Microscopic and Chemical Characterisation of Metallurgical Residues from an Early Roman Site at Haven Farm, Sutton Valence, Kent

Archaeological excavations by Archaeology South-East at Haven Farm, Sutton Valence, Kent identified the remains of three furnaces and recovered a significant deposit of Roman iron- production residues including slag and furnace construction materials. The entire assemblage was submitted to the Archaeomaterials laboratory for analysis and a sample of 29 specimens representing different materials and slag morpho-groups were selected for analysis by scanning electron microscopy and X-ray microanalysis. Results indicate that specimens are the product of optimised iron-smelting activities making use of local resources for the production of ferritic iron.

INTRODUCTION

Rome was able to take over and scale up iron production in south-east England following the invasion of AD 43. The Weald was amongst the first and largest iron-production districts established by the Romans, though there are other traces of Roman iron production in northern Kent. The excavation (sitecode: SVH11) provided evidence of early Roman iron- production that offers new insight into the earliest phases of production and its organisation. The iron produced at this and other early sites were important for the construction industry in Roman Britain and supplying the military, although the patterns of distribution remains unclear. Investigations of metallurgical residues can shed light on technological practices as well as the organization of production and socio-economic behaviour.

Approximately 25 kg of metallurgical residues from the upper fill ([219]) of furnace G29 were submitted to the UCL’s Institute of Archaeology archaeomaterials laboratories for characterisation. The first step of this analysis was to characterise the material to assess what technological practices had taken place and results demonstrated that their character was consistent with primary iron-smelting intended to produce a soft ferritic iron. Production was optimized to maximize production of ferritic iron without risking yields of steely iron.

TECHNOLOGICAL FRAMEWORK

Slag and technical ceramics are the most common finds associated with metallurgical activity, but differentiating metallurgical processes can be challenging (Dungworth 2015). Blast furnace slag is perhaps the easiest to identify by its colourful glassy chunks; copper slags can sometimes be differentiated by their green carbonate corrosion products; lead slags by the presence of red oxides and glassy matrix; and iron slags by the presence of hydrated [pg1]iron oxides (rust). The absence of rust absence or other such indicators are not, however, reliable indicators on their own. Similarly, most pre-industrial metal slags also possess microstructures dominated by olivines, with varying amounts of iron oxide and glass phases. Identification of prills included within the slag becomes the most important discriminator of metal production.

Technical ceramics tend to be made from local clays that may or may not have quartz or other tempers added to enhance strength and refractory properties. Identification of process from ceramic fragments relies to some extent on their morphology, but more on the presence of slag and metal residues clinging to them.

Identifying iron-production residues presents additional problems as in pre-industrial Europe it consisted of two main tasks: smelting and smithing. Until the late Middle Ages, iron-smelting technology relied on the direct or ‘bloomery’ process that yielded iron in the form of an unconsolidated mass of metal entangled with slag. Iron was reduced from its ore in the solid state with the remnant non-reduced compounds forming a liquid ferrosilicate slag. The resulting iron tended to be a soft ferritic variety, although carbon-steels and phosphoric iron were sometimes produced. The morphology of the slag produced during smelting reflects a system combination of thermochemistry, furnace design and operator skill. The frozen slag’s microstructure is dominated by olivine (especially fayalite, Fe2 SiO4), wüstite (FeO), glass and occasional minor phases (including metallic prills). Unlike non-ferrous metals, bloomery iron could not be cast and instead was refined, shaped and repaired through a process called ‘smithing’. The same is true for wrought iron (de-carburised cast iron). Smithing involves repeated cycles of hammering and heating, each step leading to a more refined product as well as new slag. Much of the primary smelting slag would be blasted away from the hearth by repeated hammer blows forming small drops and spheres. Other slag was formed by the interaction of oxidised iron surfaces (hammerscale) with hearth materials and fuel ash. These slags, under the right conditions, would take the shape of the smithing hearth basin and are termed ‘plano-convex bottoms’ or PCBs. However, it could also be produced in amorphous lumps with oxidised surface skins.

