Quaternary proboscidean discoveries are not rare in the NE of Catalonia, where a large part of a Mammuthus meridionalis skeleton was found in the Incarcal Quarry (Mazo et al., 2003). Palaeoloxodon antiquus is not as well represented and only some tooth fragments were discovered in Mollet Cave (Serinyà) (Solés & Maroto, 2002) and Cau del Duc (Torroella de Montgrí) (Estévez, 1979). Some Middle Pleistocene proboscidean remains have also been recovered from the nearby municipality of Besalú, but their fragmentary conservation hinders a conclusive taxonomic attribution (Galobart et al., 1996).

Until recently, Palaeoloxodon antiquus was thought to be widely distributed within the Middle and Late Pleistocene interglacial periods in Europe because it was adapted to a temperate climate and parkland or wooded environments, and that its extinction took place at the end of MIS-5 (Kurtén, 1968; Mol et al., 2007). Lately, this belief has been challenged by some recent research at sites in the Atlantic Iberia and the Netherlands, where some remains indicate that this proboscidean persisted until MIS-3 (Mol et al., 2007; Stuart, 2005). Nevertheless, evidence of its existence in Italy (Grotta Guattari) after MIS-5 is weak.

It now seems clear that Mammuthus primigenius evolved in the Eurasian steppe from its ancestor Mammuthus trogontherii (Lister, 1996). The woolly mammoth has been associated with a cold, largely treeless environment, especially in the Mediterranean fringe of the Iberian Peninsula, where it has been related to the open landscapes of the last glaciation. However, its remains have sometimes been found in temperate, partly wooded conditions. In fact, in contrast to the Western-Central European assemblages, the Iberian Mammuthus primigenius assemblages show a different pattern because it involves a mixture of temperate and cold faunas (Álvarez-Lao & García, 2012).

Most of the woolly mammoth fossils of the Iberian Peninsula come from MIS-3 and MIS-2 contexts (Álvarez-Lao & García, 2011) located in the north (Cantabrian and Catalan areas). However, some of them have also been found in the central regions of Iberia (the Tagus estuary and the province of Madrid) and one of them was even found in the south (Padul Cave) (Álvarez-Lao et al., 2009; Álvarez-Lao & García, 2012). The discoveries of scarce Mammuthus trogontherii remains from the Middle Pleistocene of the Iberian Peninsula have been reported for the Guadix-Baza basins as well as for other sites in Cantabria, Toledo and Teruel (Ros-Montoya, 2010). Excavations in EDAR Culebro 1 (Madrid) have revealed the presence of mammoths in the Middle to Upper Pleistocene transition in central Spain (Manzano et al., 2011; Yravedra et al., 2014). The EDAR Culebro 1 mammoth has been classified as Mammuthus primigenius by Álvarez-Lao & García (2012). These authors agree with the primitive features that mammoth remains from the Aldehuela, Arriaga and Butarque sites display (Jarama and Manzanares basins) (Álvarez-Lao & García, 2010; Sesé & Soto, 2002). Although there seems to be a continuous record, the most reliable scenario for a lineage that had evolved on the Eurasian steppe is that it would have dispersed into the Iberian Peninsula during cold episodes.

It is difficult to predict changes to the palaeobiogeographic limits of the Mammuthus genus during the Upper Pleistocene, and its record does not seem continuous. Some authors claim that its anecdotal presence in the Iberian Peninsula can be only explained by migratory waves, which most recently are those of MIS-3 (between 43–31 cal ka BP, including Heinrich Event 4) and MIS-2 (between 28–18 cal ka BP, including Heinrich Event 2 and the Last Glacial Maximum) (Álvarez-Lao & García, 2010; Álvarez-Lao & García, 2011).

The presence of Mammuthus primigenius is well documented in the Cantabrian passage, a geographical setting containing other typical taxa of the Mammuthus-Coelodonta faunal complex: woolly rhinoceros (Coelodonta antiquitatis), reindeer (Rangifer tarandus), saiga antelope (Saiga tatarica), wolverine (Gulo gulo) and arctic fox (Alopex lagopus) (Álvarez-Lao & García, 2010).

In Catalonia, the coexistence of woolly mammoths and woolly rhinos in MIS-3 has been well documented in the oryctocenosis of Riera dels Canyars (Gavà) (Daura et al., 2013) and Unit III of Teixoneres Cave (Moià) (Álvarez-Lao et al., 2017). Nonetheless, this association could already be present in MIS-4 of Riera de Sant Llorenç (Viladecans) (Daura et al., 2013). Another remain that could belong to a Mammuthus primigenius is a humeral diaphysis recovered from level E of Abric Romaní (Capellades) (Rosell et al., 2012).

