Critical evaluation of in situ analyses for the characterisation of red pigments in rock paintings: A case study from El Castillo, Spain.

Paint technology, namely paint preparation and application procedures, is an important aspect of painting traditions. With the expansion of archaeometric studies and in situ non-destructive analytical methods, a renewal of technological studies is being observed in rock art. In situ analyses have several limitations that are widely discussed in the literature, however. It is not yet clear whether they provide accurate information on paint technology, except under certain conditions. Here, we evaluated digital microscopic and pXRF in situ analyses for the characterisation of a large set of red and yellow paintings from the El Castillo cave, Cantabria, Spain. We have set experiments and used statistical methods to identify differences between paint components and determine factors impacting pXRF measurements. We found that the compositional heterogeneity of the paintings' environment, especially variations in secondary deposits, was responsible for most of the differences observed between the pXRF signals recorded on the paintings. We concluded that the El Castillo cave environment is not suitable for non-destructive technological studies, but that more favourable contexts might exist. Following previous works and our own results, we advocate a combination of both in situ and laboratory invasive analyses for the study of paint composition and paint technology. Our research protocol, based on the comparison of rock paintings, their substrate, experimental paintings and Fe-normalisation of the signals can improve the reliability of pXRF results. We also propose to include more systematic characterisation of rock wall heterogeneity and the use of microscopic analyses in non-destructive approaches.

1. Introduction

Palaeolithic rock art is a key feature of human evolution, providing information on humans’ language, abstract thoughts, symbolic behaviour, ontological world and social organisation. Traditionally, rock art studies have focused on chronology, theme and style. With recent developments in archaeometric techniques, however, interest in paint technology, namely paint preparation and application procedures, has increased considerably (see e.g. [1-15]). In-situ analyses performed with portable X-ray fluorescence (pXRF) and Raman spectroscopy equipment are becoming widespread for the study of paint materials and painting techniques. They allow the paintings to be conserved, unlike more conventional laboratory analyses requiring sampling. However, conservation is not the only issue that must be taken into account. The reliability of the results also needs in depth evaluation. It is therefore necessary to assess the limitations of in situ analyses more effectively, in order to determine the extent to which they provide valid knowledge on paint technology, and need to be backed up by sampling procedures.

Over the last decade, in situ spectroscopic Raman and pXRF analyses have provided interesting results on the pigment families composing various open-air rock paintings and paintings associated with portable art around the world (see e.g. [2, 12, 13, 16–29]). Successful results have also been obtained on some European cave paintings, despite great difficulties gaining access to the painted panels [1, 30–36]. Microscopic in-situ analyses have been used less frequently in recent studies, but they do give significant information on paint components and painting application techniques [12, 37, 38]. However, the identification of complex paint recipes or ‘paint pots’ with these methods remains challenging, especially for red iron oxides, the most common family of rock art pigments. How in-situ analyses can be used to identify more than a pigment family and how they can provide consistent information on paint technology remains unclear [8, 12, 20, 39]. Recent work shows that pXRF elemental composition of the paintings is not directly comparable with the pXRF elemental composition of colouring material used to make them [13]. Yet most of the time rock art pigment composition studies operate under permits that restrict analyses to non-destructive techniques and pXRF represents the only method of data acquisition allowed.

The red paintings from El Castillo (Cantabria, Spain), inscribed on the UNESCO World Heritage List in 2008, represent an ideal case study to address this issue. Various figures are painted in red on the walls of this important cave art site, including deer, bison, tectiforms, sets of negative hand stencils, juxtaposed disks, and various other abstract representations [40-42]. Uranium-series disequilibrium dates of calcite deposits on the paintings have shown that these representations extend back to at least the Aurignacian period and possibly the late Mousterian, with minimum ages of 41.4±0.6 ka for a red disk, 37.6±0.3 ka for a hand stencil, and 35.7±0.3 ka for a red disk (95.4% probability intervals; [43]). Other Uranium-series disequilibrium and radiocarbon dates have shown that the paintings were made over a long time period from the Aurignacian (or earlier) to the Magdalenian (12–17 ka; [43-45]). Recently, interdisciplinary analyses were carried out on 11 panels of red disks and micro-samples were taken from four of them [4]. The results showed that at least two paint ‘recipes’ were used, with different grinding intensities and mineralogical compositions. These two different ‘recipes’ were applied with different blowing techniques in different parts of the cave, suggesting the panels of disks were not all made at the same time. The technological and chronological relationships between the red disks and other figures remain unknown.

In this paper, we present in situ analyses of a wide range of paintings from El Castillo, including the red disks. The aims of this study were to 1) test pXRF and microscopic in-situ analytical methods on El Castillo red paintings, 2) evaluate their reliability for the characterisation of pictorial layers in a cave context, 3) search for significant compositional variations between different red figures from El Castillo. To achieve this, we carried out microscopic examination and pXRF analyses of several red paintings and one yellow painting for comparison purposes, and used experimental paintings and statistical analyses to assess the reliability of the results obtained on the El Castillo paintings. The interest and limitations pXRF analysis has already been discussed for other archaeological applications (see e.g. [46-50]) but no in-depth studies have been carried out in the critical domain of painted rock art. The approach we developed, which until now remained untried, was of critical importance in the revelation of the origin of the compositional differences observed on the El Castillo paintings that we were able to analyse.

2. Background

2.1 Why investigate variations in pictorial techniques?

Technological studies in rock paintings are useful to understand the technical and economic framework of rock art production as well as to identify possible diachronic trends in these domains (see e.g. [2, 4–6, 8–10, 14, 19, 35, 51–75]). From a technological point of view, two main aspects are expected to vary between paintings: the way the paint was prepared and the way it was applied. Variations in application procedures can reflect functional choices related, for instance, to the intention of producing fine lines or large figures, regular or irregular edges, transparencies or opacities between the line and the painted surface. Paint preparation can be driven by colour choices and functional requirements, such as the ability for the prepared paint mixture to stick to the substrate or to dry rapidly. The pictorial mixture might be more or less viscous, depending on the way it is prepared [4]. In this regard, the choice of the binder is also an important parameter. Technical choices may be limited by raw material availability, depending on a geological context for the inorganic part of the paint mixture. Restrictions in the choice of paint materials may have influenced both preparation and application techniques or, conversely, the choice of one particular technique may have guided the choice of raw materials and the use of other techniques. Finally, binders and pigments are not independent parameters. The choice of one influence the choice of the other. They are also highly influenced by a society’s overall technical and symbolic systems (e.g. use of the same binder for different purposes) and the way in which the painter has been taught to mix and apply the paint. In other words, the cultural logic driving the “chaîne opératoire” [76] may play a key role in the final appearance of the painting.

Consistency has been observed at El Castillo between these three parameters, choice of raw materials, preparation and application techniques: the paintings that are likely to have been blown with two tubes were made with a fine-grained and pure hematite pictorial mixture, while those thought to have been applied with the mouth or with a single tube contain coarse fragments of pigments and a mixture of iron oxide and clay minerals [4]. This might be due to the fact that the former was more liquid and better suited to the two-tube blowing technique, traditionally known as the airbrush technique. On the contrary, when there are few functional or raw material constraints, variations in pictorial techniques are more likely related to technical choices embedded at different degrees within social and symbolic dimensions.

Technological studies at a site or set of sites are also a powerful tool to understand the chronocultural framework behind rock art production [4, 5, 52, 63, 69]. The different choices listed above can reflect cultural traditions, especially if they are driven by arbitrary, social and/or symbolic considerations. Nonetheless, a technical choice can also relate to individual habits, whatever the influence of functional issues, raw material availability constraints, or arbitrary motivations. Simultaneous variations in pictorial techniques and other parameters, such as location in the cave, the type of designs or their style, are more likely to reflect significant variations in group habits and cultural traditions [4–6, 14, 28, 72]. Between different regions, differences in pictorial techniques could relate to differences in the properties and composition of the raw materials available in the surrounding environments. They may involve distinct cultural traditions, but also variations within a broad set of shared knowledge, depending on the geological context. At a single site or a cluster of close sites, the geological environment can be considered quite stable, except under particular climatic conditions. Variations in technical choices at a site, if significant and consistent with other parameters, likely reflect moments or phases of production. But they may also reflect abrupt changes in raw material availability caused by intense climatic or geophysical changes (erosion of an entire geological horizon in the surroundings for instance). For this reason, artificial pigments whose date of invention or first import in a region is known are more reliable chronological markers than natural local pigments [69]. The same might be true for paint preparation or application techniques that could be accurately dated at contemporaneous archaeological sites. In addition, paint recipes may vary as a function of changes in trade networks and patterns of cultural exchange. It is also possible that groups with different artistic practices came to paint in the same cave roughly during the same period. Available dating methods do not allow for the chronological precision required to investigate such sources of variation.

2.2 How to identify pictorial techniques?

Paint application techniques have often been investigated through in situ microscopic examination of paintings, with macrophotography being used [38, 52]. They are now completed by in situ microscopic images [12]. They are simple, efficient methods that allow the distribution of the pictorial mixture on the substrate to be observed, and also the particle size of the paint components to be characterised. In some cases, they have been combined with experimental programs [38, 77].

Paint preparation used to be and is still commonly investigated through micro-sampling and micro-analysis of pictorial layers (see e.g. [2, 4, 9, 10, 15, 52–56, 58, 61, 62, 64–67, 70–75, 78–81]). Micro-sampling allows various information on the pictorial layers to be collected, since samples can be observed as such in the laboratory and analysed on cross-sections or powder samples. The grain-size, crystal shape, elemental and mineral composition, along with the spatial arrangement of the paint components, can be identified and compared between different samples. These analyses give information on the raw materials that were used to make the paint and, to a lesser extent, on the way these materials were prepared. This approach has been formalised by the determination of significant compositional groups among the sampled paintings, groups referred to as ‘pots de peinture’ (‘paint pots’; [57, 58, 82]). Different ‘paint pots’ might represent different paint recipes, but also the acquisition of different raw materials. The scientific limitation of micro-sampling and micro-analyses is the variable and possibly low representativeness of the samples area. Its obvious main disadvantage is its destructiveness.