The microstructure of smithing slag is similar to that of smelting slag, but often containing much higher densities of wüstite and the presence of magnetite (Fe3O4) (Eekelers et al 2016). Magnetite is indicative of hot oxidising conditions typical of smithing hearths. Smithing slag also tends to have a higher void density, possess more inclusions of earthen materials and charcoal, and contain more metal (both corroded and uncorroded). Finally, the microstructure of smithing hearth slags tend to exhibit more layered and heterogeneous structures compared to smelting slags. This is due to the more rigorous manipulation of the material by smiths as they position their iron objects within the hearth.

No single criterion is sufficient to characterise metallurgical residues. Rather, conclusions must be drawn from a preponderance of evidence from macroscopic, microscopic, and chemical observations. The results can then be fit within a series of multiple working hypotheses at each scale to establish material and task.

MATERIALS AND METHODS

The assemblage submitted to the archaeomaterials laboratory at the UCL Institute of Archaeology were cleaned and sorted into morphological types. A sample of 29 specimens representing the four identified slag categories (tap, furnace, furnace with wood impressions and undiagnostic) and technical ceramics (Table 6). The size, shape, and weights were recorded for each specimen along with notable features in a common form. Relative magnetic response was recorded for all slag specimens. Each sample was given a unique laboratory identification with the letters ‘SV’ followed by number.

[pg2]

TABLE 6. Materials and sampling frequency

[tb] [th]Material type|Sample size[/th] [tr]Tap slag|11[/tr] [tr]Furnace slag|11[/tr] [tr]Furnace slag (with wood impressions)|3[/tr] [tr]Undiagnostic slag|2[/tr] [tr]Technical ceramic|3[/tr] [/tb]

A small section of each specimen was cut with a wet abrasive tile cutter, cleaned by light brushing under tap water and dried. Sections were mounted in a polymer resin and polished to a 1μm finish following standard metallographic practice. Each specimen was then given a 10µm conductive carbon coating to facilitate their analysis by scanning electron microscopy and X-ray microanalysis using energy dispersive X-ray spectroscopy (SEM-EDS).

Microstructural and chemical characterisation of specimens was conducted by SEM-EDS using a Zeiss Evo 25 SEM equipped with two Oxford instruments Ultim Max 65 EDS detectors. All images were collected using a backscattered electron (BSE) detector to provide atomic number contrast to specimen phases. EDS spectra were acquired using parameters listed in Table 7. All images and analytical areas were scanned at a magnification of 250x producing images of ~1.1mm[fn2]. Five area scans were made for each specimen, selected to cover a representative area of the section’s microstructural features while avoiding porosity. Spectra were manually interrogated using Oxford Instruments Aztec software to optimise peak identification and identify anomalies created by the pulse pile up correction algorithm. Elemental quantifications were made using the factory 20 kV calibration profile and measuring oxygen by stoichiometry.

TABLE 7. Operating parameters for SEM-EDS analysis

[tb] [th]Instrument Parameter|Setting[/th] [tr]Accelerating Voltage|20 KV[/tr] [tr]Current|1200 pA[/tr] [tr]Working distance|8.5 mm[/tr] [tr]Beam Standardisation|Co[/tr] [tr]Deadtime|~40%[/tr] [tr]Process time|4[/tr] [tr]X-ray counts|1.5 million[/tr] [/tb] [pg3]

All data were normalised to control for variation in beam current. Quality of the resulting data was evaluated using the US geological survey reference basalts BCR-2G (Plumlee 1998a; Wilson 1997a), BIR-1G (Smith 1998), and BHVO-2G (Plumlee 1998b; Wilson 1997b). Analyses were undertaken for each certified reference material during each analytical session to provide robust accuracy and precision measurements across the study.

