Final Report Summary - IRONWORKS (Geochemical Provenance Mapping of Pre-Roman Ironworking sites in Britain)
IRONWORKS: Geochemical Provenance Mapping of Pre-Roman Ironworking sites in Britain
Final Project Summary
http://www.ucl.ac.uk/archaeology/research/directory/ironworks-charlton
IRONWORKS is a two-year cross-disciplinary project based at the UCL Institute of Archaeology that combines innovative data analytic techniques with geochemical characterization of bloomery ironmaking slag to improve archaeological understandings of the past. The primary goal of the project is to build a multi-element provenance database of smelting slag from primary iron production sites that will enable researchers to realize the promise of advances in slag inclusion analysis and test a robust series of iron artefact sourcing hypotheses. More importantly, it gives archaeologists the chance to better explore the unique material window iron offers on past societies. The kinds of problems that can be investigated with an ever growing high quality slag chemistry database include past economic networks, relationships between producers and the quality of their products, and artefact biographies. Specific objectives of the IRONWORKS project include the development of a new slag analytical program at UCL, determining the optimal element combination for discriminating slag chemistry at site, regional, and superregional scales, comparing the utility of various data treatments, and constructing a series of multivariate geochemical maps for facilitating artefact provenance investigations.
Region of interest:
Pre-Roman Iron Age Britain was selected as the region and time period of investigation. Classical authors touted Britain as an exporter of iron metal to Gaul and the island’s archaeological record is rich with 'currency bars' and other forms of trade iron. Previous studies noted the variability in slag chemistry both within and between regions of Britain, indicating its potential as a proving ground for iron provenance research. IRONWORKS, thus, set out to gather new slag samples from sites in geologically defined and historically relevant regions across England and Wales. These totaled 133 specimens obtained from 21 sites (Figure 1—map of sampled sites in Britain).
Chemical Analysis
IRONWORKS is not unique in its efforts to establish an iron provenance program and one important goal of the project was to ensure that data quality and comparability were aligned with those of other leading researchers. This required the construction of a new method of slag chemical characterization at UCL using solution-based inductively coupled plasma optical emission and mass spectrometry (ICP-OES/MS).
The special attention given to maximizing data quality and comparability required the development of a drift monitoring and calibration program external to the typical instrument software. This program was written in the R statistical programming language and incorporated 2 unique elements. The first is a cubic spline drift correction model that estimates intensities (OES) and isotope counts (MS) changes for each analysis based on regular repeated measures of the blank and final standard. The model makes it possible to create individual calibration lines for every sample. The second element is the incorporation of advanced statistical approaches for calculating limits of quantification and detection that exceed the guidelines proposed by the International Union of Pure and Applied Chemists (IUPAC). In total, 35 elements (including the REEs) are measured with confidence based on comparison with certified and standard reference materials. Accuracies tend to be within one standard deviation of reference material values and uncertainties are around 5 % relative or less for values measured above the limits of quantification.
Data Analysis and Scale:
The large numbers of variables considered in the chemical characterization of slag necessitates the use of multivariate analysis. Linear Discriminant Analysis (LDA), in particular, is used as a means of modelling difference between slag derived from different sources. LDA produces a classification model for a set of groups and assumes both multivariate normality and equivalent variance-covariance structures in the data that characterize all groups. These conditions can never be assumed in slag chemistry and are encountered at low frequency. LDA is, nonetheless, a robust discriminator even when these assumptions are not met. To improve the method’s abilities as a classifier, a bivariate kernel density estimation (KDE—a kind of smoothed histogram) is calculated for each source. The KDE contour that contains 100% of the cases sampled for each source in bivariate projections onto the LD axes defines the source field. These fields become provenance hypotheses for comparison with iron objects characterized by their slag inclusion chemistry. In this way, objects to be sourced can become hypothetical products of no source that is included in the analysis, a single source, or multiple sources (in the case of field overlap). The bivariate LDA plots produced through the analysis are referred to as “Geochemical Provenance Maps”, reflecting the relationship between chemistry and geography, as well as the colorful rendering of multiple provenance fields that resemble common political maps. The complexities involved in the analysis are mitigated by the use of a specially constructed R-script.
Geochemical Provenance Maps for Britain
One projection of the multi-dimensional geochemical provenance map for Britain is shown in Figure 2. It can be noted that sites from the same geographic regions—like Crawcwellt West (CCW), Bryn y Castell (ByC) and Llwyn Du (LD) in northwest Wales—tend to cluster together. It can also be noted that overlap tends to be low between sites and often disappears when examining other map dimensions of the linear discriminant space. Where overlap cannot be avoided, a new analysis can be run that focuses on the potential provenance fields (See Figure 3—Geochemical provenance map of sites in northwest Wales).
