p-XRF analysis of multi-period Impasto and Cooking Pot wares from the excavations at Stromboli-San Vincenzo, Aeolian Islands, Italy

ABSTRACT This exploratory study focuses on the elemental analysis by p-XRF (portable X-Ray Fluorescence Analyser) of 62 samples of coarse wares, consisting of Bronze Age handmade burnished ware, so-called Impasto, and of Cooking ware (dated from the Roman period to Modern times). All wares originate from the site of San Vincenzo, Stromboli, and Aeolian Islands. The question addressed here is whether it is possible to differentiate between local (Aeolian) and imported (non-Aeolian) fabrics with the use of the p-XRF; 42 of the 62 samples were also subjected to petrographic analysis as a way of testing our hypothesis. Our results show that p-XRF analysis can clearly assist in distinguishing between Aeolian vs. non-Aeolian wares. Analyses can take place in the field and large quantities of sherds can be processed as a result. We suggest that no further demands should be made of the technique in providing answers to more detailed provenance questions. This is because finer separation in subgroups (as achieved recently by combined petrographic and EPMA analysis on select samples) is not possible given the nature of coarse pottery and the limitations of the technique in measuring key light elements (Na, Mg). Furthermore, for some elements (e.g Cr) accuracy is below acceptable levels in which case results for these particular elements are considered semi-quantitative.


Introduction
Stromboli, the northernmost island in the Aeolian Archipelago in the southern Tyrrhenian Sea, is in a strategic position for controlling the maritime routes, in particular the Messina Strait ( Figure 1). The new excavation at the site of San Vincenzo, directed since 2009 by Sara T. Levi in collaboration with the Soprintendenza of Messina and with the Parco Archeologico delle Isole Eolie e delle aree archeologiche di Milazzo, Patti e Comuni Limitropfi, has revealed a wide Early-Middle Bronze Age village (Capo Graziano facies. XXII-XV cent BC) with several oval stone-built huts inside a network of stone fences and terraces Ferranti, et al., 2015;Levi, et al., 2011Zhao, et al., 2015). The occupation of the site continues into historical times with evidence of burials, buildings and abundant pottery from the Classical period to the present.
The pottery from San Vincenzo (SV) is considerable and varied. It includes Impasto, Mycenaean, Black Gloss, terra sigillata, amphorae, cooking wares, glazed pottery and maiolica. This paper sets out to investigate the role of p-XRF for the purpose of preliminary characterisation and determination of provenance of the coarse pottery retrieved from the Bronze Age and later levels at the excavation site. The coarse wares are attested only by: a. Bronze Age handmade burnished ware, so-called Impasto pottery (Figure 2.1), the typical ware of central Mediterranean pre and proto-history (Levi 2010). b. Cooking ware (Figure 2.2), a wheel made pottery, sometimes glazed, produced from the archaic period onwards and mainly used for cooking. The chronology of these cooking wares at San Vincenzo is insecure, but what is known is that production during the Roman period took place in southern Italy, Sicily and Lipari (Bernabò Brea, et al., 1998).
In the course of 50 weeks of fieldwork (2009-2014) a total of nearly 4000 artefacts have been processed in our field-based laboratory at Stromboli. Focusing on the goal of the present work, more than 2000 Impasto and nearly 100 Cooking ware sherds have been individually examined and recorded (giving an average of about eight per day). Aeolian products have been satisfactorily differentiated from non-Aeolian ones primarily thanks to the petrography-based pioneering work of Williams (1980Williams ( , 1991. More specifically and  critically, Williams demonstrated the unique situation regarding prehistoric pottery, including impasto, in the Aeolian Islands: there was respectively locally (ie Aeolian) made pottery containing (local) volcanic inclusions, locally made pottery using imported clay to which (local) volcanic material was added, and imported pottery with inclusions derived from plutonic and sedimentary rocks. This complex pattern of production and circulation is what makes Aeolian pottery so interesting and why it is important to know the origin of the pottery in the Aeolian Islands. Since we cannot thin section every sherd there is a real need for an alternative approach -pXRF. Williams' early studies were complemented with subsequent further work (Cazzella, et al., 1997;Jones, et al., 2014, 235;Levi & Williams, 2003;Williams & Levi, 2008). As a result, it is now clear that the Aeolian Archipelago pottery production sites are characterised by their volcanic tempers and those in Calabria and/or Sicily by their sedimentary and metamorphic clasts.
Recently a petrographic study combined with micro-chemical analysis (by electron microprobe and ICP-MS) was carried out by Brunelli, et al. (2013) on 139 Middle Bronze Age (Capo Graziano facies) samples from various Aeolian Islands in order to further refine the above characterisation and to establish reliable reference groups of Impasto pottery based on temper. Capo Graziano vessels from different islands in the archipelago are not simply characterised by their tempers but also by their decoration in different patterns and styles . According to the Brunelli, et al., (2013) study, it has been possible to subdivide the Aeolian Capo Graziano Impasto pots into four "Temper Compositional Reference units": AI, AIV, AVIII, AX.
The AX group is characterized primarily by lava clasts (50-70%) with abundant plagioclase and less clinopyroxene, orthopyroxene and hydrous phases as shown in Table 2. The AIV group is mainly characterized by mineral phases and lava clasts with very low to negligible glass content. Minerals are plagioclase, clinopyroxene and scarce orthopyroxene. Hydrous minerals are represented by euhedral brown hornblende and subordinate green hornblende and biotite. Temper in group AI is characteristically rich of colourless fresh volcanic glass fragments and pumice, and less plagioclase and pyroxene phyric lava clasts (basaltic andesite and andesite) with cryptomicrocrystalline groundmass. AVIII unit contains large and abundant grains (up to 2 mm) of plagioclase, clinopyroxene and orthopyroxene, along with some pyroxene-plagioclase glomerocrysts, which represent a distinct feature for the group.
The question addressed experimentally in this work is whether Aeolian vs. non Aeolian coarse wares can be differentiated easily and faster using a field-based technique like p-XRF. For the purposes of the present work, 50 Impasto and 12 Cooking ware sherds were specifically selected to be analysed both with p-XRF (a total of 62) and petrography (a total of 42). We present the results of petrographic analysis first, and then those of p-XRF analysis.
Petrographic analysis 42 out of the 62 sherds have been examined by petrographic analysis and the results are shown in Table 1. Five Fabric Groups have been defined and are summarised as follows: -Volcanic Aeolian ( The composition of volcanic temper of the Volcanic Aeolian Fabric is typical of the pottery made locally in the Aeolian Archipelago. This consists of three of the groups -AX, AIV and AIpreviously identified by Brunelli, et al. (2013) and previously mentioned. The remaining fabrics (Granitic, Granitic+Micas, Micaschists and Siltstones) are characterised primarily by quartz (dominant phase) and are indicative of imported pottery (Calabria, Messina Strait area, or Sicily) due to the presence of minerals and rock fragments totally absent in the Aeolian Islands. The relative proportions of AX, AIV and A1 at SanVincenzo are 25%, 39%, 27% respectively with the remainder comprising AVIII (9%) which is the fabric exclusively produced in Filicudi and Lipari and is not attested in our dataset. Table 2 summarises the petrographic data and gives the suggested origin of each fabric. Regarding this point, the results evidence a local production in the single island, but also show an active intra-archipelago exchange network.