The structure of the resulting materials chemistry was explored using principal component analysis (PCA) of centred log ratio transformed data. The reducible iron index (RII, see Charlton et al 2010) was calculated for smelting slag to provide a proxy for the strength of the reducing atmosphere. All slag data are then compared against the database of slag chemistry collated by Sarah Paynter (2006) to evaluate fit with sites from known geological groups. In order to facilitate this comparison, all cases in the database containing analytical zeros were removed from consideration. All data analysis was conducted in the R statistical computing environment (R Core Team 2023).

RESULTS

Results are reported separately for each material type observed within the sample; technical ceramics and slag, respectively. Micrographs and detailed microstructural observations are reported in Appendix A. A table of raw chemical data is reported in Appendix B. Data quality assessments for CRM chemical analyses are reported in Tables 8, 9, and 10. Precision and accuracy tend to stay within 5% relative except when values are near the limits of detection. Limits of detection vary by element and analyte matrix, but tend to range between 0.05 – 0.15 wt%.

TABLE 8. Data quality assessment for BCR-2G using recommended values from the georem database (Jochum et al, 2005).

= mean, s = standard deviation, RSD = relative standard deviation, E = error.

[tb] [th]BCR-2G|Na|Mg|Al|Si|P|K|Ca|Ti|Mn|Fe[/th] [tr]Recommended values (wt%)|2.40|2.15|7.09|25.43|0.16|1.49|5.27|1.41|0.16|9.64[/tr] [tr]Measured (n=22)|2.32|2.17|7.24|25.61|0.15|1.54|5.16|1.43|0.17|9.87[/tr] [tr]S|0.03|0.01|0.02|0.03|0.01|0.02|0.02|0.03|0.02|0.04[/tr] [tr]RSD (%)|1.34|0.62|0.31|0.12|7.79|1.15|0.39|2.12|11.55|0.38[/tr] [tr]E|-0.08|0.02|0.15|0.18|-0.01|0.05|-0.11|0.02|0.02|0.23[/tr] [tr]%E|-3.50|0.77|2.18|0.70|-7.35|3.06|-2.07|1.27|11.44|2.39[/tr] [/tb] [pg4]

TABLE 9. Data quality assessment for BHVO-2G using recommended values from the georem database (Jochum et al, 2005).

= mean, s = standard deviation, RSD = relative standard deviation, E = error.

[tb] [th]BHVO-2G|Na|Mg|Al|Si|P|K|Ca|Ti|Mn|Fe[/th] [tr]Recommended values (wt%)|1.78|4.30|7.20|23.05|0.13|0.43|8.51|1.63|0.13|8.78[/tr] [tr]Measured (n=22)|1.61|4.39|7.25|23.46|0.10|0.45|8.21|1.72|0.14|8.77[/tr] [tr]S|0.02|0.03|0.03|0.04|0.01|0.01|0.04|0.03|0.02|0.04[/tr] [tr]RSD (%)|1.05|0.61|0.36|0.16|10.60|2.98|0.46|1.56|11.19|0.50[/tr] [tr]E|-0.17|0.09|0.05|0.41|-0.03|0.02|-0.30|0.09|0.01|-0.01[/tr] [tr]%E|-9.50|2.00|0.68|1.79|-20.28|4.53|-3.48|5.26|4.41|-0.15[/tr] [/tb]

TABLE 10. Data quality assessment for BIR-1G using recommended values from the georem database (Jochum et al 2005).

= mean, s = standard deviation, RSD = relative standard deviation, E = error.

