Geochemical and petrographic assessment of clay outcrops and archaeological ceramics from the pre-hispanic site of Aguas Buenas (cal 400–1250 CE), Central Nicaragua

Abstract This research characterizes and reconstructs clay procurement and production practices through the integration of in-situ portable XRF and petrographic analysis on ancient ceramics and clay materials recovered from the Mayales river subbasin (central Nicaragua). A particular choice for this study was the largest and arguably most significant archaeological site in the area, Aguas Buenas (cal 400–1250 CE), a pre-Hispanic indigenous agglomeration consisting of 371 human-made mounds of various shapes arranged in geometric patterns. Microanalytical approaches were applied to reconstruct the use of raw mineral resources in the production of ubiquitous pottery materials found at this site and in its immediate surroundings. The resulting compositional analysis produced geochemical and mineralogical data allowing for the characterisation of distinct, geologically-based compositional groups throughout the valley, improving on the limited geological data resolution previously available. The integrated microscopic and compositional analysis (through p-XRF) of archaeological pottery materials and raw clay samples, generates a number of hypotheses and insights about the nature of the Aguas Buenas site, and its role as a shared space amongst groups living in the Mayales river Subbasin. Additionally, this study provides a solid research framework of investigation that can be employed for more detailed and extensive future studies on pre-Hispanic human occupation in this research area or elsewhere.


Introduction
This research aims to define ceramic provenance at a micro-regional scale through the application of both chemical and mineralogical techniques including the use of chemometric approaches. This data is crucial to characterise pre-Hispanic networks of clay procurement practices in the valley of Juigalpa within the Mayales river subbasin (Chontales region, central Nicaragua), with specific attention to the archaeological site of Aguas Buenas. Aguas Buenas is composed of 371 man-made mounds [1,2], making it the largest archaeological site with architectural remains of central Nicaragua, and the most extensive pre-Hispanic site documented in Nicaragua to date. The architectural features are geometrically arranged in an ellipsis that includes six concentric arcs, with a rectangular plaza in the centre. Preliminary studies of the site evidence a long biography from 400 to 1200 CE and 1400 to 1600 CE, or the Cuisalá, Potrero and Cuapa phases in the original local chronology [3,4], which is currently under review. Recent detailed mapping and spatial analysis of the site proposed different moments for the construction of Aguas Buenas, which combines various construction preferences resulting in both circular and linear arrangements, alongside clustered structures [2]. The area around Aguas Buenas has been the subject of intensive surveying and excavations in recent years [5][6][7][8][9][10][11][12], in the process of which a total of 1671 surface structures were documented distributed across numerous clusters, and in combination with rock art panels [13]. These 1671 mounds are clustered in 47 different sites in the research area (Fig. 1).
The selected sample of pottery fragments reported here is derived from a large ceramic assemblage excavated from Mound 1 (M1) at Aguas Buenas, which provided macro-fabric, micro-chemical and petrographic variability in a particular "sealed" excavation context estimated to date between cal 400 and 800 CE, according to recent research at the site [14]. As these materials were used in the construction of the mound-both as filling and possibly also as offerings-, they represent choices made over a brief period of time, and therefore represent a suitable opportunity for assessing variability in the human use of clay and ceramic resources at the moment of construction of rows of individual mounds to form concentric arcs-one of the most defining geometric architectural features of the site. In addition, a pedestrian survey, conducted in 2016, aimed to identify clay sources, offering reference materials critical to characterize the mineral resources present in the region [15], and the possibility to compare these with collected archaeological samples. Chemical analysis by means of portable X-ray fluorescence (p-XRF) combined with principal component analysis (PCA) showed clear chemical variability in clay outcrops, dividing the valley in main geochemical groups. In order to enhance initial results presented in Casale et al. [15] and characterise human clay procurement practices as well as technological manufacturing traditions at Aguas Buenas, mineralogical and textural analyses (thin section petrography) were performed, in combination with non-destructive chemical characterisation (by means of in-situ XRF analysis).
This research paper provides the first insights into pre-Hispanic ceramic production in central Nicaragua by integrating data from a pedestrian surface clay survey with microanalytical information. The petrographic and compositional analyses are conducted on the pre-Hispanic ceramic assemblage excavated from M1 at Aguas Buenas, improving the resolution on technological practices at, and beyond, the site. This study improves the capability to establish the provenance of ceramics retrieved at Aguas Buenas, and has a wider impact on the archaeology of central Nicaragua as it provides an initial reference framework for ceramic provenance studies in this region.