Systematic sampling of rock paintings is no longer recommended for the study of rock art. The respect of these paintings is now favoured in curation policy in Europe and elsewhere. The development of transportable in situ analytical instruments has been proposed as a means to replace invasive sampling approaches in order to determine pigment composition and significant compositional differences between paints [1, 21, 27, 28, 34, 36]. Two techniques are commonly used: portable Raman spectrometry and portable X-ray fluorescence (pXRF). Raman spectrometry allows the identification of the main minerals responsible for the colour of a paint, sensu stricto the pigment [16, 21, 24, 25, 27, 32, 65]. In some cases, mixtures of minerals composing the pictorial layers have been identified [21]. The main limitations of this method are the strong fluorescence phenomenon induced by the substrate and/or deposits on the pictorial layers and the difficulty of distinguishing the minerals composing the pictorial layers from those belonging to the substrate and alteration deposits.

In situ pXRF spectrometry is the most frequently used method to identify pigment composition in rock art (Table 1). It has been successfully applied to rock shelters, cave walls and mobiliary art for the identification of the main elements composing pictorial layers. Theoretically, several families of pigments can be identified with this equipment: iron-based yellow and red pigments, black manganese oxides, white phosphates, carbonates, titanium oxides. For black paintings, the absence of manganese and iron on the spectrum is an indication in favour of the use of pigments rich in carbon. In some cases, the type of pigment or differences in pigment composition were used to support chronological attributions [5, 69, 83]. However, data acquisition, processing and interpretation remain challenging in these contexts because 1) the pictorial layers are systematically thinner than the penetration depth of the X-ray emissions; 2) they cannot be physically separated from overlaying alteration deposits or from the underling substrate; 3) rock walls are very irregular in shape and being in contact with the paintings is not always allowed nor possible; 4) the rock substrate is often heterogeneous in composition and covered by various secondary deposits (see e.g. [12, 19, 20, 27, 36]). In particular, compositional variations detected by in situ XRF analyses of red paints are often difficult to distinguish from compositional variations in the substrate and in alteration deposits, likely because the iron signal is too weak (thin pictorial layers), too variable (variations in the thickness of alteration deposits and/or pictorial layers), or because discriminant elements are below detection limits [13, 18, 20, 26, 33]. In several pioneering studies, differences between red paints were observed, but no consistent clustering was detected; only one or two figures could be differentiated within the sample being compared with no clear correlation with other parameters, such as graphic style or theme represented [12, 17, 34–36, 84]. Such random differences in composition may not relate to intentional technical choices. They could be due to undetected variations in the composition of the substrate (presence of inclusions, concretions, etc.), undetected presence of alteration deposits (Mn-enriched deposits for instance) or the presence of repaints [4]. Only one study has found consistent differences between stylistic groups of red paintings [28]. The low number of figures analysed (only 6) makes the results preliminary, although promising.

Table 1

Details of previous pXRF analyses carried out in rock shelters or caves for the characterisation of rock paintings.

Study	Location of the paintings			Instrumental parameters								Sampling
	Name of the site	Type of substrate	Rock substrate	Type of instrument	Anode of X-ray tube	Voltage	Intensity	Counting time	Detector	Filters/colimaters	Detection of light elements	Figures	Analyses of pigments	Analyses of the substrate and alterations
Chanteraud et al. 2021	Grotte aux Points, France	Cave wall	Limestone	ELIO XGLab	Rh	40 kV	100 mA	300 s	SDD		From Al (air system)	32	43	23
Trosseau et al. 2021	Font-de-Gaume, France	Cave wall	Limestone	ELIO XG-Lab	Rh	40 kV	40 μA	600 s	SDD	?			78 (including the substrate)	Yes
Huntley et al. 2021	BH15-01, Central Pilbara, Australia	Rock shelter wall	BIF	Bruker Titan S1 800	Rh	45 kV—15 kV	10.45 μA—31.55 μA	90 s	SDD	Ti-Al filter—no filter	From Al (air system)	?	21	?
Castañeda et al. 2019 [87]	Eagle Cave and Sayles Adobe, Texas	Pebbles	?	Innov-X Systems Alpha Series	Ag	40 kV	?	60 s	Si-PIN	?	No (from Ti to Bi)	70	139	115
Lebon et al. 2019	Doi Pha Kan, Thailand	Rock shelter wall	Limestone	ELIO XGLab	Rh	40 kV	100 mA	300 s	SDD		From Al (air system)	6	?	Yes
Mauran et al. 2019	8 sites, Erongo, Namibia	Rock shelter wall, stained artefacts and grinding stones	Granite	ELIO XGLab	Rh	20 kV	200 mA	300 s	SDD		From Al (air system)	34	>57	41
Bedford et al. 2018	Pleito Creek, USA	Rock shelter wall	?	Bruker Tracer III-V	Rh	40 kV	3,4μA	60 s	Si-PIN		?	1	6	4
Huntley et al. 2018	Kabi Kabi Hand Stencil Site, Australia	Rock shelter wall	Sandstone	Olympus 40 kV Delta Environmental	4 W Ta	40 kV—15 kV	100 μA—80 μA	180 s	Si-PIN		No (air system)	3	8	11
Gay et al. 2016 (2020)	Rouffignac, France; Font-de-Gaume, France; La Garma, Spain	Cave wall	Limestone	Home made—MOXTEK X-ray tube	Pa	40 kV	-	-	SDD		From Al (air system)
Rifkin et al. 2016	Apollo 11, Namibia	Plaques	?	Bruker Tracer III SD	?	40 kV- 15 kV	10 μA—55 μA	120 s	SDD	Al-Ti filter—no filter	From Al (vacuum system)	6	37	6
Wallis et al. 2016 [88]	VSTA_20140611_1 quarry	Quarry wall	Shale	Bruker Titan S1 800	Rh	45 kV—15 kV	10.45 μA—31.55 μA	45 analytical s (effective time of detection)	SDD	Ti-Al filter—no filter	From Al (air system)
Gay et al. 2015	La Garma, Spain	Cave wall	Limestone	Home made—MOXTEK X-ray tube	Pa	40 kV			SDD			10+1	57	Yes
MacDonald 2016	8 sites, Southern Canadian Shield area	Rock shelter wall	Granite, granitic gneiss	Innov-X Delta Premium	Au	40 kV—15 kV	0.1 mA	120s	SDD	?	From K	62	>62x2	yes
Sepulveda et al. 2015	6 sites, Lluta to Camarones coastalvalleys, Chili	Rock shelter wall	Igneousferromagnesian volcanic rock	Bruker Tracer III-SD	Rh	15 kV	21 μA	30 s	SDD		From Al (air system)		19	6
Beck et al. 2014	Rouffignac, France	Cave wall	Limestone	Home made	Cu	40 kV	700 μA	-	SDD	polycapillary semi-lens	No (air system)	15	26	26
Bedford et al. 2014	Three Springs, California	Rock shelter wall	?	Bruker Tracer III	?	40 kV	3,4μA	60 s	?		No (air system)	?	150 (including substrate)	Yes
Hernanz et al. 2014	Multiple sites, Spain	Shelter and cave wall	Limestone? Dolomite?	X-MET5100, Oxford instrument	Rh	45 kV	-	50 s	SDD	-	No (air system)	-	-	-
Koenig et al. 2014	10 sites, Lower Pecos Canyonlands, USA	Rock shelter wall	Limestone	Innov-X Systems Alpha Series	Ag	40 kV	-	30 s	Si-PIN	-	No (air system)	-	225	23
Lopes-Montalvo et al. 2014	Cova Remigia rock shelters, Spain	Rock shelter wall	Limestone?	Home made	Ag	30 kV	0.1 mA	200 s	Si-PIN	Al pin-hole	No (air system)	25	34	18
Pitarch et al. 2014	Los Chaparros, Spain	Rock shelter wall	Limestone	X-MET5100, Oxford instrument	Rh	45 kV	-	50 s	SDD	-	No (air system)	2	12	5+4(black dendrites)
Wesley et al. 2014	4 sites, Red Lily Lagoon area, Australia	Rock shelter wall	Sandstone	Bruker Tracer III-V	Rh	40 kV- 15 kV	15 μA—1.1 μA	180 s	Si-PIN	Cu-Ti-Al		32		Yes
Loendorf and Loendorf 2013	Picture Cave, Texas, USA	Rock shelter wall	Limestone	Bruker Tracer III–V	Rh	40 kV—15 kV	12 μm	150 s	Si-PIN	Al-Cu-Ti and Ti	Vacuum	9	?	Yes
Olivares et al. 2013	La Peña Cave, Spain	Cave wall	Limestone	X-MET5100, Oxford instrument	Rh	45 kV	-	50 s	SDD	-	Yes (detector?)		56
Roldàn et al. 2013 (2016)	Parpalló cave, Spain	Plaquettes	Limestone-sandstone	Home made	Ag	30 kV	-	-	Si-PIN	Al pinholes	No (air system)	18	40	27
Velliky and Reimer 2013 [89]	DjRt–10 and EaRu–9, Squamish Valley, British Colombia, Canada	Rock shelter wall	Granite	Bruker Tracer III–V	Rh	40 kV—15 kV	12 μm	150 s	Si-PIN	Al-Cu-Ti and Ti	Vacuum	3	15	15
Huntley 2012	BR29 Rock Shelter	Rock skelter wall	Sandstone	Bruker Tracer III-V	Rh	12 kV	20 μm	300 s	Si-PIN	Ti 0.0254μm filter	From Al (vacuum system)		2(experiment)+5(rock art)	27(experiment)
Nuevo et al. 2012	Abrigo dos Gaivões and Igreja dos Mouros caves	Cave wall	?	Home made	Ag	30 kV	30 μm		Si-PIN	-	No (air system)		10	5
Roldàn et al. 2010	Saltadora rock shelters VII, VIII and IX, Spain	Rock shelter wall	Calcareous rock	Home made	Ag	30kV	0.1 mA	200 s	Si-PIN	Al pin-hole	No (air system)	41	57	15
Appoloni et al. 2009 [85]	Jaguariaíva 1 rockshelter, Brazil	Rock shelter wall	Sandstone	Home made	Ag	?	-	300 s	Si-PIN	Ag	No (air system)	-	19 (including substrate)	Yes
de Sanoit et al. 2005	Rouffignac, France	Cave wall	Limestone	Home made	-	15kV	40 μA	-	SDD	-	No (air system)	7	12
Newman and Loendorf 2005 [86]	Musselshels site, Montana; Castle Gardens Wyoming	Rock shelter wall	Sandstone	Innov-X systems XT series model 700	-	-	-	90 s	-	-	No (air system)	-	46	-

So far, only one painting experimental program was performed to assess the pertinence of pXRF analyses [14]. An experimental approach reveals to be very useful to evaluate the informative potential of in situ pXRF results when this technique is applied to paintings and, if necessary, to design means to improve its reliability.