[tb] [th]BIR-1G|Na|Mg|Al|Si|P|K|Ca|Ti|Mn|Fe[/th] [tr]Recommended values (wt%)|43.83|1.37|5.67|8.20|22.20|0.01|0.02|9.93|0.54|0.15[/tr] [tr]Measured (n=22)|43.81|1.30|5.81|8.28|22.38|0.00|0.02|9.57|0.59|0.14[/tr] [tr]s|0.02|0.02|0.02|0.02|0.04|0.00|0.01|0.02|0.02|0.02[/tr] [tr]RSD (%)|0.05|1.17|0.40|0.27|0.16|NA|58.36|0.24|3.91|16.98[/tr] [tr]E|-0.02|-0.07|0.14|0.08|0.18|NA|0.00|-0.36|0.05|-0.01[/tr] [tr]%E|-0.05|-4.80|2.47|1.00|0.82|NA|23.73|-3.62|8.87|-4.19[/tr] [/tb] [pg5] ‌Technical ceramics

Three specimens of technical ceramic were submitted for analysis and a representative micrograph is presented in Fig. 8. All are characterised by a high concentration of quartz grains within a vitrified matrix.

[fg]png|Fig. 8 Micrograph showing typical technical ceramic microstructure at Sutton Valence (SV19). Note the high concentrations of quartz.|Image[/fg]

SEM-EDS measurements of technical ceramic chemistry are summarised in Table 6. Silica (SiO[fn2]) concentrations are high reflecting the microstructure rich in quartz. There is also an inverse association between FeO and SiO[fn2]that represents reaction with slag at high temperature.

TABLE 11. Mean oxide values expressed as wt% for technical ceramic specimens included in the Sutton Valence sample.

All data are normalised with original totals given in the final column.

[tb] [th]|Na2O|MgO|Al2O3|SiO2|P2O5|K2O|CaO|TiO2|MnO|FeO|total[/th] [tr]SV18|0.4|0.5|8.6|81.6|0.8|1.5|0.5|0.7|0.1|5.3|74.6[/tr] [tr]SV19|0.5|0.3|7.0|73.1|0.8|2.3|0.8|0.6|0.1|14.4|76.5[/tr] [tr]SV20|0.5|0.6|6.9|75.2|0.9|3.3|1.8|0.6|0.1|10.1|76.1[/tr] [/tb] [pg6]‌Slag

Slag microstructure varies as a function of chemistry, redox environment, temperature, and cooling environment. Four microstructural categories were identified within the Sutton Valence sample. The first shows relatively fine crystal structures of fayalite and wüstite in a glassy matrix (Fig. 9) and is indicative of slag tapped from a furnace. The second microstructure type is similar to the first, but with more variable crystal sizes and orientations (Fig. 10) and is indicative of furnace slag. The third microstructure is also a furnace slag, but one that has cooled over a long period of time in the presence of charcoal. This is evidenced by the large blocky fayalite crystals, globular wüstite, and the appearance of a wüstite-leucite (KAlSiO[fn6]) eutectic phase (Fig. 11). The final microstructure observed is comprised of a dominant globular/dendritic wüstite phase with some fayalite and glass (Fig. 12). Similar structures are noted in smithing slags as well as in less-reduced parts of furnace slags, possibly representing part-reduced ore.

[fg]png|Fig. 9 Micrograph showing typical tap slag microstructure at Sutton Valence (SV11). Note the uniform orientation of the fayalite lathes (medium grey) indicative of rapid cooling from a single surface. Wüstite (white) displays a fine dendritic structure with a reticulated net-like pattern across the specimen.|Image[/fg]

[fg]png|Fig. 10 Micrograph showing typical furnace slag microstructure at Sutton Valence (SV11). Note the random orientation of the fayalite lathes (medium grey) indicative of cooling from all sides. Wüstite (white) displays a fine dendritic structure with a reticulated net-like pattern across the specimen. Larger wustite globules may indicate remnant ore trapped in the slag.|Image[/fg]

[fg]jpg|Fig. 11 Micrograph showing typical furnace slag microstructure at Sutton Valence (SV15). Note the large blocky fayalite crystals (medium grey) indicative of slow cooling from all sides. Wüstite (white) displays a globular structure as well as secondary expression in eutectic composition with leucite. The presence of this latter phase indicates cooling in the presence of charcoal, possibly within the deeper centre of the furnace.|Image[/fg]

