.9‰ in goose, which is again represented by only one specimen (Fig. 4c).The highest single values of herbivores belong to infantile individuals of ibex and horse (3.6‰ and 3.5‰, respectively).In contrast to the pre-LGM sites, the δ 15 N values of mammoth (2.5 ± 0.9‰) at KG are in the range of other herbivores.Carnivores show δ 15 N mean values of 7.5 ± 2.4‰ and are thus 5.2‰ higher than herbivores.However, the arctic fox's δ 15 N values are highest but exhibit a large scatter (8.6 ± 3.2‰), while wolf (7.3 ± 1.9‰), red fox (6.7‰) and wolverine (5.9‰) show lower values.
A comparison between pre-LGM (KW + HU, LK) and LGM (KG) isotope-derived food webs shows similarities but also some differences.The main similarity between both periods is a separation between two clusters of herbivores: in KW + HU and KG, horse and hare have generally lower δ 13 C values than the cluster formed by ibex and reindeer (Fig. 4a,c).In LK the data on hare, reindeer and horse are too small to draw such a conclusion.Red deer exhibits a varying position in δ 15 N vs. δ 13 C space: in the pre-LGM period at KW + HU it falls between the fields of rhinoceros and horse on the one side, and hare on the other (Fig. 4a), while in the LGM it plots together with reindeer and ibex (Fig. 4c).The most striking difference between pre-LGM (KW + HU) and LGM (KG) is, however, the shift of the mammoth cluster mainly related to the strong δ 15 N decrease between both periods as already mentioned above.During the pre-LGM period, the cluster of mammoths does not match any of the herbivores except a single musk ox and rhinoceros bone, while the data cluster overlaps with those of horse and hare in the LGM.For better evaluation, we statistically tested the isotopic differences between pre-LGM and LGM sites for the most common herbivorous taxa, i.e. mammoth, horse, hare, reindeer and ibex using the Wilcoxon-Mann-Whitney test (Fig. 5a,b).All tested pairs belonged to KW + HU versus KG, except for ibex, for which LK versus KG was used.With the exception of hare, all taxa had significantly higher δ 15 N values in the pre-LGM than in the LGM (Fig. 5a).
In terms of absolute values, a marked δ 15 N shift occurs between both periods except for hare.In the pre-LGM period, δ 15 N mean values of all herbivores are on average 3.3‰ higher than in the LGM food web.If mammoth is excluded, the shift is still 2.0‰.
Detailed statistical analyses demonstrate that only the δ 13 C of mammoth is significantly different between both periods and exhibits the lowest δ 13 C values in the pre-LGM period of KW + HU (Fig. 5b).In KW + HU, δ 13 C mean values of herbivores range from −21.2 ± 0.2‰ in mammoth to −19.4 ± 0.7‰ in reindeer, showing a slightly broader range compared with the LGM (−20.8 in goose to −19.6 ± 0.4‰ in reindeer).The largest δ 13 C shift between both periods occurs in mammoths: from −21.2 ± 0.2‰ in the pre-LGM period to −20.3 ± 0.4‰ in the LGM.Wolf and wolverine in KW + HU exhibit δ 13 C mean values of −19.7 ± 0.4 and −19.7 ± 0.1‰, while arctic fox and red fox show slightly lower δ 13 C mean values of −20.2 ± 0.4 and −20.5 ± 0.2‰, respectively.In the LGM, carnivores exhibit δ 13 C mean values ranging from −20.0 ± 0.1‰ in arctic fox to −18.7‰ in wolverine (Fig. 4a).
In comparison to the LGM, the results of SIBER analyses for the pre-LGM confirm larger isotopic niche widths for all analysed herbivores but mammoth.Mammoth has a SEAc of 0.6‰² in the pre-LGM (KW + HU) compared with 0.9‰² in the LGM.The SEAc of the other pre-LGM herbivores range from 0.9 (ibex, LK) to 3.1‰² (horse, KW + HU), while in the LGM the same species have SEAc of 0.5 and 0.7‰², respectively (Fig. 6).When considering the isotopic niche areas of both periods, the isotopic niche of ibex is the least variable.Horse shows the largest isotopic niche in the pre-LGM and the biggest change in isotopic niche width between pre-LGM and LGM (Fig. 6).

