Archaeol Anthropol Sci (2015) 7:387–397
DOI 10.1007/s12520-014-0198-z
ORIGINAL PAPER
A (not so) dangerous method: pXRF vs. EPMA-WDS analyses
of copper-based artefacts
V. Orfanou & Th. Rehren
Received: 12 March 2014 / Accepted: 30 May 2014 / Published online: 13 June 2014
# The Author(s) 2014. This article is published with open access at Springerlink.com
Abstract Analysis of metal objects with portable and handheld X-ray fluorescence spectrometry has become increasingly
popular in recent years. Here, methodological concerns that
apply to non-destructive, surface examination with XRF instruments of ancient metal artefacts are discussed based on the
comparative analyses of a set of copper-based objects by means
of portable X-ray fluorescence (pXRF) and electron probe
microanalyser (EPMA). The analytical investigation aims to
explore issues of instrument comparability and reliability of the
non-invasive pXRF results. The different analytical approaches
produce a comparable pattern for the major element concentrations, but substantial variation is evident when it comes to the
absolute values for major and minor/trace elements.
Keywords X-ray fluorescence spectrometry . Electron probe
microanalyser . Museum artefacts . Copper-alloys
Introduction
Starting point for this study was the ever-increasing use of
portable X-ray fluorescence (pXRF) in archaeology due to its
non-destructive, non-invasive character and the possibility of
in situ examination of both artefacts and structures, as well as
its low cost compared to other means of analysis. Surface,
non-invasive analytical techniques are often favoured and
preferred by archaeologists and curators who often directly
correlate ‘archaeometric analysis’ with ‘object deformation’.
All the above make pXRF instruments the most preferable
choice for archaeologists, curators and conservators alike (Tite
et al. 2002; Henderson and Manti 2008; Cesareo et al. 2011;
Martinón-Torres et al. 2012). A growing number of researchers gain access to and use both handheld and labbased pXRF equipment for the archaeometric study of metal
artefacts which highlights the importance of this and similar
comparative studies (Angelini et al. 2006; Kantarelou et al.
2007; Karydas 2007; Dussubieux et al. 2008; Shugar and
Mass 2012; Shugar 2013; Charalambous et al. 2014).
A group of copper-based objects has been selected for
comparative analysis in order to address issues of the comparability of analytical results and their possible effect on the
archaeological interpretation of ancient copper alloys. The corrosion effect on the objects’ surfaces, and the potentials and
limitations of the different analytical techniques and methodological strategies applied that would affect any analytical results
were also taken into consideration. For this, a portable, labbased XRF spectrometer (pXRF) and an electron probe
microanalyser with an attached wavelength-dispersive spectrometer (EPMA-WDS) were used to analyse the same samples. The main aim of the present study was to explore the
relationship between non-invasive, surface (lab-based pXRF)
and quantitative invasive (EPMA) analyses on sound metal
from the objects’ core using metallographic cross sections, in
order to provide an evaluation of the reliability of surface pXRF
data on excavated copper-based artefacts. As such, it is intended
to identify trends and patterns in the results and to highlight the
particular characteristics of the two instruments that could affect
archaeological interpretations of metal object assemblages.
V. Orfanou (*)
UCL Institute of Archaeology, London, UK
e-mail: s.orfanou@ucl.ac.uk
Methodology
T. Rehren
Th.
Rehren
UCL Qatar, Education City, Doha, Qatar
e-mail: th.rehren@ucl.ac.uk
A sample of 41 small copper-based objects has been examined
qualitatively by pXRF and quantitatively by EPMA. All artefacts are part of the Archaeological Museum of Volos
388
Archaeol Anthropol Sci (2015) 7:387–397
collection and have been found at the sanctuary of Enodia and
Thavlios Zeus at ancient Pherae in Thessaly, Greece; they
largely date to the Protogeometric, Geometric and early
Archaic periods, that is the first half of the first millennium
B.C. (Béquignon 1937; Kilian 1975; Kilian-Dirlmeier 1985;
Vokotopoulou 1986, 1990; Bouzek 1997). Sampling criteria
primarily focused on the preservation of a substantial metallic
core as well as the possibility of obtaining a cut sample in
addition to the surface analysis, such as from already
fragmented objects. Even though no particular distinction
was made in regard to the typology and/or use of the artefacts,
mainly articles of personal adornment are represented in the
sample such as fibulae and pins, arm bands, sheets, rings and
spirals (Table 1).
The assemblage presented here was analysed using three
different protocols in order to better understand the nature of
the quality of surface XRF analysis as they can be obtained in
real-world conditions when analysing archaeological collections of copper-based artefacts. The first protocol included
EPMA-WDS (core metal) analysis, the other two lab-based
pXRF (substrate metal layer on a cleaned surface and the
intact corroded surface). Thus, XRF1 reports data obtained
on a cleaned area using established conservation methods to
reveal a visually metallic surface, while XRF2 reports data on
the corroded surface.
The investigation of the samples’ microstructure was conducted with reflected light (Olympus BX60 with an attached
digital camera) and scanning electron microscopy (Hitachi
S3400N); the results are used here only for illustrative purposes. The emphasis of this study lies on the comparison of
two analytical instruments (pXRF and EPMA) employed for
the quantitative examination of the sample. The instruments
are both commonly used and present different features that
would immediately affect the quality of the data provided,
such as different detection limits, area vs. spot measurements
and the potential for analysing different parts of the object, i.e.
the corroded surface, the substrate, defined here as the partly
corroded layer immediately beneath the original surface, and
the core metal (Fig. 1).