Geological context
The research area is located immediately east of the Central Nicaraguan Depression. The Nicaraguan Graben or Depression is the geologically youngest area of Nicaragua and is characterised by a mix of volcanic and sedimentary sequences of paleocene-Eocene age. The volcanic rocks in this area are mainly composed of porphyritic basalt, basaltic andesite, and andesite to dacitic pumice [16,17]. The physiography of central Nicaragua includes elevations, plains, large inland lakes, their shores and its islands, as well as the hydrological drainage ultimately leading to the Caribbean Sea [18]. The topography of the Mayales river subbasin consists of isolated hills with a plateau shape and soft slopes that are characterised by sharp endpoints. Undulating mountains consist of more recent volcanic rocks, associated with the Tertiary-Quaternary, Coyol group (Late Miocene through the Late Pliocene, possibly up to the Pleistocene); while the eroded slope corresponds to the earlier Matagalpa group (Late Oligocene until Middle Miocene) [16,18]. Several large clay deposits have been formed on these geological substrates throughout the entire region. Most of these clay deposits are typically located in the vicinity of rivers and reasonably close to Late Holocene pre-Hispanic settlements, and therefore were likely used as a direct resource for the production of pottery [15].

Materials and methods
Macro-fabric analysis, thin section petrography and chemical characterisation were applied to a selected dataset of archaeological pottery fragments. The analysis of this study is associated with the results of Casale et al. [15] that presented results of p-XRF analysis on raw clay samples collected in the Mayales river subbasin (n = 44 samples), and in the Zapatera island (n = 5 samples), situated in the Nicaraguan Graben, near the western shore of Lake Cocibolca.

Sampling of clay and pottery fragments
The ceramic assemblage was obtained through a sampling strategy applied to the entire ceramic collection excavated from Mound M1 at Aguas Buenas. In total, 874 sherds were selected for macro-fabric examination, of which 65% consisted of large fragments (>5 cm), while the remaining 35% were of smaller dimensions (< 5 cm). Most of the sherds were undecorated or featured a red or brown slip. The results of these macroscopic observations and the established diversity of ceramic paste recipes provided the main rationale for sampling for compositional analysis (see [12]). The description of the sherds was based on five characteristics that are visible with a magnifying glass (10x) and an optical microscope: inclusions, fracture, hardness, compaction, and colour of the matrix. As a result of the macroscopic analysis, 49 archaeological pottery fragments were selected for petrography and, within this group, a subgroup of 30 samples was chosen for compositional analysis. The main goal of the chemical characterisation was to specifically test the possibility of linking the composition of the, (mostly) very coarse, ceramic pastes retrieved at Aguas Buenas, to the available collection of clays, and be able to use this as a baseline and initial reference for future provenance studies and hypothesis testing.

