The two US studied in this work are only known from a 50 cm2 test-pit each, and their deposition was made over several periods of time of deposits.

US4 (5 to 20 cm thick) is composed of pinkish sedimentary breccia with small calcareous elements, Aurignacian Lithic material and burnt fauna; it was dated at 34 880 ± 260 BP (Beta-449 411), i.e., 40 631 to 39 480 cal. BP (Intcal 2020, Reimer et al. 2020) (Jarry et al. 2017; Pallier et al. 2024).

US5 (> 40 cm thick, base not reached) is a mixture of limestone blocks and fine brown silty-clay sediments with an increasingly dense chaos of limestone blocks, Aurignacian material and fauna; it was dated at 33 260 ± 210 BP (Beta-449 412), i.e., 38 954 to 37 157 cal. BP (Intcal 2020, Reimer et al. 2020) (Jarry et al. 2017; Pallier et al. 2024).

The Radiocarbon datings were made on undeterminable burnt bone fragments, and although indicating an apparent inversion attributable to post-depositional processes, they are homogeneous and belong to the early Aurignacian (Fig. 2). The early Aurignacian is the most well-known among the different stages of the Aurignacian sequence of the northern Pyrenees, and lithic industries from US4 and US5 are perfectly similar to those described in other regional context securely attributed to the early Aurignacian, like at La Tuto de Camalhot (Ariège) or La grotte ‌ ‌ ‌des Hyènes at Brassempouy (Landes) (Bon 2002; Anderson 2019). No difference was observed between the Aurignacian industries of US4 and US5, which have been noted as “sedimentary units” but belong to the same archaeological set.

A total of eight samples from US4 and ten samples from US5 were recovered for microvertebrate study from test-pit 2.2 of the “Theatre” room (Fig. 1).

Each sample represents a bucket of sediment (c. 10 l.) corresponding to a “spit” extracted during the 2014 and 2015 excavation campaigns, then sieved (2 mm mesh) and sorted. Each sample was numbered separately, keeping at each stage the information of its provenance (test-pit, US, z; Appendices 1; 2).

Even if it would have been preferable to consider each spit separately (Royer 2014), the quantity of material present in each spit was not sufficient to obtain reliable and representative information. Moreover, some samples appear clearly smaller than the others, probably linked to variations in the amount of sediment vs limestone blocks in each bucket, leading to a bias of species and anatomical representation between spits. Consequently, for the present study we have chosen to group the spits following the archaeological US.

The identification of the recovered species were made thanks to comparisons with modern specimens from the Muséum national d’Histoire naturelle (collection Mammifères et Oiseaux), as well as identification keys from the literature, notably Chaline et al. (1974), Quéré & Le Louarn (2011), Laplana et al. (2015) and Kryštufek & Shenbrot (2022) for small mammals, Bailon (1999) for amphibians, Bocheński & Tomek (2009) and Tomek & Bocheński (2009) for birds. For each taxon, we considered the number of identified specimens (NISP), the minimum number of elements (MNE) and the minimum number of individuals (MNI). Birds and bats were not precisely identified. For Arvicolines, we followed Kryštufek & Shenbrot (2022) for the nomenclature; the identification at the species level and the calculation of MNI were based on isolated and in situ lateralized first lower molars, representing the most discriminating element. Other elements were placed in unidentified Arvicolines.

For the taphonomic analysis of the small mammal remains, we followed the method developed by Andrews (1990) and Fernández-Jalvo et al. (2016). The anatomical representation was calculated following the formula: PR = [Fobs / (Fth × MNI)] × 100, with PR = percentage of representation, Fobs = observed frequency of each element in the fossil assemblage, Fth = theoretical frequency of each element in a living individual, MNI = minimum number of individuals. Concerning the surface alteration of bones, we especially considered rodent femoral heads, isolated incisors and molars for the study of the digestion. For each considered element, the percentage of digestion (PD) was calculated as follows: PD = (Ndig / Ntot) × 100, with Ndig = number of digested elements, Ntot = total number of elements. Other traces (oxids, roots, weathering, trampling, among others) were noticed but not quantified. As bone fragmentation can be subject to various peri-mortem and post-depositional factors, the percentage of fragmentation is most often high in archaeological context and not very informative, so we have chosen not to study it here in detail. For bats, shrews, amphibians and birds the low number of remains did not allow an in-depth taphonomic analysis, and only general observations were made.