During quantitative analysis with both instruments, a set of
nine elements was analysed, namely, Cu, Sn, Pb, Fe, As, Zn,
Sb, Mn and Ni (note that Mn was not analysed during XRF1;
Table 1 Object type and
frequency in the sample
Description
Count
Count (%)
Rings
Sheets
Fibulae
Arm bands
Pins
Spirals
Total
17
10
7
2
2
3
41
41
24
17
5
5
7
100
the EPMA analysis of sound metal also included oxygen. This
was found in rather low levels as expected for the examination
of sound metal and thus is not reported here). Cut samples
have been mounted in epoxy resin blocks and then polished
using standard procedures down to 0.25 μm using diamond
paste. Quantitative analyses were conducted with an EPMAWDS (JXA-8100 Electron Probe Microanalyser) at the
Wolfson Archaeological Science Laboratories, UCL Institute
of Archaeology, London. Pure elements have been used to
calibrate the instrument with the exception of arsenic for
which an indium arsenide compound was used, while its
performance was monitored by the analyses of two certified
reference materials (CRMs), namely, brass 42.23.2 and leaded
bronze 50.04.4 (Bureau of Analysed Samples Ltd.; Table 2).
Depending on the nature of each sample, multiple area and
spot measurements were taken in magnifications of 1,000× at
a working distance of 11 mm with an acceleration voltage of
20 kVand a beam current of 50 nA. The data reported here for
each sample is the average of those individual measurements.
The standard deviation of those multiple analyses is reported
for the reference materials (Table 2) to illustrate the analytical
precision of the instrument.
Non-invasive XRF analyses took place at the Metal
Conservation Laboratory of the 13th Ephorate for Prehistoric
and Classical Antiquity (EPCA) in Volos, Greece, using a
portable, though not handheld, ED-XRF spectrometer developed at the Institute of Nuclear Physics, NCSR Demokritos
(Fig. 2). The XRF spectrometer consisted of an Rh-anode
side-window, low power X-ray tube (50 W, 40 kV, 125 μm
Be window), a PIN X-ray detector and a multichannel
analyser (MCA) card. The analytical range of this portable
XRF spectrometer extends from Z=14 (silicon) up to Z=92
(uranium). The device can operate under two distinct
Fig. 1 Backscatter scanning electron image of ring AE 34 showing
corrosion of the substrate (mid-grey) and the surface (dark grey) of the
object, separated by a thin lead-rich corrosion layer (white line) which
also marks the original surface. Bright white areas in the lower left part of
the metal are lead inclusions.
Archaeol Anthropol Sci (2015) 7:387–397
389
Table 2 Summary of brass and leaded bronze standards analysed with EPMA
Zn
As
Fe
Cu
S
Sn
Bi
Sb
Ni
Mn
Pb
42.23.2 brass
Mean (n=8)
22.47
0.19
0.35
73.23
0.05
1.66
0.02
0.39
0.18
0.02
0.63
σ
CV (%)
CRM
δ abs
0.08
0.38
22.13
−0.34
0.03
13.51
0.17
−0.02
0.03
7.92
0.35
0.001
0.56
0.77
74.36
1.13
0.01
22.73
0.05
−0.01
0.11
6.87
1.63
−0.03
0.02
86.89
0.03
0.01
0.06
14.77
0.36
−0.03
0.01
4.83
0.17
−0.01
0.00
18.04
0.02
0.00
0.21
33.38
0.58
−0.06
δ rel (%)
−1.53
−10.86
0.25
1.52
−14.72
−1.87
43.75
−9.23
−8.33
2.63
−10.24
50.04.4 leaded bronze
Mean (n=10)
0.66
σ
0.05
0.06
0.02
0.13
0.01
76.53
2.60
0.17
0.05
11.02
0.49
0.10
0.05
0.53
0.04
1.18
0.05
0.04
0.01
9.07
2.94
CV (%)
CRM
δ abs
8.25
0.66
0.00
42.78
0.06
0.00
4.95
0.10
−0.03
3.40
76.11
−0.42
27.33
0.14
−0.03
4.41
11.30
0.28
43.68
0.10
0.00
8.13
0.50
−0.03
3.96
1.10
−0.08
28.33
0.03
−0.01
32.43
9.94
0.87
δ rel (%)
−0.55
6.67
−31.30
−0.55
−22.43
2.46
−3.30
−6.38
−7.57
−43.57
8.76
conditions: one unfiltered mode with the voltage set at 15 kV
and a filtered one with the voltage set at 40 kV; we used the
latter for the analyses presented here. Two laser pointers are
mounted in the spectrometer head in such a way that the
intersection point of their beams coincides with the crosspoint of the incident X-ray beam axis and the detector axis.
The beam spot at the sample position has a diameter of less
than 2 mm. The spectrometer head is attached in an X-Y-Z
position, allowing its easy movement in the X-Y directions
(for more details on the XRF technique, see also Karydas
2007).
Measurements were taken both on the objects’ intact corroded surface (‘XRF2’) as well as on an area scraped clean
where the metal substrate with its characteristic shine was
revealed (‘XRF1’). For each analysis, i.e. XRF1 and XRF2,
two spot measurements of 300 s at 2,048 channels were taken
per sample at an acceleration voltage of 40 kV and a beam
current of 30 μA. Taking into consideration issues of ancient
metal heterogeneity caused either during manufacture or by
post-depositional oxidation (Caley 1964; Charles 1973, p.
105), spot measurements of 2 mm were taken in two different
areas on the surface of each sample in order to obtain more
representative chemical compositions. Finally, a set of CRMs
was regularly analysed to test the XRF’s stability, accuracy
and precision, namely, no. 691 of the European Commission,
Community Bureau of Reference (BCR; Table 3).
Both instruments’ performance was found to be stable over
the duration of the study; accuracy and precision levels were
repeatedly found satisfactory with a coefficient of variation
(CV) for bronze (BCR 961) analysed with pXRF of 0.5 % for
Cu and 6.2 % for Sn, and δ relative of 0.03 and 1.53 % for Cu
and Sn, respectively. Both instruments have broadly similar
levels of accuracy and precision when analysing the CRMs,
with EPMA regularly having the edge over pXRF.
Nonetheless, different minimum detection limits (MDL) are
reported for the two instruments. Thus, MDL for the EPMA is
reported at approximately 100 ppm, whereas a realistic MDL
for the XRF would be placed at 0.1 %. All the same, lower
accuracy levels on the whole are expected for such low
concentrations.
