Está en la página 1de 17

Journal of Archaeological Science 43 (2014) 239e255

Contents lists available at ScienceDirect

Journal of Archaeological Science


journal homepage: http://www.elsevier.com/locate/jas

Smelting of magnetite and magnetiteeilmenite iron ores in the


northern Lowveld, South Africa, ca. 1000 CE to ca. 1880 CE
David Killick a, *, Duncan Miller b
a
Department of Anthropology, University of Arizona, Tucson, AZ 85721-0030, USA
b
Department of Geology, University of the Free State, Nelson Mandela Drive, Bloemfontein, Brandhof 9324, South Africa

a r t i c l e i n f o a b s t r a c t

Article history: The Lowveld of north-eastern South Africa is poorly suited to agriculture or pastoralism but is rich in
Received 10 November 2012 mineral resources. There is much evidence of exploitation of the copper, iron and salt resources of this
Received in revised form region during the second millennium CE. The bloomery iron smelting technologies of this region show
15 December 2013
several unusual features: (1) all known smelting sites used almost pure oxide ores, requiring the addition
Accepted 25 December 2013
of silicate flux to produce slag; (2) some used magnetiteeilmenite ores that produced slags containing up
to 25 mass % TiO2; and (3) iron ore was carried substantial distances to smelting sites. Modern blast
Keywords:
furnaces cannot use iron ores containing more than 2% TiO2, but in the bloomery process TiO2 combines
Archaeometallurgy
Technology
with iron oxides and silica to produce highly fluid slags that separate cleanly from iron metal. This
Smelting allowed ancient ironworkers to process a much wider range of iron ores than modern industry can use.
Iron The archaeological significance of this study is that we show that the geological source of ore used at any
Slag smelting site in the Lowveld can be determined by chemical analysis or thin-section microscopy of the
South Africa slag. This will allow archaeologists to include iron production in future network analyses of past regional
economic interaction in this region.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction Much of this archaeological evidence relates to exploitation of


the rich mineral resources of the northern Lowveld. Late Iron Age
The northern Lowveld (Fig. 1) forms the extreme northeast populations produced salt at six saline springs in the region, mined
corner of South Africa. It is a low plain bounded to the west by the copper ores in at least six locations, and smelted copper and iron at
north-south escarpment, above which lies the Highveld, and to the many sites, some of which are up to 50 km from any potential
east by the low Lebombo Hills in modern Mozambique. It is covered source of ore (Fig. 1). A previous synthesis (Miller et al., 2001)
by hot infertile savanna woodland, prone to drought and host to provided an introduction to the archaeology of metal production in
many endemic human and animal diseases, including malaria, tick the region. This study focuses upon the technology and spatial
fever, sleeping sickness and bilharzia. These made the Lowveld distribution of iron working. Copper working will be discussed in a
unattractive for agriculture or pastoralism, and even today it is future paper.
relatively sparsely populated.
There is little archaeological evidence for permanent human
2. Prior studies of iron production in the Lowveld
settlement in the Lowveld before 1000 CE, although its western
boundary  the escarpment around Tzaneen (Fig. 1)  has been
The greatest concentration of Iron Age mining and smelting
settled by iron-working agriculturalists since at least 350e450 cal.
activity in the Lowveld is around the modern mining town of
CE (Klapwijk and Huffman, 1996). Nevertheless, the northern
Phalaborwa (Fig. 1). Until open-pit mining began in 1965, the
Lowveld does have a rich archaeological record during the Late Iron
summit of Lolwe hill, just south of the town, was riddled with old
Age, which is defined here as the interval from about 1000 cal. CE to
shafts, trenches and adits for mining oxidised copper ores. The
1880 CE, when European miners and missionaries first settled in
lower slopes were covered with a blanket of magnetite scree, used
the region.
as ore for smelting iron. Karl Mauch, a European explorer, observed
the mining and smelting of copper here in 1868 (Bernhard, 1971;
pp. 41, 54), but all indigenous metallurgical activity seems to have
* Corresponding author. Tel.: þ1 520 621 8685; fax: þ1 520 621 2088. ceased by 1880. The abandoned mines and smelting furnaces were
E-mail address: killick@email.arizona.edu (D. Killick). described by a succession of visiting geologists (Mellor, 1906; Hall,

0305-4403/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jas.2013.12.016
240 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

Fig. 1. Map of the northern Lowveld, showing principal locations mentioned in this paper.

1912; Trevor, 1912; Schwellnus, 1937; Verwoerd, 1956; Hanekom volcanic hills around Phalaborwa (Mason, 1968), and in 1965 N. J.
et al., 1965). The first scientific analyses of metallurgical residues van der Merwe carried out further excavations on both habitation
from archaeological sites around Phalaborwa were those of and smelting sites and produced the first radiocarbon dates for the
Schwellnus (1937) and Verwoerd (1956). region (Stuiver and van der Merwe, 1968). Between 1970 and 1973
The earliest written account of an archaeological excavation at van der Merwe led an American team, which conducted an exten-
Phalaborwa appears to be an unpublished report by D. S. van der sive field survey of the wider region, including much of the
Merwe, who excavated five smelting furnaces (van der Merwe, Murchison Range and northern portions of the Kruger National Park
1936), but no trained archaeologist excavated there until shortly (Thorne, 1974). In 1977 Killick and van der Merwe revisited most of
before open-pit mining began at Lolwe in 1965. R. Mason conducted the smelting sites located by this survey to collect ore and slag
several test excavations in 1962 on terraced settlements on the samples for chemical analysis, and in 1978 and 1979 Killick carried
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 241

out further survey in the Murchison and Sutherland Ranges (Fig. 1; metasedimentary and metavolcanic rocks that originally formed
van der Merwe and Killick, 1979). These surveys introduced the around 3.0 billion years ago (Brandl et al., 2006; Jaguin et al., 2012).
regional approach to past mining and metallurgy to South Africa. These include minor BIF and host epigenetic antimony, mercury,
Excavations of smelting sites at Eiland (Fig. 1; Evers, 1975, 1982) zinc and copper deposits (van Eeden et al., 1939). The sediments
and around Tzaneen (Fig. 1; Klapwijk, 1986) revealed the distinctive were intruded on their northern side by the Rooiwater Complex, an
triangular furnaces previously noted at Phalaborwa. Comparison of ancient layered suite of igneous rocks. This contains two thin sheets
the spatial distributions of iron ore resources (see below) and of magnetite (Fe3O4) e ilmenite (FeTiO3) ore that were formed by
smelting sites showed that ores were transported over substantial oxidation and gravitational separation in a magma chamber. These
distances (van der Merwe and Killick, 1979; Miller et al., 2002). In now dip nearly vertically because of tight folding of the Rooiwater
the Soutpansberg mountains, which define the northern boundary Complex. These parallel bands are only 46 m wide and are about
of the Lowveld (Fig. 1), a remarkable historical account (Giesekke, 700 m apart. Both are exposed intermittently in schists for 14 km
1930; translated and annotated by Miller et al., 2002) notes the along strike in flat terrain to the north of the Murchison range,
regular transport of iron ore by human porters from the carbonatite directly south of Eiland in Fig. 1 (van Eeden et al., 1939). To the west
complex at Schiel (Fig. 1) to the northern mountains. Indigenous of this double band there are two reported outcrops of a single
iron smelting appears to have continued in the Soutpansberg to the magnetite band in the Rooiwater Complex. A “magnetite lens” was
end of the nineteenth century (Wessmann, 1908; Junod, 1927; mapped by van Eeden et al. (1939) 8 km due west of the west-
Stayt, 1931; de Waal, 1942). An anthropological study of the ernmost exposure of the double band, while Schwellnus et al.
Lobedu people also notes that iron ore was formerly transported (1962, p. 16) report a magnetite band “2 feet thick” in basic
from this source to Modjadji (Fig. 1), which is about 55 km south of schists in the valley of the Thabina River about 25 km SSE of Tza-
Schiel (Krige, 1941). Mining trenches and smelting sites have been neen. Millions of tons of magnetite rubble are exposed at surface, so
recorded around Schiel (Miller et al., 2002; Mathoho, 2012) but no there would have been no need for iron smelters ever to mine the
systematic survey for smelting sites north of the Murchison Range ore.
has yet been undertaken. Reynolds (1986) published chemical analyses of fourteen sam-
Subsequent studies of metal production in the Phalaborwa re- ples of ore from the bands in the central Murchison Range, five of
gion by the University of Pretoria include the work of Meyer (1986) which are reproduced here in Table 1. In his analyses TiO2 contents
in the Kruger National Park, and the intensive archaeological and range from 14 to 25 mass % TiO2, V2O5 contents from 0.75 to 1.40
oral historical study by Pistorius (1989) of metal working sites mass %, and samples from the lower band also contain up to 1.5
around Phalaborwa. This identified three different groups of metal mass % Cr2O3. The high levels of these elements mean that Rooi-
workers in the Phalaborwa region, and confirmed that there were water magnetites cannot be used in modern blast furnaces (for
two discrete periods of occupation at Phalaborwa itself (Pistorius, which the cut-off is 2 mass % TiO2) but we show below that they
1989; Plug and Pistorius, 1999). posed no problem for ironworkers using the bloomery process. In
polished section the microstructure of these ores is quite variable
3. Distribution and geology of iron ores in the northern (Reynolds, 1986), as some are more affected than others by meta-
Lowveld morphism, but all have at least two phases  magnetite and
ilmenite (Fig. 2)  and many also have exsolved lamellae of ulvö-
All iron ores in the northern Lowveld are of Precambrian age spinel (Fe2TiO4) or ilmenite within the magnetite. Crystals of
because younger strata have been stripped away by westward- pleonaste  solid solutions of magnesian spinel (MgAl2O4) in her-
moving erosion, exposing the Precambrian basement as far as the cynite (FeAl2O4)  may also occur as a minor component. Many
escarpment (which runs along the left edge of Fig. 1). Most of the samples are heavily altered, with much of the magnetite converted
bedrock exposed across the Lowveld is Archaean granite gneiss, to haematite, and the ilmenite to leucoxene.
which hosts no iron ores. Within the granite there are meta-
morphosed remnants of older rocks. There are several greenstone 3.3. The Phalaborwa Complex
belts (sedimentary and igneous rocks subsequently meta-
morphosed to the greenschist facies) and two intrusive carbonatite The other two ore sources in the Lowveld are both carbonatite
complexes. Two of the greenstone belts and both carbonatites are complexes. The Phalaborwa Complex is the eroded throat of a
potential sources of iron ore for past bloomery iron smelting. volcano that intruded through the Archaean granite 2.0 billion
years ago (Hanekom et al., 1965; Palabora Mining Company
3.1. The Sutherland Range Geological Staff, 1976; Verwoerd, 1986; Verwoerd and du Toit,
2006). The cross-section shown in Fig. 3 represents the situation
The Sutherland Range (Fig. 1), forming part of the Giyani as open-pit mining began in 1965. Almost all of the section shown
Greenstone Belt (Brandl et al., 2006), is roughly 70  20 km in here was mined away by 2002, when the pit reached a depth of
extent, composed of metavolcanic schists of ultramafic and mafic 800 m below the former summit of Lolwe Hill. The mine was then
composition (McCourt and van Reenen, 1992). These include minor converted from an open-pit to a shaft mine, and is now more than
Banded Iron Formations (BIF), which have white layers of chert 1300 m deep.
alternating with red layers of haematite and/or siderite. BIF ores The inner portion of the complex, centred on Lolwe Hill, was a
have not yet been noted on iron smelting sites in the Lowveld. It is vertical pipe of roughly concentric bands of pyroxenite, phoscorite,
likely that their silica contents are too high for bloomery smelting, and carbonatite, cross-cut by younger dolerite dykes (Fig. 3).
though systematic investigation of them has not yet been under- Phoscorite is a rock composed principally of magnetite (Fe3O4),
taken. Attempts by Killick to split the silica-rich from the iron-rich- apatite (Ca5(PO4)3(OH,F,Cl)) and olivine ((Mg,Fe)2SiO4). Two gen-
layers with hammer and chisel were unsuccessful. erations of carbonatite volcanism (the banded and transgressive
carbonatites) are recognised but have similar mineral assemblages,
3.2. The Murchison Range and Rooiwater Complex the major minerals being calcite (CaCO3, with 2e5% Mg substitut-
ing for Ca) and magnetite. Phalaborwa is unique among the world’s
The Murchison Range (Fig. 1) is a 140  20 km greenstone belt carbonatite complexes in possessing exploitable copper minerali-
that trends WSW to ENE, and is mostly composed of tightly folded sation (Verwoerd, 1986). The copper sulphide minerals are found in
242 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

Table 1
Analyses of magnetites from the Phalaborwa Complex (PC) and the Rooiwater Complex (RC). SARM1 is from Stoch (1978); PMC5, PMC9 and PMC12 are unpublished analyses
from drill cores in the Transgressive Carbonatite member of the PC, reproduced by permission of the Palabora Mining Company; RW18 and RW19 are from the upper
magnetiteeilmenite band of the RC, while RW27 and RW32 are from the lower band. RW2 is from the “magnetite lens” (all from Reynolds, 1986, Table 1). TB1 is an analysis by
Killick of an archaeological sample of magnetite ore from the farm Broederstoomdrift 534LT, Tzaneen.

