J Paleolimnol (2014) 52:265–276

DOI 10.1007/s10933-014-9792-4

NOTE

Wavelength dispersive X-ray fluorescence spectroscopy

as a fast, non-destructive and cost-effective analytical
method for determining the geochemical composition
of small loose-powder sediment samples
Johan Rydberg

Received: 28 October 2013 / Accepted: 11 August 2014 / Published online: 17 August 2014
Ó Springer Science+Business Media Dordrecht 2014

Abstract X-ray fluorescence spectroscopy (XRF) powder samples, can be a useful geochemical tool for
has been used extensively to analyze many types of many paleolimnological applications, especially
environmental samples, including lake sediments. In because lack of pretreatment ensures that samples
most cases, however, analyses have required either a can be used for further analysis.
relatively large sample mass or sample pretreatment,
e.g. lithium borate fusion, and have not taken advan- Keywords WD-XRF analysis  Method 
tage of the potential of XRF analysis as a non- Calibration  Lake sediment
destructive technique. This paper describes the devel-
opment of two completely non-destructive calibration
methods that use small, i.e. 200- and 500-mg loose-
powder sediment samples. Analytical performance of Introduction
these methods was assessed using ten different
certified reference materials and a previously analyzed When using lake sediments to study past environmen-
sediment profile, and for both methods, accuracy and tal changes, it is often advantageous to use a multi-
precision were less than ±10 % (or a few ppm) for 26 proxy approach to ensure the validity of our infer-
elements (Na, Mg, Al, Si, K, Ca, Sc, Ti, V, Mn, Fe, Ni, ences. It is also often desirable to use a high temporal
Cu, Zn, Ga, As, Br, Rb, Sr, Y, Zr, Sn, Sb, Ba, W and resolution, especially when comparing recent changes
Pb). This shows that quantitative wavelength disper- in the sediment to monitoring data. However, from a
sive X-ray fluorescence analysis, using small loose- purely practical perspective these two approaches are
somewhat contradictory because the former requires a
large amount of sample, whereas the latter tends to
Electronic supplementary material The online version of
this article (doi:10.1007/s10933-014-9792-4) contains supple- restrict it. For example, a surface-sediment gravity
mentary material, which is available to authorized users. core, sub-sampled at 0.5-cm resolution, typically
yields \500 mg of dry sample material from the
J. Rydberg (&)
unconsolidated, surface-sediment layers, assuming a
Department of Ecology and Environmental Sciences,
Umeå University, 901 87 Umeå, Sweden water content of up to 99 % and a standard corer with
e-mail: johan.rydberg@emg.umu.se an inner diameter of 60–70 mm (Glew et al. 2001;
Renberg and Hansson 2008; www.uwitec.at). This
J. Rydberg
limited sample mass should then be related to the
Institute für Geoökologie, Technische Universität
Braunschweig, Langer Kamp 19c, 38 106 Brunswick, required sample masses needed to measure frequently
Germany analyzed variables in paleolimnology: 5–20 mg for

123
266 J Paleolimnol (2014) 52:265–276

diatoms (Battarbee et al. 2001), *10–50 mg for pol- (Cheburkin and Shotyk 1996) or required develop-
len (Bennett and Willis 2001), 10–50 mg for biogenic ment of additional analytical software (Boyle 2000).
silica (Ohlendorf and Sturm 2008), 2–50 mg for car- The last two decades, however, have seen rapid
bon and nitrogen, including stable isotopes (Brodie development of commercially available XRF ana-
et al. 2011), 50–200 mg for mercury (Rydberg et al. lyzers (www.bruker-axs.com; www.rigaku.com;
2008; Yang et al. 2002), 200–500 mg for major and www.panalytical.com). These instruments are deliv-
trace elements, using ICP-MS on acid-digested sam- ered with powerful software packages that effec-
ples (Farmer et al. 1996), and 200–1,000 mg for 210Pb tively deal with inter-elemental matrix effects
dating by alpha spectrometry (Zaborska et al. 2007). through advanced post-analytical calculations (Butler
As most of these commonly used analytical techniques et al. 2008; Revenko 2011; Rousseau 2006). This
are destructive, this implies that the amount of sample paper outlines the development of two calibration
material, rather than the research question, will dictate methods for loose-powder samples, using a com-
either the number of variables measured or the tem- mercially available wavelength dispersive XRF
poral resolution of sampling. analyzer (WD-XRF) together with the built-in soft-
In this context, a non-destructive technique such as ware package, to quantify a wide range of major and
quantitative X-ray fluorescence spectroscopy (XRF) trace elements in small sediment samples, i.e. 200
provides a solution to both retain high temporal and 500 mg. The methods are intended for use on
resolution and determine the sediment geochemical fine-grained sediment samples without sample pre-
composition prior to applying other techniques. treatment, thus ensuring that samples can be reused
Quantitative XRF analyses have proven to be appli- for additional analyses.
cable to a wide variety of environmental samples, e.g.
rocks (Brown et al. 1973; Mucke and Farshad 2005),
peat and vegetation samples (Cheburkin and Shotyk Materials and methods
1996; Margui et al. 2005), and lake sediments (Bindler
et al. 2011; Boyle 2000; Koinig et al. 2003; Rydberg XRF set-up
et al. 2012). However, one inherent limitation of XRF
analysis is inter-elemental matrix effects, which in All analyses were carried out using a Bruker S8-Tiger
principle means that each element in the sample WD-XRF analyzer equipped with a Rh-anticathode
affects the quantification of all other elements (La- X-ray tube, five analyzing crystals (XS-55: W/Si
chance 1993). Quantification of an element is also multilayer crystal; XS Ge-C: Curved germanium
affected by sample mineralogy, grain size, homoge- crystal; LIF200 and 220: Lithium fluoride crystals;
neity and surface roughness (Norrish and Hutton PET: pentaerythrite crystal), different collimators and
1969). One effective way to deal with the latter set of filters, a flow proportional counter for light elements
problems is to fuse the sample using lithium borate, (Na to V) and a scintillation counter for heavy
which produces a homogenous piece of glass (Norrish elements (Cr to Pb). Analyses were performed directly
and Hutton 1969). Another alternative is to finely on loose powders, without pre-treatment, under a low-
powder the sample and then press it into a pellet using pressure (168 mbar) helium atmosphere, and with a
wax binders. Similar to the lithium borate fusion, this collimator mask that determines the size of the X-ray
produces a denser and more homogenous sample, beam to fit the size of the sample cup (i.e. 8 and 18 mm
which enhances analytical performance (Gazulla et al. in diameter for LAKE_200 mg and LAKE_500 mg,
2012). When using these procedures, however, XRF respectively). In the current set-up, the analyzer is
analysis is not truly non-destructive, and thus, it is not theoretically capable of analyzing elements from Na to
an alternative for applications in paleolimnology U, and can be equipped to analyze lighter elements,
where it is desirable to re-use the sample for other down to B. All maintenance (e.g. drift corrections,
purposes, e.g. diatom or pollen analysis. spectral alignment) and calibration-method develop-
In the past, XRF instruments capable of directly ment were conducted using the standard SpectraPlusÒ
analyzing loose-powdered samples without pretreat- and MethodWizardÒ software provided by Bruker
ment were largely limited to custom-built systems (www.bruker-axs.com).