XRF1 and XRF2 methodologies were followed in order to
monitor changes that take place on the objects’ surface during
burial as a result of endogenous (e.g. the nature of the alloy)
and exogenous corrosion processes, i.e. the objectenvironment interaction, and the way these may affect element
concentration as measured on the substrate and on the intact
surface including patina and other growth corrosion layers
(Scott 1985, 2002; Franceschi et al. 1998; Ingo et al. 2006;
p. 518; Mezzi et al. 2012, p. 953). This comparison allowed
testing the degree to which surface XRF analysis is representative of the metal’s composition prior to final deposition and
Fig. 2 The portable XRF instrument used for the surface, non-invasive
analysis
390
Archaeol Anthropol Sci (2015) 7:387–397
Table 3 Analysis of the four CRM BCR-691 discs during examination
of the material with the pXRF
BCR-691 disc
A
% Cu % Zn
% As
% Sn
% Pb
Leaded bronze
Analysed
Mean (n=11)
σ
CV (%)
77.08
0.93
1.20
5.62
0.33
5.82
0.18
0.03
14.70
6.26
0.42
6.64
10.86
0.92
8.51
Certified
BCR
78.73
6.02
0.19
7.16
7.90
δ abs
δ rel (%)
1.65
2.10
0.40
6.67
0.01
7.03
0.90
12.53
2.96
37.51
Brass
Mean (n=9)
84.08
13.33
0.23
2.00
0.36
0.61
0.54
0.01
0.23
0.13
CV (%)
BCR
0.73
82.65
4.06
14.80
6.02
0.10
11.63
2.06
36.90
0.39
δ abs
δ rel (%)
1.43
1.73
1.47
9.94
0.13
135.24
0.06
2.99
0.03
8.12
Leaded bronze
Mean (n=9)
σ
76.48
0.52
0.34
0.06
0.22
0.06
8.08
0.63
14.88
0.85
0.68
80.27
3.79
4.72
17.26
0.15
0.19
128.38
28.55
0.29
0.07
23.35
7.83
10.10
2.02
19.98
5.74
9.20
5.68
61.76
92.42
0.45
0.49
92.45
0.02
0.03
0.41
0.04
8.92
0.16
0.25
158.67
0.20
0.03
13.70
0.19
0.01
3.32
6.89
0.43
6.22
7.00
0.11
1.53
0.08
0.07
81.31
0.20
0.12
60.68
B
Analysed
σ
Certified
D
Analysed
Certified
E
Analysed
Certified
CV (%)
BCR
δ abs
δ rel (%)
Bronze
Mean (n=9)
σ
CV (%)
BCR
δ abs
δ rel (%)
σ standard deviation, CV coefficient of variation, n number of spot
analyses
corrosion. At this point though, it is worth pointing out that
removing growth corrosion layers from a bronze object’s
surface alone does not guarantee the removal of all corrosion
products until sound metal is revealed since the survival of
metal grains in a matrix rich in corrosion products (intergranular corrosion) occurs quite often (Scott 1985, pp. 50–51,
1991). Typically the substrate of metal artefacts is corroded
to variable depths. Thus, in order to reveal the sound metal
layers, part of the substrate was scraped away as well. In the
case of bronzes, metal is mostly substituted by copper oxides,
chlorides and carbonates, as well as pure copper, and the
oxides of other metals such as tin, iron, lead, etc., as, for
example, found in a group of Roman and Punic bronzes with
very similar compositions to the Pherae assemblage (Scott
1985; Mezzi et al. 2012, p. 956).
Based on the assessment of the CRM data and the fact that
the EPMA analyses were done on demonstrably sound metal
in polished cross sections, while even the substrate pXRF data
is likely to include some corrosion material, we take the
EPMA data as the ‘true’ reference point for our subsequent
comparison. On the basis of the results from the above CRMs
analyses and the CV values, we assume that this data is
accurate within a few percent of the real values for copper
and tin, and within 10 % of the absolute values for the minor
elements such as lead or iron.
EPMA analysis and comparison with pXRF results
allowed the evaluation of the efficiency and reliability of
pXRF analyses of visually sound metal (XRF1). Since both
(EPMA and XRF1) are analyses of sound metal areas from the
same objects, a direct replication of their values would indicate the same level of accuracy of each instrument. Certain
patterns and trends though that arise as a result of different
instrument features such as detection limits or spot/area measurements result in a deviation of pXRF values from the more
accurate EPMA ones. Additionally, variation in values could
result from the heterogeneity of the metal core in relation to
the substrate as well as from the various and complex degradation phenomena (Ingo et al. 2006). Overall, the above
methodology was designed to provide a ‘guide’ for the interpretation of pXRF results on copper-based artefacts and the
critical evaluation of the questions to be addressed when
invasive, and thus, more accurate analysis is not available.
Results
Quantitative examination of the samples showed the presence
of mainly bronze artefacts (here, the term ‘bronze’ is used to
describe a binary Cu-Sn alloy with additions of tin of more
than 4 % Sn) with fewer examples of unalloyed copper
(mostly showing 98–99 % Cu). A few leaded bronze samples
are present in the sample, i.e. a ternary alloy with tin and lead
in amounts to suggest their deliberate addition. Lead contents
above 4 % have been often used to identify its purposeful
addition as opposed to impurity levels or the results of
recycling operations (Tylecote et al. 1977; Pernicka et al.
1990, p. 272; Mangou and Ioannou 1998, p. 98). Here, five
objects with >4 % lead (as analysed with EPMA) have been
found, but only four of these stand out from the bulk of the
sample pointing to deliberate additions of lead (Fig. 3). It
seems that in this assemblage, the natural lead contents can
reach up to 5 wt%.
