Archaeological Chemistry - American Chemical Society

---

Chapter 25

Fingerprinting Specular Hematite from Mines in Botswana, Southern Africa 1

2

3

Adam V . Kiehn , George A . Brook , Michael D . Glascock , Jonathan Z . Dake , Lawrence H . Robbins , Alec C . Campbell , and Michael L. M u r p h y 3

4

5

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

6

1

2

Departments of Geology and Geography, University of Georgia, Athens, G A 30602 Research Reactor Center, University of Missouri, Columbia, M O 65211 Department of Anthropology, Michigan State University, East Lansing, MI 48824 P.O. Box 306, Crocodile Pools, Gaborone, Botswana Kalamazoo Valley Community College, Kalamazoo, MI 49003

3

5

6

Specular hematite, or speculante (Fe 0 ), was a valued cosmetic in Southern Africa during the Late Stone Age (LSA) and Iron Age (IA), and there are ancient mines throughout the region. Early explorers reported that it was applied to the body and hair with animal fat and was traded extensively. We analyzed specularite samples from seven prehistoric mines in Botswana, including five in the Tsodilo Hills, and one each at Dikgatlampi and Sebilong to test the feasibility o f geochemically fingerprinting sources. Most of the mines examined exploited specularite-rich schist or hydrothermal quartz veins. The heavy mineral fractions (density >2.9 g/cm ) of 73 samples, separated by heavy liquid and consisting largely of specularite, were subjected to instrumental neutron activation analysis (INAA). Multivariate statistical analyses of the compositional data suggest that specularite can be fingerprinted to specific mines or groups of mines separated by tens to hundreds of kilometers. The method appears robust enough for determining the provenance of archaeological samples throughout the region and so could provide valuable information on past trade routes and patterns. 2

3

3

460

© 2007 American Chemical Society

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

461 Archaeological occurrences and early historic accounts indicate that in Southern Africa specularite was heavily exploited and highly valued as a cosmetic from the Early Iron Age through the 19 century (1,2). It was prepared for use by grinding and mixing with grease then applied to the hair and body, giving the wearer a shimmering appearance (2, and references therein). Specular hematite or specularite (Fe 0 ) is a mineral with steel gray to black color, metallic luster, tabular or platy crystals, and a specular or micaceous habit. It can occur in igneous, metamorphic, and sedimentary geologic settings, but the most desirable crystals, such as those sampled for this study, are most commonly found in hydrothermal vein deposits and metamorphosed hematite-rich sedimentary rocks. Specularite from these deposits is chemically and geologically similar to other pigments, such as red and yellow ochre, which should allow similar methods to be applied to these materials also. Being able to provenance archaeological materials is crucial to understanding why people procured, processed and used them. In the case of specularite, very little has been done to provenance archaeological finds. In Australia, the isotopic and magnetic characteristics of sedimentary hematite have been used successfully to fingerprint sources (3-6). However, specularite in Southern Africa has not been fingerprinted despite widespread archaeological occurrence. The ability to source specularite, which is common at archaeological sites, could provide answers to such contentious questions as how prehistoric groups in the Kalahari region interacted (7-9). This study is a first attempt to geochemically characterize and fingerprint specularite deposits from mines in Botswana by applying multivariate discriminant analysis to chemical compositions determined by instrumental neutron activation analysis (INAA). th

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

2

3

Objectives and Assumptions The main objectives of the study are to determine i f specularite from prehistoric mines can be distinguished by its geochemical signature, and at what scale this is possible (mine, local, regional). To answer these questions, specularite ore was collected in Botswana from individual mines at Sebilong and Dikgatlampi near Gaborone, and from five separate mines in the Tsodilo Hills. Samples from the Tsodilo Hills mines, which are at most a few kilometers apart, provide a good test of whether specularite can be sourced to individual mines at the local scale. As the Tsodilo Hill mines are hundreds of kilometers from the Sebilong and Dikgatlampi mines, ores from these mines provide a test of how well specularite can be sourced at the regional scale. Another aim of this study is to determine which elements are most reliable and useful for fingerprinting specularite and other heavy minerals commonly associated with it. Preliminary analyses and previous literature (5, 10, 11) indicate that transition metals (TM) and rare earth elements (REE) have unique signatures associated with genetic and metamorphic processes. This study will