Current distribution and ecological requirements of species follow Aulagnier et al. (2008), Quéré & Le Louarn (2011), Lescure & Massary (2012), Nicolas et al. (2021), Kryštufek & Shenbrot (2022) and the IUCN database (http://www.iucnrredlist.org). Then several complementary methods for palaeoclimatic and palaeoenvironmental reconstructions commonly used in micromammal studies were applied:

— Climatogram: this method considers the relative proportions of rodent species (minimum number of individuals (MNI) by grouping them into climato-ecological classes according to their affinities. This method has been used by various authors (e.g. Chaline 1983; Marquet 1993; Desclaux & Defleur 1997; Barroso Ruiz et al. 2003; Cuenca-Bescós et al. 2005, 2009; Hanquet & Desclaux 2011).

— Taxonomic Habitat Index (THI): this is an approach that takes into account only the presence or absence of species (Andrews 1990, 2006; Evans et al. 1981). For each species, a total value of 1 is distributed between the different frequented habitats, with different values according to the intensity of frequentation. For each habitat type, the sum of the indices given for all the species present per level is calculated, allowing a dominant habitat to be determined.

— Climate Restriction Index: this method, developed by Hernández-Fernández (2001) and updated by Royer et al. (2020), shows some similarities with the THI. The main difference is that the score allocated to the different habitats frequented by a species is distributed equally, regardless of the proportions of habitats occupied. If 15% or more of a species’ range is present in the climate zone, then the species is considered to inhabit that zone. As with the THI, this method makes it possible to reconstruct palaeo-landscapes, avoiding potential biases of over-representation of species. It also allows ecological preferences to be considered, but with an equitable distribution between the zonobiomes.

Two additional methods are used for the reconstruction of climate parameters, based on large databases of current species distribution and climatic parameters. The use of the two following methods makes it possible to step back from the amplitudes of variation of the climatic parameters:

— Bioclimatic Model: based on the assumption that there is a correlation between climate and mammalian communities, the Bioclimatic Model uses data from the Climate Restriction Index to determine climatic parameters using discriminant analysis and multiple linear regression (Hernández-Fernández 2001; Hernández-Fernández & Pelaez-Campomanes 2005; Royer et al. 2020). This method allows an initial approach to climatic parameters, particularly temperature (mean annual temperature, mean temperature of the warmest month and mean temperature of the coldest month).

— Quantified Ecology method: was developed by ­Jeannet (2010). For each species, quantitative climatic values are assigned (mean annual temperature, minimum and maximum annual temperature, mean annual precipitation, among ­others) according to its range.

A total of 3 762 remains were studied in both units. Among small mammals, one lagomorph species, nine rodent species, one mole and one shrew species and several unidentified bats were recognised (Table 1; Fig. 3). Remains of amphibians (mainly frogs) and several species of small passerine birds (as well as three distal phalanges belonging to a small raptor in US5) have also been found.

The relative proportions of the major faunal groups and the faunal spectra are very similar in the two US, although the quantity of material appears more important in US5. Both units are dominated by Arvicolines, especially Microtus gr. arvalis-agrestis and Arvicola gr. amphibius-sapidus, and amphibians are also well represented (Table 1). Most species still inhabit the study region today (Apodemus gr. sylvaticus-flavicollis, Arvicola gr. amphibius-sapidus, Microtus gr. arvalis-agrestis, Talpa gr. europaea-aquitania, Sorex gr. araneus-coronatus, Ranacf. temporaria, Pelophylax sp.), but others are found today more in altitude or in more distant areas in the same region (Chionomys ‌ ‌ ‌nivalis ‌(Martins, 1842)), or in more eastern regions of northern and central Europe and Asia (Alexandromys ‌ ‌ ‌oeconomus ‌(Pallas, 1776)).