p-XRF analysis
During the last decade portable XRF (p-XRF) has revolutionised archaeological materials science and by extension archaeological fieldwork on account of its ability: a) to examine most archaeological inorganic materials (lithics, metals, glass, ceramics, pigments, and soils) and b) to bring technical analysis into the field (Shackley, 2011). Progress has been made in applying pXRF to the sourcing of ceramics, especially fine wares (eg. Jones & Campbell, 2016), but, unsurprisingly, it lacks the resolving power that techniques of destructive analysis can give. Although current concerns are rightly centred on defining p-XRF's limitations in the characterisation of ceramics (eg. Aimers, et al., 2012;Hunt & Speakman, 2015), pXRF has a role to play in the determination of origin so long as appropriately simple questions are asked of the data.
The present study, along with that at San Vincenzo on soils (Di Renzoni, et al., 2016), employed a NitonX3Lt-GOLDD instrument was a 50kVAgX-raytube, 80 MHz real-time digital signal processing and two processors for computation and data storage respectively. The instrument has three calibration modes suited to ceramic and soil analysis (TestAllGeo, Soils, Mining), of which the TestAllGeo (TAG) mode was chosen. Analysis time was set at 90 seconds and three measurements were taken on different spots on a cleaned, sawn surface of the sherd, prepared with a circular saw. The concentration of each of the twenty-five measured elements were averaged. Accuracy and precision were calculated by analysing six international standards GSP-2, AGV-2, DNC-1a, BCR-2, NIST2780 and NIST2709 and Edinburgh Standard Clay (this standard is described by Jones, et al., 2014, 527) and plotting the measured values against the certified values. The equation of a straight line was produced in Microsoft excel. From the regression analysis (calculation of the R 2 value) it was possible to ascertain the accuracy of measurement for each element. As shown in Table 2, not all elements were acceptable for the creation of bi-variate and multivariate PCA s using IBM-SPSS v. 21.
The Impasto and cooking wares have a coarse, heterogeneous fabric characterized by tempers of volcanic rocks (basalt and andesitic lava clast), intrusive rocks (granitic rocks), metamorphic rocks (mica-schists) and sedimentary rocks (siltstones), as described above. The Cr and the Na contents are the best markers of the local volcanic groups (Brunelli, et al., 2013). The other types of rocks (intrusive, metamorphic and sedimentary) have similar mineral constituent (mainly quartz) and the difference between the lithologies is mainly due to the presence and the composition of the micas. Of the elements -Al, Cr, Mn, K and Mg -that can best discriminate between these lithologies, p-XRF cannot determine Mg and Cr only semi-quantitatively. Table 2 presents the results of the regression analysis (R 2 ) Element determinations were deemed satisfactory if the R 2 value is >0.8. From Table 2 it is observed that Al, Mn, Ba, should not be considered when carrying out Principal Component Analysis (PCA) and that Cr, Zn and Nb should be accepted as semi-quantitative. In this way the dataset prepared for PCA was reduced to the thirteen elements shown in Table 3.

