RADIOCARBON DATES FOR THE LATE PLEISTOCENE AND EARLY HOLOCENE OCCUPATIONS OF COVA ROSA (RIBADESELLA, ASTURIAS, SPAIN)

ABSTRACT Four excavations have been performed at the archaeological site of Cova Rosa (Asturias, Cantabrian Spain): three of them in the second half of last century and the other in this decade. Although little of the archaeological material found in those excavations has been published, here we attempt the stratigraphic correlation of sections revealed by the different excavations and we present 22 new radiocarbon dates for bones and marine shells, built in a Bayesian statistical model. This has enabled the documentation of occupations that mainly took place during the Last Glacial period, in the Solutrean (middle and upper phases) and Magdalenian (archaic, lower, and upper phases), and also in the early Holocene (Mesolithic). These occupations are compared with the record at other sites in Cantabrian Spain in general and in Asturias, in particular.


INTRODUCTION
In the last decade, one of the areas in the Iberian Peninsula in which knowledge of prehistoric huntergatherer groups has advanced most is the Sella valley, particularly as regards the transition from the Upper Pleistocene to the early Holocene. This valley, in the east of the Principality of Asturias (northern Spain), is the location of one of the densest concentrations of prehistoric sites in caves and rock-shelters in the whole of the Cantabrian region (e.g., El Cierro Cave, with archaeological levels covering the period from the Mousterian to the Mesolithic, and Tito Bustillo-Dwelling Area, or "Area de Estancia," with intense occupations dated in the Magdalenian), many of which have been made knowninrecentpublications(seeÁlvarez-FernándezandJordá2018forspecificbibliography;Figure1).
Cova Rosa is located in the lower Sella valley. Despite its status as a classic site in Palaeolithic research in northern Spain, owing to the excavations performed by Professor Francisco Jordá Cerdá (FJC) in 1958, the results of later fieldwork at the site (in 1964 and from 1975 to 1979) have remained practically unpublished, even the age of the different levels that were excavated. The remains found during the latter two series of excavations are currently being studied by an interdisciplinary team based at the University of Salamanca. Researchers from the same university have again been excavating and sampling the site since 2017.
The objectives of the present paper are: first, to provide an updated stratigraphic sequence for the Palaeolithic deposit in Cova Rosa, based on a new lithostratigraphic study carried out at the site and its correlation with the stratigraphic sequences presented by past researchers; and second, to present a series of radiocarbon dates that cover all the levels in the sequence and thus place them precisely on the Upper Pleistocene chronoclimate scale.
The archaeological site of Cova Rosa is situated in the small village of Sardéu, Ribadesella Municipality (Asturias, Spain). Its UTM coordinates (ETRS89, Zone 30) are: X = 327,289 and Y = 4,812,217. It is 4 km in a straight line from the current coastline, about 3km from the neighboring caves with archaeological deposits of Les Pedroses and El Cierro and 5.5 km from the left bank of the Sella estuary ( Figure 1). Geologically, Cova Rosa is located at the eastern end of the Asturian Massif in the Cantabrian Range, in a sector with Palaeozoic materials belonging to the Cantabrian Zone of the Iberian Massif. The karstic cave formed in the limestones of the Carboniferous La Escalada Formation (Moscovian) (Navarro et al. 1986). The cave entrance is at 149 m asl, at the bottom of an endorheic depression, 73 m deep, in the extreme northwest of a blind valley, at the foot of the limestone ridge called Peña Pagadín (417 m asl) on its southern slope. It is a large rockshelter, 25 m wide and 15 m high, which gives access to the cave of the same name, whose entrance is now blocked and can only be accessed with speleological techniques. The small river flowing in the blind valley currently sinks into the karst 23 m below the rock-shelter, whereas in the Pleistocene the sink was through the rock-shelter. The Cova Rosa karst complex is a long cave system through which the San Miguel River circulates before resurging in the village of Soto. Cova Rosa currently shelters an important bat colony protected under the legal figure of Reserva Natural Parcial de Cova Rosa.