123
J Paleolimnol (2014) 52:265–276 267

Selection of certified reference materials and W, the L-lines were selected. For Cr, As, Zr and
W, the peak of the Ka-line (or La-line) overlaps with
The calibration methods are based on commercially neighboring peaks, and it was therefore necessary to
available certified reference materials (CRMs). use an alternative line (i.e. the Kb- or Lb-line).
Twenty-two CRMs were selected to cover the ele- Following line selection, the measurement parameters
mental ranges normally encountered in sediments for: (X-ray tube current, collimator, crystal, filter, peak and
sodium (Na), magnesium (Mg), aluminum (Al), background positions, number of background posi-
silicon (Si), phosphorus (P), sulfur (S), potassium tions, detector discriminator window and measure-
(K), calcium (Ca), scandium (Sc), titanium (Ti), ment time) were optimized for each of the selected
vanadium (V), chromium (Cr), manganese (Mn), iron lines (ESM Table 2). This procedure was based on
(Fe), cobalt (Co), nickel (Ni), copper (Cu), zinc (Zn), continuous scans of the CRMs (ESM Figs. 1 and 2),
gallium (Ga), arsenic (As), bromine (Br), rubidium and the parameters were chosen to maximize the
(Rb), strontium (Sr), yttrium (Y), zirconium (Zr), tin peak:background ratio, give good separation between
(Sn), antimony (Sb), barium (Ba), tungsten (W) and neighboring peaks and avoid detector saturation. For
lead (Pb) (Electronic Supplementary Material [ESM] most elements this task is straightforward, but for
Table 1). Some elements (Sc, Cr, Co, Ga, Y, Sb and elements with low concentrations, e.g. Sc, the mea-
Sn) had to be excluded from the LAKE_200 mg surement time of the continuous scans is not enough to
calibration because the low elemental concentrations allow the peak to develop. In this case, the peak and
normally found in sediment samples could not be background position was set based on the theoretical
accurately quantified using the smaller sample size. peak position. It is also worth noting that even though
The majority of CRMs represent different kinds of the Zr Kb1-peak is on the slope of the Rh-peak from
lake, estuary, marine or stream sediments, but to cover the X-ray tube, the analytical performance was better
the desired concentration ranges, it was necessary to using a single background position, compared to a
include other types of CRMs, such as rock and soil sloping background based on two background posi-
samples. tions, as done, e.g. for Ba (ESM Figs. 1 and 2).
All of the CRMs were delivered as very fine- The second step of the development procedure was
grained powders, and the only treatment carried out a full analysis of all 22 CRMs using the selected
prior to analysis was to dry them over night at 40 °C. A measurement parameters. For each of the elements, a
200 and 500 mg (±10 mg) aliquot of each CRM was regression line was established between the certified
then transferred into standard plastic sample cups (15 CRM concentrations and the measured peak intensi-
and 20 mm inner diameter for LAKE_200 mg and ties. This regression line was refined by selecting
LAKE_500 mg, respectively), covered with 2.5-lm either net or gross intensities, an offset (i.e. a
MylarÒ film at the base (www.chemplex.com). It regression line not passing through zero), or a
should be noted that for some CRMs, the minimum quadratic regression line (ESM Table 2). The calibra-
sample quantity given in the certificate was[200 mg, tion was also refined using post-analytical matrix
and for LAKE_200 mg, given concentrations should corrections, called variable alpha in the SpectraPlusÒ
be considered as informational values, rather than software, to minimize the effects of the inter-elemen-
certified (ESM Table 1). tal matrix effects, by taking into account the interfer-
ences from other elements in the sample. These
Line selection and optimization of the calibration calculations require that close to 100 % of a sample’s
method composition is known. Therefore, LOI550 and LOI950
were determined in all CRMs (Heiri et al. 2001), and
The procedure to develop the two calibration methods the sum of LOI550 and LOI950 was entered into the
can be divided in two steps. In the first step, the SpectraPlusÒ software as carbon dioxide (CO2). For
appropriate measurement line (Ka-, Kb- or La-line) the same reason, the concentrations of Na, Mg, P, Ca,
was selected for each element (ESM Table 2). In K, Ti, Mn, Fe were entered, in the form of their most
principle, the Ka-line is preferred because it gives the common oxide (ESM Table 1). In the analytical
highest peak height, and consequently, better analyt- procedure, the content of CO2 in an unknown sample
ical performance. For some heavy elements, i.e. Pb is estimated using matrix balancing, i.e. the difference