A global geochemical provenance map for iron?
The geochemical map for Britain, while incomplete, demonstrates proof of concept with extreme clarity. But could British iron be separated from iron produced in mainland Europe or regions farther afield? An opportunity to consider slag from other countries including China, Nigeria, Rwanda, and Thailand was exploited and used to produce Figure 4—A Geochemical map for the entire project (including a further 69 specimens from 16 additional sites). The new map, much like the one for Britain only, reveals strong regional patterning and good discrimination between all sites and marks it as an exceptional model for sourcing bloomery iron. More importantly, it indicates that a global database of slag chemistry would be an unprecedented tool for documenting the large scale movement of materials in the pre-industrial world.
Impact
IRONWORKS, through the combination of database construction, method development, student training, and European scholarly interaction, is creating a knowledge-based infrastructure that helps overcome longstanding obstacles to iron provenance research at institutional, European and international levels.The project has succeeded in constructing a large database of bloomery slag chemistry that will prove useful for provenancing iron objects in Britain and beyond. While providing far from complete coverage of the island’s bloomery remains, the data and applied methods reveal the enormous potential of cataloguing the chemical diversity of slag for generating and testing iron provenance hypotheses. The addition of an international set of bloomery sites further illustrates this point and marks the foundation of a new tool and service for the archaeological community.
Beyond the research itself, the IRONWORKS project involved an important knowledge transfer component. Methods and preliminary results were disseminated at conferences in the USA and Europe as well as at invited lectures at the UK’s top universities. The iron provenance strategy developed in IRONWORKS were also discussed with African archaeologists as part of an experimental smelting campaign in Sudan, at a workshop in Qatar for archaeological scientists researching ancient glass, and at a workshop on the applications of statistics in ceramic geochemical research. These actions have led to direct engagement with scholars and students based on three continents including access to unique sets of materials and promise of future collaborations in all.
The quantitative methods developed by the project were actively disseminated to students at all educational levels through lectures, tutorials, and workshops. Quantitative methods play an ever more important role in archaeological research and are a vital transferable skill for many non-academic careers. Educational opportunities created by IRONWORKS have yielded a BSc dissertation and influenced the research objectives of several postgraduate degrees at the UCL.
Final Project Summary
http://www.ucl.ac.uk/archaeology/research/directory/ironworks-charlton
IRONWORKS is a two-year cross-disciplinary project based at the UCL Institute of Archaeology that combines innovative data analytic techniques with geochemical characterization of bloomery ironmaking slag to improve archaeological understandings of the past. The primary goal of the project is to build a multi-element provenance database of smelting slag from primary iron production sites that will enable researchers to realize the promise of advances in slag inclusion analysis and test a robust series of iron artefact sourcing hypotheses. More importantly, it gives archaeologists the chance to better explore the unique material window iron offers on past societies. The kinds of problems that can be investigated with an ever growing high quality slag chemistry database include past economic networks, relationships between producers and the quality of their products, and artefact biographies. Specific objectives of the IRONWORKS project include the development of a new slag analytical program at UCL, determining the optimal element combination for discriminating slag chemistry at site, regional, and superregional scales, comparing the utility of various data treatments, and constructing a series of multivariate geochemical maps for facilitating artefact provenance investigations.
Region of interest:
Pre-Roman Iron Age Britain was selected as the region and time period of investigation. Classical authors touted Britain as an exporter of iron metal to Gaul and the island’s archaeological record is rich with 'currency bars' and other forms of trade iron. Previous studies noted the variability in slag chemistry both within and between regions of Britain, indicating its potential as a proving ground for iron provenance research. IRONWORKS, thus, set out to gather new slag samples from sites in geologically defined and historically relevant regions across England and Wales. These totaled 133 specimens obtained from 21 sites (Figure 1—map of sampled sites in Britain).
Chemical Analysis
IRONWORKS is not unique in its efforts to establish an iron provenance program and one important goal of the project was to ensure that data quality and comparability were aligned with those of other leading researchers. This required the construction of a new method of slag chemical characterization at UCL using solution-based inductively coupled plasma optical emission and mass spectrometry (ICP-OES/MS).
