The Abric Romaní archaeological site is a rock-shelter located on the northeastern side of the Cinglera del Capelló cliff in the town of Capellades, 45 km northwest of Barcelona, Spain. Its coordinates are 41° 32′ N and 1° 41′ E, its altitude is 280 m asl, and the cliff wall faces north-northeast. Its stratigraphy is made up of 20 m of well-stratified travertine sediments. Rock fragmentation and alluvial and biochemical sedimentary processes generated boundstones, gravels, calcarenites, and calcilutites interbedded with very fine archaeological levels. U-series and radiocarbon dates put the Abric Romaní chronostratigraphy between 70 and 40 kyr. The sedimentation rate is estimated at around 0.6 m/kyr (Bischoff et al., 1988). Discovered in 1909 by Amador Romaní, the site was excavated at different periods throughout the 20th century. Ongoing excavations started in 1983 under the direction of Dr Eudald Carbonell.

Except for level A, all of the archaeological units in the sequence correspond to the Middle Palaeolithic (Fig. 1). The upper levels of the sequence (A–E) were largely excavated during the previous excavations, and levels E–I were also affected by these digs, although to a lesser extent.

The different levels of Abric Romaní are the result of the accumulation of an unknown number of occupational events. The occupation models that make up the archaeological record can be distinguished as long-term occupation events or short occupation and/or non-residential events (Vallverdú et al., 2005, Vallverdú et al., 2010 and Vaquero et al., 2012a). In both models, the hearths were reused, especially in long-term events, which resulted in the preservation of these structures (Vallverdú et al., 2012 and Vallverdú et al., 2010). The main characteristic that differentiates these two occupation models in the record is that in the short-term or non-residential events (Table 1), areas where specific activities were performed (conservation areas for animal biomass, knapping areas, butchering areas) can be isolated and differentiated (Vallverdú et al., 2005). Meanwhile, levels characterized by long-term events exhibit overlapping of different areas and places where activities occurred and these cannot be clearly differentiated from one another (Vaquero et al., 2012b).

Therefore, the representation and superposition of occupancy models may involve the mixing of materials corresponding to different types of strategies. This effect is also observed in the lithic assemblage of Abric Romaní (Vaquero, 2008). The predominant structural category is that of flakes. Cores and retouched flakes are not abundant in the assemblages from the different levels (Chacón and Fernández-Laso, 2007, Martínez and Rando, 2000, Vaquero, 2008, Vaquero et al., 2004 and Vaquero et al., 2001). The predominant raw material in the assemblage is flint, compared to other materials such as quartz or limestone. Other raw materials, like slate, sandstone, quartzite and, to a lesser extent, porphyry are also present (Table 1). Most raw materials are found in nearby sheltered environments, such as the terraces of the Anoia River and the Palaeozoic formations of the Capellades strait. Flint, however, is found in various formations located within a radius of about 28 km (Table 1) (Gómez de Soler, 2007 and Morant and García-Antón, 2000). Short-duration or non-residential occupation events have been linked to the transport and abandonment of most of the retouched artefacts, notches and denticulates, whereas the exploitation of cores oriented towards the production of small flakes has been associated with long-term occupation events. The grounds for these conclusions lie in the fact that flint is the preferential raw material at the site, and it comes from reservoir outcrops not found in the immediate vicinity. Therefore, the abandonment of retouched elements has been related to less economical occupations and the exploitation of cores to occupations that tend to maximize resources (Vallverdú et al., 2005 and Vaquero et al., 2012a).