Discussion
Herbivore niche partitioning indicated by δ 13 C Niche partitioning can generally reflect a spatial separation in terms of different habitats, distinct dietary habits, temporal variation, or even temporal and spatial avoidance (Kronfeld-Schor et al., 2001;Stewart et al., 2003;Britton et al., 2012;Bocherens et al., 2015).Herbivore data in the pre-LGM period show a larger variability and thus partly broader isotopic niches in the SIBER analysis than during the LGM (Fig. 6a).This is best explained by the climatic deterioration that significantly affected the environment and consequently the stable isotope pattern of herbivores (Zimov et al., 2012;Drucker et al., 2018;Drucker, 2022).
In comparison to the δ 15 N values, the δ 13 C clustering of herbivores (with the exception of mammoth) is relatively stable over time (Fig. 5b) and can be directly linked to their dietary habits, which are determined by the regional plant communities, except for migratory mammals.The δ 13 C results point to two main groups of herbivores: horse and hare show lower, and reindeer and ibex higher δ 13 C values (Fig. 5b).Mammoth belongs to the first group during the LGM but has lower δ 13 C values than all other taxa in the pre-LGM (Fig. 5b).In (modern) arctic environments, various plant groups show different isotopic compositions (Nadelhoffer et al., 1996;Stevens and Hedges, 2004).Modern plant communities from different arctic locations include grasses, sedges, forbs, shrubs, mosses, fungi and lichen differing significantly in both δ 15 N and δ 13 C values (Barnett, 1994;Wang and Wooller, 2006;Munizzi, 2017;Drucker, 2022).Forbs, graminoids and fungi show the highest δ 15 N values.Shrubs, forbs and graminoids (grasses and sedges) exhibit the lowest δ 13 C values, while lichen shows the highest δ 13 C values (Fizet et al., 1995;Finstad and Kielland, 2011).In times when other food sources are deficient, reindeer are thought to consume high proportions of lichen, which explains their high δ 13 C values (Fizet et al., 1995;Drucker et al., 2001;Bocherens, 2003;Stevens et al., 2008;Finstad and Kielland, 2011;Bocherens et al., 2015).Consequently, previous observations of high δ 13 C values n fossil reindeer bones at various locations, such as Belgium (Bocherens et al., 2001;Germonpré et al., 2009), Germany (Stevens et al., 2009;Drucker et al., 2011), France (Fizet et al., 1995;Drucker et al., 2000Drucker et al., , 2003;;Bocherens, 2003;Drucker and Henry-Gambier, 2005;Bocherens et al., 2005Bocherens et al., , 2011;;Drucker, 2022) or Beringia, i.e.Siberia (Iacumin et al., 2000) and Alaska-Yukon (Fox-Dobbs et al., 2008), were explained by a high proportion of lichen consumption.As ibex falls into the cluster of reindeer, a diet similar to that of reindeer could also be considered for ibex during the glacial, although nowadays the diet of European Capra ibex consists mainly of graminoids (Parrini et al., 2009).However, it is known that the Siberian ibex (Capra sibirica) in the Altai Mountains preferentially feeds on lichens (Fedosenko and Blank, 2001).The observed dietary δ 13 C differences are probably related to different habitats.As carbon isotope fractionation is driven by distinct pathways of photosynthesis, C 3 and C 4 plants can be easily distinguished by their δ 13 C values (Farquhar et al., 1989).However, the presence of substantial amounts of C 4 plants in the investigated region during the last glacial can be excluded (Iacumin et al., 1997;Bocherens, 2003;Stojakowits et al., 2020Stojakowits et al., , 2021)).Rather, a range of environmental factors, such as humidity, temperature and atmospheric CO 2 partial pressure affected the carbon isotope composition of the prevalent C 3 vegetation at that time, which directly affected physiological parameters controlling isotope fractionation in plant tissues (Tieszen, 1991).Decreased temperatures, a lower atmospheric CO 2 partial pressure, and increased water stress provoke increased δ 13 C values, while high water availability, reduced light, and denser vegetation result in overall decreased δ 13 C values (Tieszen, 1991;Pataki et al., 2003;Drucker et al., 2008;Kohn, 2010;Bonafini et al., 2013).Apart from increased lichen consumption, elevated δ 13 C values could therefore be a response either to increasing drought or to reduced vegetation cover, as exemplified for fossil reindeer (Drucker et al., 2011).