Results
The Sr-Zr plot ( Figure 4) shows a clear distinction between the Aeolian Impasto and the other classes. This is furthermore observed in the multivariate view ( Figure 5a) but it is not possible to separate the Granitic from the Granitic+Micas Impasto. A similar situation occurs for the Cooking wares, with the volcanic Aeolian examples standing apart from the non-local fabrics. It is significant that the discriminatory elements -Zr, Sr, Th and Nbare among those expected geochemically to differ between volcanic and non-volcanic environments.

Conclusions and Future Work
This exploratory study has highlighted that field based p-XRF analysis can provide good separation between the imported (non-Aeolian) vessels from the locally made ones. It cannot differentiate between different groups of non-Aeolian wares. This limitation is associated with the technique's inability to determine geochemically critical elements such as Na and Mg, its low level of accuracy for some elements, Cr being one of them. This element together with Na provided the best markers for the more finely resolved fabric distinctions made as a result of Brunelli et al electron micro probe analysis.
Furthermore, regarding the non-Aeolian Fabrics -Granitic, Granitic+Micas, Mica schist and Siltstonethe ability to discriminate them chemically has been undermined by their relatively similar mineralogical composition based mainly on quartz). What differences exist between the lithologies of these fabrics lie in the micas which in turn are governed chemically mainly by elements that, as already mentioned, p-XRF either cannot determine or determines with low confidence. Adopting one of the p-XRF's other calibration modes would not have alleviated this problem.
The message from this study is clear. p-XRF's role in characterising coarse wares such as Impasto and Cooking wares begins after their fabrics have been defined, either macroscopically or petrographically, p-XRF's considerable attributes then come into play providing in the field rapid analyses of the cut (sawn) surface of large numbers of sherds; this can be viewed as a 'screening' procedure generating chemical data which is scrutinised in the light of the fabric classification to make useful, if broadly based statements about identity, for instance precisely those elucidated in this studyvolcanic vs. non-volcanic or local vs. non-local. The chemical classification should not be expected to be amenable to a more detailed level of interpretation, paralleling the outcome of the other p-XRF  investigation at San Vincenzo on soils from different excavated contexts (Di Renzoni, et al., 2016). For sure, there is room for the analytical protocol to be refined beyond that described in the present study in order to improve accuracy, to allow additional elements to be included in the data set and to calibrate with respect to the corresponding data obtained by destructive analysis such as benchtop WDXRF (Jones & Campbell, in press). We suggest that p-XRF should not be habitually compared to ICP/NAA and repeatedly be found wanting in its inability to cover all elements and /or display equivalent levels of accuracy for all elements. At the same time p-XRF should not develop a 'scatter gun' reputation, generating one-off data which cannot be reused or can only be used for internalin-house -purposes. It is time to look boldly at its greatest attribute, that is, that of non-destructive, in situ, analysis aimed at analysing large quantities of sherds; equally important is our ability to formulate appropriate questions that it can answer, satisfactorily and conclusively. In the case of impasto ware there is already a good understanding of the origin and production of these wares; so the need now is to process large quantities of newly excavated material, and to assess whether they are imported or not. This study has shown that this type of assessment (local vs non local) is indeed feasible and effective. Coarse wares are amongst the most common fabrics found in many archaeological sites so the implication of the results of this study goes beyond the remit of our work on Stromboli.