Certain trends can be seen for the three analytical approaches employed (EPMA, XRF1 and XRF2) particularly
when looking at the distribution of individual elements. The
occurrence of these characteristic patterns has to be attributed
mainly to the analytical instrument properties when it comes
to the comparison between EPMA and XRF1 results, and to
corrosion effects as opposed to the analyses of sound metal,
namely, EPMA/XRF1, and surface corrosion layers (XRF2).
Archaeol Anthropol Sci (2015) 7:387–397
391
Table 4 Means of element concentrations
measured in all 41
objects
Values in wt%
na not analysed
EPMA
XRF1
XRF2
Fe
Cu
0.16
90.51
1.55
86.82
2.82
77.37
Ni
0.07
0.24
0.26
Zn
As
0.00
0.16
0.44
0.30
0.49
0.39
Sn
7.50
8.84
13.23
Sb
Pb
0.09
1.24
0.18
1.64
0.28
5.07
Mn
0.003
na
0.10
Comparison EPMA–pXRF
value 0.07 wt%, XRF1 value 0.24 wt%, more than three times
the EPMA value).
The following discussion explores the possible reasons for
the discrepancies of the results element by element. The two
main possible reasons for errors in the pXRF analysis are
limitations of the instrument (relatively high detection limits,
errors in the calibration) and factors that affect the analysed
surfaces, such as contamination through elements introduced
from the environment during burial or the differential behaviour of the original metal components during corrosion, either
through enrichment or selective leaching.
A noteworthy observation is that in the XRF1 measurements, the concentrations for all elements except copper are
significantly higher than their equivalent EPMA values; for
iron by a factor of 10 and for zinc even more: it is below the
detection limit of the EPMA, estimated to about 0.02 wt%, but
is reported by the XRF1 as nearly half a percent (0.44 wt%).
The nominally largest differences between EPMA and XRF1
results are with copper (3.7 % difference) and tin (1.3 %;
Table 4). In relative terms, however, the discrepancy between
the two copper values is less than 5 % of the absolute value, a
level of agreement or accuracy often accepted for routine
analyses [see, e.g. Hein et al. (2002) for a more detailed
discussion of the levels of accuracy encountered in
archaeometric analyses]. For tin, the relative difference is
nearly 20 % and clearly higher than what would normally be
accepted as uncertainty (Fig. 4). The above difference noted
here could either reflect the instrumental error as, for example,
reflected in the 20 % δ relative difference for the CRM BCR691 D (Table 3) or, in fact, a remaining effect of corrosion
enrichment on the cleaned surface; the latter appears more
likely based on the fact that when analysing certified reference
materials, the pXRF values for tin were regularly lower than
the certified values, not higher.
The relative error, defined here as the difference between
the EPMA and the XRF1 values, increases as concentrations
decrease. This is a general phenomenon widely known which
can be noted also in our data here, from lead (EPMA value c
1.2 wt%, relative error in XRF1 c 30 %) to nickel (EPMA
Fig. 4 Scatter plot of the EPMA and the pXRF analyses for the tin
content; the black line indicates the fit line of the measured values; red
dotted line indicates the ideal correlation for the two axes, i.e. y=x; the
black dotted lines show the approximate 20 % difference between the two
protocols where an EPMA value of 10 % could be 12 % for the XRF1
Fig. 3 Scatter plot of tin against lead values for the EPMA analyses
Major differences are seen for all analysed elements, with
absolute differences most obvious for the two main compounds (reaching 13 wt% for copper and nearly 6 wt% for
tin) but proportionately even bigger discrepancies for the
minor elements (reaching a factor of 15 for iron). Below, we
present and discuss the differences first between EPMA and
XRF1 values, where we assume that the differences mostly
reflect the different performance parameters of the instruments, with only some influence from corrosion. Following
this, we compare XRF1 (visually sound metal) analyses to
XRF2 (corroded surfaces) analyses to explore the effect of
corrosion on the analytical results.
392
Archaeol Anthropol Sci (2015) 7:387–397
For some elements, this is further exacerbated in the XRF2
data, such as for lead (Fig. 5), but also for iron, antimony and
tin. Other elements, such as nickel, zinc and arsenic, appear
relatively unchanged in the XRF2 data compared to the XRF1
data. Only the copper concentrations are lower in the XRF1
data than that in the EPMA data, and again lower in the XRF2
data than that in the XRF1 data.
Tin
Tin is present in the sample with values approaching a 10 %
Sn bronze, i.e. the optimum bronze recipe which was also the
par excellence copper-base alloy used in early Greece
(Papadimitriou 2001, p. 597). Tin concentrations fall mostly
between 5 and 10 wt% (EPMA mean value of 7.5 wt%), a
level where the XRF data should be accurate within a few
percent relative. However, the mean surface values are significantly higher, reaching nearly 9 wt% in the cleaned surface
(XRF1) and more than 13 wt% in the corroded surface
(XRF2). Tin is known to be relatively immobile in most burial
conditions compared to copper and therefore often found
enriched in corroded bronze substrate layers. The enriched
XRF2 data reflect this relative surface enrichment. However,
Fig. 6 shows a significant amount of intergranular corrosion,
which would result in some tin enrichment even if the artefact
was cleaned to a depth where visually sound metal was
exposed (the XRF1 protocol). It is, therefore, reasonable to
assume that the relative enrichment between EPMA and
XRF1, of around 20 % of the absolute EPMA value, is at least
partially due also to corrosion effects, despite analysing a
visually sound metal surface.
Lead
A similar observation as for tin can be made for lead, even
though, here, the enrichment from XRF1 to XRF2 is far
higher. A characteristic property of lead is that it does not
mix evenly with the rest of the elements present in the alloy
(Staniaszek and Northover 1983). Instead, lead forms distinct
prills which can create difficulties in accurately determining
the lead content of bronze artefacts. During EPMA examination, only very small volumes in the order of a few cubic
micrometres are analysed; depending on operator bias in
selecting measurement spots, this can result in an over- or
under-estimation of lead concentrations. In contrast, the pXRF
analyses a surface area of about 1 or 2 mm across and, due to
its higher excitation energy compared to the EPMA, reaches
deeper into the artefact. Thus, the analysed volume is far
greater and more likely to include a representative amount of
lead. Nonetheless, a downside of the pXRF is its limited
accuracy for lead which, as determined by CRMs analyses,
was found on average overestimated by approximately 30 %
by the instrument (Fig. 7).