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

462 test how well these two groups of elements can source specularite from different mines in Botswana by comparing the results of discriminant analysis for both a T M model and a TM+REE model (10, 11). The final step, i f this preliminary work is successful, is to develop a consistent and applicable methodology and begin building a database of sources that can be used in future archaeological provenance studies. A l l provenance studies make implicit assumptions. According to the provenance postulate, the sourcing of materials requires that "there exist some qualitative or quantitative chemical or mineralogical difference between natural sources that exceeds the qualitative or quantitative variation within each source" (12). We assume that this postulate applies to the specularite sources examined here and that the inter-source variation is greater than the analytical errors of the techniques employed. Provenance studies are also limited by the fact that an artifact cannot be definitely assigned to a single source until all possible sources have been analyzed. However, this study does not attempt to source artifacts but merely to determine i f specularite ores from different mines or mining areas can be differentiated on the basis of geochemical characteristics.

Sampling Regions Samples of specularite ore were collected in 1995 and 2005 from five prehistoric mines at the Tsodilo Hills and from single mines at Sebilong and Dikgatlampi in Botswana (Figure 1). The samples were generally taken from tailings piles at or near the mines and some of the most specularite-rich fragments collected. This was done as an attempt to take samples from the tailings discarded by ancient miners that would have been most similar to the material collected and actually used and eventually deposited in archaeological contexts and to avoid analyzing samples with specularite content incomparably lower than archaeological samples. A few hundred grams for each distinct bedrock type was obtained from each mine site to help determine similarities and differences between the mines, and several samples were taken from different areas of each of the larger mines. The result of the sampling was roughly 2-4 samples per mine at Tsodilo Hills and 8-12 for Dikgatlampi and Sebilong.

Tsodilo Hills The Tsodilo Hills World Heritage Site is located in northwestern Botswana, in the Kalahari Desert, about 40 km from the Okavango River and is surrounded by relict linear sand dunes. This is one of the most interesting archaeological localities in southern Africa, in part because hills in the region are rare, the nearest to Tsodilo being about 200 km away. Evidence of human activity at the

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

463

Figure 1. Map of Botswana showing the locations ofprehistoric specularite mines examined in this study (modified from (13)).

Tsodilo Hills includes two Early Iron Age (EIA) villages, more than 4,000 Middle (MSA, c. 150k B.P. - 30k B.P.) and Late Stone Age (LSA, c. 30k B.P. l k B.P.) rock paintings in rock shelters and caves, and numerous mica schist and specularite mines (7-9, 13-16). There are four named hills at Tsodilo: Male, Female, Child, and North (7). Male H i l l rises 410 m above the surrounding landscape, the Female H i l l is 300 m high, and the other two hills are 40 m high or less. A l l are composed of compositionally mature Precambrian quartzites and schists that have been faulted and altered in many places. Despite this metamorphism, hematite concentrated on the original bedding planes still shows clear cross-bedding features as shown in Figure 2. Judging by the common occurrence of these features at the shallow mines, they may have signaled to prehistoric miners that significant specularite deposits were present. Evidence for occasional specularite mining at Tsodilo Hills dates to before 5300±160 C yr B.P. at Rhino Cave (13). Radiocarbon ages for soot deposits formed during fire-spalling indicate that there was intensive specularite mining in the period ca. A . D . 800-1025, which coincides with occupation of the E I A villages of Nqoma and Divuyu on the top of Female H i l l . 1 4

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

464

Figure 2. Hematite concentrated on paleo-crossbedding planes in quartzite at a prehistoric mine in Tsodilo Hills.

Précambrien

The large scale of production from the mines at this time implies that specularite was being traded or distributed beyond local populations. More than 1000 tons of rock may have been removed from Upper Male H i l l mine alone (7). Ostrich-eggshell and ceramic containers containing specularite have been found in Swaziland and Namibia, respectively (77, 18). The Namibian find has no known source of specularite within 100 km or more, demonstrating long distance transport (19). Glass beads and marine shells found at the Nqoma, dating to around 900 A . D . , are evidence that the people living at Tsodilo were involved in trade networks extending as far as the Indian Ocean (8).