Several species groups occurring in the study region today display a similar bones and teeth morphology, which make them difficult to discriminate without further morphometric analyses. This is notably the case for Apodemus gr. sylvaticus-flavicollis, Arvicola gr. amphibius-sapidus, Microtus gr. arvalis-agrestis, Talpa gr. europaea-aquitania and Sorex gr. araneus-coronatus. Both morphotypes were recognized at the Mas d’Azil cave for M.arvalis ‌(Pallas, 1779)/M. ‌ ‌ ‌agrestis ‌(Linnaeus, 1761) and A.  ‌ ‌ ‌sapidus ‌Miller, 1908/A. ‌ ‌ ‌amphibius ‌(Linnaeus, 1758). For the other genera (Apodemus ‌Kaup, 1829,Talpa ‌Linnaeus, 1758,Sorex ‌Linnaeus, 1758) it is difficult to say for the moment if one or two species are represented in the assemblage for each of them, waiting further morphometric analyses. However, Sorex mandibles of the Mas d’Azil appear closer to S. ­coronatus Millet, 1828. Concerning Talpidae, other European Talpa species are also known in Italy, Greece or Spain, but they are generally smaller than T. ‌ ‌ ‌europaea ‌Linnaeus, 1758 and T. ‌ ‌ ‌aquitania ‌Nicolas, Martinez-Vargas & Hugot, 2017.

We note the absence of forest species such as Sciurus ­vulgaris Linnaeus, 1758 or Glirids, occurring today in the area, as well as Mediterranean voles of the sub-genus Terricola Fatio, 1867, today represented in southwestern France notably by Microtus ‌ ‌ ‌duodecimcostatus ‌(Selys-Longchamps, 1839), M. ‌ ‌ ‌lusitanicus ‌(Gerbe, 1879) or M. gr. pyrenaicus-gerbei. We observed a very large morphological variability within the Microtus gr. arvalis-agrestis and Chionomys ‌ ‌ ‌nivalis species (Fig. 3). Moreover, three lower m1 display unusual morphological characters in the US5: specimen of Microtus size from sample US5-Sond2.2-z10-11 (Fig. 4C) with a reduced T6 and presence of a BRA4 re-entrant angle (aberrant Microtus ‌ ‌ ‌arvalis/agrestis or Chionomys ‌ ‌ ‌nivalis specimen?); cf. Arvicola sp. from sample US5-Sond2.2-z10-11 (Fig. 4B) with a relatively small size and absence of T4 (reduced to a simple “bulge” of the anterior lobe); cf. Arvicola sp. from sample US5-Sond2-z11-12 (Fig. 4A) with an exceptionally high crown and an asymmetrical wear in occlusal view (probably related to an occlusal defect resulting in abnormal growth and wear of the tooth, perhaps accentuated by digestion).

Concerning amphibians, Rana ‌ ‌ ‌temporaria ‌Linnaeus, 1758 is probably the most common frog species found in current and fossil communities. But today in the study region, we can also find several other species belonging to the genera Rana ‌Linnaeus, 1758 (R. ‌ ‌ ‌pyrenaica ‌Serra-Cobo, 1993,R. ‌ ‌ ‌dalmatina ‌Fitzinger, 1838) and Pelophylax ‌Fitzinger, 1843 (P. ‌ ‌ ‌ridibundus ‌(Pallas, 1771),P. ‌ ‌ ‌perezi ‌(López-Seoane, 1885)), for which the osteological anatomy remains poorly known. Thus, their presence in the site is not impossible, but difficult to demonstrate for now, especially considering the high rate of fragmentation of the bones in the site. No specimen belonging to the genus Bufo ‌Garsault, 1764 could have been identified at the Mas d’Azil cave, but we cannot exclude that remains belonging to this genus can be present among the amphibian indet. material.

The material is highly fragmented in both US, and almost all the molars are dislocated (leading to their over-­representation in the anatomical profile; Fig. 5), and no element was found complete, which limits certain identifications, especially for amphibians.

Regarding the anatomical representation of micromammals (Fig. 5), we see that both US display a similar pattern, with however a higher representation of cranial bones (mandibles, palatum) in the US5, whereas in US4 the isolated molars are better represented. Most skeletal elements are well represented, except for the smallest and/or most fragile elements such as phalanges, ribs and radius, most likely due to the relatively large mesh size of the sieves (2 mm). For the same reason, Murinae and other small taxa may be under-represented (molar size frequently < 2 mm). Apart from that, no sorting by size/density/shape seems to have occurred by the action of a water flow or other natural process, reinforced by the fact that no element rounded by water action was observed.