123
268 J Paleolimnol (2014) 52:265–276

between 100 % and the sum of all analyzed elements, several of those CRMs still contribute to improving
with major elements included as their most common the calibration lines and hence they were retained
oxide. Even though the CO2 value reported by the (ESM Figs. 3 and 4). In some cases, CRMs were
analyzer should never be considered as an analytical deleted based on interference (either through inter-
result on its own, it is important that the matrix elemental matrix effects or overlapping peaks) from
balancing gives an accurate estimate of ‘‘CO2’’ other elements with extreme concentrations. For
concentrations because this value is used by the example, two carbonate rock CRMs (NCS DC70306
variable-alpha function. In order to evaluate the and NIST-1d) were removed from the calibration lines
accuracy of the matrix balancing, CO2 concentrations for Sc, V and Ba, two marine sediment CRMs (NCS
were compared to conventionally measured LOI550 DC75304 and NRC/CNRC PACS-2) with extremely
data in samples from the small lake, Önsbacksdammen high Cl concentrations were removed for P, and one
(Ek and Renberg 2001). These samples have very low CRM (NIST-2710a) with very high Zn concentrations
carbonate concentrations, and hence LOI550 can be was removed from the W calibration line. For Br and
considered to be equal to the sum of LOI550 and Rb, respectively, CRMs (NIM GBW07404 and NCS
LOI950, which was used for the CRMs. For the entire DC70311 for Br; Mintek SARM-46 for Rb) were
range of LOI550 values, i.e. 24–52 %, there is reason- removed because of a large unexplained deviation
able agreement between the estimated and conven- from the calibration line established by the other
tionally measured values. This is reflected in both the CRMs.
down-core profiles, as well as by significant bi-variate The only element for which larger numbers of
correlations both between the two calibration methods CRMs throughout the calibration range were
(r = 0.872) and between the respective calibration removed, was S. The rational for this was that when
method and conventionally measured LOI550 using the XS Ge-C crystal, the exact peak position of
(r = 0.666 and 0.693 for LAKE_200 mg and S depends on the S-speciation (ESM Figs. 1 and 2;
LAKE_500 mg, respectively; Fig. 3j). As mentioned (Coulson and Zauli 1963). Because the SpectraPlusÒ
above, the accuracy of the CO2 values is far too poor to software measures the peak intensity at a single fixed
be considered as an analytical result, however the wavelength, which is optimized during the line
values are consistent between both methods and selection procedure, this peak shift implies that the
display a pattern similar to the LOI550 data, which measured peak height does not necessarily reflect the
ensures that the matrix balancing provides a reason- maximum peak height, and hence CRMs with
able approximation of the total carbon content, which different S speciation will plot along different
can be used for the variable alpha corrections. For P, calibration lines (ESM Figs. 3 and 4). The objective
the calibration line was significantly improved by of removing selected CRMs was to obtain better
using customized alpha corrections based on the accuracy based on quality control samples (QCs) and
obtained intensities for Al, Si, Ca, Ti and Fe, and the Önsbacksdammen samples, which were mea-
concentrations of carbon (C) and oxygen (O), rather sured previously using ICP-AES (Ek and Renberg
than the variable-alpha function. 2001).
During the calibration-line optimization it was also In addition to the inter-elemental matrix correction
necessary to remove certain CRMs to obtain a good and the removal of certain CRMs, the calibration lines
calibration line for some elements. In most cases the for V, Cr, Co, Ga, As, Rb, and Y were also optimized
removed CRMs had either very low concentrations by correcting the measured intensities for the influence
(close to or below the LLD; i.e. Cr, Cu, As, Y, Sn, Sb, from overlapping neighboring peaks (ESM Table 2;
W) or very high concentrations that are only rarely ESM Figs. 1 and 2). These corrections were based on
encountered in sediments, and that negatively affected the peak height of the analyzed line for the interfering
the performance in the lower range of the calibration element and the known ratio between the analyzed line
line (i.e. Mg, Ca, Cr, Cu, Zn, As, Pb; ESM Table 1). and interfering line, e.g. the measured intensity of the
CRMs with low concentrations sometimes have a Mn Ka-line is used to calculate the theoretical
peak:background ratio close to one (ESM Table 2), intensity of the Mn Kb-line, which in turn is used to
however, from the calibration lines for each element, correct the intensity of the Cr Kb-line.