Chemical and petrographic analysis
Based on the provenance postulate, mineralogical and chemical composition of a homogeneous group of ceramics is closely connected to the mineralogical and chemical composition of different clay outcrops used in a production area [19]. Different analytical techniques such as neutron activation analysis (NAA), inductively coupled plasma mass spectrometry (ICP-MS), X-ray diffraction (XRD) and X-ray fluorescence (XRF) either as Energy Dispersive XRF (ED-XRF) or Wavelength Dispersive XRF (WD-XRF) provide precise compositional results of ceramics and clays and also relating to other archaeological materials such as obsidian, glass, metals and flints [20][21][22][23][24][25][26]. However, with the exception of ED-XRF, which also exists in portable version (p-XRF), these techniques are destructive, requiring the extraction of a part of the sample. They are also time consuming and lab-based. As an alternative, p-XRF is non-destructive and expedient for use in the field. The major drawbacks, when compared with lab-based techniques, is that p-XRF has a lower precision to measure low-Z elements, due to the absence of a vacuum that is available for lab-based XRF. It also produces a lower energy X-rays, decreasing the range of elements that can be excited. For quantitative analysis, p-XRF technique has similar limitations to any other XRF devices with ideally the need of a smooth sample surface; a rather homogeneous specimen matrix; and calibration methods that are based on reference materials with matrix similar to the analysed material, allowing users to produce reliable data, in terms of accuracy, precision, and sensitivity. The portable and non-destructive nature of the instrumentation is a major advantage for further field studies with a suitable reference framework in place, however, data analysis needs to be carefully controlled and cautiously interpreted because p-XRF provides semi-quantitative data and it has been the scope of numerous research efforts to ensure reliable data [27][28][29][30][31][32]. In this research, chemical analysis was carried out by p-XRF, which has been used as an explorative and especially non-destructive technique for provenance studies [27].
Prior to the analysis, samples were broken and, subsequently, the area of the fresh cut was polished and smoothed with sandpaper to create a clean even surface. To optimize data reliability for each sample, three measurements in different areas were taken to compensate for possible heterogeneity. In general, the ceramics themselves do not show a macroscopically heterogeneous matrix, and occasionally visible larger grains were avoided. As it is shown in Section 3.1, the texture of the selected samples is mostly medium coarse, and the average size of inclusions is < 0.5 mm. These characteristics increase the potential of testing for chemical identification of the assemblage, and to compare the results with petrographic observations. The chemical analysis was conducted using a Bruker Tracer III-SD p-XRF device. The time of analysis was 120 s with 40 kV and a 14 μA. An Al-Ti-Cu filter was fitted to enhance the instrument's sensitivity in measuring mid-Z trace elements (Rb, Sr, Zr, Y, Nb), following the same methodology used by Casale et al. [15] to analyse the clay outcrops. Semi-quantitative results were obtained utilizing a custom empirical calibration for ceramic and soil materials. Quality control during analysis was monitored utilizing five rock and soil certified reference materials. These international standards are SRG-1 (Green River Shale), BIR-1 (Icelandic basalt), GSP-2 (Granodiorite Silver Plume), BCR-667 (Estuarine sediments), and CRM NIST-98b (Plastic clay). The following elements were obtained from the analysis: K, Fe, Ca, Ti, Cr, Nb, Ni, Sr, Rb, Y, Zn, and Zr. The elements Fe, Ca, Ti, Cr, Sr, Zr, Rb expressed a R 2 > 0.90, as an assessment of accuracy, and were considered for further bi-plots and statistical analysis. Precision-both repeatability and reproducibility-of the measurements was controlled at several instances by replicated analyses and was regularly assessed through the calculation of the relative standard deviation (RSD/%RSD) [33]. All selected elements generated %RSD values below 10% RSD. A table with measured vs. certified values can be found in the supplementary material.
In order to compare and interpret the chemical data of the ceramics, we created bivariate diagrams and additionally used a multivariate statistical approach as classification tools. The use of the multivariate statistical calculations allows for isolating and identifying chemical groups within the assemblage [34]. In particular, principal component analysis (PCA) was applied to compare clay raw materials with ceramics and assess possible provenances. Discriminant analysis (DA) using Squared Mahalanobis distances was further used to test the groups created as a result of the petrographic observations and compositional analysis. PCA is additionally a powerful tool to identify which elements contribute to the final composition of the sample. This technique allows for reducing the multivariable dataset to two dimensions, and they are frequently employed to identify subgroups as well as to verify groups created through petrography, stylistic features, or other approaches [35].
For optical mineralogy, thin sections of ceramic fragments were prepared and their structure and composition were observed with a polarizing microscope (Leica DM750P), which uses transmitted planepolarized (PPL) or cross-polarized (XP) light. The petrographic analysis aimed to characterise the two most important components that constitute a ceramic material: the clay matrix and the non-plastic inclusions. Moreover, it is possible to observe the characteristics of voids, pores, and the presence of slip or paint layers, as well as other types of surface treatment applied by potters [36]. The interpretation of these features in comparison with the geological context of the archaeological site from where the materials were excavated, allows us to formulate hypotheses both about provenance, as well as to discriminate materials formed from different mineral resources [37,38]. Furthermore, thin section analysis provides information on the employed technological procedures involved in the manufacturing of pottery [37]. For instance, through the observation of the birefringence of the clay matrix, it is possible to estimate the firing temperature at which the earthenware was fired. The firing process causes the sintering of the clay components and mineral/rock inclusions, with a loss of 'optical activity' of the matrix. Higher temperature induces important structural changes and the matrix becomes anisotropic, with inclusions that melt and vitrify, creating a more glassy aspect. On average, a clay matrix loses "optical activity" between 800 and 850°C. Thus, samples that yield an optical clay matrix are considered to have been fired at temperatures of < 800-850°C, while those having an inactive matrix are usually fired at temperature > 800-850°C [37]. Photomicrographs were taken with a factory-built Leica camera at the Faculty of Archaeology at Leiden University.