Since the 1990s, considerable importance has been placed on the zooarchaeological and taphonomic studies of the faunal remains recovered in the Abric Romaní sequence, many of them part of multidisciplinary research projects (e.g. Aïmene, 1997, Aïmene, 1998, Aïmene et al., 1996, Cáceres, 1998, Cáceres et al., 2012, Cáceres et al., 1998, Carbonell, 2002, Chacón and Fernández-Laso, 2007, Chacón et al., 2007, Chacón et al., 2010, Gabucio et al., 2013, Gabucio et al., 2014, Rosell et al., 2012a, Rosell et al., 2012b, Vallverdú et al., 2005 and Vallverdú et al., 2010). Faunal remains are very abundant in all of the levels of the Abric Romaní stratigraphic sequence. A total of 13 different taxa have been identified, although deer (Cervus elaphus) and horses (Equus ferus) are the most abundant animals (Table 2). Auroch remains (Bos primigenius) are also common at the current bottom of the sequence and chamois (Rupicapra pyrenaica) in the upper levels. Rhinoceros (Stephanorhinus sp. hemitoechus) remains have been documented in several levels, although they are scarce. In level E, a proboscidean femur shaft has been recovered. Although large herbivores predominate in all of the levels, some carnivore remains have also been recovered (Table 2) (Supplementary material).

Combustion structures are also common in all of the levels excavated at Abric Romaní to date. These structures are circular or oval in appearance and are distinct from the travertine substrate. They are characterized by a colour change, by the stratification of rubefied and carbonaceous facies and by sediment compaction.

The hearth areas also contain combustion waste products, such as ashes and coals, along with other remainders of anthropic activity, such as bone and stone remains (Vallverdú, 2002, Vallverdú and Courty, 2012 and Vallverdú et al., 2012). Abric Romaní has yielded fossil wood remains recovered as empty cavities in the shape and with the negative mould of the original wood surface. These remains are empty travertine imprints of completely decayed wood (Carbonell, 2002 and Castro-Curel and Carbonell, 1995). The wood totally decays leaving only an imprint in the travertine when the water process stops (Castro-Curel and Carbonell, 1995). The study of these imprints has made it possible to distinguish between firewood and worked wood (Allué et al., 2012 and Solé et al., 2014).

As mentioned earlier, the archaeological remains at the site are the result of different accumulation events. In the case of the faunal remains, each of the different transport events that generated the final archaeological assemblage cannot be isolated (Monahan, 1998 and Saladié et al., 2011). For this reason, we have used the total number of ungulate remains for each species/age classes and size-weight categories in each level with the goal of identifying the possible transport strategy that generated the entire assemblage. The minimal number of elements (MNE) and the minimal animal units (MAU) were calculated in all five assemblages considering the bone landmarks in conjunction with a repetition of regions and sides, and the anatomical position (left/right) of the bones, as described by Saladié et al. (2011). As indicated in Faith and Gordon (2007), the MNE was calculated taking into account the shaft fragments in the assemblages. We only considered the high-survival elements and did not use the low-survival elements (Marean and Clerghorn, 2003). This is because the high-survival elements are more resistant to the different attritional processes and more accurately reflect the original abundance of elements resulting from hominin transport/discard decisions. We test the correlations between the %MAU of high-survival elements of large and medium-sized animals and their bone mineral density by portions (Lam et al., 1999 and Lyman, 1994), in order to consider the possible intervention of destructive taphonomic process as well as for these elements where densities are also variable. To compare the distribution of the elements (MNE) into the size-weight groups (large and medium-sized animals) (Table 5) of each set, we applied a Chi² test. The correlation analysis between the %MAU and the Standard Food Utility Index (SFUI) (Metcalfe and Jones, 1988) was calculated by means of Spearman's rank-order correlation coefficient (r _s). In keeping with Faith and Gordon (2007), we used Spearman's r _s because the correlation is dependent on rank-order values and the MAU and SFUI should be treated as ordinal-scale measures. In addition, we tested the MAU with the Unsaturated Marrow Index (UMI) (Morin, 2007), with the purpose of identifying a possible preference in the transport of bones as a function of their marrow and grease content. As with the SFUI, we use Spearman's r _s to test the statistical correlation.

Our assessment of the correlation between MAU and SFUI results takes into account the possibility of type I and type II statistical errors due to the small size of most of the assemblages (Faith and Gordon, 2007). Type I errors are those in which significant correlations are found in an assemblage when no such correlation exists in the population assemblage, and type II errors are those in which no correlation is found in an assemblage despite the presence of a correlation in the population assemblage. Faith and Gordon (2007) noted that type II errors are very common in assemblages with an MNE lower than 100 and resulting from an unconstrained or bulk strategy.