2.3 Issues of terminology and methodology related to in-situ analyses

The impossibility of physically separating a pictorial layer from its surrounding context by pXRF reveals terminological limitations in rock painting analyses. Usually, the pictorial layer is distinguished from its substrate and overlaying secondary deposits that naturally form or accumulate on the surface of caves or rock shelters, such as coatings of calcite, soluble salts, soot, etc. [2, 19, 23, 58, 70, 90, 91]. Strictly speaking, the substrate is the bedrock on which the paintings were made, but it can also refer to the surface on which the paintings are applied, when it is not possible to separate the superficial layer of the bedrock from secondary alteration deposits underlying the paint layer. When that superficial layer is covered by secondary deposits, analyses of cross-sections will reveal them, and they will be distinguished during the analysis from the rock substrate [2, 15, 19, 23, 58, 70]. The characterisation of each of these layers by in situ pXRF analyses is not possible. The pictorial layer is commonly characterised by comparing the signal of the paintings with the signal of the surface close to them. This is what, depending on the authors, is referred to as the analysis of the substrate, i.e. the cave wall, rock wall, or unpainted rock background, [2, 5, 12, 20, 36, 39]. The signal obtained is a combination of the signal of the bedrock plus the signal of its superficial layer and its alteration deposits. In the present work, a fragment of the cave wall was sampled and its inner part (fresh rock) analysed and compared to its outer part and other locations of the cave wall in order to distinguish the composition of the bedrock from the composition of its surface as exposed in the cave.

More recently, some authors have proposed another method to separate the pXRF signal of the paint components from the signal of their ‘environment’ [5, 36]. They used the measurement from the cave wall near a painting as a blank and its signal was subtracted from the signal of the painting. The remaining signal was considered to be that of the pictorial layer alone. The terminology introduced in this study is very useful as the term ‘environment’ is neutral and can be used whatever the substrate is made of and covered with. In the present work, we choose to keep this terminology. However, we will not use cave wall signals as blank signals because we cannot be sure that the environment beneath the paintings is the same as the environment close to them. We will instead evaluate how the cave wall signal varies from place to place, how variations in the environment may affect the signal of the paintings, and how these variations affect the clustering of the measurements.

The last limitation to be taken into account is the distance of the pXRF instrument from the paintings and the irregularities of their surface. If the instrument is not in contact with the paintings, the signal is affected by the layer of air present between the two. This results in a decrease in the intensity and peak ratios of light elements (see e.g. [46]). In some studies, it was possible to position the XRF instrument in contact with the paintings (see e.g. [19, 69, 83, 84, 92]). In our case, however, as in most similar studies in Europe, we were not allowed to touch the wall and, although always in the range of few millimetres, the precise distance from the pictorial layer could not be controlled. The irregular shape of the wall was a supplementary limitation for close contacts. As a consequence, peak ratios between light and heavy elements may vary from one spectrum to another. Surface irregularities result in similar problems [93, 94]. In order to evaluate the impact of these factors on the results, we carefully analysed the spectra, and as in previous studies, we applied statistical analyses to the spectra [13, 25, 95]. This approach allowed us to evaluate the potential factors affecting the signals.

2.4 Archaeological context

El Castillo cave opens onto the side of a mountain of the same name near the village of Puente Viesgo (Cantabria, Spain) (Fig 1). Featuring rich evidence of human occupation since at least 150,000 years ago, El Castillo and the nearby caves contain remarkable stratigraphic sequences and art, whose discovery and study has played a key role in Palaeolithic archaeology. El Castillo was discovered in 1903. Excavations at the entrance exposed an 18–20 m deep stratigraphy comprising 26 alternating sterile and anthropogenic layers [96]. From top to bottom, the archaeological sequence begins with Medieval and Chalcolithic deposits overlying Azilian and Magdalenian levels [96, 97]. Solutrean, Gravettian and Aurignacian occupations overlie layer 18, attributed to the Late Mousterian [98] or Initial Aurignacian [97], under which we find levels recording the regional evolution of the Mousterian. The basal deposit contains Acheulean lithic artefacts. The karstic network beyond the entrance contains one of the most conspicuous Palaeolithic art ensembles of Western Europe, with numerous engravings, drawings and paintings covering the full range of the themes, techniques and styles of the period [40–42, 45, 99]. The paintings are distributed around various parts of the cave and organised in different panels. Some panels are polychromic, other entirely made with red paints. The latter largely dominate when all the abstract designs, such as disks and lines, are included.

Fig 1

Location of El Castillo site and map of the cave with the location of the different panels (source of the map: open source https://www.naturalearthdata.com).

3. Materials and method

3.1 El Castillo paintings

A total of 14 red disks, four red hand stencils, one yellow bison, one red tectiform, one red vulvar sign, one red digital line and one red lineal composition were analysed (Fig 2; Table 2). We selected red disks from different parts of the Galería de Los Discos: part 1, end of the corridor; part 2, middle of the corridor; and part 3, beginning of the corridor. Since access to the red disks that have been sampled and analysed in a previous study was impracticable with the pXRF instrument [4], we analysed similar disks located in the same areas. The dated disks [44] were not selected for analysis. The first one (date O-69-80) because it is located in a niche that is not accessible with the pXRF instrument and the other (O-83) because it is overlaid by a thick layer of calcite.

Fig 2

Pictures of the main paintings analysed in the present study.

Black squares indicate where pXRF analyses were performed. A: Galería de los Discos, beginning (H-1). B: Galería de los Discos, end (H-3). C: Galería de los Discos, middle (H-2). D: Panel de los Polícromos, Vulvar sign. E: Techo de las Manos.

Table 2

List of the paintings analysed, type of analyses performed and number of pXRF spectra used in the present study.

PANEL	REPRESENTATION	REF. PAINTING	pXRF analyses	Microscope	Micro-sampling (d’Errico et al. 2016)
Galería Final	DISK	LL1	-	-	X
Galería de los Discos, H-end	DISK	H3	3	-	-
	DISK	H5	1	-	-
	DISK	H11	2	-	-
	DISK	H15	3	-	-
	DISK	H14	-	-	X
	DISK	H20	3	-	-
	DISK	H25	2	-	-
	LINEAR COMPOSITION	HFIG	2	-	-
	DISK	H39	3	-	-
	DISK	H40	3	-	-
Galería de los Discos, H-middle	DISK	H53	2	-	-
	DISK	H54	3	-	-
	DISK	H58	-	-	X
Galería de los Discos, H-beginning	DISK	H93	3	-	-
	DISK	H99	-	-	X
	DISK	H103	1	X	-
	DISK	H104	2	-	-
	DISK	H105	2	X	-
Techo de los Manos	BISON	BIS	2	-	-
	HAND STENCIL	M1	2	-	-
	HAND STENCIL	M2	2	-	-
	HAND STENCIL	M3	1	-	-
	TECTIFORM	TECT	3	-	-
Sala de los Polychromos	HAND STENCIL	M4	2	-	-
	VULVAR SIGN	SIGN	2	-	-
	DIGITAL LINE	LINE	4	-	-
TOTAL PAINTINGS			53	-	-
TOTAL CAVE WALL			13	-	-
TOTAL			66	2	4

The red disks from the Galería de los Discos were compared with a figure from part 1 of the Galería, three hand stencils from the Techo de las Manos, one from the Panel de los Polícromos, two figures (bison and tectiform) from the Techo de las Manos and a line and vulvar sign from the Panel de los Polícromos (Fig 2; Table 2). The dated hand stencils and disks are older than 34 ka and may belong to the Aurignacian or, possibly, the Mousterian [44]. The selected figures are considered to be younger than the disks and hand stencils, based on the stratigraphy of the paintings and their style [44, 100]. The style of the yellow bison suggests a Gravettian age, while the red tectiform and vulvar sign are attributed to the Magdalenian.

A total of 13 pXRF measurements were taken on the substrate close to the paintings. Only the flattest surfaces were analysed. Measurements of the substrate between two paintings was used as a reference for both. That is why the number of substrate measurements is lower than the number of paintings analysed. In the Panel de los Polícromos, a red natural deposit (Wall red) was also measured. The exact location of pXRF measurements is given in (S1-S6 Figs in S1 File).

3.2 Experiments

3.2.1 First experiment (Exp 1)

The aim of the first experiment we carried out was to determine the extent to which red pigments with different compositions could be distinguished with no control on the distance of the pXRF equipment from the wall. As a substrate, we used a large block of Rupelian marine limestone from the quarry of Lugasson, Gironde, France. Experimental paint was applied on an irregular fresh fracture of the block. The absence of concretions and alteration deposits considerably decreased the compositional variations of the substrate. Two colouring materials were used:

S: Ref-Fe1: ferruginous shale from De Hoop, South Africa (home-made reference, see [101, 102]).