History of Research
Cova Rosa was discovered in 1957 by FJC. In 1958, after sieving the sediment from the spoil heaps left by unknown people who had dug a large pit, he cleaned a section and excavated a surface area 3 × 1 m in size. Based only on the archaeological remains, he distinguished two different periods in the occupation: one dated in the Upper Solutrean and the other in the Lower Magdalenian (Jordá Cerdá 1977). These remains were studied by P. Utrilla (1981) and L. G. Straus (1983), who confirmed FJC's initial attributions (Layers 8-6: Upper Solutrean; Layers 5-2: Lower Magdalenian; and Layer 1: superficial). Additionally, P. Utrilla (1981) and C. González Sainz (1989) noted some artifacts belonging to the Upper Magdalenian, although out of context. FJC excavated in Cova Rosa again in 1964. A surface area of 3 × 0.75 m was studied adjacent to the area excavated in 1958. Recently, based on the excavation logbooks and the study of the archaeological objects found (including the analysis of materials obtained by wet-sieving the sediment samples, which recovered anthracological and above all micro-vertebrate remains), the process of the excavation has been reconstructed and the different levels have been discriminated (Álvarez-Fernández et al. , 2019a. FJC differentiated a disturbed surface level (layer 1) and three geological/archaeological levels which he excavated in 12 layers reaching a depth of approximately 1.40-1.50 m: Level I (layers 2-4) attributed to the Magdalenian; Level II (layers 5-7) with very scarce archaeological material also attributed to the Magdalenian; and Level III (layers 8-12), attributed to the Solutrean. Below layer 12, FJC found "a layer of very compacted calcareous clay," approximately 20-30 cm thick, without archaeological remains. He did not reach the bedrock of the cave From 1975 to 1979 FJC andAlejandro Gómez Fuentes (AGF), supported by the geologist Manuel Hoyos Gómez (MHG), carried out an excavation over a surface area covering more than 13 m 2 of the rock-shelter (Jordá Cerdá and Gómez 1982) (Figure 2). They documented nine levels, named Cova Rosa A to I. These excavations were carried out following the method of Cartesian coordinates (Laplace 1973) and by subjecting all the extracted sediments to a triple sieve screening. Of the nine levels documented, only Cova Rosa A and Cova Rosa B were dug in the 1970s. Layers A0 and A1 were excavated in Cova Rosa A (10-35 cm thickness). They were provisionally ascribed to the Final Magdalenian-Azilian. Twelve layers were excavated in Cova Rosa B (from the top to the bottom, B1 to B12). Additionally, in order to continue the excavation with more ease, Square D4 was excavated, and the whole sequence was documented in that area, from the clays at the base to the top. Apart from an interim report on Layer A0 (Jordá Cerdá and Gómez 1982) and a study of the biotic remains in Layer B6, which were assigned to the lower Magdalenian (Álvarez-Fernández et al. 2019b), the work in the 1970s has remained unpublished.
FJC did not date, by radiocarbon or any other method, the archaeological levels that he documented in the different excavations that his team carried out in Cova Rosa during the second half of the last century.
Since 2017 the University of Salamanca has again being carrying out archaeological fieldwork in the cave with the objectives of cleaning up the excavated surfaces and the stratigraphic sections that were still intact, and of collecting different kind of samples ( Figure 2). The refreshed sections were those in the outer face (south section) of the 1958 excavation and the south and east sections in Square D4 of the 1975-1979 excavations. In addition, staff from the Archaeozoology Laboratory at the University have collected samples of bones and shells from the archaeological levels documented during the last century (in 1964 and in the 1975-1979 excavations) in order to determine the chrono-stratigraphic position of the artifacts found in the past and correlate them with the finds from the new archaeological fieldwork.