123
J Paleolimnol (2014) 52:265–276 269

Assessing the performance of the calibration concentration (n = 3–5) using the correct sample weight
method and either the certified value or the reported concentra-
tions for the deviating sample weight, respectively. Bi-
Lower limits of detection (LLD) are estimated by the variate correlations are calculated as two-tailed Pearson
SpectraPlusÒ software for each element using repli- correlations, using IBM SPSS Statistics 20 (www.spss.
cate measurements of 10 quality control samples com) and p values \0.01 were considered significant.
(QCs), i.e. commercially available CRMs not used in Only significant correlations are reported.
the development of the calibration methods (ESM
Table 3), using Eq. 1,
rffiffiffiffiffi Results and discussion
3 Ib
LLD ¼ ð1Þ
Si Tb Overall, both calibration methods performed well. The
where Si is the analytical sensitivity (Cps/concentra- linearity of calibration lines was good for both
tion), Ib is the background intensity (Cps) and Tb is the calibration methods, and the standard deviation of
background measurement time (s). the calibration lines ranged from 1 to *30 ppm for
Repeatability (i.e. agreement between several mea- trace elements and from 0.01 to 4 % for major
surements using the same analytical method), accuracy elements (Table 1; ESM Figs. 3 and 4). The LLD,
and reproducibility (i.e. agreement between several calculated as an average for all ten QCs, generally
measurements using different analytical methods), ranged from 1 to 15 ppm, with Ba (for
were assessed by repeated measurements (n = 3–5) of LAKE_200 mg) and Cr being exceptions, with LLDs
10 QCs. One of the QCs (NCS DC-73310) was also of 32 and 43 ppm, respectively (Table 1). For most
used to assess how analytical performance is affected elements the LLD was lower for LAKE_500 mg,
by sample weight, that is, if more or less than 200 or compared to LAKE_200 mg, and there was a ten-
500 mg of sample material is used for analysis (100, dency for LAKE_500 mg to perform slightly better
150, 300 and 500 mg for LAKE_200 mg and 200, 400, than LAKE_200 mg in all respects, especially for low
600 and 1,000 mg for LAKE_500 mg). The latter is concentrations (Figs. 1, 2; Table 1). For most ele-
interesting for two reasons. Firstly it speeds up the ments, the LLDs are well below the concentrations
weighing process significantly if a larger variation can normally encountered in lake sediments, however it
be accepted, and secondly, if the method is robust, it should be noted that the reported LLD is often lower
allows for analysis of samples that do not contain the than the lowest CRM included in the calibration line
required sample mass, i.e. 200 or 500 mg for (Table 1). Extra care should be taken when interpret-
LAKE_200 mg and LAKE_500 mg, respectively. ing concentrations close to the lower end of
QCs with elemental concentrations below the calibra- the calibration range for Sc, V, Cr, Co, As, Ga, Y,
tion range or LLD were not used in the performance Sn, Sb, W.
evaluation for that particular element (ESM Table 3). For the majority of elements (Na, Mg, Al, Si, K, Ca,
Reproducibility of the XRF analyses was also assessed Sc, Ti, V, Mn, Fe, Ni, Cu, Zn, Ga, As, Rb, Sr, Zr, Ba
using a down-core sediment profile from Önsbacks- and Pb), average repeatability, accuracy and repro-
dammen that was previously analyzed for several ducibility for the ten QCs are within ±10 % (Figs. 1,
elements (Mn, Fe, Co, Ni, Cu, Zn, Pb, and 2; Table 1). For Mn, Fe, Co, Ni, Cu, Zn and Pb, the
LOI550) using ICP-MS, and S using ICP-AES good performance is further validated by the good
(Ek and Renberg 2001). coherence between the values obtained using the two
Repeatability was calculated as absolute and relative WD-XRF methods and the ICP-MS or ICP-AES data
standard deviations, whereas reproducibility was calcu- from the Önsbacksdammen sediments (Ek and Ren-
lated as the absolute and relative difference in the average berg 2001; Fig. 3; Table 2). That these samples were
reported concentrations between LAKE_200 mg and analyzed without any pretreatment, except drying,
LAKE_500 mg. Accuracy and stability against varia- shows that for the fine-grained material found in the
tions in sample weight were calculated as the absolute and lake’s deep basin, it is not necessary to grind samples
relative deviation between the average reported prior to WD-XRF analysis. It should be noted,

123
 Table 1 Analytical performance, range and fit of the calibration lines, lower limit of detection (LLD), repeatability and accuracy, for the two calibration methods
270

Element Range of CRMs No. of CMRs Range of QCs No. QCs LAKE_200 mg

SD of calib.line Analyzed layer LLD Precision Accuracy

123
SD Rel. SD Difference Rel. difference

Na (%) 0.01–7.16 21 0.08–1.81 9 0.35 4 lm 0.0007 0.02 (0.01–0.07) 3 (2–6) 0.05 (0.01–0.2) 10 (2–37)
Mg (%) 0.12–3.16 20 0.04–1.65 9 0.21 7 lm 0.0003 0.02 (0–0.06) 2 (1–11) 0.06 (0–0.27) 8 (0–28)
Al (%) 0.53–29.26 21 4.92–12.41 9 1.16 11 lm 0.0003 0.14 (0.02–0.32) 2 (0–3) 0.32 (0.07–0.59) 4 (1–11)
Si (%) 4.1–89 21 23.82–38.69 8 4.4 13 lm 0.0016 0.58 (0.21–0.85) 2 (1–3) 2.01 (0.71–2.94) 5 (2–14)
P (ppm) 180–9458 19 235–1600 8 76 8 lm 1 16 (4–33) 2 (1–5) 30 (12–110) 8 (1–18)
S (%) 0.010–1.500 14 0.014–1.310 7 0.031 11 lm 0.0008 0.002 (0–0.03) 3 (2–6) 0.01 (0.01–0.43) 28 (6–50)
K (%) 0.13–7.48 21 0.86–3.27 9 0.11 31 lm 0.0014 0.03 (0.01–0.07) 2 (1–3) 0.06 (0–0.20) 2 (0–9)
Ca (%) 0.16–16.40 19 0.07–2.61 9 0.2 39 lm 0.0015 0.01 (0–0.05) 2 (1–5) 0.02 (0–0.06) 5 (1–17)
Sc (ppm) 2–28 16 5–26 8 – – – – – – –
Ti (ppm) 306–33694 21 659–10800 9 486 65 lm 14 52 (16–272) 1 (1–3) 107 (18–756) 3 (1–9)
V (ppm) 24–358 19 47–360 8 12 84 lm 8 4 (2–10) 4 (1–9) 3 (1–12) 4 (1–8)
Cr (ppm) 23–559 16 35–370 7 – – – – – – –
Mn (ppm) 270–11400 21 519–2060 8 135 136 lm 8 19 (2–55) 2 (0–3) 15 (4–94) 2 (0–7)
Fe (%) 0.32–28.16 21 0.17–8.83 9 0.27 172 lm 0.0018 0.1 (0–0.17) 2 (1–3) 0.13 (0.01–0.81) 2 (0–13)
Co (ppm) 2–97 20 8–48 8 – – – – – – –
Ni (ppm) 2–276 19 13–93 8 3.1 232 lm 1 1 (1–3) 4 (3–10) 1 (1–3) 2 (1–9)
Cu (ppm) 12–566 16 13–1230 8 12 286 lm 1 4 (2–129) 6 (3–11) 5 (0–19) 1 (0–8)
Zn (ppm) 14–797 19 52–2056 8 9.9 350 lm 2 8 (1–53) 3 (1–5) 7 (1–158) 5 (0–8)
Ga (ppm) 2–39 11 14–32 7 – – – – – – –
As (ppm) 6–512 14 19–412 7 7.4 850 lm 6 7 (2–15) 8 (4–19) 2 (0–13) 3 (0–22)
Br (ppm) 1–145 9 1–9 6 1.2 900 lm 1 1 (0–1) 18 (4–68) 0.1 (0–1) 10 (0–20)
Rb (ppm) 6–466 19 27–270 8 7.5 1.3 mm 1 3 (0–12) 2 (0–5) 4 (2–22) 4 (2–8)
Sr (ppm) 25–1100 20 24–202 8 19 1.5 mm 1 2 (0–13) 4 (1–7) 4 (1–12) 5 (1–92)
Y (ppm) 14–98 15 17–40 7 – – – – – – –
Zr (ppm) 9–500 16 111–500 7 28 2.9 mm 5 15 (3–34) 6 (2–14) 10 (0–24) 6 (4–11)
Sn (ppm) 13–370 7 18–54 3 – – – – – – –
Sb (ppm) 11–60 6 12–35 3 – – – – – – –
Ba (ppm) 42–1490 19 206–1199 9 77 18 mm 32 31 (14–104) 9 (3–13) 13 (4–63) 5 (0–7)
W (ppm) 5–179 8 34–37 2 3.7 0.33 mm 2 3 (1–5) 10 (4–15) 5 (4–7) 18 (12–25)
Pb (ppm) 6–731 16 23–1200 9 18 1.0 mm 2 2 (1–20) 3 (1–9) 6 (3–75) 7 (1–18)
J Paleolimnol (2014) 52:265–276
 Table 1 continued
Element LAKE_500 mg Reproducibility