[fg]png|Fig. 12 Micrograph showing hearth or furnace slag microstructure rich in wüstite (SV13). Note the large blocky fayalite crystals (medium grey) indicative of slow cooling from all sides. Wüstite (white) displays a globular structure as well as secondary expression in eutectic composition with leucite. The presence of this latter phase indicates cooling in the presence of charcoal, possibly within the deeper centre of the furnace.|Image[/fg]

[pg7] [pg8]

The chemistry of the slag sampled from Sutton Valence shows relatively low variance, despite the differences in observed microstructures. Slag chemistry based on slag types defined by morphology and microstructure is summarized in Table 12. The macroscopic category of furnace slags with wood impressions is retained, though its microstructures are no different from other furnace slags.

[pg9]

TABLE 12. Means and standard deviations for select oxides reported as normalised wt% for slag types identified through macro and microscopic examination.

Totals represent the mean of unnormalized data for the same groups. Excluding the FeO-rich slag mean values are within one standard deviation of each other implying no significant differences.

[tb] [th]type|Na2O|MgO|Al2O3|SiO2|P2O5|K2O|CaO|TiO2|MnO|FeO|Total[/th] [tr]FeO-rich|0.1|0.4|3.1|8.8|1.7|0.3|2.2|0.0|0.5|82.7|99.4[/tr] [tr]Furnace (n=10)|0.3|0.4|5.2|21.4|2.0|1.2|4.0|0.1|0.6|64.6|98.2[/tr] [tr]s|0.1|0.2|0.8|3.7|0.9|0.3|1.0|0.0|0.2|4.2|[/tr] [tr]Furnace- Wood impression (n=3)|0.2|0.6|6.2|22.8|2.8|1.6|4.3|0.2|0.7|60.2|100.6[/tr] [tr]s|0.0|0.2|1.3|2.5|0.9|0.7|1.3|0.0|0.2|4.9|[/tr] [tr]Tap (n=12)|0.2|0.3|5.3|21.3|1.7|1.0|3.0|0.2|0.6|66.2|98.1[/tr] [tr]s|0.0|0.1|0.8|3.0|0.4|0.3|1.2|0.0|0.1|3.9|[/tr] [/tb]

A principal component analysis was conducted for all slag to identify any evidence of chemical grouping between sites (Fig. 13). While there do appear to be some differences in the overall distribution of furnace and tap slags, overall, the multivariate chemical patterns indicate materials with a consistent composition. Only two outliers are observable, SV13 (FeO-rich slag) and SV28 that has a lower FeO content. These kinds of outliers are normal within a slag assemblage and do not reflect any significant variation in smelting recipe.

[fg]png|Fig. 13 PCA arising from centred log ratio transformed oxide values measured from the Sutton Valence slag.|Image[/fg]

[pg10]

The reducible iron index (Charlton et al 2010) was calculated for all slag specimens as a rough indicator of reduction efficiency. A value of 1 indicates perfect conformity to the standard bloomery model and a slag composed of fayalite and glass without any free wüstite. Higher values indicate, efficient reduction, production of a steely iron, and a competitive market economy. Lower values (<0.7) indicate less efficient reduction and production of iron for a small-scale economy. The mean RII for the Sutton Valence sample is 0.77. Distribution of RIIs by slag type are shown in Fig. 14.

[fg]png|Fig. 14 Boxplot of RII values by slag type.|Image[/fg]

[pg11]A PCA was also conducted to understand the fit of the Sutton Valence sample within the geologically defined slag variants described by Paynter (2006). Results (Fig. 15) reveal a close correspondence of the Sutton Valence slag to the Kent Tertiary Sands region. Sites represented by slag chemistry in the database for the Kent Tertiary Sands group include Brisley Farm (Iron Age/Roman), Hawkinge (Early Iron Age), Leda Cottages (Late Iron Age/Roman) and Westhawk (Roman). These slags tend to be richer in P[fn2]O[fn5]and MnO relative to other geological groups.