Finally, and again in keeping with the proposals of Faith and Gordon (2007), we applied the Shannon Diversity Index (evenness), where evenness is equal to E = Σ − (Pi × i ln P)/ln S. S is the number of types of elements (e.g. MNE of humeri, femora, skulls) and pi is the standardized proportion of specimens of the i-th element. Finally, we compared the Abric Romaní assemblages using information from the faunal assemblages of the Hadza campsites of Mugugu and Tispitibe, published by Lupo (2001). The Hadza are a hunter-gatherer group that has been intensely studied with regard to different social and economic aspects, and especially concerning their animal carcass discard and transport strategies (Bunn, 1993, Bunn et al., 1988, Bunn et al., 1991, Monahan, 1998, O’Connell, 1993, O’Connell et al., 1988a, O’Connell et al., 1988b and O’Connell et al., 1990). At these two sites, the faunal remains were collected from the surface and subsurface at a depth of 10 cm, the MNE was calculated using all identified parts (including the diagnostic limb bone shaft), and the assemblages were formed through the overlapping of events/occupations (Lupo, 2001).

According to the values of the correlation between the presence/absence of elements and their economic value given by Faith and Gordon (2007), two groups can be distinguished in the levels examined in this paper: •

a group composed of sets where r _s is within the range of 95% CI values obtained by Faith and Gordon (2007) for an unconstrained strategy. This group contains the assemblages of faunal remains from large animals from levels L and M (Fig. 4), and the remains the medium-sized animals of levels Jb, K, L (Fig. 4);

On the other hand, the evenness value for all of the sets of remains reveals a different distribution, clearly corresponding to two transport strategies: the bulk strategy for the large animals of level Ja and the unconstrained strategy for the medium-sized animals of level M (Fig. 5). For the large animals of level Jb, the evenness value is within the values of the bulk and unbiased strategies (Fig. 5). For the large animals of levels K, L and M (Fig. 5) and the medium-sized animals of levels Ja, Jb, K and L (Fig. 5), the evenness values fall within the range of 95% CI values for the bulk and unconstrained strategies.

Based on these results, there are two possible statistical type II errors (in which a statistically significant correlation between the %MAU-SFUI was not observed although the set was formed attending to the maximization of the utility elements transported): in the data for the large animals of levels Ja and Jb. In these cases, due to the size of the MNE, there might not have been a statistically significant correlation between the %MAU and SFUI because the evenness value indicates a bulk strategy for large animals in level Ja and a bulk and unbiased strategy for large animals in level Jb.

Faith and Gordon (2007) indicate that when MNE size decreases, especially for small assemblages (MNE 50), the evenness value of the bulk and unconstrained strategies overlaps. Therefore, at levels where there are no identified type II errors, the lack of a significant correlation between the %MAU and SFUI suggests an unconstrained transport strategy. According to these data, the predominant strategy among the Abric Romaní levels analysed here for medium-sized and large animals would be unconstrained. This would suggest that the body parts of the prey were transported from the kill/butchering site to the campsite according to their abundance in a complete skeleton (Faith and Gordon, 2007). In other words, these results suggest the transport of complete carcasses to the shelter, regardless of the size of the animal, meaning similar transport patterns were employed for deer, horses and aurochs. However, comparisons (Chi² test) between the skeletal profiles of the large and medium-sized animals indicate that there are not similarities in the anatomical distributions in all of the levels studied, which points to a possible differential treatment of carcasses based on animal weight.

The analysis between bone mineral density and %MAU indicate two results. On levels Ja, Jb, K and L has not been observed differential conservation based on the mineral content of bones (Lam et al., 1999 and Lyman, 1994). In contrast, in level M, we found a high statistically significant correlation between bone mineral density by portions and the %MAU of high-survival elements for large and medium size. This indicates that parts of bones with low mineral density are suffering a taphonomic process that is creating an underrepresentation of these anatomical parts. Taken into account the low alterations by carnivores and the high presence of anthropic activities, among which stand cremation and fire used (Costamagno and Rigaud, 2014, Gabucio et al., 2013, Gabucio et al., 2014 and Vallverdú et al., 2012), this underrepresentation could be a combination between hominin activities.