H: Blutstein, Kremer pigment, Blutstein Type 2004.

Their elemental composition is given in Table 3. These two reddish pigments were mixed with water in different proportions and applied on the substrate in different ways (Fig 3A). For the present study, we kept only preparation type 1 (H1 and S1), which resulted in the thinner and more homogenous paint layer (Fig 3B). The paints, made by mixing 0.3g of pigment in 1.5mL of water, were applied with a brush.

Table 3

Elemental composition (major elements) of the two pigments used in the experiments.

Ref.	Name	Fe2O3	SiO2	Al2O3	CaO	MgO	K2O	TiO2	P2O5	MnO	Na2O	SO3	CO2	H2O (struct.)	PF*
S	Ref.-Fe1	8.152	47.34	26.417	0.031	0.989	5.474	1.507	0.04	0.0118	1.101	-	-	-	8.84
H	Blutstein	80	7.8	3.2	2.1	1.5	1	<0.1	<0.1	<0.1	<0.05	<0.05	3.4	(strukt.)	-

*PF: ignition loss (perte au feu).

For Blutstein Type 2004, the data are those given by Kremer pigment manufacture. For Ref-Fe1, see Dayet et al. 2014; 2019. Bold values: elemental contents that could theoritically be used fo differentiate the two colouring materials with the instrumental conditions used.

Fig 3

Pictures of the experimental paintings.

A: paintings made on the Rupelian limestone (1: mixed with water, applied with a brush; 2: mixed with less water, applied with a brush; 3: powder directly applied with a finger). B: pictures of the selected experimental paintings H1 and S1 on the Rupelian limestone; C: fragment of El Castillo cave wall. D: same fragment with painting H1 (Exp2).

3.2.2 Second experiment (Exp 2)

The purpose of the second experiment was to create a reference painting comparable to the El Castillo paintings in order to help the interpretation of the archaeological results and evaluate the extent to which differences in provenience or paints’ pot could be distinguished when working at El Castillo. We applied a paint on a fragment of El Castillo limestone with the H1 pigment mixture (Blutstein+water; Fig 3C and 3D). The Blutstein pigment is rich in Fe (about 80%) and it is also composed of a little Si and Al, a composition that corresponds to the composition of the El Castillo red disks [4]. The provenience of the two pigments is, however, clearly different. The paint was applied with a brush in order to produce a homogenous paint layer.

3.3 Methods

3.3.1 Digital microscope

For the in situ microscopic observations, we used a Hirox VCR 800 digital microscope attached to the end of a metal arm mounted on a heavy tripod and driven by a crank handle in order to avoid vibrations, allow precise movement of the microscope and avoid accidental contact with the cave wall. Observations were conducted at magnification ranging between 20x and 160x.

3.3.2 Portable X-ray spectrometer (pXRF)

The pXRF measurements were carried out using a portable SPECTRO xSORT X-ray fluorescence spectrometer (Ametek), equipped with a silicon drift detector (SDD) and a low-power W X-ray tube with an excitation source of 40kV. Measurements were acquired in the air without contact with the paintings, by fixing the device on a dedicated metal stand attached to the same equipment described above for the digital microscope. The working distance varied according to the flatness of the painting surface, but was kept below 3mm. An area of 8mm in diameter was analysed. Spectra acquisition time was set to 300s. The spectrometer was internally calibrated by an automated measurement of the contents of a standard metal shutter. We took at least two pXRF measurements per painting and up to four when possible. Only the spectra showing the highest Fe peaks were kept because the lowest Fe peaks were observed when secondary deposits on the paintings were the thickest. The number of spectra per painting used in data processing is given in Table 2. Measurements were also performed on the experimental paints, first on the substrate before the application of the paintings, and then on the paintings themselves.

3.4 pXRF data processing

3.4.1 Processing of pXRF spectra

As discussed above, in situ pXRF analyses have limitations. We chose to compare the raw spectra first, in order to get an idea of how the measurement conditions impacted them. As a second step, the spectra were normalised by the total counts per spectrum, using the same range of energy (see [17, 18]). We also tested a normalisation method inspired by the Fe-ratios normalisation used in ochre provenance studies [101, 103–108]. The method reduces the differences in trace element composition between ochre samples of varying Fe content. Palaeolithic red and yellow paints are usually composed of iron-rich pigments. Using Fe-ratios may decrease the compositional differences between pictorial layers that have different thicknesses or that are covered by heterogeneous deposits. The following normalisation formula was applied:

Ix = number of counts for the energy channel considered.

bgmin = minimum of the background (number of counts) depending on the range of energy (1.7–4.3keV; 4.3–7.6keV;10.3–10.9keV).

AFe = area of K-alpha Fe peak, sum of the counts from E = 6.11 to 6.65keV.

AbgFe = approximate background under K-alpha Fe peak, bgmin multiplied by the number of channels used to calculate AFe.

Logarithms were preferred because data calculated using ratios did not show a normal distribution.

3.4.2 Calculation of peak areas

In order to assess whether the characterisation of iron-rich pictorial layers can be improved by pXRF data processing, we also used the areas of the peaks (netto counts) calculated using the fundamental parameters (direct values calculated by the PDA of the SPECTRO xSORT X-ray instrument). Considering the large differences between elements, we used the centred log ratio (clr) and the logarithm weighted by Fe netto counts (alr) [109]. We used Fe-ratios of netto counts for the same reason we used the Fe-ratios of the spectra. The calculations were performed using CODAPAK software [110]. When more than 25% of the values were missing (netto count = 0), the element was not taken into account. Missing values were replaced by 0.65 multiplied by the minimum peak area detected for the element concerned following the software recommendation.

3.4.3 Multivariate statistical data processing

Statistical analyses were used in order to improve sample comparison. We used Principal Component Analyses (PCA) in order to describe composition variability among the measurements. The variables and individuals retained in each PCA are summarised in Table 4. For the spectra, PCAs were carried out using a selected range of energies: 1.70–10.92 KeV (variables). Energies below 1.70 keV were not used because the instrumental conditions were not suitable for the detection of light elements and energies above 10.92 keV, as the peaks of heavy elements are too weak or absent. As the logarithm of null values cannot be calculated, this means that when the background reaches its minimum, there is a missing value in the Fe-ratio log data. All energies showing missing values were removed from PCA4, as were energy ranges between 7.61 and 10.39 keV because of the high contribution of the X-ray source in this part of the spectra.

Table 4

Summary of the variables (energies and elements) and individuals (measurements) used in the different PCAs.

PCA	Variables	Individuals
PCA1	Raw spectra, energies from 1,70 to 10.92	All in situ measurements, Exp 2
PCA2	Normalized spectra, energies from 1,70 to 10.92	All in situ measurements, Exp 2
PCA4	log of Fe-ratios, energies from 1,70 to 7,60; from 10,31 to 10.92; variables with missing values removed	All in situ measurements, Exp 2
PCAa	clr Netto counts, elements with less than 25% of null values	All in situ measurements, Exp 2
PCAb	alr-Fe Netto counts, elements with less than 25% of null values	All in situ measurements, Exp 2
PCAc	alr-Fe Netto counts, elements with less than 25% of null values	Reddish cave wall and Exp 2 excluded

Except for PCA6 and PCAc, we used all the measurements performed at El Castillo associated with the second experimental reference (Exp2) as individuals. Cave wall measurements were kept in all PCAs to avoid underestimating the contribution of the cave wall to the results. PCA coordinates were calculated using the FactoMineR library in R software [111], and graphs were done with the factoextra library.

4. Results

4.1 Microscopic examination

Macroscopic and microscopic examination of the two analysed disks revealed that their external part is composed of pigment drops (Fig 4A to 4C). This corresponds to an application by the blowing technique. Pigment trails were observed when fresh paint was reworked with the fingers (H103; Fig 4B and 4D). The pigment layer of disk H103 is thick and well preserved (Fig 4B and 4F). It is composed of a homogenous dark red fine-grained matrix. Small dark particles and coarse transparent inclusions were detected in the paint. The pigment layer of the disk H105 is very thin (Fig 4A and 4E). It is composed of a dark red fine-grained matrix. No inclusions were observed.

Fig 4

Macro and microphotographs of the paintings H105 (A, C, E) and H103 (B, D, F). Colours are not calibrated, which explains the differences in colour between the macro and microphotographs. The light conditions were identical.

4.2 pXRF analyses

4.2.1 First experiment

Paints H1 and S1 can be clearly distinguished from each other by their pXRF spectra (Fig 5A). Paint S1 is characterised by higher Si, K and Ti K-alpha peaks. These peaks are very weak on the spectrum of Rupelian limestone, indicating that they correspond to the pictorial layer. This is consistent with the higher content of these three elements in S1 (Table 3). The Fe peaks is higher in paint H1. This is consistent with its higher Fe content. The Ca K peaks are very similar, while the Ar K-alpha peak (present in the air) is a little higher in S1. This pattern indicates a possible difference in air thickness between the two measurements. The distance from the substrate was probably higher for S1. In these conditions, the peak heights of the light elements are probably underestimated on the S1 spectrum. In spite of this underestimation, the difference in composition between H1 and S1 remains perfectly detectable.

Fig 5

Normalised pXRF spectra of the first experiment (A) and the second experiment (Exp2, B); Normalisation by total counts.

4.2.2 Second experiment (Exp2)

The El Castillo limestone used in the second experiment is heterogeneous in composition. Its inner part is composed of higher proportions of Si, K, Ti, Fe and As than the Rupelian limestone (Fig 5B). These elements are, however, detected in lower proportions in its external altered part. The pXRF measurements of the paint made on the external part is characterized by higher K, Ti, Fe and As K-alpha peaks than this latter. The difference in Fe peaks’ intensity is the main difference we can observe between the limestone as a whole and the paint layer. These results show that the composition of the Fe-rich pictorial layers can be distinguished from the composition of El Castillo substrate, and that this difference is higher when the environment is depleted in elements such as K, Ti and As.