Macroscopic analysis
The assemblage examined included sherds representing different parts of vessels, such as bodies, necks, and rims. Analysis of technological macro traces showed a fashioning technique without the application of rotary kinetic energy, on assembled elements-coils-implying gestures of digital and hand pressure through pinching, drawing, and slight crushing. No percussion was identified, aside from some traces of discontinuous palmar beating. In particular, coils were majorly equidistant, placed alternatively from inside to outside and vice versa. Fashioning was highly homogenous within the entire sample and within the various parts of the vessels, with small variances in coil size, which ranged from 1-2 cm. Pre-forming techniques featured a higher variability than in fashioning, with both leather hard (shaving) and wet clay techniques (scraping). For finishing, the microtopography showed horizontal striations linked to brushing and/or smoothing techniques on both wet and leather hard clay. The most commonly applied surface treatment was burnishing; decoration mainly consisted of slips and paint, with a wide range of colors but mostly red, black, red on orange, red on orange on white, red on white, and brown on orange. The high degree of erosion found, together with post-depositional layer attached to the sherd walls and the low firing temperature in some sherds, made macro-trace analysis extremely challenging.
The macroscopic analysis of pastes revealed that inclusions found in the ceramic pastes were coarse, measuring approximately 0.5-1 mm, and their colours varying from white, red, and brown to black. 2% of the sherd assemblage has a highly fine grain matrix (< 0.1 mm), which can be considered as an outlier cluster. In general, the main differences were in hardness and colour. Variability in the core-margin relationship, cross-section and surface Munsell colors suggests that firing practices involved different temperatures, duration of firing, and position of the vessel (or vessel part) in relation to the fuel source-possibly an open hearth firing context. There may have been an intentional change from oxidizing to reducing firing conditions, based on the current observations. Nevertheless, partially oxidizing firing atmospheres were observed frequently, as identified by a characteristic "sandwich" colouring in the matrix [39].
A total of 24 homogenous macroscopic clusters were identified, featuring clay recipes that mainly contain feldspars, dark mica, rock fragments, and quartz. These sets of inclusions were repeated in different sizes and frequencies, resulting in a homogenous assemblage comprised of several groups. The orientation of the inclusions was subparallel, concentric, and oblique, and chaotic, which is consistent with the coiling technique for pre-forming vessels [37]. Voids were generally of plate-like, oval-sphere, and irregular shapes, and their orientation was also subparallel, concentric, oblique, and chaotic. These macro-groups defined according to their paste characteristics were then used in the secondary sample selection for thin section and compositional analysis.

Petrographic analysis
A total of 49 sherds was selected for petrographic analysis and subdivided into six petro-groups. A detailed overview of the six petrographic groups identified, and their characteristics, is provided (Table 1, and Figs. 2 and 3). Most of the inclusions are connected to a relatively homogeneous mineral composition derived from a basic intermediate igneous parent rock, and in some cases, there are samples with a more mafic (AB-N-45, AB-L-68, AB-D-45) or more andesitic (AB-H-68 and AB-S-78) composition. The ceramic assemblage does not show evidence of calcareous rock inclusions visible through the optical microscope, however, calcareous microscopic inclusions might be present in the matrix as is pointed out by the chemical analysis (see below). The most common inclusions that can be related to these mafic/intermediate rocks are plagioclase, basalt, andesite, pyroxene, olivine, and rarely biotite, as well as hornblende grains. The mineralogy of the inclusions present in the ceramic body is consistent with the geology of the area, which is mainly characterised by Oligocene and mid-Miocene rocks [16].
From a technological perspective, the interpretation of the size, shape and distribution of the inclusions within the matrix suggests that the majority of ceramics were probably not tempered by the intentional grinding of rocks. However, adding temper by re-using ceramic materials (i.e. grog) was an identified practice. Moreover, the matrices of the sherds show evidence of possible mixing of different clays, with at least two clearly distinguishable pastes noted in the same fabric ( Fig. 2, caption 11). In particular, it seems that a lighter, more calcareous fine clay matrix might occasionally have been added to, as indicated, for instance, by samples AB-C-74, AB-A-71X, AB-M-68X, and AB-D-29, however, there is no evidence of medium-coarse rock inclusions of calcareous origin.
The mineralogical characterisation of the samples, examined through thin section petrography, yielded four main compositional clusters (Chontales A, B, C and D). Both petrofabrics C and D have been subdivided to incorporate systematic variations with the respective groups. In total six petrofabrics have been characterized and described.

Chontales A
This fabric is characterised by an orange-brown optically active matrix with an estimated firing temperature lower than 800-850°C [37]. The edges of the samples feature traces of a post-depositional calcite layer and, in a few cases, also a red line below this surface, which may be a trace of finishing and/or surface treatment techniques. The aplastic inclusions show significant variations in dimension. Monocrystalline subangular to subrounded quartz and plagioclase (euhedral and non-euhedral) are the most common inclusions (150-500 μm). There are also medium-coarse sized inclusions (150-600 μm) of chert, granite, basalt and andesite, which vary between subangular and subrounded shapes. The other, much more rare, aplastic inclusions are olivine, orthopyroxene, clinopyroxene, and biotite. The grain size is up to 600 μm, and in a few cases, there are coarse subrounded inclusions of clay pellets and rare grog particles of up to 2000 μm.