In previous studies of these levels, some authors suggested the existence of the differential transport of animals based on size-weight (Chacón et al., 2007, Chacón et al., 2010, Fernández-Laso et al., 2010, Rosell et al., 2012a and Rosell et al., 2012b). It was argued that large animals (horses and aurochs) would have been processed at the kill/butchering site with the intention of facilitating their transport to the shelter, while body parts, which were energetically not worth carrying were left behind due to their low economic value. In contrast, medium-sized animals (deer) would have been transported intact and completely exploited at the campsite (Carbonell, 2002, Carbonell, 2012, Chacón and Fernández-Laso, 2007, Fernández-Laso et al., 2010, Fernández-Laso et al., 2011, Rosell, 2001, Rosell et al., 2012a, Rosell et al., 2012b and Vallverdú et al., 2005). According to these studies, differential transport decisions were linked to the animal's weight, which is not contrary to our results. However, it was argued that this pattern is visible through a greater or lesser presence of the postcranial axial skeleton. But, having evaluated the anatomical distribution (%MAU) in the size-weight categories of the different levels (Fig. 2), it became clear that the differences proposed due to the presence of the postcranial axial skeleton are not supported. The size-weight categories of all of the levels exhibit a very similar concentration of vertebrae and ribs. In levels K, L and M, the presence of these bones for large and medium-sized animals is virtually identical. In levels Ja and Jb, although there are more vertebrae and ribs among the medium-sized animals, their presence is still scarce and they never exceed 10% of the MAU.

Valensi and Psathi (2004) note that in the Middle Palaeolithic levels of the sites of Lazaret (France), Arma delle Maine (Italy), Madonna dell’Arma (Italy), Caverna delle Fate (Italy), St. Lucia Superiore (Italy) and San Francesco (Italy) the pattern of anatomical distribution is characterized by the abundance of cranial and appendicular elements and by the scarcity of the postcranial axial skeleton, regardless of the size-weight of the animals. They indicate that this pattern suggests a primary access and a transport of complete or nearly complete carcasses into the sites (Valensi and Psathi, 2004). The same anatomical distribution was observed by Moncel et al. (2004) for medium-sized animals at the Grotte de Saint-Marcel. They said that the absence of part of the axial elements, like ribs, could be the result of a differential transport. However, the absence of the epiphyses of long bones also indicates that these underrepresentations could be result of an intensive treatment of these anatomical parts by boiling or use as fuel. Moncel et al. (2004) hypothesize that the assemblage is the result of differential transport of certain anatomical parts, but the presence, even in very small quantities, of all parts of the skeleton shows that some hunted animals were carried complete to the cave. On the other hand, Romandini et al. (2014) indicate that Grotta di Fumane was used as a habitat where the processing of carcasses was finalized after having been begun at the kill site. They infer that human groups transported to the site selected anatomical parts with high nutritional value. Gaudzinski and Niven (2009) evaluate the open-air site of Salzgitter-Lebenstedt (Germany) specializing in the hunt of Rangifer tarandus. It was found that the axial skeleton is always underrepresented in this assemblage, and never exceeds 20% of the MAU. Gaudzinski and Niven (2009) define the presence of long bones as result of economic decisions made by hominins.

Given this distribution, just as we observed in our study of the faunal remains of levels J, K, L and M at Abric Romaní, it seems that the trunk would be the most commonly scarce segment in all cases of transport. However, the absence of vertebrae and ribs, together with other low-survival elements may be due in part to multiple attritional processes (Lyman, 1994 and Marean and Spencer, 1991), crushing of axial bones in order to collect the fat with boiling practices, and even intentional cremation during maintenance work on hearths or the cleaning of the living floor (Costamagno and Rigaud, 2014, Gabucio et al., 2014, Thery-Parisot and Costamagno, 2005, Vallverdú et al., 2005 and Vallverdú et al., 2012). On the other hand, the absence of bones of the trunk is not universal in these assemblages, so their presence, although scarce, indicates that at some point some carcasses arrived the shelter intact, as for example Saladié et al. (2011) indicate for level TD-6 of Gran Dolina, Domínguez-Rodrigo et al. (2009) for FLK-Zinjanthropus 22 site or Moncel et al. (2004) for the Mousterian site of Grotte de Saint-Marcel. Therefore, the sum of attrition generated by Neanderthals and other postdepositional processes has resulted in a high degree of uniformity in the skeletal profiles of the Abric Romaní levels, characterized by the scarcity of the axial skeleton as in many Neanderthal’ sites, for example Grotte de Saint-Marcel (Moncel et al., 2004) or Lazaret (Valensi and Psathi, 2004).