In the NE of Catalonia, Mammuthus primigenius remains are very scarce. The first of them were discovered by Francesc Xavier de Bolòs in the Clot del Llop site (Sant Andreu de Socarrats, Garrotxa), in the first half of the 19^thcentury, for which we lack any chronological reference (Alsius, 1915). Nonetheless, during archaeological excavations in Cau de les Goges (Sant Julià de Ramis, Gironès) carried out between the end of the 19^th and the beginning of the 20^th century, a dental remain was recovered in a Solutrean context (Cabrera, 1919; Soler, 1997). In Arbreda Cave, the record of this proboscidean is dispersed in the stratigraphic record. An almost complete molar plate was recovered during excavations carried out by Josep Maria Corominas in 1972–73, in the Mousterian levels (layer 30 of Corominas, close to 6.00 metres deep). Estévez (1979) classified some pieces from the same Corominas excavation as proboscidean, scattered throughout the Solutrean, Gravettian and Mousterian levels. Although these remains have not been reviewed, recent archaeological excavations in Beta Sector have confirmed ivory fragments in the Solutrean culture levels. Previously, Maroto et al. (1996) had studied an ivory fragment from the Archaic Aurignacian (level H), now dated close to Heinrich event 4 (following the chronology of Sánchez Goñi & Harrison, 2010), but they did not exclude the possibility that it arrived there through the exchange of raw material with northern hunter-gatherer groups. In the following sections we will present a new mammoth calf premolar from Arbreda Cave, discovered in a Mousterian context (level J) and allowing us to complete the knowledge of this slippery taxon in the NE of the Iberian Peninsula.

Arbreda Cave is situated at 42º09’04’’N and 2º44’45’’E at an altitude of 206 m above sea level. This archaeological site is located very close to the village of Serinyà (Girona province, Catalonia, Spain) (Figure 1) in a Palaeolithic complex known as Reclau Caves, where more than fourteen cavities, including large caves and little hollows, have been discovered since the 1940s. In terms of dimensions and archaeopaleontological record, the most important caves are Arbreda, Mollet, Mollet III, Pau and Reclau Viver (Soler, 1999).

The presence of these rock shelters is connected to the geological framework of the Banyoles-Besalú Basin (Julià, 1980), which is a morphodynamic hydrogeological unit that has been active in lake formation from the Lower Pliocene to the Holocene thanks to the existence of a lower Lutetian carbonate and gypsum substrate (Pallí, 1972). These karstified water-bearing rocks contain confined aquifers connecting the places where water enters, the pre-Pyrenean mountains of la Garrotxa, to the areas where it emerges, mostly in the Pla de l’Estany region (Sanz, 1985). The origin of these karst lakes is the collapse doline phenomenon, which creates funnel-shaped depressions (Roqué et al., 1999). The lakes’ carbonate sedimentation can occur in a great variety of facies, from carbonate muds to the biogenic littoral stromatolites (Brusi et al., 1996; Brusi et al., 2005).

The Reclau Caves are located at the conjunction of the Usall platform and the Serinyadell River valley, forming a terrace 200 m long, 50 m wide and with a slope of about 10 m. The Usall platform is a structural plain of lacustrine origin formed during the last part of the Lower Pleistocene (Julià, 1980) (Figure 1). Water from Usall’s springs flowed to the Serinyà area, where waterfalls built a fluvial travertine terrace between MIS-7 and MIS-6 (c. 215,000 and c. 134,000 BP) (Maroto, 2014). New U-Th analysis carried out in the tufa falls of Arbreda Cave have pushed back the oldest known datings of this geologic member, close to c. 254,000 BP (MIS-8). The topography at the Reclau site favoured the creation of cavities and rock shelters whose height may have exceeded 12 metres and whose length was about 18 metres, as in Arbreda Cave (Soler & Soler, 2016). Until now, the caves have undergone several transformations that have changed their appearance in comparison with the original morphology. On the one hand, karstification processes, secondary travertine deposits and recrystallizations have helped to remodel the geomorphology of the place. On the other hand, roof collapses and detrital sediment inputs have filled the ancient empty spaces (Brusi et al., 2005).

