The Abric Romaní rock-shelter (41°32′ latitude N, 1°41′30′′ longitude E) is located near the town of Capellades (Barcelona province, Spain) in the travertine cliff called Cinglera del Capelló at 280 m ASL (Fig. 1). The karst system is rich in natural shelters (e.g., Balma dels Pinyons, Balma de la Costa de Can Manel and Abric Agut), which have yielded archaeological finds dating from the Middle Paleolithic up to the Mesolithic (Vaquero et al., 2013). The Abric Romaní archaeological site was discovered in 1909 by Amador Romaní I Guerra (1873–1930), who carried out several field campaigns between 1909 and 1930. The excavations affected the upper part of the stratigraphy up to level J and a test well, known as Romaní well, was dug at the base of Pit 2 providing evidence of the site's complete stratigraphic sequence (Bartrolí et al., 1995). A second phase of investigations was carried out between 1956 and 1962 under the direction of Dr. Ripoll Perelló, director of the Archaeological Museum of Barcelona, but the information about these fieldworks are scarce (Bartrolí et al., 1995). The present archaeological research into the site began in 1983 under the direction of the senior author.

The stratigraphic sequence consists of 15 archaeological levels and has been dated by U-series and radiocarbon methods as between 40 and 70 ka BP (Bischoff et al., 1988 and Vallverdú-Poch et al., 2012) (Fig. 2). Below the Proto-Aurignacian level, the remaining anthropogenic occupations are attributed to the Middle Paleolithic. Interdisciplinary works point out the recurrence of Neanderthal settlements, both short-term (Carbonell, 1992, Vallverdú et al., 2005 and Vallverdú et al., 2010) and long-term (Carbonell i Roura, 2012 and Fernández-Laso et al., 2011), with abundant evidences of combustion structures, exploitation (skinning, bone breakages, bone tool making) and consumption of animal remains, and lithic production.

Paleo-environmental studies at the site document the rapid and abrupt climatic changes of the Dansgaard–Oeschger (D–O) cycles and Heinrich events from the start of MIS 3 to the Hengelo interstadial or D–O 12 (Burjachs and Julià, 1994 and Burjachs et al., 2012). In the chronological period between 57 and 50 ka BP, the climate reveals short and abrupt oscillations with vegetation characterized by dominance of Pinus followed by abrupt advances and setbacks of mesic taxa (Burjachs and Julià, 1994, Burjachs et al., 2012 and Vaquero et al., 2013). Only during short warming phases, mesothermophilous taxa developed and humidity increased (Burjachs and Julià, 1994). In level O, paleobotanical and micro-vertebrate studies document the cold and moist environment between D–O 17 and 16 (Burjachs et al., 2012 and López-García et al., 2014). Conversely in level M, pollen analysis documents the cool and dry condition between D–O 14 and 15 (Vinogradova, 2014).

The archeological levels considered in this study are characterized by large assemblages of faunal and lithic remains together with wood imprints and hearths (Chacón et al., 2013, Fernández-Laso et al., 2011 and Vallverdú et al., 2012). In level O, dated by a weighted mean about 55 ka BP, anthropogenic activities were mainly carried out in two areas located in the western part of the rock-shelter (Chacon et al., 2015 and Gabucio et al., 2014b). Analysis of the faunal remains has revealed high frequencies of red-deer (Cervus elaphus), horses (Equus ferus) and aurochs (Bos primigenius) with occasional consumption of rhino (Stephanorhinus hemitoechus), wild cat (Felis silvestris), hare (Oryctolagus cuniculus), bear (Ursus sp.) and goat (Capra aegagrus) (Gabucio et al., 2014a and Gabucio et al., 2014b). In level M, dated to between 54.9 ± 1.7 and 51.8 ± 1.4 ka BP, six main activity areas were identified, showing a diverse use of the space in comparison with the previous occupation (Fernández-Laso, 2010). The prey species most frequently found are red-deer, horses and aurochs, whereas hare, bear and lynx (Lynx sp.) were occasionally eaten. Seasonal studies of the teeth of the herbivores placed the hunts between autumn and early winter (Fernández-Laso et al., 2010) (Fig. 3 and Fig. 4).