The presence of digested elements for all taxa (rodents, bats, amphibians, birds) attests that at least part of the assemblage is due to the action of predators. The percentage and intensity of digestion of rodent elements are relatively low (Table 2; Fig. 6), considering femoral heads (15.4% and 17.1% for US4 and US5 respectively) and isolated molars (13.3% and 9.6% for US4 and US5 respectively). Values are higher for isolated incisors (62.5% and 82.4% for US4 and US5 respectively). In situ incisors and molars were too scarce to be considered (Table 2). In comparing with modern referentials (e.g. Andrews 1990; Denys et al. 2018; Royer et al. 2019; Lebreton et al. 2021), our results argue for a nocturnal raptor of category 1 (e.g. Barn owl, Short eared owl) based on femoral heads, but rather of category 3 (e.g. Eagle owls, Tawny owl, Fox) based on isolated incisors and molars. Such discrepancies were previously observed for the Snowy owl for example (Royer et al. 2019), and the differences observed between US may be related to a sample effect rather than a difference in the predator species. Moreover, considering the large size of the cave, it is not impossible that several predator species could have occupied the site simultaneously. Concerning bats, even if some remains display traces of digestion, a natural in situ accumulation is also to be considered.

Post-depositional alterations were observed in low proportion, only few root marks, weathering effects and spots of Manganese oxids.

Throughout the Quaternary climatic fluctuations, the Arize river experienced several phases of alluvial filling (periglacial context) and re-excavation (more temperate context). The sedimentary sequence of the Mas d’Azil cave reflects these environmental changes and has recorded at least two aggradation phases (Pallier et al. 2016, 2024; Jarry et al. 2021a; Pallier 2021). The oldest fluvial fillings (Mandement and Temple rooms) show an important sedimentary aggradation in this part of the cavity, which has been entirely filled in by the alluvium of the Arize during an earlier period of the Pleistocene (MIS6). Then the Arize river resumed a free flow dynamic and dug this filling, emptying a part of the rooms and galleries of the cave, and allowing humans or faunas to reach the cave. It is in this context that the Aurignacians have occupied a part of the cavity, during a more favorable climatic period of MIS3. The last fluvial sequence (Theatre/Rotonde room) dates to the last glacial period (from the end of MIS3) and covers the Aurignacian occupation levels with over several meters of deposits. This event blocked part of the Mas d’Azil cave and prevented any access to the human populations for several thousand years, between the Gravettian and the Solutrean. Human populations started to settle back in the cave in the Solutrean/Badegoulian period (MIS2), and then massively occupied it during the Magdalenian period (Pallier et al. 2016, 2024; Pallier 2021).

Concerning faunal data from the Aurignacian levels of the Mas d’Azil cave (US5-3), unfortunately the number of determined remains of large mammals is quite low, due to significant bone fragmentation and burning (Martin in Jarry et al. 2014, 2015; Jarry et al. 2017). The identified taxa are mainly equids, reindeer (Rangifer ‌ ‌ ‌tarandus ‌(Linnaeus, 1758)) and bovines, with the occasional presence of birds and carnivores (Vulpes sp., large canids). Although limited, these results confirm the idea of an open environment of grassland or steppe under a cooler climate than the present one, especially through the presence of reindeer. However, this latter is less represented than in the Magdalenian levels, characterized by more marked cold conditions.

To summarize, both sedimentary record and large mammals argue for cool, but not glacial conditions during the early Aurignacian occupations of the Mas d’Azil cave, and thus agree with the results of the microvertebrates. Moreover, it has been hypothesized that US4 may be an indurated facies of the same deposit as US5, or that US5 may be an open scree deposit into which the matrix has been subsequently inserted (Pallier 2021), which could partly explain the proximity of both microvertebrate assemblages.