Fig. 5 Scatter plot of tin against lead values the XRF1 and XRF2
analyses; see above Fig. 3 for the same plot based on EPMA data. The
lead values appear dramatically higher in this plot
Figure 8 illustrates a typical cross section through a lead-rich
artefact. The area presented in the image is about half the size of
the analytical spot of the pXRF instrument used here, while each
individual EPMA analysis would not exceed a surface area in the
image equal to about the size of the letter ‘O’ in the caption
‘BSECOMP’. During corrosion, however, lead-rich compounds
enrich on or near the original surface of the artefact (see also
Fig. 2, where the original artefact surface is marked by a thin
layer of lead-rich material, resulting in a white line in the BSE
image). Thus, XRF2 values for lead tend to be higher as a result
of corrosion processes taking place on the objects’ surface.
In addition to the overestimation of the lead content by the
pXRF discussed above, occasionally, lead traces as analysed
by the EPMA have not been caught up by the pXRF due to the
latter’s lower detection limit. The above is mostly seen not
only for EPMA lead values between 0.01 and 0.04 % but also
with larger values up to 0.5 % Pb as, for example, seen in the
Fig. 6 Photomicrograph of sheet AE 480; corrosion products have
affected the object throughout, outlining metallic grains. 100×, image
length 1.85 mm, plane polarized light (PPL) (left) and XPL (right)
Archaeol Anthropol Sci (2015) 7:387–397
393
already in the copper/bronze metal as well as the presence of
large amounts of iron in soil environments since iron is the fourth
most abundant element in the Earth’s crust (Jenkins 1989, p. 59;
Ingo et al. 2006, p. 513). The near-doubling of iron concentrations in the XRF2 data compared to the XRF1 data is most likely
due to such environmental contamination of the corrosion products during burial, either through absorption of iron ions or as
mechanically incorporated soil particles. The presence of intergranular corrosion suggests that also the XRF1 values are affected by such environmental enrichment.
Zinc and nickel
Fig. 7 Scatter plot of Pb analysed by XRF vs. certified values for the set
of BCR-691 alloys; the black line indicates the fit line of the measured
values; the red dotted line indicates the ideal correlation for the two axes,
i.e. y=x; the grey dotted lines show that a lead value of 6 % determined by
XRF could be in reality as little as 4 %
case of AE 107 (Table 5). Samples with <0.5 % (EPMA) lead
traces which were not detected during XRF1 have all shown a
metal microstructure where minute lead inclusions are rarely
seen (Fig. 9), as opposed to the large lead prills seen in AE 784
(Fig. 8) with 3 % lead (EPMA).
Iron
Iron contents, as in the case of tin, tend to steadily increase from
EPMA to XRF2 values. Iron is a common trace element in
ancient copper, and the values found by EPMA are rather typical
of Iron Age copper work as impurities from the raw material
(Craddock 1976, p. 94, 1977, p. 115; Ingo et al. 2006, p. 517).
Iron oxides typically occur in the corrosion layers of bronzes as a
result of two factors, namely, the presence of the iron amount
Fig. 8 Backscatter image of ring AE 784; lead prills (white) are widely
visible in the microstructure. Note also the intergranular corrosion
outlining the individual metal grains
Both zinc and nickel are common minor or trace elements in
most copper ores. Nickel transfers almost completely during
smelting into the metal phase, while zinc either evaporates or
enters the metal phase (Pernicka 1990). The assemblage from
Pherae is relatively low in both elements, with EPMA values for
zinc below the detection limit of around 0.02 wt% and nickel at
an average 0.07 wt%. However, pXRF values are much higher
for both elements, averaging between a quarter and half of 1 %
by weight. This cannot be readily explained by either environmental contamination nor by a corrosion-driven enrichment on
the surface, even though zinc is a relatively mobile element
most often present in soil environments with average soil
concentrations of 50 μg/g (50 ppm), which can be enriched
up to ten times depending on human or animal activity and the
decomposition of remains (Jenkins 1989, pp. 57–58). Instead, it
is most likely that these pXRF values are spurious, based on
noise in the background, peak overlaps from the neighbouring
very strong copper peaks or an over-interpretation of a weak
real signal during the quantification process. Zinc in particular
is prone to spurious signals even in wavelength-dispersive
spectrometer analyses such as EPMA or WD-XRF, due to the
proximity of the Zn Kα line to the dominant copper Kβ peak.
This problem is even stronger in energy-dispersive spectrometers with their lower peak resolution. The implications of an
almost 5 % Zn bronze, such as the ring AE 838 (4.5 % Zn,
XRF2 data), particularly for such an early period as the
Geometric would have been of considerable importance, since
early brasses and zinc-rich bronzes in Greece (typically before
the Roman period and the first century B.C.) are quite rare
(Craddock 1978, p. 1, 1998). Nonetheless, the zinc content as
measured during XRF1 for the same object (AE 838) is only
0.1 %, i.e. much closer to the EPMA data, demonstrating that
the surface analysis (XRF2) may not be reliable even when it
comes to the basic identification of the alloy type. In total, six
objects were found with pXRF values for zinc of more than 1 %
by weight, while on the basis of EPMA results, the entire
assemblage has to be considered zinc-free. Consequently, the
pXRF data for zinc have to be treated very critically before
proceeding to any definite conclusions regarding the type of
copper alloy.
394
Archaeol Anthropol Sci (2015) 7:387–397
Table 5 Summary table for the lead content (wt%) as analysed in the three protocols
Pb (wt%)
Sample
EPMA
XRF1
XRF2
Sample
EPMA
XRF1
XRF2
BE 45741
n.d.
n.d.
n.d.