SAMPLE SiO2 TiO2 Al2O3 Cr2O3 Fe2O3 FeO MnO MgO CaO V2O5 H2Oþ H2O TOTAL

Rooiwater magnetiteeilmenite ores


RW18 1.78 14.62 2.68 0.80 66.05 11.09 0.21 1.75 nd 1.41 nd nd 99.24
RW19 2.92 14.93 2.34 1.19 61.82 14.00 0.18 0.86 nd 1.32 nd nd 99.61
RW27 2.66 20.83 3.83 <0.05 48.37 22.5 0.32 1.68 nd 0.80 nd nd 100.94
RW32 2.21 21.52 4.08 <0.05 52.00 17.06 0.27 2.34 nd 0.84 nd nd 100.27
RW2 2.09 12.23 6.12 <0.05 55.11 22.97 0.19 1,78 nd 0.86 nd nd 99.52
TB1 2.23 23.22 2.26 nd 40.08 26.71 0.33 2.37 0.01 0.99 0.96 0.08 99.24
Phalaborwa magnetite ores
SARM12 0.34 0.72 0.77 <0.01 95.28 nd 0.22 2.80 1.09 0.05 nd nd 101.26
PMC5 0.34 0.76 0.75 <0.01 94.95 nd nd 1.60 0.98 0.09 nd nd 99.47
PMC9 0.31 1.43 0.41 <0.01 95.81 nd nd 0.70 0.73 0.15 nd nd 99.55
PMC12 0.49 2.66 0.47 <0.01 93.95 nd nd 1.96 0.98 0.11 nd nd 100.62

shear zones in the phoscorite and carbonatite zones, but have enclose crystals of apatite, as in Fig. 4. The low-Ti magnetite from
highest values in the transgressive carbonatite, which formed the carbonatite often contains arrays of tiny cubic crystals of
Lolwe Hill, in which prehistoric copper mines were located. magnesian spinel and may show fractures filled with calcite and
The phoscorite and carbonatite members contain vast quanti- copper sulphides (Hanekom et al., 1965; Palabora Mining
ties of magnetite iron ore, but it appears that this was not mined in
precolonial times, except when removed incidentally during
copper mining. Millions of tons of magnetite, the residual product
of erosion, formed thick scree over the lower slopes of Lolwe Hill
and the surrounding pyroxenite. This was an inexhaustible ore
source for bloomery ironworkers. The major impurity in the
magnetite is a variable amount of titanium. The lowest values of
TiO2 (02%) are in magnetite from the transgressive carbonatite
(e.g. analyses in Table 1). In the phoscorite the magnetite contains
up to 4% TiO2, and in the surrounding pyroxenite the TiO2 content
rises to as much as 8% (Hanekom et al., 1965; pp.6365). As TiO2 is
not reduced in bloomery iron smelting, all of it should end up in
the slag. The TiO2 contents of slags smelted from Phalaborwa
magnetite thus allow us to infer from which of these rocks the ore
originally derived. In polished section Phalaborwa magnetite is
easily distinguished from Rooiwater Complex magnetite because
Phalaborwa magnetite lacks euhedral crystals of ilmenite and may

Fig. 2. Polished section of sample KSME from the lower magnetiteeilmenite band of
the Rooiwater Complex e reflected cross-polarized light (XPL) in air. Subhedral
ilmenite crystals showing strong bireflectance from medium to very dark grey are Fig. 3. Vertical geological section through the central lobe of the Phalaborwa Complex
surrounded by heavily weathered magnetite (light grey) containing lamellae and before the onset of open-pit mining in 1965, as defined by drill cores. Depth levels are
irregular patches of ulvöspinel (medium grey, isotropic). The small black areas within from the summit of the former Lolwe Hill. (Redrawn after Palabora Mining Company
the magnetite are pitting from sample preparation. Geological Staff, 1976, Fig. 5.)
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 243

Company Geological Staff, 1976). The high-Ti magnetite from the sections examined the TiO2 was in solid solution (“in vaste
outer phoscorite and pyroxenite is a solid solution of ulvöspinel in oplossing”) in the magnetite.
magnetite, and in consequence has a strong pinkish brown colour
in reflected plane polarised light (PPL) (Fig 4). Parallel arrays of 3.5. Summary of the iron ore resources
thin to coarse plates of ilmenite or ulvöspinel can sometimes be
seen in some high-Ti magnetite crystals; these have formed by In summary, at three locations in the northern Lowveld huge
solid-state oxidation exsolution after crystallisation of the host amounts of iron ore could be obtained from surface scree or from
magnetite crystal. These microstructural signatures allow Phala- shallow trenches. The magnetiteeilmenite ores of the Murchison
borwa magnetite to be identified on smelting sites distant from Range are located between the two carbonatite complexes, and can
the ore body, and to be readily distinguished from the magnetitee be clearly distinguished from magnetite in the carbonatites by ore
ilmenite ores of the Rooiwater Complex. microscopy and by higher contents of TiO2 and V2O5. Neither of
these oxides can be reduced in bloomery furnaces, and thus they
should transfer quantitatively to the slag. It should therefore be
3.4. The Schiel Complex
easy to determine the geological source of ore used on any smelting
site in the region from chemical analysis of the slags. (We assume
The Schiel Complex (Fig. 1) is also a carbonatite complex. It has
that ore from the Phalaborwa Complex was not transported north
not been mined within the last 150 years, but oral histories state
across the Murchison Range, and conversely that ore from the
that this was a major source of iron ore for the Venda of the
Schiel Complex was not transported south across the Murchison
Soutpansberg, as is reflected in its Venda name Tshimbupfe, which
Range.) It has also been shown that forged iron artefacts produced
means “iron mountains” (Wessmann, 1908; Junod, 1927;
in the Lowveld can be traced to either the Rooiwater bands or to the
Schwellnus, 1937; Stayt, 1931; de Waal, 1942; Miller et al., 2002;
carbonatites by electron microprobe analysis of the slag stringers
Mathoho, 2012). It is thought to be of roughly the same age as
that they contain (Gordon and van der Merwe, 1984).
the Phalaborwa Complex but has a less well-defined annular
structure and has minimal copper mineralisation (Verwoerd,
4. Smelting sites and samples
1986; Stettler et al., 1993; Verwoerd and du Toit, 2006). The car-
bonatite, phoscorite and pyroxenite members form irregular
Table 2 lists 43 smelting sites recorded by van der Merwe and
subparallel lines and arcs in syenite. They have eroded faster than
his teams between 1965 and 1973, and by Killick between 1977 and
the syenite, and are thus obscured by soil, their presence being
1979. Slag samples were collected for analysis from many of these
indicated at surface largely by lines of magnetite rubble
sites in 1977 and 1978. This is far from a complete listing of known
(Verwoerd, 1986). Shallow trenches up to a hundred metres in
smelting sites in the northern Lowveld; for additional sites see
length from pre-European mining were still visible in the 1990s,
Evers (1981), Klapwijk (1986), Meyer (1986), Pistorius (1989) and
and there is a smelting furnace on the farm Schuynshoogte (VS2 in
Mathoho (2012).
Table 2) that was preserved to its full height of 1.2 m until the
1990s, but has since been vandalized. (Compare photographs in
4.1. Sites around the Phalaborwa Complex
Stayt (1931) and Miller et al. (2002) with that in Mathoho (2012)).
Much of the iron ore accessible to past miners derived from the
The outermost ring of the Phalaborwa Complex consists of a
phoscorite, which drill cores have shown to contain about 15%
circle of syenite pipes emplaced vertically through the Archaean
magnetite along with phlogopite mica, calcite, apatite and
granite. These are more resistant to erosion than the granite and
accessory biotite. Magnetite samples from drill cores contained
now form steep-sided hills. These hills were terraced in antiquity,
15% TiO2, averaging 3%, and contain only trace amounts of va-
and several investigators have excavated houses, smelting furnaces
nadium (Verwoerd, 1986; Mathoho, 2012). An unpublished report
and forges on these terraces and at the base of the hills (Mason,
by the mining company FOSKOR states that in the two polished
1968; van der Merwe and Scully, 1971; Meyer, 1986; Evers and
van der Merwe, 1987; Pistorius, 1989; Plug and Pistorius, 1999)
The earlier sites (10th13th centuries cal. CE) were open villages on
flat land at the bases of syenite hills. The later sites (17th19th
centuries cal. CE) are located on terraces built on the lower slopes of
these hills (Evers and van der Merwe, 1987).
The smelting sites of SPK4, SPK5 and SPK6 are located around
the base of Kgopolwe hill. Each had one or more iron smelting
furnaces. All excavated examples had a triangular plan with a
tuyère port at each apex (Evers and van der Merwe, 1987; Miller
et al., 2001). The distance from each apex to the opposite wall
was 1.2 m in the best-preserved furnace at SPK5 (Evers and van der
Merwe, 1987, Fig. 15), and the shaft height in intact examples is
about 1.0 m. Radiocarbon dates are 280  60 BP (Y-1658) for the
SPK4 furnace, 80  45 BP (Pta-339) for the SPK5A furnace and
520  60 BP (Y-1657) for a slag heap at SPK4 (Evers and van der
Merwe, 1987). A stratigraphic sequence of three house floors was
excavated by N. J. van der Merwe at SPK3 and yielded radiocarbon
dates ranging from 960  80 BP (Y-1638) to 820  80 BP (Y-1662).
These yielded a few iron artefacts, some of which are examined
below.
Fig. 4. Polished section of magnetite (sample PMC) from the Phalaborwa Complex
(reflected plane-polarized light (PPL), air). The medium grey crystals (brown in online
At the base of Matsepe Hill there are several clusters of undated
version) are titaniferous magnetite, oxidized to haematite along grain boundaries and smelting furnaces and slag heaps. Sites SPM2, SPM3 and SPM4 are
cracks. The two dark grey subhedral crystals are heavily altered apatite. unexcavated iron smelting sites. SPM2 had an intact triangular
244 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

Table 2
Smelting sites recorded by the 1970e1973 surveys (Thorne, 1974; Evers and van der Merwe, 1987) the 1978e1979 surveys (Killick, unpublished) and some other published
sources.