On a landscape scale, different soil moisture conditions on slope versus valley positions result in distinct δ 13 C differences.For instance, C 3 plants growing in valleys had on average 2‰ lower δ 13 C values than those on the slopes in an arid landscape in northwest China (Wang et al., 2005).Such small-scale differences are relevant for the local (e.g.hare) or likely non-migratory (e.g.bison, horse, ibex) mammals in our assemblage, while reindeer (Britton, 2010) and mammoth (Wooller et al., 2021) presumably migrated over greater distances and could therefore also reflect the isotopic signatures of a broader geographical range.However, it is a matter of debate whether reindeer migration took place in Europe in every region and at every time during the last glacial period (Fontana, 2017).For the investigated area, it can therefore be assumed that at least ibex lived in a habitat characterised by edaphically drier conditions which might have been the case on the exposed slopes and higher elevations north and northwest of the study sites (Fig. 1).In contrast, hare, horse and bison presumably lived in habitats with edaphically more humid conditions, e.g. in valleys or on the floodplains of the Danube and its tributaries south of the study sites (Fig. 1).At least for modern wild Przewalski's horses, a preference for riparian habitats in the Gobi Desert is confirmed by satellite telemetry.In contrast to Asiatic wild asses in the same region, Przewalski's horses have a much smaller home range along riparian zones, and prefer floodplain vegetation (Kaczensky et al., 2008).In addition, horses and bison need to drink water regularly (Caboń-Raczyńska et al., 1983, 1987;Scheibe et al., 1998;Kaczensky et al., 2008).These arguments support an edaphically more humid, most likely riparian habitat, where hares and bison lived alongside horses during the LGM (Fig. 4b).Whether the δ 13 C values of potentially migratory species, such as reindeer and mammoth, just accidentally fall within the fields of mammals of edaphically drier (ibex) and wetter habitats (horse, hare, bison), respectively, remains to be resolved with other techniques (e.g. 87Sr/ 86 Sr; Britton et al., 2011) in the future.Schwartz-Narbonne et al. (2015) reported that certain ecological niches can be occupied by multiple species, and despite the rather inhospitable conditions during the last glacial, these environments apparently allowed for a greater variety of habitats with options for for food sources for both generalists and specialists.For instance, the authors report an overlap of equine δ 15 N values with those of woolly mammoth, whereas in our study, horse, hare and reindeer of the pre-LGM food web overlap but at the same time have broader niches than in the LGM (Fig. 6a,c), suggesting availability of more diverse food sources for these taxa during pre-LGM.This is confirmed by the pollen record from Bergsee in the Black Forest (southern Germany), which is one of the few continuous records of floral biodiversity covering the LGM in the northern Alpine foreland (Fig. 2f; Duprat-Oualid et al., 2017).During the time of KW, HU and LK, i.e. in the latest MIS 3, the Bergsee record still shows up to 30% shrub and tree pollen, whereas the values decrease to below 20% in the LGM.This is in agreement with other palynological records that indicate small patches of trees during late MIS 3, while after 30k cal a BP arctic plant assemblages prevailed and pedogenesis came to an end (Stojakowits et al., 2021).The presence of shrubs and trees in the study area during the pre-LGM is evidenced by an abundance of charcoal found in the archaeological layers at KW + HU (Neugebauer-Maresch and Cichocki, 2008;Händel, 2017).Wood anatomical studies on charcoal particles from KW + HU and the adjacent site Krems-Wachtberg 2005-2015 allowed the identification of predominantly pine wood with minor contributions of spruce/larch and single specimens of fir and beech wood (Einwögerer, 2000;Neugebauer-Maresch and Cichocki 2008;Cichocki et al., 2014).
In both periods, goose exhibits a particularly high nitrogen isotope value among the herbivores, possibly reflecting an occupation of a particular habitat that differs from that of other herbivores.This could be connected to the uptake of a higher proportion of aquatic food or partial feeding in more southerly habitats due to seasonal migration.An accurate interpretation, however, is hardly possible due to the scarce data.