4.2.3 El Castillo normalised spectra

According to the pXRF analyses, all the red paints are likely composed of Fe-based pigments (Fig 6A; Table 5). The yellow paint depicting the figure of a bison, is also composed of a Fe-bearing pigment (Fig 6A). The difference in Fe peak intensities between the paintings and the cave wall is small in most cases. The pictorial layers are thin despite their bright colours. Additionally, the composition of the cave wall is highly variable (Fig 6B). As a consequence, no or few differences in pigment composition are detectable between representations of different themes or cultural attributions when doing direct comparison of the spectra (Fig 6A). Only one disk shows significantly higher Mn and As contents (H15). These results are consistent with those obtained on the disks of the Galería de los Discos [4].

Fig 6

Normalised pXRF spectra of a selection of figures with the substrate (A) and normalised pXRF spectra of all the measurements carried out on the cave wall (B).

Table 5

Netto counts of elements detected by pXRF, calculated with the fundamental parameters.

Painting	Type	Si	P	S	K	Ca	Ti	Cr	Mn	Fe	Ni	As	Sr	Y	Total
Exp2_1	Exp	11414	31206	12281	116367	4561897	20086	524	40138	19859546	20112	17506	45304	11413	24747794
Exp2_2	Exp	10139	26581	10790	108546	4309299	19137	525	29565	20350628	18346	17711	45776	15312	24962355
H3_1	Disk	1407	20338	5545	21582	2604060	1005	578	5019	213496	9223	nd	14679	4641	2901573
H3_2	Disk	1453	19178	5247	18558	2302612	149	572	3208	207113	8593	nd	9705	2407	2578795
H3_3	Disk	4173	40683	10512	38163	4809867	nd	576	4599	382034	11180	nd	20614	732	5323133
H5_1	Disk	3145	19331	5517	20763	2313247	1062	569	7213	603833	9258	1238	12500	6922	3004598
H11_1	Disk	3869	30926	8287	30885	3795766	nd	575	4614	569420	9520	nd	16246	3839	4473947
H11_2	Disk	6253	39818	10583	40110	4645304	1409	572	13362	586275	11408	nd	28088	3659	5386841
H15_1	Disk	2166	15469	4354	18057	1903752	435	567	39005	650699	7983	1145	8584	1349	2653565
H15_2	Disk	2706	25646	7142	28114	3160429	259	573	97242	811192	8566	3242	16403	3712	4165226
H15_3	Disk	2516	15150	4652	18419	1837420	824	566	61147	590116	8902	3615	12264	3314	2558905
H20_1	Disk	6597	27609	7891	33305	3320362	3531	575	8510	2052050	9286	2315	18325	4313	5494669
H20_2	Disk	12954	26093	7583	30254	3137303	5262	572	8332	487582	9538	1023	20876	3704	3751076
H20_3	Disk	6249	29035	7921	34324	3400391	4394	575	7380	1193541	10248	1421	19648	2325	4717452
H25_1	Disk	2250	17335	4926	28871	2191556	3675	594	1581	3107547	9044	3355	10228	917	5381879
H25_2	Disk	5063	22050	6848	34916	2551864	11074	584	8900	1909613	10771	4592	20784	3066	4590125
H39_1	Disk	7443	25634	6355	31068	3109413	2064	572	4841	1195462	8944	1197	26207	6092	4425292
H39_2	Disk	12236	52406	14349	57152	6189905	3728	565	11173	1040764	12690	731	34287	910	7430896
H39_3	Disk	9101	49941	13320	61919	5689740	6529	567	12269	2670121	12285	3037	29967	6846	8565642
H40_1	Disk	10412	37710	10911	61415	4384986	12386	570	13823	4768707	11023	645	52033	3609	9368230
H40_2	Disk	7343	23676	5914	45264	2839420	9567	580	5578	5096387	9968	1659	25399	2347	8073102
H40_3	Disk	5058	19269	5441	31977	2369365	5980	580	3203	3006075	8374	nd	22534	1458	5479314
Hfig_1	Figure	2291	12597	2686	15168	1408994	1307	561	4049	149666	8119	290	9436	1992	1617156
Hfig_2	Figure	4068	23292	6736	24092	2667276	1897	568	3986	207456	9358	nd	19784	3160	2971673
H53_1	Disk	2752	15405	4110	19424	1827476	1296	572	2951	1311953	7948	nd	27402	5047	3226336
H53_2	Disk	2883	15646	4564	23823	2001193	2070	577	2774	2551964	9388	nd	25703	916	4641501
H54_1	Disk	3987	32883	8106	33166	3922413	1019	576	16904	593587	10482	nd	37014	2987	4663124
H54_2	Disk	8924	45029	11957	48871	5429052	2996	568	8119	1195999	12787	nd	64158	2463	6830923
H54_3	Disk	4236	20903	5482	22080	2533117	684	572	3343	836935	9223	nd	29350	6827	3472752
H93_1	Disk	7306	55063	14287	58598	6444580	2978	574	20368	1138474	13176	nd	37431	4246	7797081
H93_2	Disk	5076	27885	7520	38198	3303969	5043	583	5781	2508381	10014	nd	22422	5022	5939894
H93_3	Disk	6034	41757	10479	45015	4694889	3579	574	11299	532241	10951	nd	29961	2415	5389194
H103_1	Disk	16739	15961	4176	28596	1860991	3026	558	5811	921510	9566	nd	30613	6022	2903569
H104_1	Disk	5358	9331	2531	16943	1210022	1279	448	3264	215116	7502	nd	14355	879	1487028
H104_2	Disk	5016	5684	1721	11044	613214	1622	442	3108	304053	7505	nd	8468	1157	963034
H105_1	Disk	10835	9390	2653	31034	1171118	5228	568	3860	1822781	10029	564	21081	1606	3090747
H105_2	Disk	19596	16487	4944	46990	1889603	7372	565	7882	2138106	10781	nd	34521	1768	4178615
Bis_1	Figure	4794	47701	12457	43118	5797214	1318	574	6866	259864	11645	nd	44518	6536	6236605
Bis_2	Figure	3624	32784	8812	29388	3915603	nd	572	4906	257948	9565	nd	28834	5063	4297099
M1_1	Figure	2236	28846	7093	25733	3477022	nd	577	3736	477754	9681	1704	13086	2873	4050341
M1_3	Figure	2547	30093	7487	28746	3467136	443	566	7717	112568	10613	nd	19003	2740	3689659
M2_1	Figure	2938	28232	7888	25205	3399824	nd	566	4045	83582	10645	nd	24184	2603	3589712
M2_2	Figure	3132	33278	8791	29980	4033382	nd	566	3557	128200	9208	nd	29703	3777	4283574
M3_1	Figure	1856	18804	5071	17241	2250116	nd	563	4229	93316	8465	nd	15349	2413	2417423
Tect_2	Figure	3107	26277	6597	26884	3213791	288	579	4236	334082	9299	nd	34806	4955	3664901
Tect_3	Figure	3396	24577	6886	27646	2951808	1248	576	6659	758405	8402	495	18784	2527	3811409
Tect_4	Figure	3893	22166	5159	29471	2672354	3621	573	17113	994284	9485	nd	12826	1721	3772666
M4_1	Figure	6735	60605	16026	51491	7325295	nd	572	6980	408303	13959	nd	61977	11479	7963422
M4_2	Figure	1438	12097	3476	11999	1405459	nd	567	3684	439688	9005	1430	17878	4226	1910947
Line_1	Line	9440	37864	10381	60789	4384686	13986	570	12212	1173130	12189	nd	48148	4086	5767481
Line_2	Line	4754	24841	6222	37937	2935625	8773	577	7264	1421399	9570	nd	26643	873	4484478
Line_3	Line	2202	10815	2790	16789	1267673	3477	352	3210	504753	5486	nd	14996	4415	1836958
Line_4	Line	5222	20078	5758	39961	2390878	12509	576	7449	1878705	11161	nd	29145	4240	4405682
Sign_1	Figure	2454	25194	6924	22884	3032249	nd	570	4094	101390	9391	nd	27760	7451	3240361
Sign_2	Figure	3128	32410	8247	29439	3851713	125	570	5332	150773	9256	nd	36135	7737	4134865
Wall Exp2	Wall	6813	63316	17521	56650	7681095	nd	560	22495	42966	17519	nd	59004	3421	7971360
Wall H5	Wall	3993	34456	10032	30924	4111481	161	566	3640	69456	9102	nd	21588	4047	4299446
Wall H11	Wall	7345	46727	12418	42161	5425281	nd	565	4038	85717	10932	nd	40233	4069	5679486
Wall H25	Wall	10816	21097	4980	19612	2436027	344	568	6739	70592	9229	nd	14079	956	2595039
Wall H39	Wall	9979	62280	16403	59268	7484469	1992	561	6129	194413	13128	nd	48649	1169	7898440
Wall H53	Wall	1962	16971	4150	15441	2060118	nd	559	1795	22703	7756	nd	16248	495	2148198
Wall H93	Wall	5325	40148	9431	42759	4625577	3725	575	20850	226301	11278	nd	25241	637	5011847
Wall H103	Wall	25660	31367	8240	43951	3791428	2928	556	7119	136956	11127	nd	48381	2602	4110315
Wall H104	Wall	11117	6584	2122	15663	836926	2107	448	6015	86655	8276	nd	33471	2639	1012023
Wall M1	Wall	3575	37142	9588	33161	4381731	nd	569	4491	48919	11228	nd	28527	nd	4558931
Wall M2	Wall	2642	25297	6998	24690	3009549	255	561	6446	49914	8996	nd	19883	5642	3160873
Wall Tect	Wall	9176	68274	17737	70792	8173522	2342	562	16821	294786	15158	nd	75775	nd	8744945
Wall Line	Wall	6157	57896	15491	53379	6980723	2008	563	7015	88391	12641	nd	53953	1130	7279347
Wall Sign	Wall	4193	34876	9321	31192	4121779	nd	566	4992	77484	8643	nd	33976	4460	4331482
Wall red	Wall red	1627	8415	4143	31530	1385205	14966	576	1E+06	754134	18799	50330	12742	31808	3655717