Stratigraphy
After examining the south section of FJC's 1958 excavation and, in more detail, the south section of Square D4, refreshed in 2019, six lithostratigraphic units were identified, which were numbered in increasing order from top to bottom, bearing in mind that further excavations in Square D4 might reveal deeper units. These six units correspond to the six main levels defined by the geologist MHG in his doctoral thesis, which were named with capital letters (from A to F) and which in total reached a maximum thickness of about 1.60 or 2 m. MHG attributed the lower level (F) to the Würm III-IV interglacial, the intermediate levels (E, D, C and B) to the Würm IV and the upper level (A) to the Postglacial period (Hoyos Gómez 1979). Figure 3 shows the correlation established between the levels described by the researchers who worked in Cova Rosa previously (FJC in 1958 and1964;FJC, AGF, andMHG from 1975 to 1979) and the units of our own lithostratigraphic sequence obtained in 2019. This has been guided by FJC and AGF's stratigraphic descriptions and MHG's lithostratigraphic sequence. In fact, the correlation shown in Figure 3 was not problematic, thanks to those prior researchers' descriptions, drawings and photographs and the stratigraphic sections currently visible in Cova Rosa. Clear erosive contacts between some of the stratigraphic units, their textural characteristics (clasts and matrix) and their colour facilitated the correlation.
From bottom to top, the lithostratigraphic units observed in the south section in Square D4, where the cave bedrock has not been reached, is as follows (Figure 4): • CR6. 8-11 cm visible. Very light brown massive silty clay with autochthonous limestone clasts. • CR5. 20-22 cm. Polyhedral autochthonous limestone pebbles and plaquettes with sharp edges arranged horizontally, in a dark grey-black silty clay matrix. • CR4. 25-23 cm. Polyhedral autochthonous limestone pebbles with sharp edges, in an abundant brown silty clay matrix.
• CR2. 25-42 cm. Polyhedral autochthonous limestone pebbles with sharp edges, in a grey silty clay matrix. The contact with the underlying layer is clearly erosive. • CR1. 24-20 cm. Light beige silts and clays with a massive appearance and very compact.
Some polyhedral autochthonous limestone pebbles with sharp edges and some very rounded ones. Limestone plaquettes also appear. The surface of the contact with the underlying layer is erosional. • CR1sup. 0-50 cm. Angular autochthonous limestone boulders and rocks, with plaques and clasts that have fragmented after their sedimentation. The matrix is formed by very compact light beige silt and clay with a massive appearance. This layer only appears in the western part of the south section, towards the interior of the cave and outside Square D4.

Dating Occupations of Cova Rosa, Spain 1057
A few remains of a strongly cemented shell-midden (CR-conch.) are adhered to the rock-shelter wall, 1.5 m above the current floor (top of CR1 and CR1sup.). Therefore, a large stratigraphic gap exists between the deposit that forms the fill in the rock-shelter and the organogenic deposit of this shell-midden; this would be the consequence of erosion after their sedimentation.
Based on its sedimentological characteristics, the lowest unit in the sequence (CR6) might correspond to the late OIS 3, which was very cold, coinciding with Heinrich Event 2, while  the detritic sequence, from CR5 to CR1sup would correspond to the coldest phases of OIS 3 (Last Glacial Maximum): the stadials GS2c, GS2b, and GS2a. In turn, the shell-midden (CRconch.) would belong to the first millennia of the Holocene, in the warmer and wetter conditions of the Greenlandian.

Method
Of the 22 samples that have been dated, three correspond to bones from FJC's 1964 excavation, ten from FJC and AGF's 1975-1979 excavations (two shells and eight bones), and nine were collected during the 2019 fieldwork (one shell and eight bones).
Shell and bone samples were chosen for radiocarbon dating as those remains have been found in abundance in the different excavations carried out in Cova Rosa. The preferred dating material was bone and, wherever possible, we selected bones for which the taxon and anatomical element could be identified and which displayed cutmarks as evidence of anthropic activity. However, in the uppermost layers excavated in 1975-1979 (A0, B1 and B6) marine mollusk shells were selected instead, either the periwinkle Littorina littorea, or the common limpet, Patella vulgata. This was because osseous remains in those layers were fragmented and altered by natural agents and post-depositional processes (the cave was used as an animal pen until quite recently). Moreover, those two gastropod species were the ones most widely consumed as food in the late glacial period in northern Spain, where they are abundant at archaeological sites located near the coast (Álvarez-Fernández et al. 2019a). In the case of the 2019 fieldwork, the only shell sample (Phorcus lineatus) came from the small remnants of a shell-midden adhered to the rock-shelter wall of the cave, where only mollusk shells have been documented. Charcoal fragments that could potentially be dated were also recovered in the 1964 and 1975-1979 excavations. Their presence was recorded in the excavation logbooks and has been confirmed in the sediment floating and screening process. They are currently being studied.
Routine radiocarbon protocols as applied at the Oxford Accelerator Unit (ORAU, Oxford, UK) were used for the dating of the samples (Brock et al. 2010). The shells 663,31,706,31,707,31,708) underwent the pretreatment for biogenic carbonates as this is applied at the ORAU (phosphoric acid dissolution). In the case of bone samples, some were dated using ultrafiltration 940,38,941,38,942,38,943,38,944,39,004,39,005,39,006) and the rest using the simpler ABA method (ABA: 268,38,269,38,270,38,271,38,272,38,273,38,274). The selection of the ABA method over ultrafiltration was based on the physical inspection of each sample and the assumption that some bones may not preserve sufficient collagen. This turned out not to be true as collagen yields ranged between 1.2-9.5% and no bone sample failed to yield collagen.
In addition, in order to place the human occupations at Cova Rosa on the chrono-stratigraphic scale for the late Upper Pleistocene currently in use, based on the Greenland Events (Björck et al. 1998), we have compared the radiocarbon dates from Cova Rosa with the NGRIP oxygen isotopic data used here as palaeoclimate proxy (Svensson et al. 2008).