SD of Calib.line Analyzed layer LLD Precision Accuracy Difference Rel. Correlation

difference
SD Rel. St.Deviation Difference Rel. Difference

Na (%) 0.34 4 lm 0.0005 0.01 (0–0.05) 2 (0–6) 0.03 (0.02–0.16) 7 (3–26) 0.02 (0.01–0.05) 2 (1–17) 1.000
Mg (%) 0.22 6 lm 0.0001 0.01 (0–0.04) 2 (0–3) 0.05 (0.01–0.28) 7 (3–29) 0.01 (0–0.2) 1 (1–28) 1.000
Al (%) 1.13 10 lm 0.0002 0.06 (0–0.20) 1 (0–2) 0.36 (0–0.49) 5 (0–9) 0.10 (0.01–0.27) 1 (0–2) 1.000
Si (%) 4.43 13 lm 0.0008 0.25 (0.08–0.52) 1 (0–2) 2.00 (0.78–2.77) 5 (2–12) 0.37 (0.05–0.72) 1 (0–3) 0.999
J Paleolimnol (2014) 52:265–276

P (ppm) 104 8 lm 0.4 8 (3–40) 2 (1–4) 83 (10–146) 11 (3–41) 85 (54–169) 21 (5–45) 0.992
S (%) 0.027 11 lm 0.0003 0.001 (0–0.02) 2 (1–36) 0.012 (0–0.42) 21 (12–48) 0.003 (0–0.062) 3 (1–67) 1.000
K (%) 0.12 31 lm 0.0006 0.02 (0–0.03) 1 (0–2) 0.10 (0.02–0.21) 4 (2–9) 0.02 (0–0.05) 1 (0–3) 1.000
Ca (%) 0.23 38 lm 0.0007 0.00 (0–0.03) 1 (0–1) 0.02 (0–0.07) 6 (1–19) 0.01 (0–0.02) 2 (0–4) 1.000
Sc (ppm) 3.2 49 lm 2.9 2 (1–3) 10 (8–24) 1 (1–3) 10 (6–20) – – –
Ti (ppm) 564 64 lm 6.2 38 (1–121) 1 (0–1) 78 (7–996) 3 (0–10) 86 (7–311) 1 (0–3) 1.000
V (ppm) 11 83 lm 3.4 2 (1–3) 1 (1–3) 3 (0–27) 3 (0–7) 3 (1–15) 4 (1–9) 1.000
Cr (ppm) 22 137 lm 40 20 (11–28) 21 (4–49) 6 (1–10) 11 (0–17) – – –
Mn (ppm) 125 134 lm 3.2 10 (1–22) 1 (0–2) 24 (2–97) 3 (0–6) 11 (5–74) 1 (0–101) 1.000
Fe (%) 0.35 169 lm 0.0006 0.04 (0–0.09) 1 (0–1) 0.15 (0.02–0.81) 3 (2–11) 0.02 (0.02–0.12) 1 (0–5) 1.000
Co (ppm) 2.6 210 lm 2.4 1 (0–2) 6 (0–10) 2 (0–4) 12 (3–23) – – –
Ni (ppm) 2.9 219 lm 0.5 1 (0–2) 1 (0–3) 2 (0–2) 3 (0–11) 1 (0–2) 2 (1–11) 1.000
Cu (ppm) 6.3 270 lm 0.5 1 (0–34) 1 (0–4) 2 (0–67) 2 (0–7) 2 (0–82) 3 (0–37) 0.999
Zn (ppm) 10 330 lm 0.9 3 (0–22) 1 (0–2) 11 (0––203) 5 (0–11) 6 (2–41) 3 (0–44) 1.000
Ga (ppm) 0.7 390 lm 0.8 1 (0––1) 4 (2–5) 0 (0–4) 1 (0–16) – – –
As (ppm) 4.1 800 lm 2.6 2 (0–3) 4 (0–5) 2 (1–6) 5 (1–11) 4 (0–11) 6 (1–21) 1.000
Br (ppm) 0.9 830 lm 0.5 0.3 (0.2–0.5) 15 (1–28) 0.4 (0–1.2) 13 (0–42) 1 (0–1) 13 (2–41) 1.000
Rb (ppm) 4.8 1.2 mm 0.4 1 (0–3) 1 (0–1) 2 (0–21) 2 (0–8) 2 (0–8) 2 (0–5) 1.000
Sr (ppm) 9.2 1.4 mm 0.3 1 (0–2) 1 (0–4) 2 (1–4) 2 (1–9) 6 (0–14) 4 (0–74) 0.999
Y (ppm) 5.2 1.6 mm 0.6 0 (0–2) 2 (0–6) 5 (2–9) 18 (6–33) – – –
Zr (ppm) 17 2.7 mm 2.9 4 (1–16) 1 (1–6) 8 (1–29) 3 (0–13) 9 (4–24) 4 (1–10) 0.997
Sn (ppm) 6.1 7.8 mm 4.2 3 (2–14) 12 (7–23) 6 (3–8) 15 (13–15) – – –
Sb (ppm) 6.1 8.9 mm 4.8 1 (1–3) 5 (4–12) 3 (0–6) 18 (1–19) – – –
Ba (ppm) 27 16 mm 13 13 (3–36) 3 (1–12) 28 (4–174) 6 (1–16) 5 (1–171) 5 (0–13) 0.996
W (ppm) 5.3 0.33 mm 1.0 1 (1–1) 3 (2–3) 4 (3–6) 14 (9–20) 1 (1–2) 3 (2–6) 0.997
Pb (ppm) 16 1.0 mm 0.7 1 (0–6) 1 (0–3) 9 (3–90) 10 (3–27) 6 (1–15) 3 (0–10) 1.000