4.2.4 PCAs on raw and normalised spectra

The PCA of the raw counts from all spectra provides similar results: the composition of the cave wall is highly variable and the paintings are composed of Fe-bearing pigments (PCA1; Fig 7A). However, the XRF signal of several paintings cannot be distinguished from the signal of the cave walls. This is partly due to the contribution of Fe and Ca peaks in Comp2, but also to global differences in intensity between the measurements according to Comp1 (positive contribution for the whole spectrum). In addition, results obtained on a single figure feature substantial discrepancy. Several factors can explain these patterns: mass absorption effects (Fe versus Ca-rich matrix, difference in the thickness if the pictorial layer), the presence of a layer of air, and differences in the thickness of the layer of air (see e.g. [46]). Field observations show that the paintings with a very low pXRF signal are frequently thin and/or covered by a thick calcite layer (mostly figures and hand stencils, Fig 7A). This is in agreement with a combination of mass absorption and “air” effects. The contribution of Ar K-alpha peak to the results was not as significant, as we would expect if the thickness of the air layer is the main factor involved.

Fig 7

Results of PCA1 and PCA2 carried out on the raw spectra (A) and normalised spectra (B). On the right: coordinates of individuals; on the left: coordinates of variables.

Two normalisation methods were designed to counterbalance the overall differences between the spectra. The PCA of the spectral data normalised by total counts allows more effective separation between the signal of the paintings and the cave walls (PCA2; Fig 7B). The experimental painting (‘Exp2’), the reddish part of the cave wall (‘Wall red’), one measurement from disk ‘H104’ and the cave wall near this disk are separated from the cluster of other measurements. Fe peaks account for a high proportion in Comp1, in contrast with the signal of the background and the source. For Comp2, we can observe an opposition between Ca peaks on one hand, and Ti, Cr, Mn and Fe peaks on the other. However, there are irregularities in the contribution of the base and the summit of the peaks. This suggests that ‘Wall-red’ is rich in transition metals from Ti to Cr, ‘Exp2’ is enriched in Fe, while ‘H104-2’ (disk ‘H104’) and ‘Wall-H104’ are anomalies caused by a difference in background intensity. This is consistent with a difference in the proportions of light (Ca) and heavy (Fe) elements and suggests that this data processing is impacted by matrix effects.

When Fe-ratio logarithms are used to normalise the data, the multivariate analyses allow higher discrimination between clusters of measurements (PCA4, Comp1 and 3; Fig 8). More paintings cluster separately from the cave wall, with the measurements on the ‘Signal’, hand stencils ‘M1’, ‘M2’ and ‘M3’ being the exceptions. The cave wall points are not aligned anymore, which suggests that the difference in global intensity between the spectra was significantly smoothed. The signals of ‘Exp2’, ‘Wall red’, and disk ‘H25’ become clearly separated from the rest of the measurements. The variability in Comp1 is explained exclusively by variations in Fe contents (summits of K peaks) and accounts for almost 80% of the total variability (Fig 8). Comp2 is explained by variations is the shape of Fe peaks. As we do not know what these differences might reflect in terms of elemental composition, we did not use this component. Comp3 only accounts for 3% of the variability, relating to variations in Fe and other elements contents: Mn (summits and other parts of K peaks) and As (summit of the K-alpha peak). The experimental paints were characterised by the highest Fe contents, with small Mn and As proportions. This is consistent with the composition of the Blutstein pigment. The results of PCA4 suggest that the ‘Wall red’ measurement is the richest in Mn and As, which is consistent with its spectrum (Fig 6B). The distinction of the disk H25 from other disks and figures could be due to higher Fe and As contents.

Fig 8

Results of PCA4 of ‘Fe-normalised’ spectra (log of Fe ratios).

On the right: coordinates of individuals; on the left: coordinates of variables.

4.2.5 PCA on peak areas (fundamental parameters, netto counts)

The PCA of netto count clr coordinates allows most of the wall cave measurements to be distinguished from the paintings, except ‘Wall-H104’ and ‘Wall-M2’ (PCAa; Fig 9A; raw data given in Table 5). For the former, the difference could be due to lower netto counts of Ca, while the latter suggests higher netto counts of Ni and Cr peaks. The measurement ‘H3-3’ (disk H3) is not separated from the cave wall. This could be due to higher netto counts of S and Ca peaks. This data processing is apparently subject to different biases, linked to the very small differences in Ca peak intensities between the paintings and the cave wall, but also possibly to an underestimation of matrix effects (Fe-rich matrix) and/or the impact of the X-ray source on the overall signal (Ni, Cr). One disk (H15) is partially separated from the other paintings, likely because of its higher Mn contents (Fig 6A).

Fig 9

Results of PCAa and PCAc of clr coordinates (A) and alr Fe-ratios (B) of netto counts (peak areas). On the right: coordinates of individuals; on the left: coordinates of variables.

PCA of the Fe-ratio alr coordinates allows complete separation of the painting and cave wall points (PCAb; Fig 9B). The ‘Wall-red’ measurement was separated from all the other points, as well as from the experimental painting. In most cases, all the points measured on a painting are clustered together (‘Line’, ‘HFig’, disk ‘H105’ or disk ‘H40’), but there are several exceptions (‘Tectiform’, ‘Bison’, disk ‘H11’ or disk ‘H3’; Fig 9B). Comp1 accounts for 82% of the variability and is characterised by a positive contribution of all elements excepting Ti/Fe, the contribution of which was almost null. Comp2 is characterised by a positive contribution of Ti/Fe (high) and Si/Fe (low) and a negative contribution of Mn/Fe and Y/Fe. We noticed that the removal of the Ni and Cr (elements that were problematic in PCAa), and P and S (elements that are highly correlated with Ca) alr coordinates did not change the main results of PCAb. The separation between the paintings and cave wall measurements seems to be due to differences in the Fe-ratios of all the elements except Ti, with a higher contribution of Si. This pattern means that the Fe proportion is significantly higher in the paintings, that of Si significantly lower, and the proportion of Ti is likely variable in both. The proportion of Mn and Y to Fe is also variable. The high dispersion of measurements from the cave wall suggests that the high variability in Ti, Mn and Y within the paintings relates to variations in the cave wall beneath them.

4.2.6 Focus on the relation between cave wall and painting signals

In order to test the hypothesis that most of the variability we observed within the paintings was due to high heterogeneity in cave wall composition, a final PCA was performed with Fe-ratio alr coordinates, but removing the individuals with extreme coordinates in PCAb (Exp2, Wall-Exp2 and Wall-red; PCAc; Fig 10). When the relationship between the measurement for a painting and its corresponding substrate is represented, the influence of the former on the result is quite clear: the higher Comp2 for the cave wall, the higher it is for the corresponding painting.

Fig 10

Results of PCAc of alr Fe-ratios of netto counts (peak areas), measurements ‘Exp2’, ‘wall-Exp2’ and ‘wall-red’ excluded.

On the right: coordinates of individuals; on the left: coordinates of variables. The lines represent the vector of the ‘compositional’ distance between a painting and its substrate.

5. Discussion: Towards an understanding of the results from El Castillo in situ analyses

Differences in elemental composition among pigments can theoretically be interpreted as 1) variations in rock substrate composition; 2) variations in the nature and thickness of alteration deposits, including calcification layers, above and beneath the pictorial layer; 3) differences in pigment composition; 4) variations in pigment composition induced by secondary chemical processes, such as element migrations and recrystallization; or 5) repaints on the original paintings [4, 12, 36]. However, the reproducibility of pXRF analyses depends on the analytical conditions: variations in these conditions may also lead to variations in raw spectral data.

The first problematic result we obtained from the El Castillo pXRF analyses was a systematic relationship between the variations in Ca content and the overall intensity of the signal (PCA1, raw spectral data): higher Ca peak intensity corresponded to lower overall spectra intensity. This phenomenon can be explained by the thin layer of air between the instrument and the wall: its presence leads to a decrease in the signal of light elements (Ca in our case). By contrast, the signal of heavy elements (Fe in our case) is almost not impacted. Another consequence of this phenomenon is that when spectral data are normalised by total counts, high variations in the background are observed (PCA2, normalised spectral data). In addition, this phenomenon can have an impact on the calculation of the peak areas when using the fundamental parameter method: an anomaly in Ca and Fe contents was clearly involved in the presence of an outlier within the signals of the substrate in PCAa (‘Wall-H104’). Another bias was observed in the El Castillo pXRF raw signals: the influence of the Fe-matrix and/or the influence of the signal of the source (PCAa). However, when using Fe-normalisation (PCA4, b and c), all the biases virtually disappear. They may still have an impact on the results, but if there is one, it is not detected anymore. The single bias identified in PCA4, relating to the presence of variations in Fe peak shapes could be avoided by excluding the involved component (Comp2 not taken into account).

With these analytical biases addressed, the El Castillo results can now be discussed on a sounder basis. Considering 1) the high variations observed between the measurements for the substrate (experiments and all data processing), 2) the small difference between the cave wall and painting signals (all data processing except PCAb and c), 3) the absence of any differences between the yellow bison and red representations (all data processing), 4) the lack of clustering for measurements done on the same painting (all data processing) and the relationship between the signal of a painting and its corresponding cave wall measurement (PCAc), we hypothesise that most of the differences in elemental composition detected between the paintings come from variations in their environment. The composition of the rock substrate and/or the deposits on its surface, below and above the paintings, vary from place to place in the cave. The measurements on the fragments of the cave wall show that there are large differences in composition between the inside of the rock substrate and its surface. The surface is depleted in K, Ti and As. This could suggest that the variable results we obtained mostly depend on variations in the nature and thickness of the coating deposits.