[fg]jpg|Fig. 15 PCA comparing the Sutton Valence slag sample with geologically defined slag groups in Britain. Sutton Valence (SV) aligns closest to the Kent Tertiary Sands slag groups.|Image[/fg]

A final PCA was conducted to compare the Sutton Valence slag with that from other sites within the Kent Tertiary Sands group (Fig. 16). Results show that despite the chemical affinity between slags from different groups, individual sites can still be discriminated. This is due to differences in resource acquisition and smelting practice at each site.

[fg]jpg|Fig. 16 PCA comparing Sutton Valence slag to other slag from Kent Tertiary Sands sites.|Image[/fg]

[pg12]

DISCUSSION

Morphological, microstructural, and chemical characterisation of 29 specimens of Sutton Valence slag reveal an iron smelting system that yielded slag of a consistent character and, by extension, a consistent iron product. Furnaces were made of a clay prepared with a silica-rich sand temper that would have made the walls or lining very refractory to sustain the temperatures of up to 1450°C. Furnaces were operated such that about 50% of the slag was tapped with the remaining 50% being removed from the furnace by raking or later cleaning. Some of the furnace slag seems to have filled a depression at the base of the furnace where it was allowed to cool in situ in the presence of hot charcoal. Here, fayalite and wüstite phases grew larger because of slow cooling and some of the fuel ash from the charcoal helped form a distinctive leucite-wüstite eutectic phase.

An RII value averaging 0.77 indicates a smelting practice that aimed to maximise production of ferritic iron without risking the production of a harder steel that would be more difficult to forge. This is in keeping with most developed bloomery industries and consistent with Roman practice. What this means in terms of actual yield of iron metal relative to the slag is impossible to assess without more detailed analyses of ore and technical ceramic. If the FeO-rich slag is taken as a proxy for potential richness of the original ore used, then we can apply a simple yield model to indicate the approximate weight of iron per unit of slag (see Rehren 2001; Severin et al 2011). Here, an ore of about 85 wt% FeO is used to produce a slag of about 65 wt% FeO and 35 wt % non-reduced compounds (NFC). Thus, 200 kg of ore should produce 170 kg of oxides and 30 kg of NFC. If we assume 5 wt% of the NFC derives from fuel ash then 200 kg of ore results in 100 kg of slag. The system contains 170 kg of FeO from the ore, less 65 kg of FeO from the slag, resulting in a remainder of 105 kg or FeO or [pg13](multiplying by the stoichiometric conversion factor 0.777) 81.6 kg of iron. Assuming 50% losses from forging (Crew 1991), this value would become ~41 kg of consolidated iron per 100 kg of slag. That is, the ~25 kg of smelting resides excavated from Sutton Valence furnace fill [219] represents the production of about 10 kg of iron. Considering that a common nail of about 6cm weighs a maximum of 6g, this same amount of iron represents more than 1600 nails (data from Manning 1985). While a degree of caution must be taken with such calculations and their extrapolation, the values are reasonable and can be used to compare with other sites.

The smelting slag from Sutton Valence fits well with the Kent Tertiary Sands geological group (see Paynter 2006). This is in keeping with the site location and access to resources within the area. That the slag chemistry can be discerned from that of other sites in the region may have important implications for the long-term establishment of a database of Roman slag chemistry for use with the provenance of iron and modelling production/distribution networks (Charlton 2015). Major and minor oxide concentrations would be aided by the additional measurement of trace elements in future investigations.