The table also shows how coherent the two calibration methods are (reproducibility)
271

123
272 J Paleolimnol (2014) 52:265–276

however, that for more coarse-grained sediment, e.g. absolute difference between the various sample
from nearshore locations, grinding would be required weights is\1 ppm. For the remaining elements, there
to obtain accurate results. For a smaller number of is an effect on the reported concentrations, both when
elements, either the relative standard deviations (Br the sample weight is larger or smaller than 200 or
and Sn) or the relative deviation from the certified 500 mg, for LAKE_200 mg and LAKE_500 mg,
concentration (Co, Br, Y, Sn, Sb and W) is larger than respectively. This is likely attributable to the thickness
±10 % (Figs. 1, 2; Table 1). This, however, is largely of the analyzed layer, in relation to the sample
a result of low elemental concentrations, for which a thickness, which, with a fixed sample diameter, will
small absolute deviation translates into a large relative be directly correlated to the sample weight. For lighter
deviation. The absolute standard deviation for Br and elements, the analyzed layer will be thinner, compared
Sn is only ±1 and 3 ppm, respectively, and the to heavy elements, because light elements emit
absolute deviations from the certified concentrations secondary X-rays of lower energy that are extin-
for Co, Br, Y, Sn, Sb and W fall within the range guished faster in the sample as compared to heavy
0.1–6 ppm (Table 1). Even if the analytical perfor- elements (Table 1), and as long as the sample
mance is generally good, it is important to recognize thickness is greater than the analyzed layer, variations
that individual QCs have reported concentrations that in the sample weight will not affect the analytical
show a larger deviation from their certified concen- result. However, as soon as a reduction in sample
tration. This is likely attributable to differences in the weight causes the sample thickness to be less than the
sample matrix, i.e. mineralogy, organic content and analyzed layer, reported concentrations will be
speciation of certain elements, that aren’t fully com- affected. For the very heavy elements, for which the
pensated for by the inter-elemental matrix corrections analyzed layer may already be greater than the sample
made by the variable alpha function. To fully resolve thickness for the intended sample weight, an increased
this issue, without using different fusion techniques, it sample weight will affect the reported concentration.
would be necessary to develop a large number of W and Pb are less affected by changes in sample
calibration methods for very narrow ranges of sample weight, compared to lighter elements Ba and Zr,
matrix types, and it would be necessary to know the because W and Pb are determined using L-lines that
approximate composition and mineralogy of the emit X-rays with a lower energy, and consequently
sample prior to analysis in order to select the have a thinner analyzed layer than the K-lines used for
appropriate calibration method. It is also important Ba and Zr (Table 1).
to recognize that for some elements (V, Cr, Co, Ga,
As, Rb, Y), there are overlaps between the analyzed Problematic elements—Cr, P and S
peak and other neighboring peaks (ESM Figs. 1 and
2). Even if the peak-overlap correction reduces this The calibration methods for Cr, P and S do not perform
effect, the correction is only valid as long as the as well as for the other analyzed elements. For Cr there
concentration of the overlapping element is within the is a need to recognize that repeatability is rather poor,
range of the used CRMs. and that a single measurement probably is insufficient
In general, light elements are not sensitive to to correctly quantify the Cr concentration in the lower
variations in sample weight, and regardless of the range of the calibration line (Figs. 1, 2). For P, it is
sample mass used, both calibration methods report mainly the reproducibility between LAKE_200 mg
similar concentrations for elements lighter than As for and LAKE_500 mg that is of concern, and when
LAKE_200 mg and Sr for LAKE_500 mg (Fig. 4). In comparing data obtained by the two methods, it is
addition, the concentrations for W and Pb, using necessary to recognize that LAKE_200 mg overesti-
LAKE_500 mg, are stable across the whole range of mates (?12 %) and LAKE_500 mg underestimates
tested sample weights (200–1,000 mg). The apparent (-11 %) the P concentrations, compared to the
sensitivity to variations in sample weight for Br is certified values (Fig. 2; Table 1). However, even if
associated with the low Br concentration in the QC the absolute values differ between the two calibration
(i.e. 2 ppm), and similarly for the repeatability, the methods, the good bi-variate correlation (r = 0.992)