The results of PCAb support this hypothesis. The measurements of a single painting formed a cluster when Ti content was high (high Comp2 coordinate). Ti being mostly part of the inner part of the substrate, this means that when the deposit layer is thinner, variations between close measurements are smaller. The composition of this alteration layer is another important factor explaining the dispersion of the measurements. In particular, Mn rate plays a role in the variability within cave wall and painting signals according to PCAb and to the significantly higher Mn content in the reddish part of the cave wall (‘Wall-red’). This is in agreement with previous observations on the composition of disk H58 (see micro-sample CAST-ADN2 in [4]): the pictorial layer shows very variable Mn content. Natural migration of Mn at the surface of the cave wall probably occurred at places, as commonly observed in karst systems [112, 113]. Y contents also appear responsible for large variations between the analyses of a single painting (PCAb). It may have a similar behaviour to Mn or be part of a specific phase coating at places the rock substrate (clay minerals?). As a consequence, Mn and Y contents cannot be used to discriminate paint pots. The behaviour of other elements is hard to establish without more detailed data on the substrate and its alteration deposits.

Regarding the identification of paint pots within El Castillo paintings, the identified clusters isolate only three paintings: the experimental painting; disk ‘H15’ and disk ‘H25’. All the results suggest that the experimental painting was mainly distinct from the others because of higher Fe content. This is in agreement with its composition, thickness, and the absence of deposits on its surface. Disk ‘H15’ is richer in Mn. We have just shown that we cannot use this element to discriminate between paint pots with any reliability. This leaves us with disk ‘H25’ isolated by PCA4. The result of the PCA suggests that it features a higher As to Fe proportion. This element could not be used in PCAb (because of the absence or very low As peak in most spectra). But when we look at the netto counts, the high As content in ‘H25’ is noticeable. ‘H25’ is the only disk we analysed from a vertically-aligned cluster, perpendicular to the main horizontal alignment of disks. It is possible that it was not made with exactly the same paint pot as the horizontally-aligned disks, including ‘H3’, ‘H5’, ‘H11’, ‘H20’ to its left, and ‘H39’ and ‘H40’ to its right. As is potentially a trace element that can discriminate El Castillo red paintings. Although our results do not allow us to go any further in the identification of paint pots, they nevertheless provide another line of evidence supporting the hypothesis that almost-pure Fe oxide was used to make the disks in the H-sector, as micro-analyses previously suggested.

The combination of pXRF and microscopic results adds some more lines of evidence. Pigments from disks H103 and H105 have a different thickness and the presence of small black and transparent inclusions in H103 also indicates a difference in composition. They were clearly separated by PCA4 and PCAb and c (Fig 10). By increasing the number of microscopic observations, we may get a better idea of the impact of thickness and composition of pictorial layers on pXRF measurements.

6. Conclusion: Evaluation of in situ analyses and implications for future studies

To sum up the discussion of the above, our results show that secondary alteration deposits covering the cave walls are highly variable at El Castillo cave. This considerably affects the pXRF spectra of the paintings. Data processing of both the spectra and peak areas identified only one element possibly reflecting differences in paint preparation. At first sight, pXRF analyses appear to be of little interest in the study of cave painting technology. However, El Castillo limestone is rich in elements such as Si, Ti, K and even As, that are precisely those that allowed the two red pigments used in our experiments to be identified as different paints. It remains possible that El Castillo cave constitutes a special case and that, elsewhere, the composition of the rock substrate may turn out not to represent such a limitation for the discrimination of red paint pots. By identifying the composition of the alteration deposits on the cave wall without any sampling, we were able to establish which paintings are the most affected by these deposits. This information will be precious if further micro-sampling of the paintings is made possible.

The El Castillo case illustrates once again that all forms of analyses are precious for the understanding of paint preparation techniques in rock art: 1) analysis of the paintings themselves, but also analysis of the substrate and alteration deposits; 2) in-situ analyses as presented here, but also laboratory analyses of micro-samples. In sum, we do advocate use of pXRF analyses, bearing in mind that they are not a replacement for the conventional laboratory analyses that remain the main way to validate the hypotheses put forward on the basis of in situ surface analyses. We are in favour of a more systematic evaluation of the impact of cave wall heterogeneity on in situ pXRF analyses. For instance, at Font-de-Gaume, the composition of the substrate and alteration deposits is not discussed [36]. There is no guarantee that the ‘blank measurements’ that were taken offset the heterogeneity of the cave wall. Differences in Mn content in the red paints of the bison could be due to heterogeneous Mn contents in alteration deposits, as observed at El Castillo. Similarly, the absence of comparison between the paintings at Le Peña and their environment entails that we cannot be sure of the origin of the differences observed between them [34]. We have shown that pXRF measurements have the potential to generate more accurate hypotheses if cave wall measurements are evaluated with more care.

It is worth highlighting, however, that we also identified significant analytical biases in the pXRF results that do not depend on the site nor on the instrument used. Considering the large variations in Ca content between the measurements, the thin layer of air between the incident X-ray beam and the surface being analysed have an important effect on the spectra. This effect may not be entirely corrected by the fundamental parameters method. When contact between the instrument and the paintings is not possible, the use of a flow of Helium would avoid such a bias. Unfortunately, such equipment is difficult to transport inside a cave. For this reason, the step-by-step method of enhanced data processing that we applied in this study presents considerable advantages. Multivariate analyses of the spectra are key for the identification of all types of biases and Fe-normalisation smooths them out considerably. Some authors prefer to rely on semi-quantitative data [5, 20, 36, 83], but this ‘blind’ method was not suitable in our case, considering the large variations in Fe and Ca contents. Matrix effects are too strong in such conditions [46, 114, this study]. Future research in rock paintings would greatly benefit from the creation of dedicated references allowing the calibration of pXRF instruments for the quantification of elements in Fe-bearing paints on limestone substrates, following on from the calibration tests carried out recently for Fe-based pigments [101].

The final conclusion we can draw from our study is that the interest of microscopic analyses is underestimated. They do take time, which is why we were not able to do them systematically, but when we consider the difficulty of identifying, whatever the context, paint preparation techniques by pXRF analyses, and the difficulties in getting micro-sampling authorisations, the interest of more systematic in-situ microscopic examination of rock paintings becomes evident.

Location of pXRF analyses.

(PDF)

Click here for additional data file.

Counts of pXRF measurements (raw spectra).

(XLSX)

Click here for additional data file.

18 Oct 2021

Submitted filename: Review for dayet et al.docx

Click here for additional data file.

6 Dec 2021

Reviewer #1: The work described in this manuscript demonstrates the limitations of handheld XRF measurements applied to the analysis of cave rock art.

The conclusions -- essentially that pXRF measurements have some value in surveying the elemental composition of paint, substrate, and alteration deposits but, in the context of rock art, cannot be relied upon to provide definitive, quantitative answers -- will not come as a surprise to XRF experts. The limitations of pXRF have been established previously for other archaeological materials, notably glasses and ceramics. In these studies, the use of reference standards, improved data collection protocols, and improved processing has improved data quality considerably.

As the authors point out, the situation for understanding rock art is made more challenging by common prohibitions against sampling or destructive analyses, leaving only a few nondestructive, field-based techniques like pXRF available to researchers.

A significant issue continues to be non-experts who do not appreciate the limitations of pXRF and thus ultimately over-interpret results (as scientists, authors, and reviewers). This situation certainly is not unique to XRF but can be seen for other analytical techniques including EDS, XPS, etc. Although advances in technology have made the mechanics of pXRF data collection easy, the critical interpretation of data leaves something to be desired in many cases.

The work described in this manuscript is a valuable contribution to the rock art field because it clearly shows the sources of common pXRF errors and the special circumstances and challenges related to heterogeneous, ancient paintings in cave environments.

Recommend publication after minor changes:

-- pXRF should be defined earlier, at its first appearance

-- some information and parallels from the literature describing XRF analysis of glasses and ceramics should be incorporated in this manuscript

Reply

We thank this reviewer for his/her appreciation of our work and for underling its methodological interest.

We have introduced the two changes suggested by this reviewer.

As regard to the first suggestion of change, the manuscript was changed as follows:

Line 34: “In-situ analyses performed with portable X-ray fluorescence (pXRF) and Raman spectroscopy equipments are becoming widespread for the study of paint materials and painting techniques.”

As regard to the second suggestion, our article is dedicated to rock art issues. Although we develop pXRF data treatments, this was not the main aims of our paper. We nonetheless added one sentence:

Line 84: “The interest and limitations pXRF analysis has already been discussed for other archaeological applications (see e.g. Shackley 2011; Speakman et al. 2011; Vazquez et al. 2012; Speakman and Shackley 2013; Wilke 2017; Tykot 2021) but no in-depth studies have been carried out in the critical domain of painted rock art.”

Reviewer #2: Dear colleagues,

It was a pleasure to read this article by Dayet et al. on their p-XRF analysis at the El Castillo cave. This article is well-written, easy to follow and present a very important contribution to the use of p-XRF for rock art analysis. The processing of the data with different methods with the aim to understand biases that may be introduced by this method is very welcome at a time when such a method is now widely used without the required cares. Although focusing on the rock art case, the results presented here are relevant for the study of archaeological materials with p-XRF in general.

All data acquired through this study are available in supplementary materials, which allow any researcher to reproduce the processing. I thank the authors for this.

In conclusion, I recommend this article to be accepted with minor revisions listed below.