Chontales B
Fabric B is characterised by a dark reddish brown optically inactive matrix, suggesting a firing temperature above 850C. The non-plastic inclusions are well-sorted and characterised by a bimodal distribution of fine grains with one range of 150-250 μm and a second range of < 150 μm. Monocrystalline plagioclase and quartz are the most common inclusions. They are characterised by subangular and subrounded shapes. There is a limited number of rock inclusions such as granite, basalt, and chert. Andesite is very rare in these samples. The size of rock inclusions ranges from 150 to 250 μm, with a few isolated inclusions exceeding 500 μm. The other aplastic inclusions are opaque minerals, orthopyroxene, clinopyroxene (augite), and olivine. Their quantity is much lower (< 5%) in comparison to quartz and plagioclase, which comprise 70% of the total inclusions. As with Chontales A, there is a low number of clay pellets and very coarse grog (1000-2000 μm).

Table 1
Summary of the petrographic analysis.
An orange-brown to dark reddish brown non-calcareous, slightly optically active. A dark brown matrix to dark greyish-brown, non-calcareous.
From optically inactive to slightly active.

Metamorphic Rock Fragments
n/a n/a n/a n/a n/a
K-feldspars and plagioclase, angular and sub angular.

Chontales c
Petrographic Fabric C can be defined as the major group in the study. It contains, however, systematic variations that yielded a further subdivision of the ceramics attributed to this petrographic group.

Chontales C-1
This fabric is composed of a non-calcareous orange-yellowish brown slightly optically active matrix, suggesting a firing temperature lower than 800-850C. The outer surface shows traces of a post-depositional layer of calcite. The aplastic inclusions are tri-modal, ranging from 150 to 600 μm. This is a significant size variation, which could be interpreted as an absent or poorly executed sorting strategy. The most common mineral inclusions are quartz, feldspars (k-feldspar and plagioclase crystals) and chert with angular, subangular, subrounded and rounded shapes. There are also minerals of volcanic origin such as orthopyroxene, clinopyroxene (either augite or diopside), olivine, and a very low quantity of large crystals of biotite and hornblende. The rock inclusions are composed of granite and andesite. Basalt occurs in minor quantities and dimensions. A main discriminator is the presence of clay pellets and grog in smaller dimensions (300 and 400 μm) when compared to fabrics A and B.

Chontales C-2
This subgroup consists of a more darkish brown to greyish-brown non-calcareous matrix. The matrix ranges from optically active to inactive, suggesting different firing temperatures from 600 to 800C. Once again, a post-depositional calcite layer covers the surface of the samples a characteristic shared with groups A and C-1. Non-plastic inclusions vary in size, ranging between 150 and 600 μm. Quartz and plagioclase are the main inclusions, followed by granite. There are large coarse inclusions of chert in variable shapes, from subangular to rounded. The other rock inclusions present are basalt and andesite, and these vary from fine to coarse (150-650 μm). Less frequent aplastic inclusions are orthopyroxene and clinopyroxene (augite and rarer diopside), opaque minerals, olivine and (very rare) hornblende and biotite. There are a few isolated cases of large clay pellets (up to 1300 μm), with the average size range being between 300 and 400 μm. The group does not present grog grains.

Chontales D
A further group of samples that shares both technological, matrix and mineralogical features but also exhibit a few differences is group D. Therefore, a subdivision in Group D-1 and D-2 can be defined where D-2 is considered to be the result of mixing multiple raw clay sources, based on the presence of mixing lines in samples (Fig. 2, caption 11), with one of these sources being highly similar as that of the source of Group D-1. The optical activity of the ceramic matrix of these two groups suggests that the vessels were fired at a relatively low temperature < 800-850 C.

Chontales D-2
The fabric is characterised by a light yellowish-brown matrix, which can be identified as a little more calcareous when compared to groups A, B and C. The distribution of the inclusions is trimodal, characterised by medium and coarse inclusions, which range from 250 to 1000 μm and some very coarse grains measuring between 1200 and 1600 μm. There is a high abundance of coarse chert grains (300-1200 μm) of subangular and subrounded shapes. The other rock inclusions are granite, basalt and andesite from a subangular to subrounded form, which are marked by a significant variation in size (200-1000 μm), suggesting a natural presence of these rocks in the raw sources rather than the addition of temper. Quartz and K-feldspars are the most common inclusions and have a significant variation in dimensions. The quartz pieces vary between 100 and 650 μm and are mostly angular to subangular. The K-feldspars have dimensions between 500 and 1500 μm, while plagioclase, being the most abundant type, measures 150 μm on average. Sanidine is present in minor quantities with dimensions between 700 and 1000 μm. The other inclusions that are present in minor quantities are opaque minerals, orthopyroxene, augite, biotite, and olivine. As in groups A and B, and D-1, grog fragments are also present and vary in dimension (400-500 μm and 1100-1600 μm). In contrast to groups A and B, clay pellets of very coarse size were not identified.