AE 827
0.17
n.d.
0.51
1309
AE 838
n.d.
n.d.
n.d.
n.d.
0.07
n.d.
AE 289
AE 929
0.24
0.33
0.13
n.d.
0.95
1.27
M 1217.1
n.d.
0.04
1.16
AE 97
0.33
n.d.
n.d.
M 1217.2
M 8100
n.d.
n.d.
n.d.
n.d.
1.40
0.11
1308
M 495
0.38
0.43
0.14
0.50
0.82
2.03
AE 666
n.d.
n.d.
0.10
M 3367.1
0.50
0.30
0.88
M 1844
AE 37
0.01
0.02
n.d.
n.d.
n.d.
0.13
AE 107
AE 899
0.50
0.57
n.d.
0.16
1.46
0.90
AE 760
0.02
0.07
0.04
M 3367.2
0.61
0.34
17.72
M 3234
0.03
n.d.
0.28
M 1739.1
0.89
0.95
3.54
M 3234
0.03
n.d.
0.30
AE 98
0.99
0.03
2.93
AE 113
M 1739.2
0.04
0.04
n.d.
0.31
n.d.
0.56
M 1314.1
AE 564
1.66
1.98
1.61
5.02
7.44
15.19
AE 624
0.06
0.04
0.74
AE 459
2.44
2.22
10.54
AE 606
0.07
0.96
3.88
M 1314.2
2.79
3.55
17.76
AE 810
AE 103
M 1739.3
AE 480
AE 827
0.08
0.09
0.10
0.12
0.17
0.05
n.d.
n.d.
0.33
n.d.
0.31
0.98
0.85
10.12
0.51
AE 784
AE 34
AE 507
1310
AE 506
3.00
5.86
7.40
8.21
9.49
5.52
5.35
15.51
6.52
17.68
21.85
11.29
31.62
13.00
25.05
n.d. not detected
Other trace elements
Values for the rest of the trace elements such as arsenic and
antimony tend to be higher in surface analyses (XRF2) as well
Fig. 9 Photomicrographs of AE
97, 103, 107 and 827 whose
traces of lead were not detected
during XRF1 analyses; in all
samples, a metal grain structure
with several sulphide inclusions
(grey) but very few minute lead
ones (black) is visible; PPL,
500×, images length 35 μm
as on clean metal analyses with the pXRF (XRF1) relative to
the values detected with the EPMA. The levels for antimony
with an average of 0.09 and for arsenic with an average of
0.16 % and a maximum value of 0.64 % As suggest their
Archaeol Anthropol Sci (2015) 7:387–397
Fig. 10 Five-point box plot for lead values in all three analytical
approaches
presence as an impurity from the copper ores. The apparent
profile in the increase from the EPMA data to the pXRF data,
however, differs for the two elements; for arsenic, the main
increase is from the EPMA to the pXRF values, with little if
any further increase from XRF1 to XRF2. This suggests that
the pXRF data is mostly spurious and a result of the instrument’s performance. In contrast, the antimony data increases
significantly from XRF1 to XRF2, suggesting a real enrichment of antimony in the corrosion products which has also
been observed in a separate study on corrosion behaviour of
archaeological metal (Rehren and Prange 1998). Despite the
lower absolute concentrations of antimony compared to arsenic, we trust the XRF data for antimony more, due to its better
excitation characteristics and fewer peak overlap problems for
the Sb Kα peak compared to the arsenic Kα peak and its
overlap with the lead Mα peak.
Traces of manganese were found during EPMA analysis
with both mean and median values of 0.003 % and a maximum of 0.1 % Mn. Manganese was not analysed during
XRF1 analysis, but it was found during surface analysis
(XRF2) with average and maximum values of 0.1 and
0.23 % Mn, respectively. This increase in the manganese
content of several times of the EPMA values has to be attributed, as in the case of iron, to the surface enrichment as a result
of corrosion processes which tend to increase surface trace
element concentration and environmental contamination from
the soil system as well (Jenkins 1989, p. 59).
Discussion
The comparison between EPMA and pXRF results has raised
a number of issues regarding data quality, instrument comparability and surface analysis reliability. This contributes to the
395
discussion of the relative merit of invasive and surface analyses, as well as on the importance to understand data quality
for any method used, for an enhanced understanding of ancient copper-based assemblages. Consequently, the interpretation of any surface pXRF results, either on scraped or intact
surfaces, needs to take into consideration possible surface
enrichment during long-term burial in certain elements which
depends on the compositions of the soil and the objects
themselves. For instance, iron, manganese and zinc which
are present in most soil environments could be enhanced
further by human activity (Jenkins 1989). These elements
have all been detected in higher levels during surface pXRF
analysis but do not reflect the original composition of the
alloys used as analysed with EPMA.
In addition, phenomena of differential corrosion of copper
or tin in ancient bronzes and preferential depletion of the main
alloy components, i.e. copper and/or tin in regard to the
present analyses, have to be addressed since they immediately
affect the quality of surface pXRF results by providing
(mostly) higher values for tin (Scott 1985; Meeks 1986, p.
137). Depending on which layer of the corroded surface is
analysed, either tin or copper can be significantly enriched.
Moreover, lead values also tend to increase in surface analyses
(Fig. 10) but not in a consistent or predictable manner (see
Table 5). Significantly, even surfaces apparently cleaned until
solid metal is exposed can still include deep-rooted intergranular corrosion, resulting in distorted results. Overall, the above
evidence points to the importance of sound metal analysis
(EPMA or XRF1) not only for the minor and trace elements
concentrations but for an accurate determination of the alloys’
nature as well.
In addition to corrosion processes and soil contamination,
the detection limit of the pXRF has also to be considered,
particularly when dealing with elements at impurity levels.
This can be illustrated by looking at several XRF1 analyses
which did not detect any lead content, even though the EPMA
detected lead due to its lower detection limit for this element
(as seen in Table 5).