In keeping with these proposals, the distribution of high-survival bones has to be studied to determine whether the carcasses were carried to the shelter complete, as suggested by the combination of non-correlations of %MAU and SFUI and the evenness values, relying on the proposals of Faith and Gordon (2007), or if the carcasses of animals of different weight groups were subjected to different transport strategies, as indicated by the comparison of their anatomical distribution using the Chi² test. Thus, considering the different frequencies of the %MAU, it becomes clear that the primary differences among the size-weight categories lie in the abundance of the appendicular bones. For example, in levels Jb, L, K and M, although the frequency of large and medium-sized animal skulls is similar, there are fewer long bones of large animals than of medium-sized animals (Fig. 2) (Supplementary material). This would support the hypothesis of reduced transport of the long bones of large animals from the kill/butchering site to the campsite in these levels. In level Ja, there seem to be fewer differences in this regard, particularly in terms of the frequency of the proximal appendicular bones (femora and humeri) and intermediate appendicular bones (radii-ulnae and tibiae). However, the representation of metapodials does seem to have a relationship to size-weight, as those from medium-sized animals are much more abundant than those from large animals, especially in level K (Fig. 2) (Supplementary material).

Based on these observations, it seems that the anatomical representation of the animals documented in Abric Romaní may be related to the sum of different carcass transport patterns, with animal size being one of the determining factors. Some researchers, such as Monahan (1998) and Lupo (2001), have argued that the superposition of different transport strategies makes them less visible and more difficult to relate to concrete transportation/abandonment decisions. In fact, Lupo (2001) suggests that transport/abandonment decisions are more accurately reflected at a single-event kill/butchering site than in camps where overlapping masks possible differences between one event and another.

It seems that the same superposition would characterize the skeletal profiles of the most abundant animals at Abric Romaní: deer, horses and aurochs. However, the remains of a rhino in level M (one femur diaphysis) may reflect the transport of only a few bones of a carcass.

This variability of different transport models used by Neanderthals is also observed in some hunter-gatherer groups (Binford, 1978, Binford, 1981, Bunn, 1993, Bunn and Kroll, 1986, Bunn and Kroll, 1988, Bunn et al., 1988, Monahan, 1998 and Schoville and Otárola-Castillo, 2014). Therefore, different specific conditions may affect each hunting event differently. The studies conducted by Gifford-Gonzalez (1993) and Monahan (1998) indicate that some of the variables that affect the transport decisions of these groups are the weight of the animals, the number of people involved in the hunt, the number of carriers and the possibility of calling more people to help, the number of hunted animals, the anatomy of the animal itself, the cost of processing and transporting a carcass, competition with other carnivores in the environment, and the distance to the camp. Ethnographic studies of the Hadza indicate that there are even more variables in play, such as the hunger of the hunters and the toolkit they are carrying (Bunn, 1993, Bunn et al., 1988, Bunn et al., 1991, Monahan, 1998, O’Connell, 1993, O’Connell et al., 1988a, O’Connell et al., 1988b, O’Connell et al., 1990 and Oliver, 1993). More recently, Schoville and Otárola-Castillo (2014) asserted that, for the Hadza, the most important factors in deciding which body parts are transported are the number of carriers, the size of the animal and the distance from the camp.