123
J Paleolimnol (2014) 52:265–276 273

Fig. 1 Box plot showing the repeatability for LAKE_200 mg measurements for each of 2–9 QCs). Boxes represent 25, median
(white boxes) and LAKE_500 mg (grey boxes) for all 30 and 75 %-percentiles and whiskers are 10 and 90 %-percentiles
elements (calculated as relative standard deviations for 3–5

Fig. 2 Box plot showing the repeatability for LAKE_200 mg concentration, based on 3–5 measurements for each of the 2–9
(white boxes) and LAKE_500 mg (grey boxes) for all 30 QCs). Boxes represent 25, median and 75 % percentiles and
elements (calculated as relative deviations from the certified whiskers are 10 and 90 % percentiles

shows that the relative difference in P concentrations profile from Önsbacksdammen shows that there is
between the QCs are captured by both LAKE_200 mg reasonable coherence between the two WD-XRF
and LAKE_500 mg (Table 2). Also, the down-core P methods and the ICP-AES data for a wide range of S
profile in the sediment of Önsbacksdammen shows concentrations (0.3–3.2 % S; Fig. 3; Table 2; Ek and
that there is good coherence between the results Renberg 2001). One way to improve the analytical
obtained using LAKE_200 mg and LAKE_500 mg performance for S would be to use a measurement
(Fig. 3). method that determines the peak height by continuous
As mentioned in the methods section, speciation of scans. However, because it is preferable to use a single
S has an effect on the exact S Ka peak position, which peak-position measurement for the other 29 elements,
affects the quantification of S in the samples (ESM which requires much shorter analytical time, this
Figs. X, Y, Z and W; Coulson and Zauli 1963). cannot be done using the SpectraPlusÒ software. It is
However, even though the QCs indicate a lower therefore necessary to keep in mind that S speciation
accuracy for S (i.e. ±26 and 21 % for LAKE_200 mg might affect the reported concentrations when inter-
and LAKE_500 mg, respectively) the down-core preting S data obtained using the presented calibration

123
274 J Paleolimnol (2014) 52:265–276

Fig. 3 Comparison between the analytical results from LAKE_200 mg, LAKE_500 mg and previously published ICP-MS/ICP-AES
data for the sediments of the small lake Önsbacksdammen (Ek and Renberg 2001)

Table 2 Bi-variate Correlations LAKE_500 mg versus LAKE_500 mg versus LAKE_200 mg versus

correlation coefficients LAKE_200 mg ICP-MS ICP-MS
between LAKE_200 mg,
LAKE_500 mg and S 0.997* 0.977* 0.971*
previously published ICP-
Mn 0.963* 0.778* 0.845*
MS/ICP-AES data for 19
sediment samples from the Fe 0.994* 0.980* 0.975*
small lake Co – 0.894* –
Önsbackadammen (Ek and Ni 0.933* 0.845* 0.922*
Renberg 2001)
Cu 0.997* 0.989* 0.997*
Zn 0.998* 0.980* 0.986*
Pb 0.993* 0.928* 0.963*
CO2 versus LOI550 0.872* 0.693* 0.666*
* p \ 0.01

methods. If there is a suspicion that the difference in S concentration, continuous spectra obtained by the
reported S concentration between two samples is a semi-quantitative GeoQuantÒ application can be used
consequence of differences in S speciation rather than to determine if there is a shift in the peak position.

123
J Paleolimnol (2014) 52:265–276 275

Fig. 4 Plots show the relative deviation in the reported intended sample mass, and the black (filled) circles represent,
concentration if the sample mass deviates from 200 or from left to right for each element, 100, 150, 300 and 500 mg for
500 mg, for LAKE_200 mg (a) and LAKE_500 mg (b), LAKE_200 mg, and 200, 400, 600 and 1,000 mg for
respectively. The open circle with error bars represents the LAKE_500 mg
mean and standard deviation for five replicates using the