Conclusion
The use of pXRF is often the only way for archaeologists and
conservators to obtain analytical results from artefacts held in
major collections, where invasive sampling is not an option
(e.g. Charalambous et al. 2014). As shown by the analytical
results discussed here, all pXRF data obtained have to be
interpreted with caution. The instrumental limitations of such
portable equipment as well as the issues relating to the inherent changes in composition that affect ancient surfaces have to
be taken into consideration. Importantly, even though most
elements other than copper generally appear too high in pXRF
analyses, the assessment of reliability has to be done for each
396
element separately and individually, as each element will have
its own specific problems. Based on the data presented here, it
is argued that it is possible to obtain reasonable quality data
regarding the nature and approximate concentrations of the
main components present in copper-based alloys such as tin
and lead, provided that a metallic surface large enough for the
spot of the instrument can be exposed (approx. 1–3 mm).
Even though absolute values may vary for the major elements
between the pXRF and EPMA-WDS, trends and patterns in
the results can be in agreement, providing thus some information on the nature of the alloys used. For trace elements,
though, pXRF data can be misleading and their interpretation
has to consider surface corrosion phenomena in addition to the
instrument’s specifications. This is particularly true for corroded surface (see XRF2), for which data obtained in this
study appears to be generally not reliable or useful.
Notwithstanding, in most objects analysed here even by
looking at XRF2 results, it is safe to argue that all objects are
copper-based (as already indicated by their green corrosion
surfaces) and that tin and occasionally lead are the main
alloying agents, even if it is not possible to give an accurate
estimation of the original alloy composition. Furthermore,
XRF1 results on the objects’ scraped surfaces are to be better
trusted and taken as relatively more realistic in distinguishing
between bronze and leaded bronze objects. Accordingly, on
the bases of the pXRF data, it is possible to sort the assemblage into rough groupings of low, medium and high concentrations for the major elements. For such a categorization,
XRF1 values are much closer to the EPMA values and, thus,
more reliable.
For most elements in the periodic table up to about arsenic,
EPMA analysis based on WDS provides more accurate results
with much lower detection limits than the pXRF, while the
latter is more sensitive for heavier elements due to its higher
excitation energy. In any case, analyses of sound metal exposed by an experienced conservator should be clearly preferred over corroded surfaces since, as shown here, the former
produces results much closer to the original alloy compositions of ancient metal objects. The benefits of analysing sound
metal (XRF1) as opposed to the corroded surface (XRF2) are
unambiguous.
Despite the above limitations, analysis of ancient metal
objects with portable and handheld XRF instruments is and
will continue to be popular not only due to its non-invasive
nature but also because it is much less time consuming and
more cost effective which makes it widely applicable in archaeological research.
Acknowledgments This study would not have been possible without
the invaluable help of Eleni Asderaki, Honorary Head of the Conservation Department of 13th EPCA, Volos, and Kevin Reeves, Wolfson
Archaeological Science Laboratories, UCL Institute of Archaeology,
who we warmly thank. We also wish to acknowledge Dr. Arg.
Intzesiloglou and P. Arachoviti, 13th EPCA, Volos, for providing access
Archaeol Anthropol Sci (2015) 7:387–397
to material and their support during analyses, and Nikos Zacharias,
University of Peloponnese, for organising the 3rd Symposium on Archaeological Research and New Technologies where preliminary results
of the study were presented.
Open Access This article is distributed under the terms of the Creative
Commons Attribution License which permits any use, distribution, and
reproduction in any medium, provided the original author(s) and the
source are credited.
References
Angelini E, Grassini S, Corbellini S, Ingo GM, De Caro T, Plescia P,
Riccucci C, Bianco A, Agostini S (2006) Potentialities of XRF and
EIS portable instruments for the characterisation of ancient artefacts.
Appl Phys A Mater Sci Process 83:643–649
Béquignon Y (1937) Recherches archéologiques à Phères de Thessalie.
Les Belles Lettres, Paris
Bouzek J (1997) Greece, Anatolia and Europe: cultural interrelations
during the Early Iron Age. Stud Mediterr Archaeol 122
Caley ER (1964) Analysis of ancient metals. Pergamon Press, London
Cesareo R, Bustamante A, Fabian J, Calza C, Dos Anjos M, Lopes RT,
Alva W, Seclen M (2011) Portable equipment for a non-destructive
analysis of pre-Columbian metal artefacts from the Royal Tombs of
Sipán by energy-dispersive X-ray fluorescence spectrometry. X-Ray
Spectrom 40(1):37–46
Charalambous A, Kassianidou V, Papassavas G (2014) A compositional
study of Cypriot bronzes dating to the Early Iron Age using portable
X-ray fluorescence spectrometry (pXRF). J Archaeol Sci 46:205–216
Charles JA (1973) Heterogeneity in metals. Archaeometry 15:105–114
Craddock PT (1976) The composition of the copper alloys used by the
Greek, Etruscan and Roman civilizations: 1. The Greeks before the
Archaic Period. J Archaeol Sci 3:93–113
Craddock PT (1977) The composition of the copper alloys used by the
Greek, Etruscan and Roman civilizations: 2. The Archaic, Classical
and Hellenistic Greeks. J Archaeol Sci 4:103–123
Craddock PT (1978) The composition of the copper alloys used by the
Greek, Etruscan and Roman civilizations: 3. The origins and early
use of brass. J Archaeol Sci 5:1–16
Craddock PT (1998) Zinc in classical antiquity. In: Craddock PT (ed)
2000 years of zinc and brass. British Museum, London, pp 1–6
Dussubieux L, Deraisme A, Frot G, Stevenson C, Creech A, Bienvenu Y
(2008) LA-ICP-MS, SEM-EDS and EPMA analysis of eastern
North American copper-based artefacts: impact of corrosion and
heterogeneity of the reliability of the LA-ICP-MS compositional
results. Archaeometry 50(4):643–657
Franceschi E, Macciò L, Rosa L (1998) The corrosion of metallic artifacts
within different environments: archaeological objects and laboratory
simulations. In: Mourey W, Robbiola L (eds) Metal 98. James and
James, London, pp 92–93
Hein A, Tsokalidou A, Iliopoulos I, Mommsen H, Buxeda i Garrigos J,
Montana G, Kilikoglou V (2002) Standardisation of elemental analytical techniques applied to provenance studies of
archaeological ceramics: an inter-laboratory calibration study.