Sebilong and Dikgatlampi Sebilong and Dikgatlampi are located in southeastern Botswana, approximately 40 km west-southwest and 60 km north-northwest of Gaborone, respectively. The topographies of both areas consist of plains and flat alluvial valleys interrupted by sharp hills and tablelands. The Sebilong mine is high on a cliff face at the edge of a plateau in the Sesitajwane Hills between Thamaga and Moshupa (20). The plateau is composed of the Mannyelanong Formation of the early to mid-Proterozoic Waterberg Group. This formation consists mainly of

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

465 cross-bedded red quartz sandstones and is brecciated. Large grained specularite is concentrated in and near quartz veins located in these altered zones (21). Fire-spalling techniques were used to extract the specularite, and extensive mining is evidenced by tailings that form a scree from the workings near the top of the 80 m high cliff to its base where slag from iron-smithing has been found (20). The site has not been excavated but thermoluminescence dates place intensive workings at the 14 C. A . D . while historical accounts indicate that mining had ended by the 19 century, with evidence for iron smelting at the base of the cliff dating to the latter part of this period (20). In the area around the village of Thamaga, there are several hills and ridges similar to the one exploited at Sebilong that may also have been mined for specularite in prehistoric times. The Dikgatlampi workings are located in one of a series of low hills rising from a plain 8 km southeast of the village of Lentsweletau. There are roughly two dozen pits in a two-hectare area. The pits exploit specularite found in brecciated zones of the sandstone and conglomerate that form the hill. As at Sebilong, these rocks belong to the Mannyelanong Formation of the Waterberg Group (20, 21). Excavations in one of the pits revealed pottery and metal tools thought to date to the 17 Century, but direct dates have not been obtained. Although there is no clear evidence of iron working or prehistoric settlement, earlier mining of the site cannot be ruled out (20, 22). th

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

th

th

Research Methods Laboratory Methods Hand samples, ranging from about 50-300 grams each, were crushed using a Plattner mortar then ground to a monominerallic grain size with an alumina mortar and pestle. The heavy mineral fraction was obtained by gravity separation in Tygon tubing filled with a sodium polytungstate solution with density greater than 2.9 g/cm . Once the samples had completely settled the tubing was clamped to separate the light and heavy fractions. Each fraction was rinsed several times with deionized water and the sodium polytungstate solution was recovered by vacuum filtration. The heavy mineral fraction was washed from the filter paper with ethanol or acetone and then allowed to dry at room temperature. The samples were inspected and described at 10-45x binocular magnification before I N A A analysis. Generally, the remaining heavy mineral fraction weighed a few grams, but ranged from 0.5-15 grams depending on the size of the original hand sample and its relative hematite content. The goal of the heavy mineral separation was to concentrate the specular hematite in order to reduce spurious variation introduced by non-hematite minerals. We recognize that in trying to source archaeological samples they would have to be concentrated in the same way before analysis. 3

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

466 Instrumental neutron activation analysis was conducted at the University of Missouri Research Reactor (MURR). Samples of approximately 50-100 mg were subjected to long and short irradiations using the same methodology used on pottery and other materials with appropriate reference standards (23). The analysis provided abundances of 33 elements for 72 samples. It is important to note that the heavy mineral separation process only removes the majority of the light minerals but not all of them. Also, any heavy minerals other than hematite will remain in the sample. The main goal of the separation process was simply to concentrate the heavy minerals, especially hematite, and remove the light minerals so that our tailings samples more closely approximated the specularite-rich archaeological samples. Although the remaining light minerals also contribute to the elemental signature of the samples, the specific elements examined in this study are generally found at concentrations at least an order of magnitude higher in the hematite than in the light minerals (10). Thus, we assume that the contribution of the light minerals to the geochemical signature is negligible for the elements of the subcomposition in consideration.

Data Analysis Statistical Procedures The I N A A data were subjected to discriminant function analysis (DFA), which uses the variance of variables within predetermined groups of observations in the data set (e.g., individual mines) to develop linear functions of the variables that can best discriminate between the groups. The centroid of each group is calculated using the geometric means of the values with the discriminant functions as the axes. Each observation is then classified based on its Mahalanobis distance to the group centroids with the nearest group centroid as the most probable candidate for group membership. The accuracy of a model can be evaluated by examining the percentage of misclassified observations in this, the initial classification. Cross-validation of the D F A model is conducted by casewise deletion, reestimation of functions, and classification. In other words, for each observation in the data set, that observation is omitted and the discriminant functions are reestimated using the full data set minus that observation. Then that observation is classified based on the re-estimated functions. The accuracy of the crossvalidation can be used to evaluate the reliability of the D F A and the potential impact of group outliers. In essence, the cross-validation process is the same process used to determine provenance of an unsourced archaeological sample where the discriminant functions are developed independently of the sample and then used to determine its most likely source.