Food resources and trophic levels
δ 15 N values of the most abundant herbivores during the pre-LGM show a large inter-taxon variability (Fig. 5a) leading to a broad niche partitioning (Fig. 6a).In contrast, niche breadth shrinks considerably during the LGM and isotopic niches cluster closer together (Fig. 6a).Most likely this is the result of a narrower range of food resources during the cold and arid LGM compared with the climatically more favourable previous periods.
Apart from hare, the δ 15 N values for all other common herbivore taxa decrease towards the LGM (Fig. 5a).Several previous studies have already documented such a δ 15 N shift at around the same time for horse, reindeer, bison or red deer (Drucker et al., 2003;Richards and Hedges, 2003;Stevens and Hedges, 2004;Stevens et al., 2008).Our data allow further narrowing of the time frame for the δ 15 N shift to a period approximately between 29 and 24k cal a BP for Lower Austria.
At the pre-LGM sites KW + HU, δ 15 N values of all carnivores are on average 2.8‰ higher than those of the herbivores, whereas during the LGM, the average difference between carnivores and herbivores is 5.2‰, suggesting a difference of at least one trophic level and increased trophic enrichment in the LGM.In the δ 13 C-δ 15 N space, two clusters of carnivores are observed at KW + HU (Fig. 4a): wolf and wolverine (δ 13 C = −19.7‰,δ 15 N = 8.8‰ for both) exhibit 0.5‰ higher δ 13 C and about 1.1‰ higher δ 15 N values than arctic and red fox (δ 13 C = −20.2‰and δ 13 C = −20.5‰,δ 15 N = 7.50‰ and δ 15 N = 7.2‰, respectively).Assuming a trophic enrichment factor of 1.1 ± 1.1‰ for δ 13 C and 3.2 ± 1.8‰ for δ 15 N (Krajcarz et al., 2018), prey isotope values for fox result in the range of all common herbivores except ibex, reindeer and mammoth.This may suggest that foxes at KW + HU fed mainly on carrion and small mammals like hare.Likewise, wolf and wolverine could have fed on all abundant herbivores except for hare and ibex.This is not in conflict with a previous study that inferred horses and possibly mammoth as prey items for Pleistocene wolves based on statistical analyses (Bocherens et al., 2015).
However, our dataset is biased by prehistoric humans' selection and therefore lacks certain species, e.g.micromammals that may have been an important prey item especially for foxes (Baumann et al., 2020).Red and arctic foxes are opportunistic predators that can change trophic behaviour and adapt to a new diet whenever necessary (Baumann et al., 2020).Foraging on micromammals on the one hand and scavenging on large mammals (potentially even other carnivores) on the other could explain the high δ 15 N variability of arctic fox in the LGM.In foxes, it has been observed that specialisation on micromammal prey or carrion of large herbivores also leads to inter-specimen δ 15 N differences of up to several per mill at other Palaeolithic sites (Baumann et al., 2020), similar as in the LGM at KG.