Site label Coordinates 1:50,000 map sheet Farm name Source Comments

Phalaborwa
SPK4 310746E 235600S 2331CC Phalaborwa Laaste 24LU Evers and van der Merwe (1987) Iron furnace
SPK5 310746E 235600S 2331CC Phalaborwa Laaste 24LU Evers and van der Merwe (1987) Iron furnace
SPK6 310746E 235600S 2331CC Phalaborwa Laaste 24LU Evers and van der Merwe (1987) Iron furnace
SPM1 310513E 235957S 2331CC Phalaborwa Wegsteek 30LU van der Merwe survey 1972/73 Copper furnace
SPM2 310513E 235955S 2331CC Phalaborwa Wegsteek 30LU van der Merwe survey 1972/73 Iron furnace
SPM3 310513E 235952S 2331CC Phalaborwa Wegsteek 30LU van der Merwe survey 1972/73 3 iron furnaces
SPM4 310513E 235952S 2331CC Phalaborwa Wegsteek 30LU van der Merwe survey 1972/73 Slag heap
Mashishimali hills and Murchison range
PQ1M 305623E 240313S 2430BB Mica Square 150 KT van der Merwe and Killick (1979) 7 iron furnaces
IH1M 305409E 240244S 2430BB Mica Hope 149 KT van der Merwe survey 1972/73 Iron furnace
IH2M 305445E 240233S 2430BB Mica Hope 149 KT van der Merwe survey 1972/73 Slag heap
IH4M 305358E 240313S 2430BB Mica Hope 149 KT van der Merwe survey 1972/73 Iron (?) furnace
IL2M 305343E 240424S 2430BB Mica Lillie 148 KT van der Merwe survey 1972/73 Iron furnace
IL4M 305358E 240422S 2430BB Mica Lillie 148 KT van der Merwe survey 1972/73 Slag heap
IL6M 305309E 240410S 2430BB Mica Lillie 148 KT van der Merwe survey 1972/73 Slag heap
IL8M 305115E 240428S 2430BB Mica Lillie 148 KT Killick survey 1978 Village/slag
IL12M 305017E 240411S 2430BB Mica Lillie 148 KT van der Merwe survey 1972/73 Iron furnace
IL17M 305134E 240423S 2430BB Mica Lillie 148 KT Killick survey 1978 Iron furnace
IL18M 305108E 240433S 2430BB Mica Lillie 148 KT Killick survey 1978 Slag heap
GG1M 305328E 235125S 2330DD Mulati Granville 767 LT van der Merwe survey 1972/73 Slag heap
GG2M 305205E 235118S 2330DD Mulati Granville 767 LT van der Merwe survey 1972/73 Slag heap
GQ1M 303445E 235149S 2330DC Gravelotte Quagga 759 LT Killick survey 1978 Iron furnace
GM3M 303036E 235848S 2330DC Gravellotte Maranda 675 LT van der Merwe survey 1972/73 Slag heap
GM5M 303017E 235738S 2330DC Gravelotte Maranda 675 LT Killick survey 1977 Slag heap
GM7M 302901E 235729S 2330CD Letsitele Maranda 675 LT van der Merwe survey 1972/73 Slag heap
GM9M 302904E 235724S 2330CD Letsitele Maranda 675 LT van der Merwe survey 1972/73 Slag heap
GM10M 303040E 235808S 2330DC Gravelotte Maranda 675 LT Killick survey 1977 Slag heap
GM11M 302926E 235809S 2330CD Letsitele Maranda 675 LT Killick survey 1977 Slag heap
GR3M 302816E 235639S 2330CD Letsitele Rooiwater 673 LT van der Merwe survey 1972/73 Slag heap
GR4M 302735E 235648S 2330CD Letsitele Rooiwater 673 LT van der Merwe survey 1972/73 Slag heap
GR6M 302804E 235431S 2330CD Letsitele Rooiwater 673 LT van der Merwe survey 1972/73 Slag heap
GN1M 302758E 235229S 2330CD Letsitele The Neck 565 LT Evers and van der Merwe (1987) Iron furnace
GN2M 302832E 235322S 2330CD Letsitele The Neck 565 LT van der Merwe survey 1972/73 Iron furnace
North of Murchison
EIL1 303933E 234254S 2330DA La Cotte Mabete 726 LT Evers (1982) Iron furnace
EIL2 303936E 234253S 2330DA La Cotte Mabete 726 LT Killick survey 1978 Slag
EIL3 304126E 234202S 2330DA La Cotte Mabete 726 LT Killick survey 1978 Ore and slag
EIL4 304219E 234036S 2330DA La Cotte Eiland 725 LT Killick survey 1978 Ore and slag
EIL5 304144E 234135S 2330DA La Cotte Eiland 725 LT Killick survey 1978 Slag
EIL6 304134E 234054S 2330DA La Cotte Eiland 725 LT Killick survey 1978 Iron furnace
GAZ1M 302705E 231935S 2330AD Hildreth Ridge State Land Killick survey 1979 Slag
GAZ2M 302652E 232058S 2330AD Hildreth Ridge State Land Killick survey 1979 Slag
Schiel complex
VS1M 302241E 230751S 2330AB Levubu Schuynshoogte 29LT van der Merwe survey 1972/73 Iron furnace
VS2M 302520E 230908S 2330AB Levubu Schuynshoogte 29LT Miller et al. (2002) Iron furnace
VD1M 302613E 231052S 2330AB Levubu Tabaan 55LT van der Merwe survey 1972/73 Slag

three-port furnace about 1.5 m high. SPM3 had three similar excavated around the Phalaborwa Complex. A very similar well-
smelting furnaces in fair condition encircled by a large ring of slag. preserved iron smelting furnace (IL-17M; Fig. 6) was excavated by
SPM4 was a slag heap without any visible furnaces. We selected Killick in 1978 on the adjacent farm Lillie, and yielded a radiocarbon
these sites as a reference group that can safely be assumed to have date of 250  45 BP (Pta-2408).
smelted magnetite ore from the Phalaborwa Complex. Each of the sites listed so far had visible furnace remains above
ground level, which explains the relatively recent radiocarbon
4.2. Sites on the Mashishimali Hills, south of the Murchison Range, dates obtained from them. The surveys of van der Merwe’s group
and near Tzaneen from 1970 to 1973 and of Killick in 1978 and 1979 found many other
sites that consist only of slag heaps. None of these are dated, but
The Mashishimali Hills (Fig. 1) are 20e30 km WSW of the slags from several of them are included in this study because they
modern town of Phalaborwa. They are of particular interest because offer opportunities to test our prediction that chemical analysis of
they host the largest known ancient iron smelting site in South slags can establish which source of ore was used. Coordinates for
Africa (PQ1M, on the farm Square). This is at least 20 km from any these sites are provided in Table 2. They include multiple sites from
potential source of iron ore (van der Merwe and Killick, 1979). The the farms Lillie and Hope in the Mashishimali Hills, and from the
site has seven furnaces surrounded by an estimated 180,000 kg of farms Rooiwater, Maranda and The Neck, all three of which are in
slag (Fig. 5), which is an order of magnitude larger than on any the Murchison Range just west of the town of Gravelotte (Fig. 1).
other smelting site in the northern Lowveld. There is a single Two smelting sites on the farms Granville are in the eastern
radiocarbon date on charcoal from the slag heap of 220  35 BP Murchison Range, north of the road linking Gravelotte to Phala-
(Pta-2402). The furnaces have the same three vertical slits for tu- borwa. Further west, Killick was shown several smelting sites
yères, but a more rounded plan than the triangular furnaces plan around Tzaneen by J. Witt and M. Klapwijk and provided with
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 245

Fig. 5. Plan of the iron smelting site PQ1M on the farm Square in the Mashishimali Hills (after van der Merwe and Killick, 1979, Fig. 3).

samples of slag and ore (see below). The locations of these are before being made into glass discs. The chemical analyses were
shown on Fig. 1. There are many other iron smelting sites around recalculated to remove the effect of this dilution. This worked well
Tzaneen; see Klapwijk (1986, Fig. 1) for the locations of an addi- for slag analyses but did not produce acceptable analyses for many
tional 36 sites in this area. of the ore specimens. Most analyses with totals outside the range
99.0e101.0% were discarded. Duplicate discs were prepared for all
4.3. Sites north of the Murchison Range and at the Schiel Complex samples and if both totals were acceptable the results were aver-
aged. Adsorbed water (H2O) was estimated by weighing before
On the north side of the Murchison Range there are six smelting and after heating overnight to 100  C. Samples were then heated in
sites (EIL1 through EIL6) on the farms Eiland and Mabete, near an a muffle furnace at 850  C to convert all ferrous iron to ferric iron,
important ancient saltworks (Evers, 1975, 1981, 1982). EIL1 is a and stored in a desiccator until preparation of the discs. Bound
furnace excavated by Evers (his 2330 DA:2), which produced a water (H2Oþ) and carbon dioxide were measured by a CO2/H2O
radiocarbon date of 270  40 BP (Pta-2505; Evers, 1982). EIL6M is analyser on a separate portion, and ferric iron by titration using
also a well-preserved furnace; the other four are just slag heaps or another weighed portion dissolved in HF and HNO3 in platinum
surface scatters. GAZ1M and GAZ2M are far to the north of Eiland, dishes.
near the Anderson Mission, and on the eastern side of the Middle X-ray diffraction (XRD) analysis was conducted on selected
Letaba River (Fig. 1). No systematic survey has been done around powdered samples at the University of Arizona in 1997, but the
the Schiel Complex (Fig. 1), but three smelting sites (VS1M, VS2M results proved difficult to interpret. This is because of extensive
and VD1M) were recorded by van der Merwe, and more recently solid solution series in most of the minerals present in these slags,
Mothoto (2012) has recorded additional smelting sites in the vi- so that their diffraction lines rarely matched those of the end
cinity. VS2M is the standing furnace that is depicted in Junod members of the series in XRD reference materials. Forty specimens
(1927), Miller et al. (2002) and Mathoho (2012). of ores and slags were prepared as polished epoxy mounts for study
in reflected light, and a further twenty specimens of slag as
5. Analytical methods 24  48 mm polished thin sections. Initial petrographic and
reflected-light examinations were mostly conducted by Killick and
Major- and trace-element compositions of selected slag, ore and Miller at the University of Arizona in 1997. The metallographic
sand samples were obtained by wavelength-dispersive X-ray analyses of iron objects were done by Miller. In 2012 Killick made
fluorescence (WD-XRF) in the Department of Geochemistry, Uni- more detailed studies of some of the slag and ore sections by optical
versity of Cape Town. Killick conducted the analyses, under the microscopy, and measured the composition of silicate and oxide
supervision of A. Duncan and J. Willis, between 1977 and 1979. phases in selected thin sections, using a Cameca SX-100 electron
About 500 g of washed sample was powdered in a rotary agate mill microprobe and natural mineral standards.
to pass a 400 mesh screen. To avoid cross-contamination, a load of
pure quartz was milled between each sample. Where more than 6. Results
one slag sample was analysed from a site, the samples are distin-
guished by suffixes (a through d). Major elements were measured 6.1. Slags from sites around the Phalaborwa Complex
on lithium metaborate glass fusion discs by the method of Norrish
and Hutton (1969). Because of their high iron content, samples of Samples SPM 2a, 2b, 2c, 3a, 3b, 4a, and 4b in Table 3 were all
ore and slag powder were diluted 1:1 with pure powdered SiO2 from the iron smelting sites around Matsepe hill. Comparison of
246 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

Fig. 6. Smelting furnace IL17M on the farm Lillie in the Mashishimali Hills. (Photo: D.
Killick).

these results with those for Phalaborwa magnetites in Table 1 Fig. 8. Partial phase equilibrium diagram and liquidus surfaces for iron oxide, silica
and titania under reducing conditions (redrawn after Fig. 3.292 in Verein Deutscher
showed that the slags had higher contents of TiO2. This indicates Eisenhüttenleute, 1995).
that the magnetite selected for smelting must have derived from
the outer edges of the phoscorite or from the pyroxenite of the
Phalaborwa Complex (Hanekom et al., 1965). (The low-TiO2 mag- rapid cooling, and a couple of slag sections had very thin skins of
netites in Table 1 were all from the transgressive carbonatite.) The magnetite. Since these slags were definitely not tapped, we think it
CaO contents of five of the seven samples were in the range 10e12 likely that the temporary seals on the vertical slits above the tu-
mass %, which is quite high for slags from bloomery iron smelting. yères were broken open while some of the slag was still liquid.
Four polished blocks and a polished thin section from SPM3 and Fig. 8 shows the experimentally determined equilibrium dia-
SPM4 were examined. All were dominated by small dendrites of gram (mass percent) for iron oxides, titania and silica under
wüstite (Fe1-xO) in a matrix of tiny parallel-sided laths of silicates reducing atmospheric conditions. The phase diagram correctly
and a dark glass crowded with crystallites. Almost all of the wüstite predicts the sequence of crystallisation seen in these slags, though
crystals were densely speckled with tiny cubes of a darker phase. kirschsteinite crystallises rather than the predicted fayalite
Examination of these by electron microprobe showed that they had (Fe2SiO4) because of the higher calcium contents in the slag. Note
higher titanium contents than their wüstite hosts, so they were also that both of the ternary eutectics in the iron-rich corner of the
tentatively identified as ulvöspinel, but they were too small for phase diagram are below 1200  C, very close to the temperatures of
quantitative analysis by microprobe. Polygonal crystals of titanif- the two eutectics in the binary FeOeSiO2 phase diagram. Because of
erous magnetite were the second phase to crystallise, and formed their high levels of CaO (5e12%) and Al2O3 (4e12%) we did not try
overgrowths on some wüstite dendrites (Fig. 7). The silicate phase to reduce the compositions of the slags in Table 3 to the three
identified by microprobe in a thin section from SPM3 (see Table 4) components of the FeOeTiO2eSiO2 phase diagram. As these oxides
was kirschsteinite (CaFeSiO4) with some monticellite (CaMgSiO4) in
solid solution. The microstructures were very fine, which suggests

Fig. 9. Typical slag microstructure from PQ1M (reflected PPL, air). Primary rounded
dendrites of wüstite (white) have overgrowths of darker subcubic titaniferous
Fig. 7. Microstructure of a slag from SPM4 (reflected PPL, air). White grains (A) are magnetite, which also forms dark spots within some wüstite dendrites (through solid-
metallic iron, the rounded grains (B) are wüstite, the angular grains (C) are titaniferous state exsolution). The matrix between the oxide phases is composed of subhedral grey
magnetite, and the dark laths (D) are fayalite or kirschsteinite. laths of fayalite and/or kirschsteinite in glass.
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 247

derived from ore from the lower magnetiteeilmenite band of the


Rooiwater Complex (compare with samples RW-18 and RW-19 in
Table 1 and with Reynolds, 1986). One of the three polished blocks
prepared from slags from IL8M had large subhedral and skeletal
ulvöspinels and late-forming silicates, which corresponded well to
the bulk chemical analysis, but the other two did not. Both of these
had microstructures very much like those described above for
PQ1M, which suggested that both derived from smelting Phala-
borwa ores. Each also contained a large (ca. 10 mm) grain of partly
dissolved magnetite. One of these had the regular array of tiny
spinel crystals, characteristic of magnetites from the central car-
bonatite member of the Phalaborwa Complex. The other (Fig. 10)
had exsolved fine laths of ulvöspinel along the octahedral (111)
planes of the host magnetite, which is characteristic of the higher-
TiO2 magnetite from the phoscorite and pyroxenite members of the
Phalaborwa Complex (Hanekom et al., 1965).