Dating Occupations of Cova Rosa, Spain 1059
Validity Analysis The radiocarbon results have been subjected to a validity test (Mestres 1995(Mestres , 2000(Mestres , 2003(Mestres , 2008 to determine that they meet the basic requirements for such dates. These requirements are of chemico-physical, analytical and archaeological types. The chemico-physical requirements refer to the capacity of the material to provide a valid radiocarbon date. The 22 samples from Cova Rosa are all organic materials (18 bone fragments and 4 mollusk shells) and therefore satisfy those conditions perfectly.
The analytical requirements that radiocarbon dates should meet refer to their accuracy (elimination of contamination, chemical treatment and measurement of the 14 C content) and precision (standard deviation) and are related to laboratory quality control. The values related to the accuracy of the results include % collagen or yield, %C and C/N ratio for bone samples; the latter is often regarded an indicator than no exogenous C was present in the dated sample. In the case of the Cova Rosa bones, the C/N ratio ranged between 3.1-3.3, the collagen yield ranged between 1.2-9.5% and the %C in the combusted gelatine was between 40.2-44.5; all values fall well within expected ranges.
The requirement of precision specifies that the standard deviation of the radiocarbon results should be as small as possible. This is met by the results provided by the ORAU, as the deviations are smaller than or equal to 110 years, which is considered reasonable for this time in the Pleistocene. Two are smaller than 50 years, seven are between 50 and 65, eight between 80 and 90, and five are larger than 90 radiocarbon years.
Finally, in archaeological terms, a radiocarbon date should be representative of the event that it is attempting to determine chronologically. That is to say, a close correspondence should exist between the physical age of the dated sample and the archaeological age of the context or event being dated. It should therefore fulfill two essential conditions: first, the material itself or its presence in the archaeological deposit should be a product of the biological or technical activity of the human group that created the deposit (association), and second its formation should be contemporaneous with the human group that created the archaeological deposit (synchronicity). As regards their association, we can be reasonably certain that the dated samples (bones and shells) are loosely linked to the anthropic activity that created the different levels in the deposit. Similarly, all the samples correspond to organisms with a short life (red deer and marine gastropods), except for two bone samples from an unknown species, and therefore they can be considered synchronic with the anthropic activity that deposited them at the site. However, two of the results does not appear to represent a time that agrees with the age of its stratigraphic position, and these anomalies will be discussed below.
All the calibrated dates that appear in the text are given with confidence intervals of 68% or 95%. The comparisons with dates from other sites are made after recalibrating them with the same calibration curve as used in the present study. Table 1 presents the dates obtained in the different archaeological fieldwork seasons (1964, 1975-1979, and 2019) and Table 2 and Figure 5 the calibration of these dates, in reverse stratigraphic order. Calibration was performed using IntCal20 and Marine20, and for the shell samples we used also a regional marine radiocarbon reservoir correction of −117 ± 70 for the late Pleistocene samples and −105 ± 21 for the Early Holocene ones as established by Soares et al. (2016). Further to this, to verify the validity of this local offset we dated a marine shell and a bone sample from the same Lower Magdalenian context (B6). We obtained two dates, OxA-38273: 15385 ± 80 (bone) and OxA-31708: 15810 ± 60 (shell) which after calibration and correction for the global and local reservoirs agree well with the aforementioned value.  The results are discussed below, within the periods they belong to, in decreasing chronological order.