Petrographic surveys of the area around the Abric Romaní rock-shelter document the abundance of diverse lithic raw materials (Chacón et al., 2013 and Vaquero et al., 2012). Limestone, sandstone and quartz nodules, in primary and secondary positions, are present within a radius of 5 km from the site. In this area, there is a scarcity of chert pebbles, and the main lithological formations are located further away at San Martí de Tous (465 m ASL, ≥ 12–15 km), Valldeperes (365 m ASL, ≥ 20–25 km) and Panadella-Montemaneu (709 m ASL, ≥ 25–28 km). In the archaeological assemblages, chert is the raw material most commonly used; patination has, however, considerably affected the surfaces of the flakes, limiting the information available as to procurement strategies. The only lithological type for which a broad comparison of the gathering activities for the two levels can be made is that from Panadella-Montemaneu, because patination turns this raw material light brown, which is very different from the light and dark grey-blue tones of rocks from the San Martí de Tou and Valldeperes outcrops (Vaquero et al., 2012).

The chert lithic assemblage from level O of the Abric Romaní rock-shelter is composed of 1965 items (Table 1). Flakes and fragments of flakes make up the bulk of the assemblage, whereas there are relatively few cores and retouched tools (Table 1). There are also many chips, but they are not included in this study. The assemblage suffered significant post-depositional alteration of the chert surfaces, making difficult to identify the different raw materials. In the chert collection, the raw material is principally indeterminate (85%) and the lithological type Panadella-Montemaneu (15%) is present with a little amount. The main causes of indetermination are patination (93.6%), thermal fractures (4.5%), concretion (1.6%) and rounding (0.3%). In this study, the indeterminate raw material is referred to as O whereas the raw material Panadella-Montemaneu is referred to as PAN.

The technological analysis documented that both tested and untested chert nodules were brought into the rock-shelter. The large number of cortical pieces and chunks indicates that the decortication phases occurred at the site (Table 2 and Table 3). Of the cores, only 5 have a rounded and weathered cortex, distinctive of secondary high-energy deposition, suggesting that the nodules were gathered from primary outcrops or in natural sections in which the chert formations were exposed.

An examination of the assemblage of cores and flakes determined that the Levallois recurrent centripetal method had been systematically used (Table 2 and Table 3) (Fig. 3). The recovery of cortical and semi-cortical flakes confirms that Levallois cores were configured and exploited in the rock-shelter. The configuration of the cores volume was maintained through the detachment of predetermining Levallois and trimming striking platform flakes. The convexity of the cores was instead created by core-edge flakes and core-edge dos limité flakes. The small number of core-edge dos limité flakes indicates that they were by-products of the flaking sequence, not the intended products. Although several preferential Levallois cores were identified, there are very few preferential Levallois flakes, whereas Levallois centripetal flakes are numerous (Table 3). This characteristic suggests that the use of the preferential Levallois modality was probably linked to a technical shift at the end of the reduction sequence rather than to a primary flaking strategy.

An analysis of secondary chaînes opératoires confirms that the main concept of flake production was based on the centripetal exploitation of the core volume. In fact within Levallois, recurrent centripetal cores were recovered several hierarchized centripetal and centripetal cores (Table 2). Generally speaking, the hierarchized centripetal method has been interpreted to lie between the discoid and Levallois technologies, due to the dynamic processes involved in the flaking methods (Vaquero and Carbonell, 2003). The appearance of this kind of core in secure Levallois contexts might be explained as an adaptation of the knapper to an unexpected change in the angle of the striking platform rather than a variability of the methods. In fact, knapping errors or inclusions in the raw materials may have modified the convexity of the core, and general reconfiguration might be very difficult in a later phase of core exploitation. Centripetal cores could re-enter in the same sphere of adaptability. In fact, one core might be a fractured Levallois core that was re-used for a short sequence of convergent exploitation of two flakes whereas for the other the knapper took advantage of an unprepared side of the core that had an angle that facilitated the detaching of three flakes. In cores-on-flakes, there is one example in which the centripetal method was used, and the overall morphology of the blank reminds one of the Levallois recurrent centripetal methods (Table 2). The other core-on-flake instead shows a simpler approach with the detachment of one flake. The remaining cores of the secondary chaînes opératoires lack any preparation of the flaking surface (Table 2). The unidirectional and simple cores could therefore be considered opportunistic, since the exploitation sequence is very short and few flakes were detached – which were probably made to meet an urgent need.