The special case of woolly mammoth
In some cases, altering environmental conditions may lead to changing isotopic niches of certain species, such as the woolly mammoth.In the pre-LGM food web, mammoths reveal the highest δ 15 N values of all herbivores and appear on a similar level as carnivores.This phenomenon is already known from other pre-LGM sites in Eurasia, e.g. from France (Bocherens et al., 2005;Drucker et al., 2015), north-eastern China (Ma et al., 2017(Ma et al., , 2021)), Belgium (Bocherens et al., 1997(Bocherens et al., , 2001)), Beringia (Bocherens et al., 1994;Fox-Dobbs et al., 2008), Germany (Bocherens et al., 2011;Drucker et al., 2015) and Yakutia (Bocherens et al., 1996;Szpak et al., 2010;Iacumin et al., 2000Iacumin et al., , 2010)).These high δ 15 N values in mammoth collagen could indicate a special diet and thus a distinct dietary niche, a particular habitat, or mammoth-specific physiological or metabolic processes.Results from compound specific isotope analyses on collagen amino acids (phenyl-alanine and glutamate) of Pleistocene mammoths from Yukon, Canada, showed, however, that the high δ 15 N values of collagen come from high δ 15 N values of the plants consumed rather than from metabolic processes (Schwartz-Narbonne et al., 2015;Naito et al., 2016).Several studies show that mammoths mainly fed on forbs and graminoids, i.e. grasses and sedges, which tend to have higher δ 15 N values (Stewart et al., 2003;Wang and Wooller, 2006) compared with shrubs and lichen (Wang and Wooller, 2006;Finstad and Kielland, 2011;Kristensen et al., 2011).Moreover, remains of grasses and sedges were also found in the jaws of fossil mammoths from Beringia (Guthrie, 2001;Drucker, 2022).A similar diet can also be reconstructed for woolly rhinoceros and steppe bison, which show δ 15 N values partly similar to those of mammoths in the pre-LGM food web in our study (Fig. 4a) and may suggest a special dietary niche.In addition, the high δ 15 N values in combination with the low δ 13 C values of pre-LGM mammoth collagen could also indicate a freshwater plant food source.Kirillova et al. (2016) found considerable amounts of freshwater plants in MIS 3 mammoth faeces from northern Russia and concluded that mammoths may also have occupied freshwater, riparian and watershed biotopes.
Our pre-LGM data imply that mammoths occupied an ecological niche where high δ 15 N values and lower δ 13 C values were prevalent and that was not used by other herbivores to the same extent.δ 15 N values of plants are controlled by inorganic primary nitrogen sources (NO 3 − , NH 4 + and N 2 ) in the soil as well as their respective isotope discrimination, caused by incorporation or dissimilation (Högberg, 1997;Evans, 2001;Werner and Schmidt, 2002).In addition, great amounts of mammoth dung may have functioned as organic fertiliser in soils which would have resulted in higher δ 15 N values of plants (Georgi et al., 2005;Bogaard et al., 2007;Fraser et al., 2011) in the preferred ecological habitat of mammoths during pre-LGM time.This interpretation, however, seems very unlikely for our dataset, as other herbivores do not show elevated δ 15 N values.
Accordingly, ecological patterns of mammoth in the pre-LGM period support both distinct feeding habits and the occupation of a specific habitat that is entirely different from habitats of other herbivores (the exception in our study sample is a single musk ox and rhinoceros specimen) (Figs. 4,5).Changing climate in different geographical or temporal zones towards the end of the Pleistocene, however, may have led to the disappearance of such special niches and specific adaptations of mammoth populations (Drucker et al., 2003;Schwartz-Narbonne et al., 2015).This is in accordance with Nadachowski et al. (2018) who reported significant changes in mammoth territories across Europe between 29 and 14 ka and an almost complete disappearance of mammoths during the LGM around 21-19 ka for the North European Plain.Drucker et al. (2018) show that the distinct ecological niche of the woolly mammoth changed shortly after the LGM in Europe around 18k cal a BP.δ 15 N values of mammoths from Mezhyrich in the Eastern European Plain decreased to the level of other herbivorous species and are in accordance with the mammoth nitrogen isotope values from Kammern-Grubgraben in this study (Figs. 4,5).Based on our data, however, the decline of mammoth δ 15 N values already occurred around 22k cal a BP and thus earlier than previously reported.One possible cause for these changes is the climatic deterioration of the LGM that started around 30k cal a BP and caused large-scale climatic and environmental changes on both global and regional scales (Luetscher et al., 2015;Kämpf et al., 2022).Alternatively, pre-LGM mammoths could have spent most of the time in other regions with distinctly higher isotope values of food sources and just seasonally migrated into the study area.As a consequence, the pre-LGM mammoths may have been more migratory than mammoths during the LGM.