Results and Critical Assessment
In an attempt to assess the coherence of the chrono-stratigraphic sequence in a probabilistic way, we incorporate the new results from Cova Rosa into a Bayesian statistical framework. We used the OxCal 4.4 platform (Bronk Ramsey et al. 2009a) to combine the radiocarbon likelihoods with the relative stratigraphic information available from the site. We use a sequence in which nested are four sequences/phases representing the distinct archaeological periods (Solutrean, Lower Magdalenian, Upper Magdalenian, Mesolithic).
We used an outlier detection analysis (Bronk Ramsey 2009b) that measures probabilistically the degree to which determinations agree within the overall model structure and its components. We used a t-type outlier model, and assigned 5% chances for each  Figure 5 Calibration of the dates obtained at Cova Rosa with the IntCal20 and Marine20 curves (Heaton et al. 2020;Reimer et al. 2020). In black the terrestrial samples, in blue the marine ones. The NGRIP oxygen isotope record, showing palaeoclimatic variations (Svensson et al. 2008), is reproduced at the top of the graph. The provenance of each of the radiocarbon dates is indicated next to their OxA code.
Dating Occupations of Cova Rosa, Spain 1063 determination to be an outlier. Each phase in the model is separated by two boundaries, to better account for the time required for observed time lag between archaeological phases.
The Bayesian results are shown in Figure 6 and Table 3. No outlier was identified in the dataset, and with the exception of OxA-31801 which showed a slightly elevated chance of being an outlier (23%) all other dates agreed well with the model (>95%).

Solutrean
Eight dates were obtained for the Solutrean units.  Figure 6).

Magdalenian
Thirteen dates have been obtained for the Magdalenian levels.  ca. 20,000-19,500 cal BP interval, a slightly more temperate time of amelioration in Stadial GS2b followed by a later fall in temperature. These two dates correlated with the samples obtained in 1964 in Level I, Spit 4 (Square E/F 4-6); both OxA-31800 and OxA-31799. They are associated with characteristic artifacts like bladelet cores/nucleiform endscrapers. In this level, while the most abundant raw materials are local (quartzite and local flint types), others are from more distant sources in northern Spain (Flysch-type flint) and Aquitaine ("grain de mil" type) (Álvarez-Fernández et al. 2020b). The radiocarbon results correspond archaeologically to a period called archaic Magdalenian in the regional classification.
These dates for CR2 are also correlated with two others for Level B in the 1975-1979 excavations: OxA-38274, from Square G3, Spit 11; and OxA-38269, from Square D4, Spit 5. These results can also be linked to OxA-31706 from Square G4 in Level A0, which however appears to represent secondary contamination from Level B that could not be identified during the excavation of Level A0.
The other three samples, OxA-31708, OxA-38273 and OxA-38268, also correspond to Unit CR2. They come from Level B in the 1975-1979 excavations but are about a millennium younger than the previous dates (ca. 19,200-18,500 cal BP). They come from the top of the level. The first two samples were taken in Square G3, both of them in Spit 6 (one was a shell of L. littorea and the other a Cervus elaphus bone). Detailed information about the use of biotic resources is available for this level, in which hunting specialised in red deer, and shellfishing and fishing were practiced. The palaeoenvironmental data from the anthracological study indicates open vegetation with a predominance of shrub taxa whereas the small mammal assemblage suggests a cold humid climate (Álvarez-Fernández et al. 2019a).
The third of these dates (OxA-38268) was obtained from a sample from Square D4, Spit 2. Information about the lithic assemblage comes from a preliminary study of this square. It is characterised by an abundance of geometric microliths, backed bladelets and, above all, triangles shorter than 1 cm in length. The calibrated interval of this date would correspond to the warmest time in the middle of GS2b and the slight cooling immediately afterwards. In sum, Unit CR2 can be attributed to the early Magdalenian, probably in two stages; one assigned to the archaic Magdalenian and the other to the "Cantabrian" lower Magdalenian.  (17,200-16,110 or 17,950-15,980 cal BP, at 68.2% or 95.4% respectively) and based on a limited number of dates (n=2) only lasts for a millennium or so.