Analyst 127:542–553
Henderson J, Manti P (2008) Improving access to collections for sampling. In: Saunders D, Townsend JH, Woodcock S (eds)
Conservation and access. Contributions to the London Congress,
15-19 September 2008. ICC, London, pp 115–119
Ingo GM, De Caro T, Riccucci C, Angelini E, Grassini S, Balbi S,
Bernardini P, Salvi D, Bousselmi L, Çilingiroglu A, Gener M, Al
Jarrah O, Khosroff S, Mahdjoub Z, Al Saad Z, El-Saddik W,
Vassiliou P (2006) Large scale investigation of chemical
Archaeol Anthropol Sci (2015) 7:387–397
composition, structure and corrosion mechanism of bronze archaeological artefacts from Mediterranean basin. Appl Phys A Mater Sci
Process 83(4):513–520
Jenkins DA (1989) Trace element geochemistry in archaeological sites.
Environ Geochem Health 11(2):57–62
Kantarelou V, Karydas AG, Zarkadas C, Giannoulaki M, Argyropoulos V
(2007) Micro-XRF analysis of high tin bronze mirrors at the
Museum of Ancient Messene in Greece. In: Argyropoulos V, Hein
A, Abdel-Harith M (eds) Strategies for saving our cultural heritage.
Athens, TEI of Athens, pp 93–99
Karydas AG (2007) Application of a portable XRF spectrometer for the
non-invasive analysis of museum metal artefacts. Ann Chim 97:
419–432
Kilian K (1975) Fibeln in Thessalien von der mykenischen bis zur
archaischen Zeit. Beck, München
Kilian-Dirlmeier I (1985) Fremde Weihungen in griechischen
Heiligtümern vom 8. bis zum Beginn des 7. Jahrhunderts v. Chr.
Jahrbuch des Römisch-germanischen Zentralmuseums, Mainz, 32,
215–54
Mangou H, Ioannou PV (1998) On the chemical composition of
Prehistoric Greek copper-based artefacts from Crete. BSA
93:91–102
Martinón-Torres M, Li XJ, Bevan A, Xia Y, Zhao K, Rehren Th (2012)
Forty thousand arms for a single emperor: from chemical data to the
labor organization behind the bronze arrows of the Terracotta Army.
J Archaeol Method Theory 19:1–29
Meeks ND (1986) Tin-rich surfaces on bronze—some experimental and
archaeological considerations. Archaeometry 28(2):133–162
Mezzi A, De Caro T, Riccucci C, Parisi EI, Faraldi F, Vassiliou P, Grassini
S (2012) Analytical methodologies for the investigation of soilinduced degradation of Cu-based archaeological artefacts. Surf
Interface Anal 44(8):953–957
Papadimitriou GD (2001) Η εξέλιξη των κραμάτων χαλκού στον
ελλαδικό χώρο μέχρι το τέλος της γεωμετρικής εποχής:
κραματικές προσμίξεις και τεχνολογική εξέλιξη. In:
Bassiakos Y, Aloupi E, Facorellis Y (eds) Archaeometry issues in
397
Greek prehistory and antiquity. Athens: Hellenic Society of
Archaeometry, Soc Messenian Archaeol Stud 587–608
Pernicka E (1990) Gewinnung und Verbreitung der Metalle in
prähistorischer Zeit. Jahrb Römisch-Germanischen Zentralmuseums
Mainz 37:21–129
Pernicka E, Begemann F, Schmitt-Strecker S, Grimanis AP (1990) On the
composition of metal artefacts from Poliochni on Lemnos. Oxf J
Archaeol 9:263–297
Rehren Th, Prange M (1998) Lead metal and patina: a comparison. In:
Rehren Th, Hauptmann A, Muhly J (eds) Metallurgica Antiqua,
Deutsches Bergbau-Museum, Bochum, pp 183–196
Scott DA (1985) Periodic corrosion phenomena in bronze antiquities.
Stud Conserv 30(2):49–57
Scott DA (1991) Metallography and microstructure of ancient and historic metals. The Getty Conservation Institute, LA
Scott DA (2002) Copper and bronze in art: corrosion, colorants, conservation. The Getty Conservation Institute, LA
Shugar AN (2013) Portable X-ray fluorescence and archaeology: limitations of the instrument and suggested methods to achieve desired
results. In: Armitage RA, Burton JH (eds) Archaeological chemistry
VIII. American Chemical Society, Washington D.C, pp 173–193
Shugar AN, Mass JL (eds) (2012) Handheld XRF for art and archaeology.
Leuven University Press, Leuven
Staniaszek BEP, Northover JP (1983) The properties of leaded bronze
alloys. In: Spinall AA, Warren SE (eds) The Proceedings of the 22nd
Symposium on Archaeometry, 30 March-3 April 1982. Bradford,
W. Yorks: Schools of Physics and Archaeological Sciences,
University of Bradford, 262–272
Tite MS, Merriman N, Rehren Th, Schadla-Hall T, Walker Tubb K (2002)
Invasive sampling vs. object integrity. Pap Inst Archaeol 13:1–5
Tylecote RF, Ghaznavi HA, Boydell PJ (1977) Partitioning of trace
elements between ores, fluxes, slags and metal during the smelting
of copper. J Archaeol Sci 4:305–333
Vokotopoulou I (1986) Vitsa: the cemeteries of a Molossian settlement (in
Greek). TAP, Athens
Vokotopoulou I (1990) The tombs of Aeneia (in Greek). TAP, Athens