The special attention given to maximizing data quality and comparability required the development of a drift monitoring and calibration program external to the typical instrument software. This program was written in the R statistical programming language and incorporated 2 unique elements. The first is a cubic spline drift correction model that estimates intensities (OES) and isotope counts (MS) changes for each analysis based on regular repeated measures of the blank and final standard. The model makes it possible to create individual calibration lines for every sample. The second element is the incorporation of advanced statistical approaches for calculating limits of quantification and detection that exceed the guidelines proposed by the International Union of Pure and Applied Chemists (IUPAC). In total, 35 elements (including the REEs) are measured with confidence based on comparison with certified and standard reference materials. Accuracies tend to be within one standard deviation of reference material values and uncertainties are around 5 % relative or less for values measured above the limits of quantification.
Data Analysis and Scale:
The large numbers of variables considered in the chemical characterization of slag necessitates the use of multivariate analysis. Linear Discriminant Analysis (LDA), in particular, is used as a means of modelling difference between slag derived from different sources. LDA produces a classification model for a set of groups and assumes both multivariate normality and equivalent variance-covariance structures in the data that characterize all groups. These conditions can never be assumed in slag chemistry and are encountered at low frequency. LDA is, nonetheless, a robust discriminator even when these assumptions are not met. To improve the method’s abilities as a classifier, a bivariate kernel density estimation (KDE—a kind of smoothed histogram) is calculated for each source. The KDE contour that contains 100% of the cases sampled for each source in bivariate projections onto the LD axes defines the source field. These fields become provenance hypotheses for comparison with iron objects characterized by their slag inclusion chemistry. In this way, objects to be sourced can become hypothetical products of no source that is included in the analysis, a single source, or multiple sources (in the case of field overlap). The bivariate LDA plots produced through the analysis are referred to as “Geochemical Provenance Maps”, reflecting the relationship between chemistry and geography, as well as the colorful rendering of multiple provenance fields that resemble common political maps. The complexities involved in the analysis are mitigated by the use of a specially constructed R-script.
Geochemical Provenance Maps for Britain
One projection of the multi-dimensional geochemical provenance map for Britain is shown in Figure 2. It can be noted that sites from the same geographic regions—like Crawcwellt West (CCW), Bryn y Castell (ByC) and Llwyn Du (LD) in northwest Wales—tend to cluster together. It can also be noted that overlap tends to be low between sites and often disappears when examining other map dimensions of the linear discriminant space. Where overlap cannot be avoided, a new analysis can be run that focuses on the potential provenance fields (See Figure 3—Geochemical provenance map of sites in northwest Wales).
A global geochemical provenance map for iron?
The geochemical map for Britain, while incomplete, demonstrates proof of concept with extreme clarity. But could British iron be separated from iron produced in mainland Europe or regions farther afield? An opportunity to consider slag from other countries including China, Nigeria, Rwanda, and Thailand was exploited and used to produce Figure 4—A Geochemical map for the entire project (including a further 69 specimens from 16 additional sites). The new map, much like the one for Britain only, reveals strong regional patterning and good discrimination between all sites and marks it as an exceptional model for sourcing bloomery iron. More importantly, it indicates that a global database of slag chemistry would be an unprecedented tool for documenting the large scale movement of materials in the pre-industrial world.
Impact
IRONWORKS, through the combination of database construction, method development, student training, and European scholarly interaction, is creating a knowledge-based infrastructure that helps overcome longstanding obstacles to iron provenance research at institutional, European and international levels.The project has succeeded in constructing a large database of bloomery slag chemistry that will prove useful for provenancing iron objects in Britain and beyond. While providing far from complete coverage of the island’s bloomery remains, the data and applied methods reveal the enormous potential of cataloguing the chemical diversity of slag for generating and testing iron provenance hypotheses. The addition of an international set of bloomery sites further illustrates this point and marks the foundation of a new tool and service for the archaeological community.
Beyond the research itself, the IRONWORKS project involved an important knowledge transfer component. Methods and preliminary results were disseminated at conferences in the USA and Europe as well as at invited lectures at the UK’s top universities. The iron provenance strategy developed in IRONWORKS were also discussed with African archaeologists as part of an experimental smelting campaign in Sudan, at a workshop in Qatar for archaeological scientists researching ancient glass, and at a workshop on the applications of statistics in ceramic geochemical research. These actions have led to direct engagement with scholars and students based on three continents including access to unique sets of materials and promise of future collaborations in all.
The quantitative methods developed by the project were actively disseminated to students at all educational levels through lectures, tutorials, and workshops. Quantitative methods play an ever more important role in archaeological research and are a vital transferable skill for many non-academic careers. Educational opportunities created by IRONWORKS have yielded a BSc dissertation and influenced the research objectives of several postgraduate degrees at the UCL.