The emergence of the early Aurignacian culture, to which are attributed the human occupations of the US4 and US5 of the Mas d’Azil, would have been contemporaneous of cold and dry climatic conditions across the European continent (Heinrich Event 4), leading notably to a change in the large mammal faunas (e.g. Discamps et al. 2011) and a geographic expansion of the human populations (Banks et al. 2013). However, anthracological data (Allué et al. 2018) have shown the persistence throughout MIS3 to MIS1 of Pinus ‌ ‌ ‌sylvestris L. type forests in northeastern Spain, and in southwestern Europe in general (Fletcher et al. 2010). This indicates that in this region, the climatic changes were not so extreme as to prevent the development of trees, and that the conditions were mildered compared to those of more northern regions. But what are the indications given by the small faunas concerning the MIS3 climatic changes in the region?

Few references are available for the Pyrenean region (either in France or in Spain). We avoided older works with few materials, for which the chrono-stratigraphic context can be dubious and for which sieving and sorting of microvertebrate remains were not systematic, or even absent. Moreover, we focused on cave sites, allowing for the recovery of large amounts of microvertebrate remains (contrary to most of open-air sites).

In the considered area, most studies on small vertebrates dating to the MIS3 focus on older cultural periods, i.e., the Mousterian (Marquet et al. 1998; López-Garcia et al. 2012; Discamps & Royer 2017; Fernández-García et al. 2018, 2020, 2022) or the Mousterian-Aurignacien transition (López-García et al. 2011; Álvarez-Vena et al. 2021). The synthesis of Royer et al. (2016) on micromammals from southwestern France includes few sites in the study area (central-eastern Pyrenees), which are dated to more recent periods. Very few studies cover the Aurignacian chrono-cultural period.

To our knowledge, only one other study deals with small vertebrate assemblages dating to MIS3 in the Ariège region, at the site of Portel-Ouest. Only few micromammal material was collected from the former Upper Palaeolithic Ensemble (levels B-B1) and was attributed to Alexandromys ‌ ‌ ‌oeconomus,Stenocranius ‌ ‌ ‌gregalis ‌(Pallas, 1779), Microtus ‌ ‌ ‌arvalis-agrestis,Microtus (Terricola) subterraneus (de Selys-­Longchamps, 1836), Arvicola ‌ ‌ ‌amphibius and Chionomys ‌ ‌ ‌nivalis (Marquet 1993; Marquet et al. 1998). However, the level B was subsequently dated to MIS 2 by the Electron Paramagnetic Resonance method (23.1 ± ٣.٥ ka; Tissoux 2004) and thus associated to the Gravettian, while the level B1a was dated to MIS 3 (36.6 ± ٥.٤ ka; Tissoux 2004) but associated to a Chatelperrionan industry. The intermediate level B1, which could have potentially corresponded to the chrono-cultural period of the Aurignacian, has apparently not been dated. Some of the underlying levels were also dated to MIS 3 (Ensemble F, c. 44-45 ka; Tissoux 2004) but associated to Mousterian industries.