Mesolithic
The sequence registers a large gap of several millennia, from˜15,500 to 7500 (Table 3, Figure 6), before a new occupation is identified through the date obtained for a shell in one of the remnants of the shell-midden adhered to the north wall of the rock-shelter.  (Straus and Clark 1986) and El Mirón (e.g., Straus and González 2003).
The Solutrean is represented at other well-dated sites in the Sella valley; for example, in Levels H1 and H2 at El Cierro (Álvarez-Fernández et al. in prep.) and Level 3 at El Buxu, with industry attributed to the upper Solutrean (Menéndez et al. 2016). In northern Spain, the evolution of this period, with an older phase known as the middle Solutrean and a more recent phase or upper Solutrean, can be followed in Sala I in the cave of Las Caldas, in Asturias (Corchón 2017a). Based on the chronological data, these two phases may be represented in Cova Rosa. From the typological point of view (Schmidt 2015), examples of points with flat retouch and laurel-leaf points may indicate the existence of either the middle or the upper phase of the Solutrean. Index fossils of the upper Solutrean, such as shouldered and concave-based points, are also present (López-Tascón et al. in prep.).
The dates obtained for Unit CR2 are coherent with the archaic or lower Magdalenian within the regional panorama. The result of the sample for CR1bis indicates the cave was occupied in the upper Magdalenian, a period that is well documented at the mouth of the River Sella. Antler harpoons with round cross-section are characteristic of this period. In addition to the harpoon recovered at Cova Rosa and others found, for example, in Level 1 at Tito Bustillo (Moure 1990), a further example has recently been described from Sector IV at Les Pedroses (Martínez Villa 2019). It is a well-known period also in the rest of northern Spain; e.g., in Level 1 at Coímbre B (Álvarez-Alonso and Jordá 2017), Level IV at La Pila, in Cantabria (Gutiérrez Saez 1998) and Level Almp/Lamc/Slnc at Santimamiñe, in Biscay (López et al. 2020). The Mesolithic date for the shell-midden at Cova Rosa agrees with dates obtained in numerous similar deposits, mainly in the east of Asturias (Fano 2019). One of these is the shell-midden at the proximate site of El Cierro (Álvarez-Fernández et al. 2020a), which is assigned to the Asturian culture and dated ca. 9500/8000 cal BP.

CONCLUSIONS
Cova Rosa is one of the few sites that have been excavated over a large surface area in northern Spain. Across several field seasons in 1958,1964,[1975][1976][1977][1978][1979], and 2019, we calculated that about 23 m 2 of cave fill has been excavated ( Figure 2). Apart from isolated instances, the results of previous fieldwork have remained practically unpublished. Here we determine a new stratigraphic sequence that correlates with the levels identified by the previous researchers (FJC, AGF, and MHG) (Figure 3). Using radiocarbon dating we precise date the sequence using 22 samples collected in 1964, 1975-1979, and 2019. Based on the direct dates and the study of the archaeological material the cave was occupied in the different phases of the Solutrean and the Magdalenian periods, in the late Upper Pleistocene (after the Last Glacial Maximum to the start of Greenland Interstadial 1). It was also occupied in the first part of the Holocene (at the start of the middle Holocene or Northgrippian). It is interesting to note that erosion has removed most of the materials that would correspond to much of the GI 1 (late glacial period), the whole GS 1 (Younger Dryas) and all the early Holocene or Greenlandian. Our new data are comparable to those of similar sites in the Sella valley, e.g., El Cierro (Álvarez-Fernández et al. 2016) and Les Pedroses (Martínez Villa 2019), and in other regions of Cantabrian Spain, e.g., La Riera (Straus and Clark 1986), in Asturias; El Miron (Straus and González 2003), in Cantabria; and Antoliñako koba (Aguirre 2014), in the Basque Country). This increasing number of recently excavated and dated sites provides useful insights into the occupation of Cantabrian Spain during a period of both extreme climatic instability and cultural variability.