Conclusions Acknowledgments The WD-XRF instrument was funded by

the Kempe foundation through a grant awarded to Professor
Richard Bindler, Department of Ecology and Environmental
Results presented here show that with a modern, Science, Umeå University. Financial support was also granted
commercially available WD-XRF and its built-in from the faculty of Science and Technology at Umeå University.
software, it is possible to develop quantitative calibra- I also thank Lars Lidén and Anne Wenger at Bruker-AXS for
tion methods using small loose-powder samples. The support with technical and software questions.
reported calibration methods displayed reasonable
analytical performance, with LLDs generally well
References
above concentrations normally encountered in lake
sediment, and accuracy and precision within 10 % for Battarbee R, Jones V, Flower R, Cameron N, Bennion H,
most elements. It is important to recognize, however, Carvalho L, Juggins S (2001) Diatoms. In: Smol J, Birks
that the inter-elemental matrix effect can affect the HJ, Last W, Bradley R, Alverson K (eds) Tracking envi-
accuracy negatively, especially if the sample matrix is ronmental change using lake sediments. Springer, Neth-
erlands, pp 155–202
very different from that of the used CRMs. Because of Bennett KD, Willis KJ (2001) Pollen. In: Smol J, Birks HJ, Last
the low analytical costs (*10 € or $14 per sample), W, Bradley R, Alverson K (eds) Tracking environmental
lack of pretreatment and short analysis time (25–35 change using lake sediments. Springer, Netherlands,
samples per 24 h), XRF analysis is cost-effective, even pp 5–32
Bindler R, Segerstrom U, Pettersson-Jensen IM, Berg A,
when considering the relatively high initial cost to Hansson S, Holmstrom H, Olsson K, Renberg I (2011)
acquire the XRF instrument. All in all, this technical Early medieval origins of iron mining and settlement in
note shows that WD-XRF analysis on small (200 and central Sweden: multiproxy analysis of sediment and peat
500 mg), loose-powder samples is well suited for records from the Norberg mining district. J Archaeol Sci
38:291–300
routine geochemical analysis for paleolimnological Boyle JF (2000) Rapid elemental analysis of sediment samples
and other environmental applications, when it is by isotope source XRF. J Paleolimnol 23:213–221
advantageous to use a truly non-destructive analytical Brodie CR, Casford JSL, Lloyd JM, Leng MJ, Heaton THE,
technique. With further modifications, it will be Kendrick CP, Zong YQ (2011) Evidence for bias in C/N,
delta C-13 and delta N-15 values of bulk organic matter,
possible to adapt the methods presented here for other and on environmental interpretation, from a lake sedi-
ranges of sample matrices, e.g. samples rich in organic mentary sequence by pre-analysis acid treatment methods.
matter, such as plants and peat. Quat Sci Rev 30:3076–3087

123
276 J Paleolimnol (2014) 52:265–276

Brown GC, Hughes DJ, Esson J (1973) New XRF data retrieval Margui E, Hidalgo M, Queralt I (2005) Multielemental fast
techniques and their application to USGS standard rocks. analysis of vegetation samples by wavelength dispersive
Chem Geol 11:223–229 X-ray fluorescence spectrometry: possibilities and draw-
Butler OT, Cook JM, Harrington CF, Hill SJ, Rieuwerts J, Miles backs. Spectrochim Acta B 60:1363–1372
DL (2008) Atomic spectrometry update. Environmental Mucke A, Farshad F (2005) Whole-rock and mineralogical
analysis. J Anal At Spectrom 23:249–286 composition of Phanerozoic ooidal ironstones: comparison
Cheburkin AK, Shotyk W (1996) An energy-dispersive minip- and differentiation of types and subtypes. Ore Geol Rev
robe multielement analyzer (EMMA) for direct analysis of 26:227–262
Pb and other trace elements in peat. Fresenius J Anal Chem Norrish K, Hutton JT (1969) An accurate X-ray spectrographic
354:688–691 method for the analysis of a wide range of geological
Coulson CA, Zauli C (1963) The K-alpha transition in com- samples. Geochim Cosmochim Ac 33:431–453
pounds of sulphur. Mol Phys 6:525–533 Ohlendorf C, Sturm M (2008) A modified method for biogenic
Ek AS, Renberg I (2001) Heavy metal pollution and lake acidity silica determination. J Paleolimnol 39:137–142
changes caused by one thousand years of copper mining at Renberg I, Hansson H (2008) The HTH sediment corer. J Pa-
Falun, central Sweden. J Paleolimnol 26:89–107 leolimnol 40:655–659
Farmer JG, Eades LJ, Mackenzie AB, Kirika A, BaileyWatts TE Revenko AG (2011) Development of X-ray fluorescence ana-
(1996) Stable lead isotope record of lead pollution in Loch lysis in Russia in 1991–2010. J Anal Chem 66:1059–1072
Lomond sediments since 1630 AD. Environ Sci Technol Rousseau RM (2006) Corrections for matrix effects in X-ray
30:3080–3083 fluorescence analysis: a tutorial. Spectrochim Acta B
Gazulla MF, Vicente S, Orduña M, Ventura MJ (2012) Chemical 61:759–777
analysis of very small-sized samples by wavelength-dis- Rydberg J, Gälman V, Renberg I, Lambertsson L, Martinez
persive X-ray fluorescence. X-Ray Spectrom 41:176–185 Cortizas A, Bindler R (2008) Assessing the stability of
Glew JR, Smol JP, Last WM (2001) Sediment core collection mercury and methylmercury in a varved lake sediment
and extrusion. In: Last WM, Smol JP (eds) Tracking deposit. Environ Sci Technol 42:4391–4396
environmental change using lake sediments vol 1: basin Rydberg J, Rosén P, Lambertsson L, De Vleeschouwer F, To-
analysis, coring and chronological techniques. Kluwer masdotter S, Bindler R (2012) Assessment of the spatial
Academic, Dordrecht, pp 73–105 distributions of total- and methyl-mercury and their rela-
Heiri O, Lotter AF, Lemcke G (2001) Loss on ignition as a tionship to sediment geochemistry from a whole-lake
method for estimating organic and carbonate content in perspective. J Geophys Res 117:G04005
sediments: reproducibility and comparability of results. Yang HD, Rose NL, Battarbee RW, Boyle JF (2002) Mercury
J Paleolimnol 25:101–110 and lead budgets for Lochnagar, a Scottish mountain lake
Koinig KA, Shotyk W, Lotter AF, Ohlendorf C, Sturm M (2003) and its catchment. Environ Sci Technol 36:1383–1388
9000 years of geochemical evolution of lithogenic major Zaborska A, Carroll J, Papucci C, Pempkowiak J (2007) Inter-
and trace elements in the sediment of an alpine lake: the comparison of alpha and gamma spectrometry techniques
role of climate, vegetation, and land-use history. J Paleo- used in Pb-210 geochronology. J Environ Radioact
limnol 30:307–320 93:38–50
Lachance GR (1993) Correction procedures using influence
coefficients in X-Ray-fluorescence spectrometry. Spectro-
chim Acta B 48:343–357

123