Twenty-five kg of metallurgical residues from Sutton Valence (SVH11), context [219] were submitted to the UCL Institute of Archaeology’s archaeomaterials laboratory for characterisation. Twenty-nine specimens of these, representing the frequency of morphological types in the were selected for invasive analysis by SEM-EDS to understand the metallurgical processes that produced them. Results showed that all materials were part of an iron-smelting system that made use of furnace with refractory walls or lining. Slag chemistry fits well with the local geologically defined smelting group (Kent Tertiary Sands) but also shows differences relative to slag chemistries from other sites in the region. RII values suggest that furnace operation was optimised to yield high volumes of soft ferritic iron without risking the production of steel. A yield estimate based on a single analysis of FeO- rich slag suggests that the slag from this context represents about 10 kg of consolidated iron metal. Characterisations of materials from a single context offer little in terms of explaining the organization of production and exchange of iron in Roman Britain. However, the early date of Sutton Valence means these materials can make a substantive contribution to broader explorations of the circulation of materials in Roman Britain. A small sample of these materials should be retained for future iron provenance studies using ICP-MS for analysis of trace and rare earth elements. It was unfortunate that nothing resembling either smithing slag or ore were observed amongst the materials investigated. Should such be revealed by any future archaeological work in or around SVH11, these should be retained to provide a better understanding of ironworking practices and generate more precise yield calculations.

Charlton, M F, 2015, The last frontier in ‘sourcing’: the hopes, constraints and future for iron provenance research, Journal of Archaeological Science, 56, 210–220,<https://doi.org/10.1016/j.jas.2015.02.017>.

Charlton, M F, Crew, P, Rehren, T, & Shennan, S J, 2010 Explaining the evolution of ironmaking recipes – An example from northwest Wales, Journal of Anthropological Archaeology, 29, 352–367, <https://doi.org/10.1016/j.jaa.2010.05.001>.

Crew, P, 1991, The experimental production of bar iron, Historical Metallurgy, 25, 21–36. Dungworth, D, 2015, Archaeometallurgy: Guidelines for Best Practice.

Eekelers, K, Degryse, P, & Muchez, P, 2016, Petrographic investigation of smithing slag of [pg14]the Hellenistic to Byzantine city of Sagalassos (SW-Turkey), American Mineralogist, 101, 1072–1083, <https://doi.org/10.2138/am-2016-5390>.

Jochum, K P, Nohl, U, Herwig, K, Lammel, E, Stoll, B, & Hofmann, A W, 2005, GeoReM: A New Geochemical Database for Reference Materials and Isotopic Standards. Geostand Geoanalyt Res 29, 333–338 <https://doi.org/10.1111/j.1751- 908X.2005.tb00904.x>.

Manning, W H, 1985, The Iron Objects, in: Inchtuthil: The Roman Legionary Fortress, Britannia Monograph Series, Society for the Promotion of Roman Studies, 289–299.

Paynter, S, 2006, Regional variations in bloomery smelting slag of the Iron Age and Romano-Bristish periods, Archaeometry, 48, 271– 292,<https://doi.org/10.1111/j.1475-4754.2006.00256.x>.

Plumlee, G, 1998a, Basalt, Columbia River, BCR-2, US Geological Survey, Denver. Plumlee, G, 1998b, Basalt, Hawaiian Volcanic Observatory, BHVO-2, US Geological Survey, Denver.

R Core Team, 2023, R: A Language and Environment for Statistical Computing. Rehren, T, 2001, Meroe, Iron and Africa, MittSAG 12, 102–109.

Severin, T, Rehren, T, & Schleicher, H, 2011, Early metal smelting in Aksum, Ethiopia: copper or iron? European Journal of Mineralogy, 23, 981–992, <https://doi.org/10.1127/0935-1221/2011/0023-2167>.

Smith, D B, 1998, Icelandic Basalt, BIR-1.

Wilson, S A, 1997a, The collection, preparation, and testing of USGS reference material BCR-2, Columbia River, Basalt, US Geological Survey Open-File Report 98-xxx.

Wilson, S A, 1997b, Data compilation for USGS reference material BHVO-2, Hawaiian Basalt, US Geological Survey Open-File Report.

[pg15]