Josep Maria Corominas first became involved in the archaeological works of Arbreda in collaboration with Josep Canal, José M. de Bedoya, Miquel Oliva and Pere Comas. In 1972, he started a 9-metre deep survey (called Alpha Sector) that revealed a long prehistoric stratigraphy. Since 1975, a team led by Narcís Soler has been following up on previous works by applying Laplace’s method and enlarging the surface excavated by Corominas.

Fieldwork was carried out using standard archaeological methods. The surface of the site was divided into one-metre squares and finds were plotted three dimensionally. Identified bones, fragments of unidentified bones (>3 cm), stone tools and other lithic fragments (>1 cm) were mapped in situ prior to removal. Sediments were washed and sieved using 5, 1 and 0.5 mm mesh screens.

Arbreda Cave is divided into four sectors as a result of the history of work at the site: Alfa Sector, Beta Sector, Beta East Sector and Gamma Sector. Beta Sector has been excavated in two phases, from 1975 to 1987 and from 1996 until the present time. It has a surface of approximately 20 m² and a long archaeo-stratigraphic record, from the Holocene to the Mousterian. The documentation taken in the field work is introduced in a data base and a GIS is used to create virtual models in 2D or 3D that help track the archaeological levels.

The elephantid remain analysed in this study was found previously by researchers of Arbreda Cave team in 2011 in square C5 (Beta Sector) nearly 6.75 metres deep.

Terminology of the morphological description follows Maglio (1973). Upper premolars are indicated by upper-case letters, lower ones by lower-case. The single most anterior and posterior plates, which are not fully attached to the root (talons/talonids), are indicated as ‘x’. Laws’ nomenclature for elephantid teeth and his age classes are also used (Laws, 1966).

To provide a quantitative and biometric description, we followed criteria detailed in Maglio (1973) and Lister (1996). Length has been taken parallel to the occlusal surface (A of Maglio). Width has been taken in plate 5 and height in plate 6. Because of the curvature of the dental piece, lamellar frequency is the average between those values taken from lingual and buccal sides. A sliding calliper in millimetres was used to take all measurements.

Systematic palaeontology and morphometric description

Specimen C5 EC135 2302 is identified as a left lower dp3, which is a M₂ in the Laws system. This proboscidean tooth is almost complete, even though it shows only the crown plates and interplate cement. Although all plates are already fused, which means that the tooth was fully grown, it preserves only a thin layer of primary dentine beneath the enamel-dentine junction. All plates are eroded on their lateral sides (lingual and buccal) because no enamel is preserved at the lateral edges of the plates and erosion has exposed the dentine. All interplate gaps were filled with cement with the sole exception of the last gap, where the tooth was slightly broken. The lack of any fractured surface eliminates the possibility that the root disappeared due to biostratinomic factors. So, it is clear that the line in the base of the tooth reflects the smooth base of the crown.

This piece possesses eight plates, but the first and the last plates can be considered talonids. The first mesial plate is the most worn out, even though it seems to have a less prominent length than the others because the mesial enamel of this plate has also disappeared. Although we can report its existence, the distal plate has been partially broken. In conclusion, we can describe these two as talonids and the plate formula is x-6-x.

Metrically, Arbreda’s tooth has reduced dimensions, which is demonstrated by its position in the lower range dimension values of length of M. primigenius dp3 specimens shown in Table 3. Moreover, it has the lowest width value of the sample, which is due to erosion suffered by the tooth. The result of the lamellar frequency calculation is close to other Mammuthus primigenius remains. Other metric values such as height, enamel thickness or hypsodonty index match well with typical Mammuthus primigenius values.

The degree of wear matches rather well with group III of Laws (1966). Generally, wear is not heavy and it is deeper in the first three mesial plates, shallower in the following two plates, and incipient in the sixth and seventh plates. We cannot know whether the distal talonid was worn or not because of the fragmentation. However, enamel examination tells us that the tooth had reached the full growth stage and the final fusion of all plates, which is said to take place in Laws’ group IV.