It is worth noting that only 2 retouched artifacts were recovered – a denticulate and a notched tool (Table 6). A previous study documented the presence of higher frequencies of serrated tools in the eastern area of the rock-shelter (Picin et al., 2011), indicating that there was a different location at the site where domestic tasks involving flake resharpening or tool discarding were carried out.

The chert lithic assemblage of level M of Abric Romaní rock-shelter contains 1953 items (Table 4). Flakes and fragments of flakes make up the bulk of the collection, whereas there are relatively few cores and retouched tools (Table 4). There are also many chips in the assemblage, but they are not included in this study. In the assemblage, the raw materials are mostly indeterminate (85%) while the ascription to the chert formations of San Martí de Tous (6%), Valldeperes (7%) and Panadella-Montmaneu (2%) is very low. The main causes of the indeterminacy are patination (84.4%), thermal fractures (10.7%), concretion (4.4%) and rounding (0.2%). Because of this problem, the indeterminate items were grouped with those from San Martí de Tous and Valldeperes and all analyzed as a single group called M. The lithic remains from Panadella-Montmaneu (PAN) are considered separately because they are less affected by patination, and could be precisely identified by macroscopic examination.

The large numbers of cortical flakes, cores and chunks (Table 5) show that chert nodules were brought into the rock-shelter, tested and then, if their flaking properties were poor, discarded after few flakes had been detached. The assemblage contains no artifacts having the weathered cortex that is typical of secondary depositions in high-energy water, suggesting that the raw materials were collected from primary outcrops or exposed sections.

A technological study of the chert assemblage has demonstrated that bifacial discoid technology was the most commonly used method (Table 4) (Fig. 4). The fact that several naturally core-edge flakes and cortical core-edge flakes were identified corroborates the theory that the cores underwent decortication and configuration in the rock-shelter. The convexities of the core's volume were maintained through the detachment of core-edge removal flakes and pseudo-Levallois points (Table 5). During the reduction phases, a large number of centripetal flakes were produced (Table 5). It is worth noting the significant number of pseudo-Levallois points in the assemblage, which suggest that they were not only by-products of the discoid method but were also the intended artifacts, which were transported. Occasionally, after knapping errors, or achievement of the maximum convexity of the cores, some technical flakes were stroked to translate the striking platform (Table 5) in order to continue the production by means of a peripheral striking platforms or turning the discoid reduction towards a polyhedral strategy (Slimak, 2003). In level M, these blanks were detached in the same direction of the striking platform and are characterized by an axial crest. As an example of the use of this technical expedient is the recovery of few polyhedral cores and a discoid core in which could be observed the creation of a second striking platform and, when the core started to become rounded, a short sequence of bidirectional flakes were produced.

The pattern of detaching few more flakes, once the maximal exploitation of the core was achieved, is commonly seen in the Abric Romaní sequence (Chacón, 2009, Picin, 2014 and Vaquero, 1997). In level M, it was also detected in the cases of 4 discoid cores and 1 orthogonal core. The flexible approaches taken by the knappers in attempting to continue flake production before discarding the cores indicate how important the raw materials were to Neanderthals. In this mind-set, discoid is the technology that best allows cores to be the homogenously reduced and the production of flakes to continue uninterrupted. The fact that just a few small flakes were detached is a clear indication of the strict economic patterns followed in the strategies for procuring and using chert nodules. Moreover, the discoid method was applied whatever the dimension of the initial core. For example, the use of discoid exploitation in two fragments shows that the concept was rooted in the Neanderthals’ technical behavior and was also utilized in short reduction sequences.

In secondary chaînes opératoires, the concept of centripetal exploitation is maintained with the recovery of 4 hierarchized centripetal cores (Table 4). In these artifacts, the bifacial and alternating flake production was abandoned as result of the volumetric division between a hierarchized flaking surface and a surface of preparation of the striking platform. This shift in how flake production was organized may have been constrained by the morphology of the initial nodule or by a change in the overall shape of the core during reduction. Such a change was probably due to a knapping error or an invasive detachment. In fact, 3 cores have flat morphology of the surface of the preparation of the striking platform. Although these cores have been heavily reduced, they each have a very short lateral edge for performing a bifacial reduction. Hierarchized centripetal exploitation may therefore have been a necessity rather than the initial concept chosen for the reduction sequence. The other hierarchized centripetal core, however, has a sub-pyramidal shape, which suggests a longer reduction sequence, with a hierarchical sub-division of the volume of the core.