6.3. Slags from sites of the Murchison Range and near Tzaneen
Fig. 10. A partially-reduced piece of Phalaborwa ore within a slag from site IL8M
(reflected PPL, air). The matrix of low-Ti magnetite (light grey) in the lower half of the Samples GR3Ma, GR4Mb, GR6Ma, GM5M, GM7M and GM11M in
frame has exsolved small laths of ulvöspinel (medium grey) on (111) crystallographic Table 3 all had TiO2 contents above 20 mass %. The silica contents
planes. The magnetite has been reduced to metallic iron (white) around its exterior, were unusually low for bloomery slags, ranging from a low of 7.44
but the ulvöspinel laths were not reduced, and two can be seen within the iron rim at
to a high of 19.29%. The range of V2O5 measured in these slags was
upper left. Reduction of the magnetite by CO has formed large bubbles of CO2 (the
black cavities) that were trapped in the slag e at the top of the frame e when it 0.79e1.28%  significantly higher than in the slags from around the
solidified. Phalaborwa Complex (0.07e0.24%  see Table 3). The elevated
levels of TiO2 and V2O5 showed that these slags were produced by
smelting ores from the magnetiteeilmenite bands of the Rooiwater
have opposite effects e CaO would lower the melting temperature, Complex. The nearest outcrops of these bands are only 510 km
while Al2O3 would increase it e we expect liquidus temperatures in due north from each of these sites (Fig. 1).
these slags to be little different from those indicated in Fig. 8. Six polished blocks and two thin sections of these slags were
examined. All but one of these were cut from dense black blocks
6.2. Slags from sites of the Mashishimali Hills that evidently had solidified at the base of the furnace. The
exception, from GR3M, was from a rounded nodule that evidently
Four samples from the estimated 180,000 kg of slag at PQ1M in had solidified in charcoal above the base. All appear to have been
the Mashishimali Hills (Fig. 1) produced acceptable bulk chemical very fluid when molten, as the microstructures of all but the nodule
analyses (Table 3). These were taken from different slag piles on the were homogenous, with no flow boundaries noted. Two types of
site (Fig. 5). All had very similar chemical compositions. TiO2 varied microstructure were recorded.
from 4.13 to 4.73 mass %, and CaO from 9.7 to 13.7%. We examined In the first type 7590% of the area of the slide was taken up by
eight polished blocks and five thin sections of slag from PQ1M, each pale to dark pink subhedral cubic crystals of a spinel (Fig. 11). These
taken from a different portion of the slag pile. All were very similar were unzoned and were isotropic in reflected XPL. The small
samples that had abundant dendrites of wüstite and titaniferous amount of interstitial space was occupied by quenched feathery
magnetite, with the latter often seen as darker overgrowths on the dendrites of a silicate (visible only at high magnification) in glass. In
former. These occurred with small grains of iron metal and minute transmitted light these dendrites had a strong reddish brown
laths of a silicate in a glass (Fig. 9). Microprobe analysis of the sil- colour, which made it impossible to estimate birefringence. In only
icates in one thin section (PQ1Mb in Table 4) showed that they were two of these sections (one each from GR3M and GM1M) were
kirschsteinite. The microstructures were very fine and some pol- crystals of this phase large enough for analysis by electron micro-
ished sections showed thin skins of magnetite, suggesting that the probe. When analyses of this phase (Table 4) were recalculated by
vertical slits in the furnace wall above each tuyère were opened standard mineralogical techniques, the best match was achieved
while some of the slag was still molten. for 6 oxygens, when the number of cations closely approached 4.0.
The chemistry and microstructures of all of these samples were This suggests that this mineral is a calcium-iron pyroxene, possibly
essentially identical to most of the slags derived from magnetite related to the augite/ferroaugite/titanaugite complex, though the
from the phoskorite or pyroxenite zones of the Phalaborwa Com- TiO2 contents are much higher than in naturally-occurring augites.
plex, which is 20 km east of the site. There was very little glass.
IL6M is a slag scatter (no furnace) on the farm Lillie in the The compositions of spinels were determined by electron
Mashishimale Hills. As the slag contains only a half percent of microprobe in three thin sections, from GR3M, GR4M and GM1M
titania (Table 3) the ore from which it derived must have come from (Table 5). They showed no evidence of zoning under backscattered
the central carbonatite at Phalaborwa. SEM, and the TiO2 contents varied from 28 to 33 mass %. These
IL8M is on the southern flank of Ga-Mashishimale hill, 9 km to spinels are therefore solid solutions in which the dominant mole-
the west. In 1978 it was a rapidly eroding scatter about 50 m across cule is ulvöspinel (64.27% FeO, 35.73% TiO2). In terrestrial rocks
of diverse materials, including burned house daub, slag and a spill solid-state oxidation exsolution is inevitably seen in iron-titanium
of bronze. It is the only site in our sample to have provided evidence spinels (Haggerty, 1991), but rapid quenching to ambient temper-
for use of ores from both the Phalaborwa Complex (29 km ENE) and ature has preserved a metastable single phase in these slags.
the magnetiteeilmenite band of the Rooiwater Complex (minimum The second type of microstructure had elongated skeletal laths
distance 31 km NW). The bulk slag sample analysed from this site of ilmenite (FeTiO3) with skeletal cubes of pink ulvöspinel, in a
had 15.01 mass % TiO2 and 1.47% V2O5 (Table 3). This must have matrix of the same feathery silicate dendrites (Fig. 12). Ulvöspinel
248 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

Carmichael, 1967). Some of the late-forming spinel was probably


titaniferous magnetite rather than ulvöspinel, but it is difficult to
distinguish these in reflected light. This would also explain the
relatively high levels of ferric iron in some of these slags (Table 3).
The thin section from GR3M was composed of several small
flows fused to each other. One of these contained a mineral that
formed thin jagged laths like ilmenite, but which had much lower
reflectance and was not bireflectant. Electron microprobe analysis
of this phase (Table 5) showed a consistent composition with about
70% mass TiO2. We have been unable to identify this mineral.
We did not obtain acceptable chemical analyses for slag samples
from sites GG1M and GG2M in the eastern Murchison Range
(Table 2 and Fig. 1), but the four polished sections and one thin
section prepared from these slags were all dominated by crystals of
ulvöspinel solid solution, as large subhedral cubes and octohedra,
or as skeletal dendrites with octahedral facets. The space between
these was occupied by a silicate phase and glass. The microstruc-
tures were very similar to the first type described above from the
Fig. 11. Microstructure of slag from the Murchison Range (site GR4M, reflected PPL,
western Murchison range, so we deduced that the ores from which
air). The subhedral cubic and octahedral crystals are ulvöspinel; the interstitial ma- these slags derived came from one of the two bands in the Rooi-
terial consists of very fine feathery silicate dendrites in glass (not resolved at this water Complex, the nearest outcrop of which is about 15 km NW of
magnification). these sites.
In 1978 Killick was shown some smelting sites on the farms
Hasivona 561LT and Sivurahli 560LT, both of which are south east of
was the first phase to crystallise, followed by ilmenite. Electron
Tzaneen (Fig. 1), by local residents J. Witt and M. Klapwijk, who also
microprobe analyses of ilmenites in a thin section from GM1M
provided some slags for analysis. We do not have bulk chemical
(Table 5) gave an average composition very close to the composi-
analyses for these slags, but we examined a polished thin section
tion of pure ilmenite (52.65 mass % TiO2, 47.35% FeO), with the high
and two polished blocks from each site. All had very similar mi-
totals suggesting the presence of a small proportion of ferric iron.
crostructures, dominated by large subhedral cubes and dendrites of
Note that the initial crystallisation in these slags is in agreement
a spinel, with a small amount of late-forming ilmenite in jagged
with the FeOeTiO2eSiO2 phase diagram (Fig. 8), in that ulvöspinel
laths and small amounts of interstitial dark brown glass containing
is the first phase to crystallise. But continuing precipitation of
faint feathery dendrites of a silicate. Microprobe analysis of a thin
ulvöspinel should drive the composition of the remaining melt
section from Sivurahli (Table 5) showed that the spinel was close to
towards the fayalite field, not towards the ilmenite field. The
the composition of ulvöspinel and that the ilmenite was also near
presence of ilmenite in some of these slags probably means that the
its ideal formula (FeTiO3). This thin section was very similar to that
assumption on which Fig. 8 is based e that the atmosphere is
previously described for site GM1M in the central Murchison
strongly reducing e was violated in the late stages of solidification
Range.
of these slags. If the furnaces were opened while the slags were
Killick also collected a block of magnetite from a smelting site on
partially molten, the furnace atmospheres would become more
the farm Broederstroomdrift 534LT, just northeast of Tzaneen
oxidising, which would favour the simultaneous crystallisation of
(Fig. 1). Chemical analysis of this block (Table 1, analysis TB1)
ilmenite and titaniferous magnetite, as in many igneous rocks (e.g.

Fig. 12. Microstructure of slag from the Murchison Range (site GR3M, reflected PPL, Fig. 13. Microstructure of ore from site TB1 near Tzaneen (reflected PPL, air). The large
air). The bright skeletal laths showing bireflectance are ilmenite; the darker subhedral featureless crystals are ilmenite. The speckled areas around them were originally
cubes and octohedra are ulvöspinel; the grey matrix consists of feathery dendrites of titaniferous magnetite, but have been largely oxidised to hematite (white). The black
silicates in glass (not resolved at this magnification); and the black areas are voids from areas between former magnetite grains are a microcrystalline mixture of haematite
gas bubbles entrapped as the matrix solidified. and iron hydroxides, resulting from weathering of the magnetite.
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 249