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

Downloaded by UNIV OF SYDNEY on May 4, 2015 | http://pubs.acs.org Publication Date: August 16, 2007 | doi: 10.1021/bk-2007-0968.ch025

467 The actual analyses were conducted using SPSS 13.0. The results presented were obtained by entering all variables together and using the within groups covariance matrix options. The stepwise entering of variables was also conducted in parallel to each test because it allows the exclusion of variables if they do not significantly contribute to the model's variance and allows the evaluation of each element's utility for fingerprinting hematite at a given scale. However, the presented methodology was chosen because of a lack of clear reasons based in statistical theory for the alternative, because the alternatives had identical cross-validation results, and because the groups generated by simultaneous entering of variables had a nearly identical form but were slightly more discrete than by stepwise entering of variables. Previous work using subcompositions, or selected groups of elements from the total composition of the rock, in geochemical data analysis has suggested that raw compositional data should not be examined using multivariate methods such as discriminant function analysis (24). The reasons for this include the constrained nature of the data. Several transformations have also been suggested in the geochemical statistical literature, and for the purposes of this study the additive log-ratio transform ( A L R ) is used. A L R was one of the first transformations suggested by Aitchison (1986) and serves to unconstrain the data while preserving the elemental ratios regardless of the subcomposition or denominator chosen. The A L R transformation process consists of taking the selected group of elements and dividing each element by an arbitrarily chosen denominator, in most cases one of the elements of the group that is present in but variable across all samples. The logarithm, natural or base 10 depending on personal preference, is then taken of each ratio. The group of elements used in this study consists of 11 transition and rare-earth elements and Fe was chosen for the denominator, which essentially normalizes each transformed subcomposition by its relative hematite content. Additionally, the small degree of analytical error associated with most I N A A analyses does not significantly affect the log-ratios in this study because most elements in the subcomposition occur at levels several orders of magnitude lower than Fe.

Data Preparation Data preparation began by excluding any possibly unreliable and irrelevant data from the set of 33 elements. The heavy mineral solution could have imparted excess sodium (Na) and was thus ignored. The mortar and pestle used could have contaminated the aluminum (Al). Based on the hardness of the material and relative contribution to the sample this is likely not a significant problem; however the role of A l was monitored closely during the data analysis. Previous literature on the geochemistry of specularite (10, 11) and preliminary

In Archaeological Chemistry; Glascock, M., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2007.

468 analyses in this study suggested that the most likely and reliable elements for fingerprinting specularite sources would be transition metals and other ion species that could easily substitute for F e in the crystal lattice. The ratios of these elements would be affected by initial composition and subsequent petrogenic and metamorphic processes which would hopefully leave a unique signal for each source. One drawback to this transformation is that a log-ratio cannot be taken for elements not present or below detection limits. Sophisticated multiple imputation techniques, simple replacement, and variable omission are strategies for dealing with these zeroes (24). In this preliminary study, elements not present in all samples were simply omitted from the analysis as most of the relevant transition metals were present in all samples. Due to these restrictions on missing abundances, 20 elements (As, Ba, Ca, Ce, Cs, Dy, Eu, Hf, K , Lu, Nd, N i , Rb, Sr, Ta, Tb, U , Yb, Zn, Zr) were excluded from the analyses. Elements with ionic species commonly associated with quartz and mica minerals such as Na, Zr, and H f are more likely to co-vary with those minerals rather than hematite. The ratio of these latter elements to Fe-substituting elements would be dependent on the relative amounts of hematite and other minerals in the sample rather than just the rock-forming processes. For this reason, elements that commonly substitute into quartz and other abundant minerals in the samples were excluded from the discriminant analysis. Some rare earth elements may also play a role in fingerprinting the sources, and their importance was tested by comparing discriminant analysis models with and without La, Sm, and Th. Samples that were very low in hematite (