Implications for environmental changes between the pre-LGM and the LGM
In general, substantial environmental changes may lead to changes in the nitrogen isotope pattern at the base of terrestrial food webs (Stevens et al., 2008;Bocherens et al., 2014;Schwartz-Narbonne et al., 2015).This is evident in our study, where mean δ 15 N values for herbivores are generally lower in the LGM than the pre-LGM periods, except for hare (Fig. 5a).Plant δ 15 N values depend on various factors, such as soil δ 15 N, pedogenesis, nutrient availability, soil acidity and nitrogen cycling (Högberg, 1997;Drucker et al., 2003;Stevens and Hedges, 2004;Stevens et al., 2008).These parameters are to a great extent controlled by climate factors, which is why plant and soil δ 15 N values indirectly correlate with climatic factors, such as mean annual precipitation and mean annual temperature (Handley et al., 1999;Amundson et al., 2003, Craine et al., 2015).The δ 15 N values of modern soils show a latitudinal variability, where cold and/or wet ecosystems at high latitudes are the most depleted and hot and/or arid zones the most 15 N enriched (Amundson et al., 2003).However, δ 15 N values of animal collagen in the last glacial may also have been influenced by temperature-induced thawing of permafrost affecting soil moisture and soil activity (Stevens et al., 2008).On the other hand, low temperatures hamper nitrogen cycling processes such as remineralisation, nitrification and denitrification leading to lower δ 15 N values (Drucker et al., 2003).Results similar to ours were reported by Stevens et al. (2008), who found significant changes in δ 15 N collagen of reindeer over the last glacial period.Before 24k cal a BP, reindeer had significantly higher δ 15 N values (mean: 4.2‰) than after 24k cal a BP, when δ 15 N values gradually decreased (mean: 2.9‰) until 16k cal a BP when, finally, δ 15 N slightly increased again as conditions became wetter (Stevens et al., 2008).Higher δ 15 N values in our study (Figs. 4,5) would thus correspond to higher soil moisture and increased soil microbial activity before the LGM when compared with the LGM.
The pre-LGM sites fall into a time period with severe climatic fluctuations best known from Greenland ice core records (Fig. 2h,i,j).For instance, δ 15 N values of air enclosed in bubbles and δ 18 O values from ice cores of the North Greenland Ice Core Project (NGRIP) provide temperature records and are inversely correlated with Ca 2+ records representing dust deposition (Mayewski et al., 1997;Andersen et al., 2004;Kindler et al., 2014;Rasmussen et al., 2014).The pre-LGM sites were occupied around the short-termed Greenland Interstadials 5.2 and 5.1, during which seasonal melting of continuous permafrost may have occurred.As such, a predominance of tundra gley soils has been reported for this time interval in the loess-palaeosol sequence of the Nussloch site in southern Germany (Fig. 2g; Antoine et al., 2009;Moine et al., 2017;Maier et al., 2021).At Krems-Wachtberg 2005-2015, the sediment sequence of this time interval partly shows periglacial features (Terhorst et al., 2014), which was also confirmed for the main archaeological layer AH 4 (Händel et al., 2021b).The slightly broader range of δ 13 C values of bone collagen from this time period may also indicate an ameliorated climate, as it reflects a more diversified diet in a mosaic environment.This is in agreement with pollen data from the Alpine region and the Black Forest showing a predominance of forest-tundra vegetation for this time interval in contrast to the subsequent period of the LGM (Heiri et al., 2014;Stojakowits et al., 2021;Kämpf et al., 2022).
Considering the δ 15 N value shift as an indication of environmental and climatic changes, decreased δ 15 N collagen values from the LGM site KG would accordingly correspond to cold and arid conditions as continuous permafrost prevailed and large amounts of water were stored in permafrost soils and adjacent glaciated areas (Stevens et al., 2008).Low temperatures but relatively stable conditions can be deduced from the NGRIP record (Fig. 2h,i,j), and loess records from KG point to relatively constant environmental conditions during the main occupation phase (Reiss et al., 2022).Low soil moisture, hardly any soil activity and only incipient tundra gley formation in the predominantly homogeneous loess (Antoine et al., 2009) are assumed for the period between 21 and 19 ka (Maier et al., 2021).Moreover, Drucker et al. (2012) attributed lower δ 15 N collagen values of reindeer, red deer and horse to continuous permafrost and/or to the proximity of glacier fields, and the highest δ 15 N collagen values to regions of discontinuous permafrost.