EIL2Mb in Table 4). The other (EIL2Ma in Table 4) is the same un-
identified pyroxene previous noted in slags GR3M and GM1M from
the Murchison Range. The spinel (Table 5) is an ulvöspinel with
minor hercynite (FeAl2O4) in solid solution, and is very close in
composition to those already noted in slags GM1M, GR3M and
GR4M from the Murchison Range.
The thin silicate laths, though small, were better crystallised
than in other high-TiO2 slag samples. In thin section two different
silicates were noted. One was strong red-brown in PPL and had
inclined extinction in XPL; the other was colourless in PPL and had
third-order birefringence and straight extinction in XPL. These are
consistent with a clinopyroxene and an olivine respectively.
In 1978 two samples of slag from smelting sites to the east of the
Middle Letaba River e roughly on a line between Schiel and
Modjadji (Fig. 1) e were given to Killick by J. Witt of Tzaneen with
the co-ordinates of the sites from which they were collected. These
were analysed by WD-XRF (Table 3). The slag from site GAZ1M has
only 2.5 mass % TiO2, so we infer that this was produced with ore
Fig. 14. Microstructure of slag from VS2M (reflected PPL, air). Dendrites of wüstite
from Schiel, which is about 23 km distant in a straight line. The
(white) are surrounded by minute grey laths of fayalite or kirschsteinite in a darker sample from the other site (GAZ2M) had 23.7% TiO2, and must
grey glass. The darker speckles within the dendrites and the overgrowths on them are therefore have used ore from the upper band of the Rooiwater
titaniferous magnetite. Complex, the nearest outcrop of which is a straight-line distance of
73 km south of this site. This is the greatest distance between ore
source and smelting site yet documented in the Lowveld.
showed 23.22% TiO2 and 0.99% V2O5. The microstructure (Fig. 13)
consisted of coarse crystals of unaltered ilmenite surrounded by
6.5. Slag from a site on the Schiel Complex
crystals of titaniferous magnetite that had been largely oxidised to
haematite. This ore sample clearly derived from the magnetitee
Slag sample VS2M in Table 3 was from the well-preserved
ilmenite bands of the Rooiwater Complex. The nearest outcrop is
furnace depicted in Miller et al. (2002) on the farm Schuyn-
that reported by Schwellnus et al. (1962, p. 16) in the Thabina River
shoogte. This site is within 1 km of the ancient magnetite mines of
about 25 km south of Tzaneen, but we have not sampled this and
the Schiel Complex. As expected from the known composition of
there are no published analyses or micrographs of it. The ore
the Schiel magnetite, the TiO2 content of this slag was very low. This
sample and slags could derive equally well from the upper
particular sample was more diluted by undissolved silica than were
magnetite band of the Murchison Range north of Gravellotte, the
most other slags, so the original TiO2 content of the ore would have
nearest outcrop of which is about 40 km east of Tzaneen (Fig. 1).
been about 1.0 mass %. One thin section and one polished block
were examined. The thin section was dominated by fine dendrites
6.4. Slags from sites between the Murchison Range and the Schiel of wüstite in a matrix of tiny skeletal laths of a silicate and glass
Complex (Fig. 14). It was therefore richer in iron than the bulk chemical
analysis. Almost all of the wüstite dendrites had exsolved tiny cubes
Two slags from sites at Eiland (Fig. 1) were analysed (Table 3). of a darker phase, and a few had overgrowths of pink titaniferous
Slag sample EIL2M had 12.66 mass % TiO2 and 0.55% V2O5, which magnetite. Electron microprobe analysis of the very small laths
suggested that the ore came from the isolated magnetite lens 33 km (Table 4) showed that they were complex silicates rich in titania
SW of EIL2M at the contact between the Rooiwater Complex and and alumina. The polished block had much less wüstite and more
the Archaean granite (van Eeden et al., 1939; Reynolds, 1986). The titaniferous magnetite and silicate laths. It also had many eutectic
one analysis available from this lens (Table 1, analysis RW2) has the intergrowths (Fig. 15) of iron oxide in kalsilite (KAlSiO4). We as-
lowest TiO2 content of any magnetiteeilmenite sample yet pub- sume that abundance of the latter mineral in this section reflected
lished from the Rooiwater complex. (Samples from the lower band the presence in the sand used as flux of a substantial amount of
have 1416% TiO2 but their V2O5 content is much higher at 1.08e potassium feldspar from the syenite hills around the Schiel
1.37% - see Reynolds, 1986, Table 1). Sample EIL6M has 19.71 mass % Complex.
TiO2 and 0.95% V2O5, and thus used ore from the lower magnetitee
ilmenite band of the Rooiwater Complex, which is 25 km south of 7. Metallographic analyses
the site.
A slag sample published by Evers (1982) from an intact furnace There was very little iron preserved on any of the excavated
(his 2330DA:2; our EIL1M) only 200 m distant from EIL2M has 14% sites. The soils of the region are extremely acid, and preservation
TiO2. Thus all three furnaces at Eiland from which we have analyses depends strongly on a favourable microenvironment, like burial in
used ores from the Rooiwater Complex magnetite-ilmentite bands, an alkaline ash deposit. This is illustrated by the fact that the iron
but from three different locations within them. artefacts from Nagome, a 19th century site, were completely oxi-
We examined a thin section and three polished sections from dised, while the iron bead and iron strip from SPK3 (11th12th
these sites. In polished section the dominant phase is a spinel in centuries cal CE) still contained metal. We examined an iron bead
subhedral cubes and octahedra, with variable but always subordi- and an iron strip from SPK 3. Both were made from bloomery iron.
nate wüstite, metallic iron and short laths of a silicate as a late- They showed elongated stringers of slag, composed of wüstite in a
forming phase in the interstices between the oxide minerals. dark glass, transversely fractured by working below the glass
Microprobe analyses of the late-forming silicates in one of these transition temperature, which is generally between 500 and 700  C
samples (EIL2M) showed that at least two different minerals are (Babcock, 1977, p.26). Etched in nital, the bead (SPK3.1) consisted of
present. One is an olivine near to the end-member fayalite (analysis annealed equiaxed ferrite with a grain size of ASTM 67 (ASTM,
250 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

African iron smelting and also within the totality of bloomery


smelting technologies in the Old World. Most African bloomery
smelters used laterite ores, which are products of intensive
weathering on old stable land surfaces. The dominant iron minerals
in laterites are iron hydroxides, which lose water in the early stages
of smelting, creating open porosity that allows carbon monoxide
access to the interior. This promotes rapid reduction to iron
throughout lumps of ore (Killick, 1990; Killick and Gordon, 1988).
But laterite ores cannot form on eroding landscapes like those of
the Lowveld, so ironworkers there had to learn how to smelt the
available magnetite and magnetiteeilmenite ores. It is not yet clear
when they first did this. The oldest date for iron slag in a securely
dated context in this region is at Silverleaves (Klapwijk and
Huffman, 1996), which is on the escarpment near Tzaneen (Fig. 1)
and is associated with a radiocarbon date of 1680  55 BP (Pta-901),
but it is not known what ore was used to produce it.
In global context, magnetites were rarely used in bloomery
furnaces because of their lack of porosity. Reducing magnetite is
Fig. 15. Slag from VS2M with inferred higher potassium and titanium content than
like peeling an onion; reduction proceeds from the exterior of the
that in Fig. 14 (reflected PPL, air). Initial crystallisation of rounded dendrites of wüstite grain to the interior, with shells of reduced iron spalling off to allow
(white) was followed by subhedral cubes of titaniferous magnetite light (grey), and carbon monoxide to react with the next layer of oxide (Killick and
then by and eutectic intergrowths of wüstite (white) and kalsilite (black). The last Gordon, 1988). This is well illustrated in Fig. 10, which shows an
phase to crystallize was fayalite, which is the large parallel-sided light grey lath in the
incompletely reduced piece of magnetite in a slag sample from
upper half of the picture.
IL8M.
The reduction of lumps of magnetite to iron metal is therefore
1981), with intergranular pearlite islands. The overall carbon con- much slower than reduction of similarly-sized lumps of hydroxide
tent was about 0.1e0.2%. Elongated grains on the outer perimeter ores like laterites or bog iron. To ensure complete reduction as the
and near the cut ends were evidence of cold work subsequent to the ore descends through the furnace stack (which in the case of
final anneal. The bead had been cut out of strip and bent or bloomery furnaces is often less than one metre tall), magnetite
hammered to shape cold. The average Vickers microhardness was must be charged to the furnace as very small grains. Thus almost all
HV 137 (200 g for 10 s, n ¼ 9, range 112e156), the relatively high published instances of the use of magnetite ores in hand-blown
hardness reflecting extensive internal strain from cold work. A bloomery furnaces have noted the use of “black sands”, recovered
raster EDAX analysis detected only iron. from streams or beaches, with an average grain size of 1e2 mm. The
The iron strip (SPK3.2) consisted of very fine grained ferrite magnetites in these sands originated as accessory minerals in
(about ASTM 10), due to annealing after heavy cold work. There was igneous rocks. These were freed by erosion and concentrated by
prominent banding with variation in grain size and carbon content, streams or wave action because of their relatively high specific
probably due to variations in phosphorus content. The overall gravity.
carbon content was low, around 0.1%, but was difficult to assess Most of these records are from Africa (Cline, 1937; Brown, 1995;
because of the inhomogeneity. The Vickers microhardness was David et al., 1989; Ige and Rehren, 2002; Iles and Martínon-Torres,
relatively high for a ferritic microstructure, probably due to phos- 2009). Other recorded instances are from Macedonia (Photos et al.,
phorus in the segregation banding, with an average of HV 142 1984), Japan (Rostoker et al., 1989), Taiwan (Chen, 2000) and Korea
(n ¼ 5, range 134e154). A raster EDAX analysis in the SEM detected (Park and Rehren, 2011). The sands were typically beneficiated by
only iron as a major element. A spot analysis of a glassy inclusion washing or winnowing to raise their iron content, but still con-
indicated an iron silicate glass composition with about 4% phos- tained enough aluminosilicate minerals to form a slag. Slag is
phorus, 3% calcium and 2% potassium. required in most bloomery furnace designs for two reasons: (1) it
The microstructures and fabrication technology of these iron transports the small grains of iron to the base of the furnace, and
artefacts were typical of indigenous bloomery in southern Africa then drains away, pulling the iron grains together to form the
(Miller, 2002). No larger items of iron were recovered in excava- bloom; and (2) it protects the small grains of iron from reoxidation
tions. Six iron tools recovered during surface survey have been as they pass through the oxidising zone at the mouths of the
studied by Gordon and van der Merwe (1984). These had inho- tuyères.
mogeneous air-cooled microstructures of ferrite or air-cooled steel, The use of magnetite ores with up to 20% TiO2 was noted in an
with carbon contents up to 1 mass % C and numerous slag in- exploratory study of iron production at the 19th century site of
clusions. Two of the six objects had slag inclusions with high levels Marothodi near Rustenburg on the margins of the Bushveld Com-
of TiO2, indicating that they were smelted from the magnetitee plex, although the source of these ores is unknown (Hall et al.,
ilmenite ores of the Murchison Range. The lack of evidence for heat- 2006). The only other cases, to our knowledge, where massive
treatment of steel in these hoes conformed to the usual practice in magnetite ores were used in bloomery furnaces were in Korea in
the southern African Iron Age (Miller, 2002). the early first millennium CE (Park and Rehren, 2011) and in the
19th century iron industry of upper New York State, USA (Gordon
and Killick, 1992, 1993). The technology of the Korean bloomeries
8. Discussion has not yet been published in detail. The New York furnaces
employed water-powered bellows and preheated blast, and hence
8.1. The use of magnetite and magnetiteeilmenite ores are not directly comparable to the technology discussed here.
In the Lowveld bloomery furnaces, massive magnetite or
The Late Iron Age iron smelting technology of the northern magnetiteeilmenite ores containing less than 3 mass % silica were
Lowveld has several unusual aspects, both within the context of used. These had to be crushed to sand size, and silica or
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 251

Table 3
Chemical analyses of iron-smelting slags from the Lowveld, made by wavelength-dispersive X-ray fluorescence (WD-XRF) analysis of glass fusion discs (method of Norrish and
Hutton (1969). The samples are labelled by site; suffixes (a, b, c, d) indicate analysis of two or more samples per site.

Sample SiO2 TiO2 Al2O3 Fe2O3 FeO MnO MgO CaO Na2O K2O P2O5 V2O5 NiO CO2 H2Oþ H2O- TOTAL