Deciduous Palaeoloxodon antiquus teeth are not abundant in the fossil record, but Guenther (1977) presents dP3 measurements for this species, which is larger than Mammuthus primigenius. Palaeoloxodon antiquus upper dP3 are larger than those of Mammuthus primigenius, which are larger than lower teeth of the same species. In the absence of published straight-tusked elephant dp3 samples, we expected that their dimensions would be higher than those M. primigenius dp3 in correspondence with the upper teeth of the former proboscidean. So, we defend that the Arbreda tooth, which can only be fitted with Mammuthus primigenius dp3, must belong to this species. Maschenko (2002) pointed to a lack of correlation between second generation molar dimensions and position in either the maxillar or the jaw. In line with studies by Haynes (1991), who said that maxillary teeth are wider, higher and larger than corresponding mandibular teeth, there is a trend, seen in our sample, where the lower values (< 50 mm) for lengths correspond to lower teeth (with the exception of Berelekh) and the higher values for widths to upper ones (Table 3).

Soler et al. (2012) declared that C5 EC135 2302 was a germ tooth because it lacked roots. We argue against this hypothesis because it is incoherent with the studies of the mammoth tooth ontogenesis. Classical studies (Haynes, 1991; Maschenko, 2002) and new technologies applied to mummified woolly mammoths (Rountrey et al., 2012) have shown that root genesis started before the cementum covering and the presence of wear. The possibility of a germ tooth must be rejected. Equally, the phase of wear and its corresponding age range are not consistent with a possible root resorption case, especially when the distal part of the occlusal surface had almost not been worn.

Level J has a singular geochemical footprint due to the increase of phosphorous levels below 6.3 meters dept (Kabiri, 1993). This enrichment is related to biogenic inputs like bat guano or the urine of larger mammals, such as cave bears or hyaenas. The phosphorous alterations have dissolved big travertines and affected faunal and lithic archaeological remains (Soler et al., 2010). Therefore, it is possible that a low pH level may have produced a differential conservation process and the root dentine could have been dissolved. For example, the presence of authigenic phosphates, like crandallite, seems to indicate lower values of pH in some moments of the lower Upper Pleistocene sequence, including level J (Kehl et al., 2014). This phenomenon has been observed in the aristocratic Gallic burial site in Clémency (Luxembourg), where only the enamel of the suid teeth have been conserved (Méniel, 1993). We argue that further taphonomic analysis must be done in this line of research.

The Arbreda tooth came from a young animal estimated to be between 1 and 2 years old. The tooth falls between groups III (because of the degree of wear) and IV (because of the full grown crown) of Laws (Metcalfe et al., 2010). Laws (1966) assigns age class III with an average age of about 1 year (African elephant years), at the same time that Craig correlates it with approximately 1.7 years (Haynes, 1991). Maschenko (2002) published evidence that suggested an eruption and wear of woolly mammoth teeth earlier than Loxodonta africana and Elephas maximus. This author claimed that dp3 are completely formed at between eleven to fourteen months and that all plates have experienced wear. So, we attempt to establish an age of approximately one to one year and a half. In addition, Arbreda seems to be close to ZIN 28284 (1) dp3 of Kostionki 14, which is 10–12 months old (Maschenko, 2002), or with MK 1027 dp3 of Krems-Wachtberg, which is 6–12 months old (Fladerer, 2003). We suggest that the age of C5 EC135 2302 is around one year old, perhaps one year to one and a half at the most. It is possible that in that stage, dp2 was still present in the dental series showing deep wear, at the same time that dp4 was erupting, but was not worn Haynes (1991) uses the term Stage B to describe an age range where dp2 and dp3 are both worn.

Although it is evident that this piece had been brought inside the cave by some biogenic agent (carnivore-scavenger), we cannot ascertain if it was a carnivore (hyaena-wolf) or a human (Neanderthal). Recent studies situate Ursus spelaeus at the herbivore level in the trophic net (Münzel et al., 2014). Despite the abundance of ursine remains, it is impossible to correlate this animal with carnivore-scavenger behaviours. Moreover, it has been demonstrated that brown bears (Ursus arctos) do not transport carcasses into caves and do not generate skeletal accumulations in their dens (Sala & Arsuaga, 2013). More archaeological interventions in this level are needed to solve the taphonomic questions. In any case, mammoths were present in the palaeoenvironment at the time of level J’s depositional context, given that Neanderthal humans had not developed the long-distance provision nets, like those that anatomically modern humans would develop in the Upper Palaeolithic (Ortega, 2002).