Chemical analysis
The results of the chemical analysis for the 30 sherds are integrated with the petrographic grouping highlighted in the previous paragraph and shown in Table 2 and Table 3. The scatterplot that compares CaO and Zr (Fig. 3) displays how the values in Zr tend to be heterogeneous throughout the dataset, while there is a correlation between the values of Sr and CaO, which can be associated with the carbonate component  of the clays. In contrast, Zr accumulates particularly in coarse and heavy inclusions as a result of high-energy environments, such as river sedimentations [40]. As illustrated in the charts (Fig. 3), the petrographic groups are not unequivocally reflected in the chemical composition through these single element observations. In particular, Chontales A (AB-G-71, AB-A-60, AB-P-71, AB-S-78) is characterised by a high correlation in Sr and CaO and a low correlation in Zr (Fig. 3). Samples AB-G-71 and AB-P-71 were also grouped together in the petrographic analysis. The values of Sr and CaO can be associated with the presence of feldspars such as k-feldspar and plagioclase, or their weathering products [41] since the carbonate content was exceedingly low in this type of ceramics. These two samples also contain low presence of basalt inclusions. Cr can generate issues in the analysis with p-XRF [30], and it was not included in further statistical calculation (see further in the discussion). Chontales B has a composition that can be connected to Chontales A, and only the Cr values are around the detection limit of the p-XRF.
Samples belonging to group Chontales C-1 show lower values in Sr and CaO, which can be related to high levels of sediment weathering, the mobility of these elements, or rather, the presence of plagioclase, as well as carbonates. The high values in Fe are confirmed by the presence of orthopyroxene, basalt and iron-rich inclusions in samples AB-N-45, AB-L-78, and AB-L-68. These three samples were already grouped together in the petrographic analysis. A similar composition is also shown by the sherds in group Chontales D-1, which are characterised by a high correlation of Fe, Ti, and Ca. The first two elements are related to the presence of pyroxene minerals while calcium is associated with the plagioclase. Groups Chontales C-1 and C-2, in contrast to the previous groups (D-1 and D-2), are enriched in Sr. The lower value in Fe for samples AB-T-60 and AB-T-83Q can be linked to presence of granite and the low number of rock inclusions (e.g. andesite) and the total absence of basalt in contrast with other ceramic samples. For sample AB-E-68X, this low value can be connected to the matrix, which is characterised by a yellowish colour and considered an iron poor clay.