Phalaborwa
SPM2a 30.79 4.42 6.58 7.42 33.34 0.19 2.65 10.09 0.94 2.20 0.42 0.18 0.07 0.58 0.30 0.13 100.3
SPM2b 23.60 7.62 4.63 2.12 42.39 0.28 3.48 12.05 0.30 1.87 0.66 0.12 n.d. 0.30 0.06 0.06 99.54
SPM2c 53.52 2.31 11.28 4.57 14.47 0.12 2.02 5.73 1.81 3.02 0.26 0.07 0.24 0.26 0.10 0.10 99.88
SPM3a 36.66 4.70 7.74 8.33 27.86 0.22 2.71 6.72 1.14 2.22 0.35 0.16 0.30 0.30 0.10 99.51
SPM3b 22.73 6.45 4.55 6.13 40.56 0.28 3.60 11.34 0.51 1.84 0.58 0.22 0.06 0.48 0.24 0.10 99.67
SPM4a 21.34 6.99 4.86 5.64 40.58 0.25 3.44 12.31 0.58 1.46 0.61 0.24 0.05 0.36 0.26 0.09 99.06
SPM4b 19.19 4.25 4.79 7.25 45.51 0.23 3.60 11.80 0.62 1.71 0.55 0.16 0.06 0.40 0.18 0.20 100.5
Mashishimali hills
PQ1Ma 20.10 4.39 4.76 9.50 43.34 0.28 3.77 9.71 0.47 1.37 0.60 n.d. n.d. 0.56 0.40 0.42 99.67
PQ1Mb 20.25 4.68 4.28 6.83 43.40 0.24 3.60 12.33 0.44 1.45 0.58 n.d. n.d. 0.72 0.22 0.25 99.27
PQ1Mc 21.89 4.73 4.48 6.14 40.62 0.26 3.67 13.68 0.44 1.53 0.69 n.d. n.d. 0.66 0.20 0.36 99.35
PQ1Md 17.85 4.13 4.12 9.79 45.52 0.24 3.50 10.59 0.41 1.27 0.50 n.d. n.d. 0.36 0.16 0.29 98.73
IL6M 32.45 0.55 7.00 15.79 36.86 0.08 0.76 3.08 1.03 2.13 0.25 0.03 0.17 0.32 0.36 0.36 101.22
IL8M 10.21 15.01 4.86 10.01 50.04 0.31 1.70 4.32 0.24 0.69 0.13 1.47 0.13 0.40 0.40 0.27 100.19
Murchison range
GR3Ma 17.16 20.90 5.37 2.62 40.59 0.30 1.20 7.54 0.12 0.83 0.42 0.91 0.06 0.48 0.26 0.17 98.93
GR4Mb 11.83 23.87 3.97 8.89 41.83 0.37 1.56 4.47 0.05 0.63 0.28 0.79 0.11 0.18 0.06 0.08 98.97
GR6Ma 12.94 21.13 5.20 8.53 42.23 0.30 0.99 6.10 0.13 0.53 0.23 0.79 0.01 0.46 0.40 0.09 100.06
GM5M 7.44 25.18 3.97 5.53 48.99 0.36 1.42 3.36 0.06 0.60 0.13 1.28 0.08 0.22 0.08 0.07 98.77
GM7M 19.29 21.31 5.15 2.42 37.4 0.75 1.14 9.13 0.19 0.95 0.30 0.81 0.08 0.40 0.28 0.22 99.82
GM11M 17.55 21.86 5.24 2.70 38.55 0.41 0.93 9.58 0.21 1.04 0.32 0.96 0.09 0.38 0.2 0.06 100.08
North of Murchison
EIL2M 19.36 12.66 4.86 9.14 41.56 0.40 1.92 6.40 0.22 0.52 1.10 0.55 0.03 0.36 0.52 0.25 99.85
EIL 6M 10.43 19.71 3.86 13.76 45.31 0.25 0.72 2.88 0.17 0.47 0.08 0.95 0.02 0.24 0.18 0.33 99.34
GAZ1M 27.51 2.50 6.22 6.97 40.68 0.51 2.43 7.39 0.10 3.02 1.46 0.15 0.00 0.18 0.16 0.06 99.34
GAZ2M 12.37 23.67 4.58 5.40 42.49 0.39 1.02 6.45 0.13 0.74 0.36 0.70 0.01 0.22 0.24 0.09 98.86
Schiel complex
VS2M 41.37 0.49 10.37 1.66 36.68 2.01 0.82 0.65 0.23 0.96 1.75 0.02 0.38 0.68 0.64 0.48 99.19

aluminosilicate flux then had to be added to form slag. At all Pha- 2002) and we assume that they were also used in earlier times,
laborwa sites and at PQ1M ‘Square’ (van der Merwe and Killick, as there is no historical record of the use of carved wooden or fired
1979) pieces of crushed magnetite 510 mm were recovered with ceramic pot bellows in South Africa. Slag was not tapped from these
a magnet from around the furnaces, and the large syenite boulders furnaces.
and bedrock around Kgopolwe hill (the locus for sites and samples In the hills around the Phalaborwa Complex (Pistorius, 1989,
with a prefix of SPK) are pockmarked by grinding hollows, from Evers and van der Merwe, 1987), at Eiland (Evers, 1975, 1981,1982)
which magnetite grains can still be extracted with a magnet. The and in the foothills around Tzaneen (Klapwijk, 1986) all iron
final size for ore presented to the furnace is not known, but we smelting furnaces recorded to date have a triangular plan, with a
assume that it was probably less than 2 mm, as we have rarely seen vertical slit at each apex. In the Mashishimali Hills (van der Merwe
pieces of incompletely reacted ore in the slags that we have studied and Killick, 1979) and in the Soutpansberg (Miller et al., 2002) they
in polished section (although one of these is shown in Fig. 10). have a circular plan. Another variable aspect of furnace design is the
presence or absence of a small cylindrical pit, sometimes found
8.2. The smelting furnaces covered by a flat slab of stone, in the centre of the floor of the
furnace. Where present, most of these are empty, but finds of hu-
Archaeological evidence shows that two types of smelting man finger bones in several such pits around Phalaborwa (Plug and
furnace were employed in the Lowveld in the second millennium Pistorius, 1999) suggest that their function was to provide protec-
CE. A forced-draft shaft furnace with three tuyères was used to tion against magical spells cast by others.
smelt iron, while a much smaller domed furnace, blown through a
single tuyère, served for smelting copper (Evers and van der 8.3. The use of sand flux and the effect of wood ashes
Merwe, 1987; Miller et al., 2001). All intact iron smelting furnaces
recorded to date, whether triangular or circular in plan, had shaft As all the potential ores in this region are essentially pure iron
heights of about one metre and had three vertical slits 120 apart, and titanium oxides, it was obvious from the beginning of this
sometimes extending the full height of the shaft, separated by a study that a silicate flux must have been added to the furnace
distance of 1.0e1.2 m from the opposite wall. These slits were charge to produce slag. The FeOeSiO2 phase diagram (Levin et al.,
clearly for insertion of tuyères, and there is good evidence on some 1964, Figs 80 and 81) shows that pure FeO melts at 1377  C un-
furnaces that the slits above the tuyères were plastered shut when der atmospheric conditions. In practice a temperature at least
the furnace was in use. Since used tuyères on these sites are typi- 100  C above this would be required to provide a free-flowing slag,
cally slag-wetted for 1520 cm from the nose, we assume that a and it would be very difficult to maintain a temperature around
single iron bloom accumulated in the centre of the furnace, and that 1500  C in a hand-powered bloomer furnace. But if the FeO is fluxed
the purpose of the slits was to allow the solid bloom to be levered with between 20 and 40 mass % SiO2, liquidus temperatures are
out, using three poles, without damaging the furnace walls. It is lowered into the range 1178e1205  C (Levin et al., 1964, Figs 80 and
clear from the map of PQ1M (Fig. 5) that these furnaces could be 81). With 100  C superheat to ensure free flow this translates to
used for many smelts. Bag bellows made of leather were used in the 12001300  C, a range of temperature easily attained under
Soutpansberg in the late 19th century (Junod, 1927; Miller et al., reducing conditions in a bloomery furnace.
252 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

Table 4
Electron microprobe analyses (in mass %) of silicates in thin sections of slags from selected sites.

Sample SiO2 Al2O3 TiO2 FeO MnO MgO CaO Na2O K2O P2O5 Total Comment

Phalaborwa
SPM3 32.11 0.22 0.38 33.06 0.27 5.61 26.65 0.23 0.12 0.90 99.55 Kirschsteinite
Mashishimali hills
PQ1Mb 32.44 0.24 0.38 33.12 0.33 6.76 24.92 0.15 0.21 0.66 99.21 Kirschsteinite
Murchison range
GR3M 39.53 8.48 7.22 18.62 0.21 2.85 22.55 0.17 0.01 0.45 100.09 Pyroxene
GM1M 41.48 7.05 4.53 22.53 0.21 1.73 22.25 0.11 0.01 nd 99.90 Pyroxene
North of Murchison
EIL2Ma 39.53 9.76 5.67 19.69 0.21 2.34 21.61 0.21 0.16 nd 99.18 Pyroxene
EIL2Mb 31.13 0.20 0.57 58.05 0.74 6.68 2.34 0.02 0.04 nd 99.77 Olivine
Schiel complex
VS2M 26.31 13.41 8.58 33.32 0.26 4.73 12.67 0.11 0.06 0.18 99.63 Unidentified

The composition of the slags (Table 3) differs from that of the % e in both groups. We cannot explain why this should be the case.
magnetite ores (Table 1) mainly in the much higher SiO2, Al2O3 and Magnetite is associated with calcite in the carbonatite and phos-
CaO contents of the slags. Silica and alumina could potentially have corite members of the Phalaborwa Complex, but the relatively high
entered the slag from chemical attack of the tuyères and furnace TiO2 levels in the slags from adjacent sites (those with SPM prefixes
walls by liquid iron oxides, but inspection of the analyses in Table 3 in Table 3) show unequivocally that the magnetite ore use to pro-
will show that the silica to alumina ratios are in most cases duce them came from the pyroxenite member of the Phalaborwa
significantly higher than the 3:1 ratio in illite, the dominant clay Complex, in which calcite is not abundant (Hanekom et al., 1965). It
mineral in Lowveld soils. Crushed quartz could have been added, is possible that the differences in the CaO contents reflect the use of
and some was noted around the furnaces at PQ1M (van der Merwe different species of charcoal to smelt ore from the Phalaborwa and
and Killick, 1979), but another possible flux is the granitic sand that Rooiwater Complexes, but we have not identified the tree species
is ubiquitous in Lowveld stream beds. In 1978 Killick collected used to produce the charcoal at each site.
samples of sand for chemical analysis from the bed of the Selati
River. These sands derive from the Archaean granite gneiss that 8.4. The high titanium slag system
forms the bedrock over most of the northern Lowveld.
In Table 6 we compare the compositions of a Rooiwater ore Many of the slags reported here have from 12 to 25 mass % TiO2.
(from Table 1), two analyses of sands from the Selati River, and a We can find no published record of such high levels of TiO2 in
slag from the Mashishimali Hills (from Table 3). This suggests that bloomery iron smelting slags. These slags also have very unusual
addition of these sands can explain the elevation of SiO2, Al2O3, K2O microstructures, with up to 90% by area in polished section con-
and Na2O concentrations in the slag, and the corresponding dilu- sisting of crystals of ulvöspinel, a mineral that is seen in geological
tion of TiO2 and iron oxides. However, it cannot explain the high contexts only as exsolved oxidation lamellae within crystals of
levels of CaO in the slags (Table 3). The sands contain more K2O magnetite (Haggerty, 1991). One of our more interesting findings is
than CaO, but all the slags contain much more CaO than K2O. The that slags with 20e25 mass % TiO2 do not seem to have been any
most likely explanation for this inversion is that most of the CaO less fluid than those with less than 10% TiO2. Slags that solidified
derives from the ash content of the charcoal fuel. The preferred tree among charcoal pieces have smooth glossy well-rounded surfaces,
for making charcoal in the Lowveld today is leadwood, Combretum while those from the pool of slag at the furnace base form dense
imberbe. When burned this produces copious quantities of white blocks without visible boundaries between flows. The high-TiO2
ash. We analysed one sample of ash from this species and found slags produced from the Rooiwater ores are black in freshly broken
that it was about 90 mass % CaO when H2O and CO2 were removed section, while those with lower TiO2, deriving from Phalaborwa and
from the analysis and the total normalised to 100%. Slags from the Schiel ores, are usually medium grey. No evidence for the tapping of
vicinity of the Phalaborwa Complex tend to have higher CaO con- liquid slag from the furnaces was noted.
tents than those derived from the magnetiteeilmenite ores of the It appears that in bloomery furnaces TiO2 is essentially equiva-
Rooiwater Complex (Table 3). This is a real difference, as the totals lent to SiO2  neither is reduced, and both form highly fluid slags by
of (SiO2 þ TiO2 þ Al2O3) vary within the same range e 35e45 mass reaction with FeO. The minimum liquidus temperature in the

Table 5
Electron microprobe analyses (in mass %) of oxide phases in thin sections of slags from selected sites.

Sample MgO Al2O3 TiO2 Cr2O3 FeO MnO SiO2 CaO V2O3 NiO Total Comment

Mashishimali hills
PQ1Mb 1.45 4.63 24.83 0.07 66.60 0.22 0.87 0.57 0.24 0.01 99.52 Spinel
Murchison range
GR3M 0.82 2.55 32.77 0.16 63.00 0.46 0.08 0.20 0.44 0.02 100.51 Spinel
GR4M 1.10 2.95 27.67 0.18 67.26 0.32 0.08 0.03 0.68 0.01 100.26 Spinel
GM1M 0.86 2.50 32.96 0.21 63.23 0.36 0.08 0.04 0.83 0.02 101.09 Spinel
GM1M 0.86 0.29 52.63 0.08 46.62 0.41 0.04 0.11 0.38 0.01 101.43 Ilmenite
GR3M 5.37 1.54 71.37 0.41 17.31 0.28 0.28 0.87 3.99 0.02 101.44 Unidentified
Near Tzaneen
Sivurahli 1.39 3.86 33.98 0.21 60.11 0.50 0.05 0.06 0.87 0.01 101.03 Spinel
Sivurahli 1.35 0.19 52.00 0.10 45.14 0.63 0.39 0.46 0.14 0.01 100.32 Ilmenite
North of Murchison
EIL2M 1.25 3.48 30.50 0.32 62.75 0.41 0.15 0.19 0.64 0.02 99.70 Spinel
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 253

Table 6
Chemical analyses of a magnetiteeilmenite ore (from Reynolds (1986) and Table 1), two sands from the bed of the Selati River (Killick, unpublished) and slag from site IL8
(Table 3). This comparison is for illustration only; it is not suggested that this ore produced this slag.