Other artifacts were also found in the assemblage of cores (unidirectional, centripetal and orthogonal cores), in which the exploitation patterns have been interpreted to be opportunistic, due to the use of chunks or core fragments for producing a few flakes and the use of one side of the striking platform. Even if some chert chunks or fragments do indeed have poor flaking qualities, in any case, the production of just a few flakes indicates a general determination to use as much as possible of the raw material before discarding it. In cores-on-flakes, this approach is even more accentuated because the starting blanks are very small. They were, however, exploited as much as was possible, as is demonstrated by one core-on-flake in which bifacial reduction was changed to unidirectional, in order to continue with production. In the analysis, some of the Kombewa-type flakes have probably been classed as debris because of their small size, and only further refitting studies might disclose their association with cores-on-flakes reductions. The technological attribution of the Kombewa-type flakes, according to Bourguignon and Turq (2003), shows that the centripetal concept and the use of a unidirectional pattern were maintained when the available volume was insufficient to continue with the reduction, or the striking platform was constrained by the morphology of the flake.

In the chert assemblage, there are few retouched tools in comparison with the number of flakes, and they are mainly notched tools and denticulates (Table 6). This is a common feature in the Abric Romaní sequence, since the ratio of stone tools to the number of complete flakes is also low in other levels, except for level J, which is slightly richer in stone tools (Picin, 2012). Serrated tools are the commonest stone tools in Abric Romaní, whereas scrapers are rare and so far have mostly been recorded in level E (Picin et al., 2011).

A general overview of the dynamics of Neanderthal settlements reveals their preference for, and selection of, natural shelters located close to, or in mountain ranges. These locations were ideal for making full use of different ecological zones where, between the valleys and the surroundings at higher altitudes, there were abundant biotic and abiotic resources. The north-eastern of Iberian Peninsula is characterized by different territorial physiographic entities: the Catalan Costal Depression and Costal Range, the Catalan Transversal Range, the Catalan Central depression (including the Ebro basin), the Pre-Pyrenees and Pyrenees. Paleoecological analyses of the late Middle Paleolithic records in these territories reveal vegetational variability of the landscapes with pine forests with or without junipers, junipers forests, mixed pine-oak forests, grasslands and xerophytic steppes (Burjachs and Julià, 1994, Burjachs et al., 2012, González-Sampériz et al., 2005 and Renault-Miskovky and Burjachs, 1992). During the MIS 3, the presence of pine forest remained stable from the cost to the Pyrenees with the lowering of the Pinus sylvestris tree boundary – that nowadays reaches 800 m ASL – at about 300 m ASL (Burjachs and Julià, 1994). During stadial phases and Heinrich events are recorded an increase of steppe vegetation (Asteraceae, Poceae and Artemisia) whereas, during interstadial periods, are documented an expansion of mesothermophilous arboreal taxa (Quercus sp., Olea-Phillyrea, Rhamnus, Betula, Corylus, Ulmus, Fagus) and, in lower percentages, other thermophilous taxa (Pistacia, Cistus, Syringa, Juglans, Hedera, Coriaria, and Erica). In some areas, the influence of the Mediterranean climate supported the development of refugia for temperate mesothermophilous vegetation during cold stages (Burjachs et al., 2012 and González-Sampériz et al., 2005).

In the Eastern Pyrenees, within the Abric Romaní rock-shelter, other archaeological sites have been identified that have provided evidence of both long-term and short-term Mousterian occupations. The Levallois method has been identified at the Roca del Bous rock-shelter in level N10 in a high mobility context in which chert nodules were gathered from outcrops distant about 15 km (Mora et al., 2008), whereas at Cova Gran level S1B, the chert raw material was collected in the neighbourhoods of the sites (Martínez-Moreno et al., 2010). At Cova Gran, paleoecological information are still unavailable while at Roca del Bous, the anthracological analysis indicates only the occurrence of Juniperus sp. and P. sylvestris-nigra, two arboreal taxa common in the Pre-Pyrenees during stadial and interstadial phases (Piqué, 1988).