Discussion
The correlation of the petrographic groups with the compositional analysis through the observation of single element values and element ratios was not unequivocal in attributing discrete sources. It is important to highlight that when such coarse grain ceramics are analysed together with the temper added during the manufacturing process, as in this case study, nonplastic inclusions may have a critical impact on the chemical composition results. The coarse grain size can obscure the presence of other minerals or the composition of the clay matrix which in extremis could influence the compositional results. It is thus essential that any potential problem is taken into account in discussing the p-XRF results. Multiple readings on each of the samples were collected in order to mitigate part of this potential variation, as the aim of the analysis is to delineate systematic differentiations within the ceramic assemblage.
The results of the clay survey of Casale et al. [15] demonstrate how the majority of the suitable clay outcrops are located near the banks of rivers, as a result of fluvial sedimentation. It is likely that differential fluvial and alluvial depositional processes throughout the valley, which vary from seasonal streams to perennial rivers (such as the Mayales river), may have had an impact on, for example, the accumulation of Zr and by extent heavy minerals (such as zircon) [40]. This results in a variable elemental composition within the raw materials employed for the preparation of the potterys. The scatter-plot of Ca and Sr (Fig. 3) shows that the two elements are partly correlated as expected, suggesting that the presence of Sr in the ceramics can be associated to both a microcrystalline (carbonate) composition of the raw materials, but more significantly relates to the feldspar component (and particularly plagioclase) of the fabric groups [40].
In order to successfully compare and integrate the analytical results of both clay and ceramic materials, we applied a multivariate statistical methodology for group assessment and variability. PCA and DA were calculated in SPSS using Fe 2 O 3 , CaO, TiO 2 , Sr, Zr and Rb. Cr and Nb were not included to not mislead the interpretation of the calculation. The first has a high variability through the assemblage while the latter is present in very low quantity (on average < 5 ppm). DA were determined to assess the validity of the ceramic groups created through the petrographic observations. Results demonstrated that 79.2% of the specimens were grouped correctly as with the petrography. Group C-1 has one sample (AB-G-71) that was assigned to C-2 and C-2 has one sample (AB-A-60) assigned to C-1. The same occurred with D-1 and D-2, where AB-A-71Y was assigned to D-2 and AB-H-68 was connected to D-1. Overall, the DA gave a positive result demonstrating that only five samples out of 30 were re-grouped differently, and four of that, were re-grouped between subgroups of the same main groups. Geochemical similarities are expected between subgroups of the same petrographic groups, due to the presence of similar mineral inclusions.
A PCA bivariate biplot was calculated including all samples, both reference clay data (n = 44) from Casale et al. [15] and representative sherds (n = 30), in order to understand possible connections between the regionally available clay raw materials and ceramics from Aguas Buenas. An important result of the statistical calculation is that, in contrast with the two-elements plots, when more significant elements are analysed (Fe 2 O 3 , CaO, TiO 2 , Sr, Zr, Rb), the petrographic groups identified show a consistent internal geochemical composition. Table 4 shows the results of the compositional analysis for the clay samples, the results of the grouping made by Casale et al. [15], and the ceramics linked to the possible clay sources, following chemometric approach through PCA. Fig. 4 graphically illustrates the comparison through PCA, in which both soil and ceramic samples are shown according to their compositional groups for clays and petrography for ceramics. The majority of the ceramic samples cluster together parallel to petrographic grouping but diverging from some of the clay outcrops analysed. Overall, petrographic and chemical data of the ceramics provides consistent observations and confirm the presence of different resource areas. In order to further test the reliability of the groups identified in PCA, Squared Mahalanobis distances were calculated to assess predicted and statistically assigned groups. Results indicated that 75% of the ceramics and clays were assigned to the same clusters as those evidenced in PCA. Generally, some overlap exists between Chontales A, D-1 and D-2 arguing for chemically related clay outcrops which can however be clearly separated through optical microscopy. Chontales B is clearly discerned from the rest of the ceramics, both mineralogically as well as chemically. Chontales C1 and C2 on the other hand overlap more significantly and can be considered chemically consistent which suggests the difference between C-1 and C-2 would be rather production related.
A positive attribution can be made for six clays outcrops (D12.1, E5.1, M3.2, M3.1, C12.1 and N4.1) which cluster with 13 ceramics in three major groups. These results propose that some of the ceramics retrieved in Aguas Buenas were likely produced with local clay readily available in the surrounding environment. The outcrop M3. Generally, both ceramics and clays yielded low-calcareous crystalline matrices associated with basalt and andesite parent rocks. Fig. 4 shows the resulting geographic distribution of clay samples that were associated with Aguas Buenas`ceramics, together with the location of the other mounds in the valley, and the locations that yielded surface ceramic materials. Table 4 lists these associations of the ceramic samples with the corresponding mineral resources. Regarding the mineral resources, Group 1 (G1) includes only the (clay) samples from the Zapatera Island, yielding systematic higher values of TiO 2 , Cr, and Fe 2 O 3 , in contrast to the other groups. DA considers this group independent from the rest of the assemblage with 100% of correctness. The Zapatera Island is a shield volcano, which is characterized by basaltic/andesitic Table 3 Average and standard deviation for the petrographic groups.   Table 4 p-XRF results for clay samples expressed in wt.% and ppm, together with the geographical coordinates (UTM, WGS84), extension and depth expressed in m for each clay outcrop, list of the clay samples clustered in groups according to PCA [15], and list of the petrographic groups of archaeological ceramics clustered to clay outcrops according to PCA.   [15]) and the six petrographic groups for ceramics. At the bottom Geographic distribution of clay samples associated to Aguas Buenas (red dots) combined with the distribution of mounds (smaller black dots) and different intensities of surface ceramics.
for instance, has the finest matrix in the assemblage. The fine size of the inclusions can be the result of sieving practices that eliminated coarse grains affecting the final geochemical composition and workability of the matrix, which strongly diverge from the rest of the assemblage. Recent observations have connected these finds as part of the Sulaco valley technical complex in northeastern Honduras (R. A. Joyce, personal communication, October 2019), which would further explain not only the wide difference in petrographic and geochemical values, but also in aesthetic perceptions. Petrographic groups Chontales C (C-1 and C-2) as well as Chontales D (D-1 and D-2) show a similar internal chemical composition as demonstrated by PCA and DA, supporting the petrographic observations. Looking at the extension of the clay outcrops, the clay deposits that yielded the highest similarity with the archaeological ceramics (M3.2, M3.1, D12.1, C12.1, N4.2, E5.1) are also the largest in extension [12,15]. It is indeed likely that populations in pre-Hispanic times quarried those sources, although a more precise identification of specific outcrops along this river system, or its tributary streams and creeks, is currently not possible.
Furthermore, the combination of data regarding clay outcrops location and specific archaeological site location in the valley provides interesting data that clarify on human-landscape interaction and mobility. Each clay outcrop has archaeological sites in the vicinity, often within less than 2 km (Fig. 4). However, these sites were not contemporaneous to Aguas Buenas, dating to a later period [14]. This enables further studies on the ceramic materials retrieved from these sites, and to understand possible connections with Aguas Buenas.
Consequently, the differences in sizes between documented exposed clay areas seem to be relevant in clay procurement choices. If we consider a long-term use of these natural resources, then it is possible that inhabitants of the area looked for the largest and perhaps the most sustainable sources available, in order not to rapidly exhaust them [42]. Large quarries were not employed for massive production, since we have no evidence for such manufacturing practices in the history of the Subbasin. Therefore, we can also propose that the selection of larger deposits ensured the continuity of socially learnt clay procurement practices over time. Such use strategies were likely shared amongst different communities of practitioners.
The optical microscopy analysis outlines the presence of different paste groups and, in some cases, the technological choice of mixing more than one clay for producing vessels. Clay mixing can be due to several reasons, from simple habit to liquidating older supplies [43], but mixing can also aid in achieving specific desired features such as plasticity, firing resistance and/or dimension of the pots [42,[44][45][46]. In order to control these features, potters can mix different clay sources for improved results. The compositional and petrographic analyses show that there are some pots consistently made with more fine grains (such as AB-T-78A, AB-T-60, AB-T-83Q), and others with coarser grains (for instance Chontales D-1 and D-2).
Clay outcrop M3.1 is the closer one to Aguas Buenas (< 3 km), has the largest surface area (100 × 100 m), and also matches other samples, although to a lesser extent. Outcrop N4.2 is located at less than 1 km distance from outcrop M3.1 and approximately 5 km from Aguas Buenas, suggesting the possibility of mixing more clays to reach a better workability of the paste, as was highlighted by the petrographic observation specifically for groups Chontales D-1 and Chontales D-2. Therefore, clays from M3.1 and N4.2 may have been employed by the population and strategically mixed with other clays, or tempered, reflecting different technological choices and production practices.