Sample SiO2 TiO2 Al2O3 Fe2O3 FeO MnO MgO CaO Na2O K2O P2O5 V2O5 CO2 H2Oþ H2O- Total

Ore RW27 2.66 20.83 3.83 48.37 22.50 0.32 1.68 n.d. n.d. n.d. n.d. 0.80 n.d. n.d. n.d. 100.99
Sand A 84.69 0.14 7.57 0.22 1.09 0.01 0.29 0.92 2.41 2.16 0.02 0.00 0.20 0.16 0.03 99.91
Sand B 76.91 0.11 11.88 0.24 1.18 0.02 0.30 1.34 2.99 4.12 0.04 0.00 0.20 0.21 0.03 99.56
Slag IL8M 10.21 15.01 4.86 10.01 50.04 0.31 1.70 4.32 0.24 0.69 0.13 1.47 0.40 0.40 0.27 100.06

binary system TiO2eFeO is 1300  C, near the limit for average 8.6. The mass balance and yield
temperature across the entire profile of a bloomery furnace, so
some SiO2 must be added to FeOeTiO2 to obtain slags that melt in Unfortunately it is not possible (contra van der Merwe and
the optimal range for the bloomery furnace (11001200  C). Liq- Killick, 1979) to calculate the mass of magnetite required to pro-
uidus temperature as low as 1145  C occur in the system TiO2e duce a given mass of slag, nor to infer the yield of iron from our
FeO(n)eSiO2 under reducing conditions (Fig. 8). Comparison of the chemical analyses of ores and slags. This is because we cannot know
chemical compositions of ore (Table 1) and slag (Table 3) compo- how much slag was produced in each smelt. Since the magnetites
sitions shows the ironworkers of the Lowveld invariably added from the carbonatite complexes are not self-fluxing, the amount of
siliceous flux. The actual melting temperatures of these slags are slag formed was determined by the quantity of aluminosilicate flux
affected by oxides not accounted for in the phase diagram, espe- that the smelters chose to mix with the crushed ore. Too little flux
cially CaO and Al2O3, but it seems safe to say that they would be would adversely affect the fluidity of the slag, and thus the
under 1250  C. consolidation of the bloom; too much would lower the yield of iron
metal. The ratio of magnetite to flux in the furnace charge cannot be
inferred from the composition of the slag.
8.5. The presence of ferric iron

In all of these slags metallic iron coexists with ferric iron (Fe3þ), 8.7. Provenance of iron ores
which is not possible under equilibrium conditions. Table 3 shows
that ferric iron (determined by titration) accounts for between 2 We have shown that chemical analysis of slags from archaeo-
and 16 mass % of these slags. Some of this may be due to subsequent logical sites in the Lowveld provides unequivocal evidence of the
corrosion, but few specimens showed any signs of it, either in hand sources of the iron ores used to produce the slags. The Phalaborwa
specimen or when examined in thin section or polished blocks. We and Schiel Complexes are essentially point sources at the scale of
have noted mineralogical and chemical evidence in some slags for a the region shown in Fig. 1, while the main exposure of the
change of atmosphere from reducing to oxidising while the slags magnetite-ilmentite bands in the Rooiwater Complex is only about
were partially molten, almost certainly because the slits above the 14 km long, with two smaller outcrops between this and the
tuyères were in some cases opened to extract the bloom while hot. western escarpment. The ore, being very dense, was not moved any
We suggest that most of the ferric iron is therefore in the last significant distance from outcrop by alluvial processes, so it is easy
minerals to crystallise and in the glass. to establish the minimum distance over which ore must have been
The electron microprobe cannot distinguish between the carried.
different valence states of elements, and thus valence must be Ore was in several instances carried more than 20 km to
assumed when calculating microprobe data as oxides. By conven- smelting sites, and in one case (GAZ2M) more than 70 km. Oral
tion, iron in electron microprobe analyses is calculated as FeO, but histories also mention ore being carried as far as 55 km from the
various post-hoc schemes for deriving the proportion of ferric iron Schiel Complex (Giesekke, 1930; Krige, 1941). In the tradition
in particular minerals from microprobe analyses have been pro- recorded by Giesekke (1930) the use of human porters is specif-
posed. Since ulvöspinel solid solutions are by far the most abundant ically mentioned. PQ1M, the largest smelting site in the region, has
minerals in our titanium-rich slags, we also checked whether there an estimated 180,000 kg of slag (van der Merwe and Killick, 1979).
could be ferric iron in some of the spinels. We calculated the pro- From the site plan (Fig. 5) it appears that this was produced in a
portion of ferric iron from the microprobe analyses of spinels in relatively short span of time. We cannot accurately estimate the
Table 5 according to the widely-used recalculation scheme of quantity of ore brought from the Phalaborwa Complex, 20 km away,
Carmichael (1967) and obtained the following results: but it clearly must have exceeded 200,000 kg. With no historical
evidence for the use of pack animals to transport ore in the Low-
veld, this represents several thousand man-journeys. We do not
know why the scale of iron production is so much larger at PQ1M
GR4M 55.84% FeO 12.70% Fe2O3 than at any other site in the region. There is no obvious explanation
GR3M 60.35% FeO 2.95% Fe2O3 to be found in distribution of resources  both fuel wood and clay
GM1M 60.96% FeO 2.62% Fe2O3 would have been available much closer to the ore source. There is
Sivurahli 60.99% FeO 0.98% Fe2O3
unfortunately too little supporting archaeological and historical
data for us to decide whether: (a) the Phalaborwa ore deposit was
These results suggest that in at least one specimen (GR4M) the an open resource, owned by no-one, or (b) access to the ore deposit
spinel is a solid solution of titaniferous magnetite in ulvöspinel. In reflected alliance with, or a negotiated agreement between, a group
the other three cases the calculation shows that the iron is almost asserting ownership of this ore source and a group resident in the
all ferrous, and thus that most of the ferric iron in the bulk analyses Mashishimali Hills.
must be in another phase (magnetite?) or in the glass. Neither of We have demonstrated that the source(s) of ore for iron
the silicate phases in these slags can contain much ferric iron smelting sites in the northern Lowveld can be established by
because their totals in microprobe analysis are close to 100% chemical analysis of the slag. This provides archaeologists in this
(Table 5). region with the means to study the regional organisation of iron
254 D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255

production. Slag inclusions trapped in iron artefacts produced in techniques to unfamiliar ores. Whether iron was ever exported
this region can also be related to the source of ore used to produce from the Lowveld west to the Highveld, or east to Mozambique, is a
them (Gordon and van der Merwe, 1984). The main barrier to question for future researchers to investigate. But within the
reconstructing the regional organisation of iron production is that Lowveld future archaeological studies can use the composition of
of dating the sites. Many of the radiocarbon dates cited in this ore or slag fragments on smelting sites as evidence for the distances
article are less than 300 BP, and thus fall into a well-known “black travelled to procure ores.
hole” in the radiocarbon calibration curve. (After calibration at two
sigma, all dates in the range 30050 BP fall in the unhelpful cal-
Acknowledgements
endar age range of ca. 16401900 CE). The fact that most of the
radiocarbon dates for iron production in the northern Lowveld are
Most of the sites reported here were discovered by teams led by
300 BP or less reflects the preferences of archaeologists for exca-
Professor Nikolaas van der Merwe between 1970 and 1973. In the
vating well-preserved furnaces. Iron production in the Lowveld
mid-1970s both of the present authors were undergraduate stu-
goes back at least to the 11th12th centuries cal. CE (as shown by
dents in geology and archaeology at the University of Cape Town
the iron samples from SPK 3, described above) and to at least 350e
(UCT), and were encouraged by van der Merwe to take up the study
450 cal. CE on the western escarpment (Klapwijk and Huffman,
of archaeometallurgy by working on this material, which we have
1996). Radiocarbon dating has useful precision for investigating
done intermittently over the last thirty five years. We are both most
the prehistory of iron production in the Lowveld before ca. 1640 cal.
grateful to him for his mentorship, support and friendship. This
CE (see Fig. 2 in Miller et al., 2001), but all more recent sites will
work could not have been carried out without the generous help of
need to be redated by archaeomagnetic dating, thermolumines-
Jurgen Witt and Menno Klapwijk, both of Tzaneen, and of Charles
cence or optically-stimulated luminescence dating of furnace
More, Ike Lombaard and Jan Scholtemeyer, all of Phalaborwa. Each
ceramics.
of them shared with us their deep knowledge of indigenous mining
and metal working in the northern Lowveld. We also wish to thank
9. Conclusions
the Department of Archaeology, UCT for access to specimens in
their care, and Professor Nikolaas van der Merwe, Professor W. J.
This study reports the direct smelting of bulk magnetitee
Verwoerd, Dr Julius Pistorius, Dr Udo Küsel, Professor Thilo Rehren
ilmenite ores, with slags containing up to 25% TiO2 as the principal
and Professor Robert Heimann for valuable discussion. We are most
waste product. Although these ores are of potential interest today
grateful to the Phalaborwa Mining Company and to FOSKOR for
as sources of titanium and vanadium, they cannot be used to pro-
allowing us to use data from unpublished reports. For assistance in
duce iron in modern blast furnaces. In the highly reducing atmo-
the analysis of samples we thank: the Director, Electron Microscope
sphere of the blast furnace, titanium 4þ oxides are reduced to
Unit, UCT; the Department of Geochemistry, UCT (and especially Dr
mixtures of titanium nitride (TiN), titanium carbide (TiC) and Ti2þ
Andy Duncan and Dr James Willis); and at the University of Arizona,
and Ti3þ oxides, which in combination make blast furnace slags
Gary Chandler for assistance with electron microscopy, Dr Ken
unacceptably viscous (McGannon, 1971). Even the magnetites of
Domanick for the electron microprobe and Mr. Phil Anderson for X-
the Phalaborwa Complex, which have much lower titanium con-
ray diffraction. Archaeological fieldwork by van der Merwe’s team
tents than those of the Rooiwater Complex, are mostly unusable
was funded by a Ford Foundation Foreign Area Fellowship and by
today. A few million tons with TiO2 content below 2 mass % have
National Science Foundation Grants GS-281 and GS-2744. Analysis
been sold, but hundreds of millions of tons of powdered magnetite
of samples was partly funded by National Science Foundation Grant
with higher TiO2 concentrations reside in waste dumps around the
SBR-9602033 to Killick, and partly by multiple grants to Miller from
modern copper mine. Yet the Late Iron Age metalworkers of the
the National Research Foundation, South Africa, Anglo American
Lowveld had no difficulty in producing pure iron and highly fluid
PLC, De Beers PLC, and AngloGold PLC. We are most grateful to all
slags from titaniferous magnetite ores by employing the bloomery
these institutions for their support and to the three referees whose
process.
detailed criticisms and suggestions significantly improved this
Although there is evidence of the bloomery smelting of
paper. Opinions expressed in this report, and conclusions arrived at,
magnetite sands elsewhere in sub-Saharan Africa, and occasional
are those of the authors and are not to be attributed to any of the
reports from elsewhere (Japan, Korea, Taiwan, Macedonia and the
supporting agencies.
eastern United States), this is the first published study to report the
smelting of high-titanium iron ores. This finding serves to
emphasize the point that though the bloomery smelting process References
was less productive than the blast furnace process, it was far more
versatile. Bloomeries can smelt iron ores with high levels of ASTM, 1981. Standard Methods of Estimating the Average Grain Size of Metals. In:
ASTM Standards e Part II. E 112. American Society for Testing and Materials,
phosphorus, arsenic, antimony, aluminium and nickel (Rostoker Philadelphia, PA, pp. 186e220.
and Bronson, 1990), none of which are suitable for modern blast Babcock, C.L., 1977. Silicate Glass Technology Methods. Wiley, New York, NY.
furnaces. To this list we can now add iron ores high in titanium and Bernhard, F.O., 1971. Karl Mauch, African Explorer. Struik, Cape Town.
Brandl, G., Cloete, M., Anhaeusser, C.R., 2006. Archaean greenstone belts. In:
vanadium. Johnson, M.R., Anhaeusser, C.R., Thomas, R.J. (Eds.), The Geology of South Africa.
There is little evidence of any human occupation of the northern Council for Geoscience, Pretoria, pp. 9e56.
Lowveld in the first millennium CE, but in the second millennium Brown, J., 1995. Traditional Metalworking in Kenya. In: Cambridge Monograph in
African Archaeology, vol. 38. Oxbow Books.
(the Later Iron Age) settlers appear to have been attracted to this Carmichael, I.S.E., 1967. The iron-titanium oxides of sialic rocks and their associated
region by its wealth of inorganic resources, mostly metal ores and ferromagnesian silicates. Contrib. Mineral. Petrol. 14, 36e64.
saline springs from which salt could be obtained (Evers, 1981). We Ph.D. thesis Chen, K.T., 2000. Ancient Iron Technology of Taiwan. Harvard
University.
do not believe that settlers were specifically attracted to the Low-
Cline, W.W., 1937. Mining and Metallurgy in Negro Africa. George Banta Publishing
veld by the vast quantities of magnetite and magnetiteeilmenite Company, Menasha, WI.
ores in the region. As we shall show in a separate study (in prep- David, N., Heimann, R., Killick, D.J., Wayman, M., 1989. Between bloomery and blast
aration), the saline springs and copper ores of the Lowveld were the furnace: Mafa iron-smelting technology in North Cameroon. Afr. Archaeol. Rev.
7, 185e210.
main attraction. But the settlers would have needed iron, and de Waal, J.B., 1942. Ysterbewerking deur die Bawenda en die Balemba in Sout-
through experiment they successfully adapted bloomery smelting pansberg. Tydskr. vir. Wet. Kuns 3, 45e53.
D. Killick, D. Miller / Journal of Archaeological Science 43 (2014) 239e255 255