Review for « Critical evaluation of in situ analyses for the characterisation of red pigments in rock paintings: a case study from El Castillo, Spain », Dayet et al.

Dear colleagues,

It was a pleasure to read this article by Dayet et al. on their p-XRF analysis at the El Castillo cave. This article is well-written, easy to follow and present a very important contribution to the use of p-XRF for rock art analysis. The processing of the data with different methods with the aim to understand biases that may be introduced by this method is very welcome at a time when such a method is now widely used without the required cares. Although focusing on the rock art case, the results presented here are relevant for the study of archaeological materials with p-XRF in general.

All data acquired through this study are available in supplementary materials, which allow any researcher to reproduce the processing. I thank the authors for this.

In conclusion, I recommend this article to be accepted with minor revisions listed below.

Major comments

-Lines 97 to 116: This paragraph is very interesting and reflects the choices that are constrained by the use of raw materials and technics of applications. However, I would emphasis that a consideration is missing here: the type of binders chosen. Indeed, your binder has to be chosen in adequation to your raw materials, its texture and granulometry, and to the application technique. Although you mention the paint materials at large, I feel it may be needed to emphasise that both the colouring materials and the binders are of importance. I would also suggest elaborating a little more on “technical facts” (predictable gestures) and “degrees of facts” (unpredictable gestures linked to social behaviours): distinctions from Leroi-Gourhan’s theory on the “chaîne opératoire”. Indeed, the materials chosen, their preparation and the application techniques may be related to specific spiritualities or rituals and gestures adapted to fit the use of such materials/techniques.

Reply:

We thank the reviewer for underling the potential role of binders and cultural choices as factors influencing the paint production and application. We have changed our text as following:

Line 109: “In this regard, the choice of the binder is also an important parameter. Technical choices may be limited by raw material availability, depending on a geological context for the inorganic part of the paint mixture.”

Few sentences later we have added the following sentence:

Line 114: “Finally, binders and pigments are not independent parameters. The choice of one influence the choice of the other. They are also highly influenced by a society’s overall technical system (e.g. use of the same binder for different purposes) and the way in which the painter has been taught to mix and apply the paint. In other words, the cultural logic driving the “chaîne opératoire” (Leroi-Gourhan 1964) may play a key role in the final appearence of the painting.”

-Lines 118 to 137: The changes in the raw materials/paint recipes may also be related to trade networks and to exchanges of believes/behaviours. We may also consider that several groups with different behaviours came to paint in the same cave at the same “period”. The dating methods do not allow a very precise chronology for such very old periods, with nomadic groups moving in large areas. Hence, different paint recipes may only be related to several groups painting in the same cave with different techniques.

ReplyWe added a couple of sentences at the end of the paragraph to incorporate the point made by this reviewer.

Line 149: “In addition, paint recipes may vary as a function of changes in trade networks and patterns of cultural exchange. It is also possible that groups with different artistic practices came to paint in the same cave roughly during the same period. Available dating methods do not allow for the chronological precision required to investigate such sources of variation.”

-Although you were not able to analyse the same figures that d’Errico et al. 2016, I feel like links between the two are missing. A table may be added that summarize similarities and differences between your study and d’Errico et al. 2016.

Reply

This link is given in table 2. In this table we establish a connection between the disks from which come the micro-samples analysed in our 2016 papers and the disks studied in the present work.

-You included a yellow paint as part of your study but you stated several times that you will focus on the red paintings (line 79 for example). But at the end, you mention that you used the yellow painted figure to determine if it is distinguishable from the red paintings. I feel that this information should come at the beginning of the article and you may also state why you chose only one yellow paint: accessibility, thickness of the layer, chronology, other reason? This yellow painting is also not visible in your PCA figures. Was it considered? If so, I would suggest marking it with a different bullet in your figures. It will help the reader.

Reply

Only one yellow painting was analysed because yellow paintings are rare in the cave and most of them are very thin. In the revised version of the manuscript, we mention to the yellow painting in the introduction, as requested by this reviewer.

Line 81: “. To achieve this, we carried out microscopic examination and pXRF analyses of several red paintings and one yellow painting for comparison purposes, and used experimental paintings and statistical analyses to assess the reliability of the results obtained on the El Castillo paintings.”

All the figures were modified accordingly: the yellow painting is now represented with a different symbol on the binary diagrams.

-At lines 645-646, you quickly mention the variations within one figure but it is not visible in the rest of your paper and in your figures. I feel it would be useful to have more information regarding the reproducibility of the measures on one single figure to better appreciate the issues of differentiation between several figures. I would suggest adding a quick paragraph on this and maybe emphasising them for a couple of relevant examples in the figures.

Reply

In order to show that we did take into account this parameter, one sentence was added in ther result section at the beginning of sub-section dealing with PCAs.

Line 520: “In addition, results obtained on a single figure may feature substantial discrepancies. Several factors can explain these trends:”

I would like to emphasis a few more references for your study (for p-XRF analysis, and for the recognition of paint pots and their chronological implications):

- MacDonald, B. L., et al. (2019). "Hunter-Gatherers Harvested and Heated Microbial Biogenic Iron Oxides to Produce Rock Art Pigment." Scientific Reports 9(1): 1-13.

- MacDonald, B. L. (2015). Methodological developments for the geochemical analysis of ochre from archaeological contexts: Case studies from British Columbia and Ontario, Canada. Anthropology, McMaster University. PhD: 209.

- Trosseau, A., et al. (2021). "In-situ XRF study of black colouring matter of the Palaeolithic figures in the Font-de-Gaume cave." Journal of Analytical Atomic Spectrometry.

- Castañeda, A. M., et al. (2019). "Portable X-ray fluorescence of Lower Pecos painted pebbles: New insights regarding pigment choice and chronology." Journal of Archaeological Science: Reports 25: 56-71.

- Bonneau, A., et al. (2021). "Characterization and dating of San rock art in the Metolong catchment, Lesotho: A preliminary investigation of technological and stylistic changes." Quaternary International.

- Bonneau, A., et al. (2017). "A pigment characterization approach to selection of dating methods and interpretation of rock art: the case of the Mikinak site, Lake Wapizagonke, Quebec, Canada." Archaeometry 59(5): 834-851

- Sepúlveda, M. (2021). "Making visible the invisible. A microarchaeology approach and an Archaeology of Color perspective for rock art paintings from the southern cone of South America." Quaternary International 572: 5-23.

Reply

All the references were added.

Minor comments

Text:

- There are several lists of references for which I would suggest adding e.g. at the beginning. Although your review of the literature is very good, I was able to spot some more references (see above). To avoid very long reference lists, here are the lines with the reference list to which I would add e.g. at the beginning:

o Line 31

o Line 92

o Line 147

DONE.

- Lines 66 – 67: I would suggest presenting the U-series dates as follow, which is a more conventional way to report the age:

o 41.4±0.6 ka

o 35.7±0.6 ka

DONE.

- Line 163 – 175: on the use of in-situ Raman spectroscopy, I would suggest emphasising the possibility of burnt marks if the system is not used correctly. Although, the instruments are getting safer and the spot sizes are getting smaller, Raman spectroscopy remains a “risk of damage” if not used correctly.

We agree with the point made by this reviewer but think that adding this is out of the scope of our paper, which specifically focuses on advantages and disadvantages of pXRF. If we were to follow the logic he/she proposes, we would also have to discuss the possible damages produced by other techniques such as portable FTIR or close microscopic analysis of the wall. We think that this would be out of place in an article like ours.

- Line 180 – 181: amorphous carbon cannot be “identified” with p-XRF, it can be assumed if no Mn or Fe is present in the spectrum. It is indeed hair-splitting but this sentence may give false information to non-initiate readers.

The text was modified as follow:

Line 196: “Theoretically, several families of pigments can be identified with this equipment: iron-based yellow and red pigments, black manganese oxides, white phosphates, carbonates, titanium oxides. For black paintings, the absence of manganese and iron on the spectrum is an indication in favour of the use of pigments rich in carbon.”

- Line 330: you mentioned a red natural deposit in Sala de los Polícromos that you analysed. Is there any photo related to this natural deposit? How is it called in your data set?

We have changed out text to explain how the natural deposit is called in our dataset. The text was modified as follow:

Line 346: “That is why the number of substrate measurements is lower than the number of paintings analysed. In the Sala de los Polícromos, a red natural deposit (Wall red), was also measured

- Line 363: “paint pots” rather than “pot paints”.

DONE.

- Lines 707 – 711: I would suggest removing the references in your points. This conclusion presents the results of your study, which indeed confirm previous ones. However, as it is written, it looks like these conclusions are drawn from previous studies and not from this paper. You may emphasis more your own contribution to these previous studies. This paper is a really important one with the several ways of processing data and their critical evaluation.

DONE.

Tables and figures:

- Table 2: I would suggest aligning numbers of “p-XRF analyses” and “microscope” columns in the centre. It will make it more understandable.

DONE.

- Table 3 caption: I would suggest repeating the article in which the compositional data are extracted for Ref-Fe1.

DONE.

- Fig 10: The lines are difficult to understand and “hide” a part of the graphic. I would suggest adding a table on the right of the graphic to list the correlations between the paintings and the wall analyses.

It is true that the lines partially hide a part of the graphs but we found no other way to make the reader sees the links between points.

Submitted filename: Response to the Reviewers.docx

Click here for additional data file.

19 Dec 2021

Critical evaluation of in situ analyses for the characterisation of red pigments in rock paintings: a case study from El Castillo, Spain

PONE-D-21-22143R1

Dear Dr. Dayet,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Andrea Zerboni, Ph.D.

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

31 Dec 2021

PONE-D-21-22143R1

Critical evaluation of in situ analyses for the characterisation of red pigments in rock paintings: a case study from El Castillo, Spain

Dear Dr. Dayet:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Prof. Andrea Zerboni

Academic Editor

PLOS ONE