At Arbreda Cave, the archeological evidence from the Mousterian levels indicates that both the Levallois and discoid methods were used on local rocks (metamorphic rocks, quartz and flint), whereas the transported artifacts made from exogenous raw materials – from locations up to 100 km away in the eastern Pyrenees – were mainly Levallois flakes (Duran and Soler, 2006). Paleobotanical analyses are available only for level I and indicate the presence of two interstadial stages, characterized by temperate and humid environment separate by a cold and dry phase. P. sylvestris is the dominant arboreal taxon whereas, during warm phases, Acer sp., Rhamnus sp. and Prunus sp. are recorded (Renault-Miskovky and Burjachs, 1992). The study of micro-mammals confirms this interpretation for the presence of species associated with open forests (Myotis myotis and M. myotis-blythii) and humid environments (M. myotis-blythii and Microtus agrestis) (López-García and Cuenca-Bescós, 2010).

In the nearby Ermitons Cave, the small lithic collections from the Mousterian levels IV–VI are characterized by high frequencies of retouched tools and the use of Levallois technology on local raw materials (Maroto, 1993). The paleoclimatic evidences are based on the micro-mammals analysis and the discovery of the species Apodemus sylvaticus, Pliomys lenki and Glis glis indicates temperate environmental conditions (Maroto, 1982).

The use of discoid technology has been detected in the uppermost levels (S5–S7) of the Tragó Cave (Martí et al., 2009), at the Gabasa Cave (Santamaría et al., 2008) and in level G of the Fuentes de San Cristóbal Cave (García-Antón et al., 2011). The lithic analysis of the assemblages from Tragó Cave and Gabasa Cave show that the discoid method was used in a pattern of reduced mobility, with knapping of local raw materials located near to the sites (Martí et al., 2009 and Santamaría et al., 2008). At Tragó Cave, paleoecological data are still missing whereas at Gabasa Cave pollen analysis on hyena coprolites reveals a mosaic glacial landscape with big frequencies of Pinus, Juniperus, Poaceae and Artemisia (González-Sampériz et al., 2005). In level G of the Fuentes de San Cristóbal Cave, the raw material most frequently used is flint gathered from within a radius of about 9 km (García-Antón et al., 2011). Paleobotanical analyses show the occurrence of pines, junipers associated with Buxus sempervirens, Viburnum opulus and Quercus sp. suggesting an arboreal Mediterranean refugium (Allué, 2002 and Llácer, 2005).

Cave sites with short occupations and small fragmented lithic assemblages have also been documented, and are characterized by a combination of Levallois, discoid and expedient knapping methods. At Cave 120 level IV–V, a few Levallois artifacts were found, together with flakes produced opportunistically (Agustí et al., 1991). Pollen and anthracological analysis suggest mild climate for the presence, within pines and junipers, of Quercus sp., Corylus avellana, Fraxinus sp., Phillyrea sp. and Arebutus unedo in level IV, and Ulmus sp. in level V whereas micro-mammals document the presence of Glis glis and Pliomys lenki (Agustí et al., 1991). At Teixoneres Cave level III, the reduction sequences are fragmented including Levallois flakes and cores, some by-products of the discoid method, and ordinary flakes (Rosell et al., 2010). Paleobotanical studies document woodland areas composed of pines and oaks with high frequencies of Coryolus and presence of Buxus sempervirens suggesting warm and humid environmental conditions (López-García et al., 2012).

All this archaeological evidence shows a general pattern in the Northeast of the Iberian Peninsula during the late Middle Paleolithic, in which the use of Levallois technology is associated with chert and longer distance mobility patterns during climatic ameliorations. Discoid technology is, on the other hand, more closely linked to the use of local raw materials and a lower degree of mobility. The fact that there are several examples of coexistence between the two methods in the region could be interpreted as an adaptation to the size and quality of the metamorphic rocks or else to careful economic management of the raw materials. This feature is particularly evident in the pre-Pyrenean and Pyrenean areas, where good quality chert sources are more scattered. Furthermore, even when the stone outcrops were located close by, analyses of the lithic assemblages generally show hyper-exploitation of the cores, accompanied by the use of expedient methods (unifacial and polyhedral) in which chunks or core fragments were re-used for short flake production sequences. The recycling patterns also included retouched tools and core-on-flakes with the clear intention of producing small blanks (Duran and Soler, 2006, Martí et al., 2009, Santamaría et al., 2008 and Vaquero, 2011).