Evers, T.M., 1975. Recent iron age research in the eastern Transvaal, South Africa. Miller, D.E., Malaudzi, M., Killick, D.J., 2002. An historical account of bloomery iron
South Afr. Archaeol. Bull. 30, 71e83. working in the Lowveld, South Africa. Hist. Metall. 36, 112e121.
Evers, T.M., 1981. The iron age in the eastern Transvaal. In: Voigt, E.A. (Ed.), Guide to Norrish, K., Hutton, J.T., 1969. An accurate X-ray spectrographic method for the
Archaeological Sites in the Northern and Eastern Transvaal. Transvaal Museum, analysis of a wide range of geological samples. Geochim. Cosmochim. Acta 33,
Pretoria, pp. 64e109. 431e453.
Evers, T.M., 1982. Two later iron age sites on Mabete, Hans Merensky Nature Palabora Mining Company Geological Staff, 1976. The geology and the economic
Reserve, Letaba District, NE Transvaal. South Afr. Archaeol. Bull. 37, 63e67. deposits of copper, iron and vermiculite in the Palabora Igneous Complex. Econ.
Evers, T.M., van der Merwe, N.J., 1987. Iron age ceramics from Phalaborwa, north Geol. 71, 177e192.
eastern Transvaal Lowveld, South Africa. South Afr. Archaeol. Bull. 42, 87e106. Park, J.-S., Rehren, T., 2011. Large-scale 2nd to 3rd century AD bloomery iron
Giesekke, E.D., 1930. Die Eisenindustrie der Bawenda. Die Brücke, pp. 5e9. smelting in Korea. J. Archaeol. Sci. 38, 1180e1190.
Gordon, R.B., Killick, D.J., 1992. The metallurgy of the American bloomery process. Photos, E., Koukouli-Chrysanthaki, H., Gialoglou, G., 1984. Iron metallurgy in Eastern
Archeomaterials 6, 141e167. Macedonia: a preliminary report. In: Scott, B.G., Cleere, H. (Eds.), The Crafts of
Gordon, R.B., Killick, D.J., 1993. Adaptation of technology to culture and environ- the Blacksmith. Ulster Museum, Belfast, pp. 113e120.
ment: bloomery iron smelting in America and Africa. Technol. Cult. 34, 243e Pistorius, J., 1989. Die Metaalbewerkers Van Phalaborwa. Ph.D. thesis. University of
270. Pretoria.
Gordon, R.B., van der Merwe, N.J., 1984. Metallographic study of iron artefacts from Plug, I., Pistorius, J., 1999. Animal remains from industrial iron age communities in
the eastern Transvaal Lowveld, South Africa. Archaeometry 26, 108e127. Phalaborwa, South Africa. Afr. Archaeol. Rev. 16, 155e184.
Haggerty, S., 1991. Oxide textures: a mini-atlas. In: Lindsley, D.H. (Ed.), Oxide Reynolds, I.M., 1986. Vanadium-bearing titaniferous iron ores of the Rooiwater
Minerals: Petrologic and Magnetic Significance, Reviews in Mineralogy, vol. 25, Complex, north-eastern Transvaal. In: Anhaeusser, C.R., Maske, S. (Eds.), Mineral
pp. 129e219. Deposits of South Africa. Geological Society of South Africa, Johannesburg,
Hall, A.L., 1912. Geology of the Murchison Range and District. In: Union of South pp. 451e460.
Africa Geological Survey, Memoir, vol. 6. Pretoria. Rostoker, W., Bronson, B., 1990. Pre-industrial Iron: its Technology and Ethnology.
Hall, S., Miller, D., Anderson, M., Boeyens, J., 2006. An exploratory study of copper Archeomaterials Monograph no. 1. Archaeomaterials Publishers, Philadelphia.
and iron production at Marothodi, an early 19th century Tswana town, Rus- Rostoker, W., Bronson, B., Dvorak, J.R., 1989. Smelting to steel by the Japanese tatara
tenburg District, South Africa. J. Afr. Archaeol. 4, 3e35. process. Archeomaterials 3, 11e25.
Hanekom, H.J., van Staden, C.M., Smit, P.J., Pike, D.R., 1965. The Geology of the Schwellnus, C.M., 1937. Short notes on the Palaboroa smelting ovens. South Afr. J.
Phalaborwa Igneous Complex. In: Geological Survey of South Africa Memoir, Sci. 33, 904e912.
vol. 54. Pretoria. Schwellnus, J.S.I., Englebrecht, L.N.J., Coertze, F.J., Russell, H.D., Malherbe, S.J., van
Ige, A., Rehren, T., 2002. Black sand and iron stone: iron smelting in Modakeke, Ife, Rooyen, D.P., Cooke, R., 1962. The Geology of the Olifants River Area, Transvaal.
south western Nigeria. IAMS 22, 19e21. Government Printer, Pretoria.
Iles, L., Martinón-Torres, M., 2009. Pastoralist iron production on the Laikipia Stayt, H., 1931. The Bavenda. Oxford University Press, Oxford.
Plateau, Kenya: wider implications for archaeometallurgical studies. J. Archaeol. Stettler, E.H., Coetzee, H., Rogers, H.J.J., Lubala, R.T., 1993. The Schiel alkaline
Sci. 36, 2314e2326. complex: geological setting and geophysical investigation. South Afr. J. Geol.
Jaguin, J., Gapais, D., Poujol, M., Boulvais, P., Moyen, J.-F., 2012. The Murchison 96, 96e107.
greenstone belt, South Africa e a general tectonic framework. South Afr. J. Geol. Stoch, H., 1978. The Preparation and Certification of a Reference Sample of
115, 65e76. Magnetite Ore. National Institute for Metallurgy Report, Randburg, South Africa.
Junod, H.A., 1927. The Life of a South African Tribe. MacMillan, London. Stuiver, M., van der Merwe, N.J., 1968. Radiocarbon chronology of the Iron Age in
Killick, D.J., 1990. Technology in its Social Setting: Bloomery Iron Working at sub-Saharan Africa. Curr. Anthropol. 3, 178e196.
Kasungu, Malawi, 1860e1940. Ph.D. thesis. Yale University. Thorne, R.M., 1974. Archaeological Survey in South Africa: Conceptual, Methodo-
Killick, D.J., Gordon, R.B., 1988. The mechanism of the bloomery process. In: logical and Practical Problems: The Phalaborwa Complex, a Case Study. Ph.D.
Farquhar, R.M., Hancock, R.G.V., Pavlish, L.A. (Eds.), Proceedings of the 26th thesis. University of Missouri, Columbia, MO.
International Symposium on Archaeometry. University of Toronto Department Trevor, T.G., 1912. Some observations on ancient mine workings in the Transvaal.
of Physics, Toronto, pp. 120e123. Journal of the Chemical. Metallurgical Min. Soc. S. Afr. 12, 267e275.
Klapwijk, M., 1986. A late iron age furnace excavation on the farm Longridge, Agatha, van der Merwe, D.S., 1936. A Preliminary Survey of Places of Archaeological Interest
north-eastern Transvaal, South Africa. South Afr. Archaeol. Bull. 41, 22e26. in the Northern Transvaal. Unpublished typescript, File 9/3. University of the
Klapwijk, M., Huffman, T.N., 1996. Excavations at Silver Leaves: a final report. South Witwatersrand Archaeological Research Unit, South Africa.
Afr. Archaeol. Bull. 51, 84e93. van der Merwe, N.J., Killick, D.J., 1979. Square: an iron smelting site near Phala-
Krige, E.J., 1941. Economics of exchange in a primitive society. South Afr. J. Econ. 9, borwa. South Afr. Archaeol. Soc. Goodwin Ser. 3, 86e93.
1e21. van der Merwe, N.J., Scully, R.T.K., 1971. The Phalaborwa story: archaeological and
Levin, E.M., Robbins, C.R., McMurdie, H.F., 1964. Phase Diagrams for Ceramists. ethnographic investigation of a South African Iron Age group. World Archaeol.
American Ceramic Society, Columbus. 3, 78e196.
Mason, R.J., 1968. Transvaal and Natal Iron Age settlement revealed by aerial van Eeden, O.R., Kent, L.E., Brandt, J.W., 1939. The Mineral Deposits of the Murchison
photography and excavation. Afr. Stud. 27, 1e14. Range East of Leydsdorp. In: Union of South Africa Department of Mines,
Mathoho, N.E., 2012. Indigenous Iron Production South of the Luvuvhu River, Memoir, vol. 36. Pretoria.
Limpopo Province, South Africa. M.A. dissertation. University of Cape Town. Verein Deutscher Eisenhüttenleute, 1995. Schlackenatlas/Slag Atlas, second ed., 2
McGannon, H.E., 1971. The Making, Shaping and Treatment of Steel, ninth ed. United vols. Verlag StahlEisen, Düsseldorf.
States Steel Corporation, Pittsburgh. Verwoerd, W.J., 1956. Sekere produkte van primitiewe koper-, yster-, en bronss-
Mellor, E.T., 1906. The Geology of the District about Haenertsburg, Leydsdorp and meltery in Oos Transvaal met besonder verwysing na Phalaborwa. Tegnikon 9,
the Murchison Range. In: Geological Survey of the Transvaal, pp. 21e52. 91e104.
Pretoria. Verwoerd, W.J., 1986. Mineral deposits associated with carbonatites and alkaline
McCourt, S., van Reenen, D.D., 1992. Structural geology and tectonic setting of the rocks. In: Anhaeusser, C.R., Maske, S. (Eds.), Mineral Deposits of South Africa.
Sutherland greenstone Belt, Kaapvaal Craton, South Africa. Precambrian Res. 55, Geological Soc. S. Afr., 2173e2191. Johannesburg.
93e110. Verwoerd, W.J., du Toit, M.C., 2006. The Phalaborwa and Schiel complexes. In:
Meyer, A., 1986. ’n Kultuurhistoriese Interpretasie van die Ystertydperk in die Johnson, M.R., Anhaeusser, C.R., Thomas, R.J. (Eds.), The Geology of South Africa.
Nasionale Krugerwildtuin. Ph.D. thesis. University of Pretoria. Council for Geoscience, Pretoria, pp. 291e299.
Miller, D.E., 2002. Smelter and smith: metal fabrication technology in the southern Wessmann, R., 1908 (1969). The Bawenda of the Spelonken (Transvaal): a Contri-
African Early and Late Iron Age. J. Archaeol. Sci. 29, 1083e1131. bution towards the Psychology and Folklore of African Peoples (L. Weinthal,
Miller, D.E., Killick, D.J., van der Merwe, N.J., 2001. Metal working in the northern Trans.). Negro Universities Press, New York.
Lowveld, South Africa, AC 1000-1890. J. Field Archaeol. 28, 401e417.

También podría gustarte