Based on this information, we know that the Hadza use a variety of different transport strategies for all animal size-weights (Bunn, 1993, Bunn et al., 1988, Bunn et al., 1991, Lupo, 2001, Monahan, 1998, O’Connell, 1993, O’Connell et al., 1988a, O’Connell et al., 1988b and O’Connell et al., 1990). This behaviour results in bone accumulations created by multiple transport events (Lupo, 2001). Hadza in these two camps (Lupo, 2001) and in general (Bunn, 1993, Bunn et al., 1988, Bunn et al., 1991, Monahan, 1998, O’Connell, 1993, O’Connell et al., 1988a, O’Connell et al., 1988b and O’Connell et al., 1990) employed highly variable transport strategies for the animal carcasses brought back to camp, including the transport of complete carcasses or only a few elements. The result of the analyses of Mugugu and Tispitibe camps indicates an unconstrained transport strategy (Table 7). For this reason, the Hadza can be used to determine how a highly variable animal carcass transport pattern is reflected in the faunal record like a palimpsest result of an unconstrained transport strategy (Table 7).

The presence of most of the parts of the skeleton in all of the analysed sets suggests that the trunks, particularly of medium-sized and large animals, were wholly or partially transported to the shelter at some time. This may be consistent with the transport of complete carcasses and therefore with an unconstrained transport strategy. The result of the analyses between the %MAU and UMI, indicates that there exists a correlation between the transport of bones and their riches in marrow and grease (Morin, 2007), except for the large size of level K and L and medium size of level Jb, which may be due to a small size of the sample. On the assemblages with high significant statistical correlations could be supported the hypotheses of a preferential transport of bones in function of the marrow riches, that is, long bones. At the same time, differences in the anatomical representation indicate that this model was not the only one used, or not necessarily the most common one used, since the skeletal profiles analysed are different by size-weights.

Carnivory and meat-sharing are central features of current hunter-gatherers, so these behaviours are critical in the evolution of hominin social relations (Stiner, 2009). The stable isotope analyses of the Neanderthal remains of Vindija, Jonzac and Marillac, among others (Fizet et al., 1995 and Richards et al., 2000), together with a great number of zooarchaeological researches (Daujeard and Moncel, 2010, Gaudzinski and Niven, 2009, Moncel et al., 2004, Patou-Mathis, 2000, Rendu, 2010, Rendu et al., 2012 and Romandini et al., 2014), indicate that Neanderthals’ primary food source consisted of large herbivores (Bocherens, 2011, Bocherens et al., 1999, Richards et al., 2000 and Richards et al., 2008). According to Hayden (1981), cooperation in hunting and meat-sharing is closely related to the importance of this resource to a particular group. As shown by Stiner (2009) for Qesem cave, Middle Palaeolithic hominins delayed the consumption of their prey long enough to transport the animal carcasses back to camp.

Back at home base, the Neanderthal group must have shared. It is assumed that this circumstance would have been the same in levels Ja, Jb, K, L and M of Abric Romaní. The transport of numerous animal parts necessarily entailed cooperation between different individuals. The number of carriers was possibly variable, and this conditioned the transport strategies chosen. However, although many other factors could have played a key role in these decisions, they unfortunately do not have archaeological visibility.

The Neanderthal groups who occupied Abric Romaní during the formation of levels Ja, Jb, K, L and M developed variable transport strategies for medium-sized and large animals. The end could result in transport processes is a mixed skeletal profile, in which some of these strategies have low archaeological visibility. However, certain guidelines can be established. It seems that the postcranial axial skeleton was transported only in a few events and once inside the Abric Romaí was highly sensitive to destruction by taphonomic processes, among those are the produced by the Neanderthals. However, the decision to transport or discard at the kill/butchering site would not have been made in relation to the size-weight difference between large and medium-sized animals. The long bones and the skull bones were commonly transported, but in this case, there are differences in the distribution linked to the size-weight category, so the size of the animal itself would have been a factor of these different anatomical profiles. Therefore, evidence of an unconstrained strategy is possibly an artefact caused by the combination of different types of transport, such as in the levels of Abric Romaní. One of these transport types is the transport of the bones richest in marrow and grease, as the high statistical correlations of MAU and UMI indicate. These conclusions can be extrapolated to other Middle Palaeolithic assemblages with similar skeletal profiles, in which human activity is the main cause of accumulation and subsequent attritional processes, are scarce. It would be desirable to extend this study with larger samples in order to test our conclusions.