Conclusions
We compared δ 13 C and δ 15 N of faunal collagen from the Early Gravettian/pre-LGM sites HU, KW and LK with the Early Epigravettian/LGM site KG (AH 2).Food webs from both periods reveal a characteristic structure with differences in trophic ecology between herbivores and carnivores.A δ 13 C clustering between the herbivorous species hare and horse on the one hand and ibex on the other hand points to niche partitioning among herbivorous groups associated with distinct habitats along a local humidity gradient.During the pre-LGM period, the potentially migratory mammoth does not cluster in one of these groups, but in the LGM it overlaps with hare, horse and bison.Reindeer, on the other hand, clusters with ibex in both periods, albeit being considered as migratory.These ambiguities related to the migratory behaviour of mammoth and reindeer can only be resolved by applying other isotope techniques such as 87 Sr/ 86 Sr in future studies.
The general δ 15 N shift of 3.3‰ between all herbivores of two investigated time periods (KW + HU, KG) can be attributed to climatically induced changes in the environment.This shift lowers to 2.0‰ if mammoth is excluded.Previously reported, strikingly high nitrogen isotope values of mammoths can also be confirmed for the Early Gravettian/pre-LGM sites in this study, but at the same time, we show that the disappearance of this particular niche occurred during the LGM before 23-22k cal a BP, and thus earlier than previously established for central Europe.The isotopic niche widths expressed by SIBER analyses indicate larger habitat diversity for herbivores during the pre-LGM compared with the LGM, with the exception of ibex and mammoth.Accordingly, the herbivore clusters tend to increasingly overlap at the LGM site KG.As climatic conditions became more extreme in the course of the LGM, and woody plants declined, certain habitats disappeared and different species had to share ecological niches.This decrease in habitats should also have resulted in a greater homogeneity of food sources as indicated by restricted isotopic ranges of non-migratory herbivores during the LGM compared with the pre-LGM period.Also, with the advances of Alpine and Baltic ice shields during the LGM, the ranges for migratory mammals, in particular mammoth, most probably became more restricted, which might have led to a greater isotopic similarity to more local mammals during the LGM.

Figure 1 .
Figure 1.Location of the study sites in Lower Austria.The study sites are indicated by red dots.A digital elevation model in 10 × 10 m resolution (www.data.gv.at/katalog/dataset/land-noe-digitales-hohenmodel-10-m#resources) served as base map.

©Figure 3 .Figure 4 .
Figure 3. FTIR spectra of (a) bone powder and (b) extracted bone collagen of the same samples from HU, KW, LK and KG, respectively.Note that the y-axis is arbitrary.[Color figure can be viewed at wileyonlinelibrary.com]

©Figure 5 .
Figure 5. Box plots of most common herbivore δ 15 N (a) and δ 13 C (b) values of KW + HU (dark grey) and LK (light grey) in comparison with KG (white).Only taxa with n > 3 are shown.Boxes represent 25 and 75 percentiles, open squares means, horizontal lines medians, whiskers standard deviations and black diamonds outliers.Stars indicate probability values that the null hypothesis ('values in both time periods are not different') can be rejected (***p < 0.001, **0.001 < p < 0.01, *0.01 < p < 0.05).

Figure 6 .
Figure 6.Standard ellipse areas (SEA) (solid ellipses) and convex hulls (dashed lines) of different herbivores from (a) Early Gravettian/pre-LGM food webs KW + HU and LK (ibex) (33-29k cal a BP) and (b) Epigravettian/LGM food web KG (24-20k cal a BP).Lower panels are density plots of SEA of (c) Early Gravettian/pre-LGM food webs KW + HU and LK (ibex), and (d) Epigravettian/LGM food web KG.Black circles are mode values of the SEAb values, i.e.SEA calculated using a Bayesian approach.Grey boxes represent 50, 75 and 95% credibility.Red crosses are SEA values corrected for sample size.For the number of specimens, see Fig. 5. [Color figure can be viewed at wileyonlinelibrary.com]

Table 1 .
List of species and a-2002ed material sampled fromKrems-Hundssteig 2000-2002, Krems-Wachtberg 1930, Langenlois A and Kammern-Grubgraben with indication of the find layer and the minimum number of individuals (MNI) at the respective sites.The MNI is stated at the beginning of each newly listed taxon.The abbreviation 'n.d.'