The increasing aridity, cooling and continentality during the late Middle Pleistocene and Late Pleistocene allowed the development of a new biome (the mammoth-steppe or tundra-steppe) which has no modern large-scale analogue. This biome expanded its geographical range in every glacial period and appeared for the first time in MIS-12, between 480 and 400 ka BP. The reduction of the taiga forest belt pushed the origin of Artic faunas further south and south-west in the same way that species of steppe origin dispersed into northern and western regions of the Palaearctic. The mixing of these faunas with distinct environmental origins created a characteristic assemblage called the Mammuthus-Coelodonta faunal complex. Together with the woolly mammoth and the woolly rhinoceros, the most representative genera were Ovibos, Rangifer, Saiga, Alopex, Bison and Equus (Kahlke, 2014; Guthrie, 1982). In contrast, faunal compositions in which herbivores of temperate character were the most representative were dominant during the Last Glacial Maximum in the Iberian Peninsula. In these assemblages, cold-adapted faunas occur in low frequencies, mainly in the regions of Cantabria and Catalonia. This demonstrates the transitional character that these areas had during MIS-3, between the Eurasian steppe biome and the Iberian temperate refuge (Álvarez-Lao & García, 2012; Álvarez-Lao et al., 2017).

Although level J’s chronology is not known precisely enough, this article confirms a very complex outlook for migration cycles of cold-adapted faunas in the Iberian Peninsula. As Álvarez-Lao & Garcia (2011) have argued, in the late Middle Pleistocene and Upper Pleistocene, cold-adapted faunas spread into the Iberian Peninsula by crossing over two points in the west and in the east of the Pyrenees. These authors established three wide-ranging, scattered episodes during which large cold-adapted mammals entered the Iberian Peninsula: between 200 and 100 ka BP and frequent occurrences between 42 and 31 cal ka BP (MIS-3) and between 25 and 18 cal ka BP (MIS-2). Arbreda cannot be situated within any of these ranges because level J has a chronology roughly between 70 ka BP and 44 ka BP. If the upper part of level I indicates an age of c. 45–41 cal ka BP (López-García et al., 2015), we must accept an older age for level J, with chronologies in early MIS-3 or MIS-4. Precisely, the oldest occurrence of the second Álvarez-Lao & Garcia episode is the dP2 of Unit III of Teixoneres Cave, which has a chronology from more than 51,000 ¹⁴C BP to 44,210 cal BP (Talamo et al., 2016). Evidence from Arbreda and Teixoneres pushes back the initial chronology of the second dispersal episode.

In level J of Arbreda, the dominance of red deer (Cervus elaphus) among herbivores can indicate a forested environment, even though it has been demonstrated that the red deer is a very adaptable eurythermic animal (Altuna, 1995). The auroch (Bos primigenius) prefers landscapes with sparse forests and forest steppes, wetlands and proximity to water courses (Auguste & Patou-Mathis, 1994). Although Equus sp. is able to live in varied conditions, its presence is mostly indicative of open meadows in proximity of the site (Arribas, 2004). Nowadays, beavers (Castor fiber) live in riverbanks and lake shores surrounded by forested areas (Cuenca-Bescós et al., 2017). So, the fauna association suggests a mosaic landscape with a high number of forested environments and a humid atmosphere around the cave, but also open landscapes not far from the site. Like many Iberian woolly mammoth assemblages, level J shows a dominance of temperate species and a low percentage of cold taxa, like M. primigenius. This assemblage does not reflect the typical composition of the Eurasian mammoth fauna and supports the idea of occasional arrivals of mammoths in the Iberian Peninsula.

The comparative morphometric study confirms that the C5 EC135 2302 molar belongs to a mammoth and not to a straight-tusked elephant. The archaeological and chronological context leaves few doubts that we are talking about a tooth of a woolly mammoth calf (Mammuthus primigenius).

There is evidence of this species in the Iberian Peninsula from the late Middle to early Late Pleistocene; however, most of the occurrences took place in MIS-3 and MIS-2. The Arbreda tooth demonstrates a complex trend of mammoth distribution on the south side of the Pyrenees in the Late Pleistocene. The stratigraphic position, uranium-thorium datings at the base of the known stratigraphic column and the chronology of the upper part of level I suggest an occurrence out of the three-episode proposal published by Álvarez-Lao & García (2011) and close to the Teixoneres Unit III occurrence (Álvarez-Lao et al., 2017). Thus, the second dispersal episode of woolly mammoths in the Iberian Peninsula could have started earlier than previously thought, at least in the early MIS-3.

The palaeoenvironmental context is not a typical cold-adapted fauna association like those from Western Central Europe, dominated by reindeer. It is similar to the Iberian assemblages that were developed in a mosaic landscape where the proximity of different ecosystems offered the possibility of cohabitation in a reduced zone of fauna species that had different habitat preferences.