The syntheses provided by López-García et al. (2014, 2015) and Fernández-García et al. (2016) allow us to make some comparisons with other sites contemporaneous of the US4 and US5 of the Mas d’Azil cave (c. 40-37 ky cal. BP), such as Teixoneres cave, Level II (c. 60-30 ky BP) and Arbreda cave, Level I (40-32 ky BP), both located in northeastern Spain. In the considered levels of both sites, the mid-European species Microtus ‌ ‌ ‌arvalis and M. ‌ ‌ ‌agrestis are dominant, while “forest species” (Apodemus ‌ ‌ ‌sylvaticus ‌(Linnaeus, 1758), Glis ‌ ‌ ‌glis (­Linnaeus, 1766), Eliomys ‌ ‌ ‌quercinus ‌(Linnaeus, 1766)) are present in much lower proportions, which would indicate cold, open humid conditions. Chionomys ‌ ‌ ‌nivalis is also represented in both sites. Palaeoclimatic reconstructions (López-García et al. 2012, 2015) and isotopic analyses (Fernández-García et al. 2022) indicate colder mean annual temperatures than today, but higher mean annual precipitations. The results that we obtained at the Mas d’Azil cave fit well with these data for the temperatures (but not for precipitation). However, one major difference is the permanent presence of Mediterranean vole species in the Spanish sites (e.g. Microtus ‌ ‌ ‌duodecimcostatus,Microtus (Iberomys) cabrerae ‌Thomas, 1906), together with Glirids (e.g. Eliomys ‌ ‌ ‌quercinus), while both are absent in the Mas d’Azil assemblages. The absence of M. ‌ ‌ ‌duodecimcostatus would be in line with the hypothesis of Royer et al. (2016), who stated that this species was absent from southwestern France during a large part of the Late Pleistocene and became widespread only from the beginning of the Holocene in the region. Moreover, the Mas d’Azil is located at the border of the current distribution of the species, which could also explain its absence in the site. Microtus(Iberomys)cabrerae is no longer recorded North of the Pyrenean Mountains, thought its distribution extended further North in the past (López-García & Cuenca-Bescós 2012; Stoetzel 2022). Concerning the absence of Glirids, Fernández-García et al. (2016) indicate that the Mediterranean coast of the Iberian Peninsula has experienced the most pronounced forest development, which could explain a better representation of typical forest species compared to the Mas d’Azil cave, which is located in a more continental area of the Pyrenean foothills. More generally, Fernández-García et al. (2016) noticed differences in the faunal communities from MIS4 and MIS3 sites from northeastern Spain, and those from contemporaneous sites located in southwestern France, which generally show a colder and drier palaeoecological signal. However, in this region, Royer et al. (2016) highlighted the absence of typical “cold” species such as Dicrostonyx ‌ ‌ ‌torquatus ‌(Pallas, 1778) between 50 and 38 ky BP. Moreover, these authors have shown that despite changes in faunal distributions in response to abrupt and brief climatic events between MIS3 and MIS1 (Heinrich events), the rodent communities were not mandatory impacted on a regional scale. In southwestern France, the MIS3 recorded the concomitant presence of ­Apodemus ‌ ‌ ‌sylvaticus, Eliomys ‌ ‌ ‌quercinus, Arvicola ‌ ‌ ‌sapidus, Microtus gr. arvalis-agrestis, ­Alexandromys ‌ ‌ ‌oeconomus, Spermophilus sp., and Stenocranius ‌ ‌ ‌gregalis is also well represented during MIS3 and MIS2. If most of these species are also represented in our study assemblages, E. ‌ ‌ ‌quercinus, Spermophilus sp. and S.gregalis are not. But it has to be noticed that no site located in the eastern part of Pyrenean Mountains has yielded E. ­quercinus (Royer et al. 2016). The absence of taxa related to cold steppic habitats such as Spermophilus sp. and S.gregalis may be linked to the occurrence of more temperate climatic conditions, although cooler than today, in the Aurignacian levels of the Mas d’Azil cave. Lagurus ‌ ‌ ‌lagurus ‌(Pallas, 1773) is rare in Late Pleistocene sites from southern France but was recorded in the neighbouring site of Portel-Ouest, although in an older, Mousterian, level (Marquet 1993; Marquet et al. 1998), which may traduce more continental conditions. For the moment, it is difficult to decipher whether all these differences are due to geographical differences (Atlantic vs Mediterranean, coastal vs continental, plains vs mountains) and/or chronological discrepancies.

Moreover, it is acknowledged that the altitude gradient favors specific diversity, notably through the geographical proximity between ecotones, but also through the existence of habitat heterogeneity (Grytnes & McCain 2007). The presence of species with different ecological affinities can be explained by the site’s proximity to a variety of habitats (grasslands, riparian forests, rocky slopes).

Appendix 1. —Detailed faunal list of the small vertebrates from the US4, in separating each spit. Abbreviations: MNI, minimum number of individuals; NISP, number of identified specimens.

Appendix 2. — Detailed faunal list of the small vertebrates from the US5, in separating each spit. Abbreviations: MNI, minimum number of individuals; NISP, number of identified specimens.

Appendix 3. — Distribution of the taxonomic habitat indices (THI) used for the micromammals of Aurignacian levels of the Mas d’Azil cave. Abbreviations: B, Boreal forest; D, Deciduous forest; HM, Humid meadow; M, Mediterranean; Mo, Mountain; S, Arid steppe; TU, Tundra (modified from Hanquet 2011).