Conclusions
Since 2009, the detailed archaeological investigation of the Aguas Buenas monumental site by the Proyecto Arqueológico Centro de Nicaragua (directed by Geurds), and its relatively unique form and dimensions, has opened a range of questions concerning the sites' construction, associated practices and wider importance in the Mayales subbasin. As part of this corpus of studies, we have here analysed practices related to clay procurement and pottery-making, using mineralogical and geochemical methods. The integration of both optical microscopy with non-destructive XRF analysis provided the necessary basis for building an interpretative framework for provenance. The petrographic analysis demonstrates that, even when excavating a single mound (M1) and within a similar volcanic geological substrate, sherds show considerable mineralogical differences, leading to clear divisions into different provenance groups. The further use of the p-XRF as a connected chemical field method, confirmed the presence of systematic compositional groups. Results are promising and suggest the existence of ceramic materials that were manufactured with raw clays from diverse local origins as well as materials produced with clays from outside the Rio Mayales Subbasin. The statistical analysis indicates that six clay outcrops, compositionally matching the ceramics from Mound M1 at Aguas Buenas, are located near other clusters of mounds, zones of highintensity surface ceramics, but mostly with larger clay deposits also being available nearby. In particular, the results show that, even though there are suitable clay outcrops within a 1 km radius of the site [15], raw materials are chosen from outcrops between 3 and 5 km away. This is relevant when considering the people that built Aguas Buenas, and their relationship to the surrounding landscape. The exploitation of several raw material sources simultaneously, for example, implies a detailed knowledge of the territory, as well as pointing to several different groups engaged with pottery-making activities.
The methodological framework entailed a combination of petrographic and compositional analysis of ceramic materials and clay samples, generating information to support a number of conclusions about the nature of the pre-Hispanic Aguas Buenas site and its role amongst people living throughout the Mayales river Subbasin. Also, it shows variability in pottery production operational sequences, from clay procurement -andesitic vs. basaltic clays-, clay preparation-tempering, mixing, and sieving practices-to surface treatment-smoothing and slipping gestures.
It is likely that additional systematic and large-scale archaeometric analysis based on, for example, more high resolution compositional techniques (such as for example XRD, INAA, and ICP-MS), applied to cultural materials from sites spanning the entire Subbasin, will further clarify the extent of connectivity between groups in central Nicaragua. XRD analysis, for instance, allows to characterize the presence of different clays in the same ceramic body, clarifying further on the technology and clay preparation strategies. Ultimately, this research provides a first framework of investigation that can be employed in future extensive studies on the Mayales river Subbasin, and the chosen methodology holds potential for its application in other regions as well.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.