Rapid simultaneous multi-element determination of soils and environmental samples with polarizing energy dispersive X-ray fluorescence (EDXRF) spectrometry using pressed powder pellets

Abstract

A rapid simultaneous multi-element analysis method for soils and environmental samples has been established using polarizing energy dispersive X-ray fluorescence (EDXRF) spectrometry. The pressed powder pellet technique was adopted because it is simple and requires no specialized skills for sample preparation. The analytes examined were: Na₂O, MgO, Al₂O₃, SiO₂, P₂O₅, K₂O, CaO, TiO₂, MnO, Fe₂O₃, V, Cr, Co, Ni, Cu, Zn, Rb, Sr, Y, Zr, Nb, Sn, Sb, Cs, Ba, La, Ce, Pr, Nd, Pb and Th. Twenty-six reference samples were used for the calibrations. Compton scatter radiation was used as an internal standard to compensate for variations in the sample matrix, particle size, packing density and operating characteristics of the instrument. The correlation coefficients of the linear regression lines were >0.98, with the exception of Na₂O, MgO, Al₂O₃, SiO₂, P₂O₅, Co, Pr and Nd. The values obtained using the proposed EDXRF spectrometry were compared with the values obtained using other methods. Out of the 31 analytes examined, the results for 12 compounds (CaO, K₂O, TiO₂, MnO, Fe₂O₃, Ni, Zn, Rb, Sr, Cs, Ba and Ce) obtained by the proposed EDXRF spectrometry compared favorably with those determined through conventional wet chemical methods. In contrast, the results for eight compounds (Na₂O, MgO, Cr, Co, Zr, Sn, Sb and Pr) obtained by the proposed EDXRF spectrometry exhibited poor agreements with those obtained using chemical methods. Among these analytes, the poor agreements of Cr, Zr and Sn were attributable to incomplete dissolution and/or volatilization losses during the chemical treatments based on an acid attack. It can be concluded, therefore, that there is a high possibility of underestimating the concentrations of these elements if wet chemical methods are used. Although the results of the other 11 analytes were not as good as those of the first group, they still appeared to be of practical use, considering the time consuming and potentially hazardous wet chemical pretreatment.

Key words:

• energy dispersive X-ray fluorescence spectrometry
• environmental samples
• simultaneous multi-element analyses
• soil samples

Introduction

In recent years, both the number of soil samples and the types of elements to be analyzed have been rapidly increasing. Although techniques such as inductively coupled plasma-mass spectrometry (ICP-MS), inductively coupled plasma-atomic emission spectrometry (ICP-AES) and atomic absorption spectrometry (AAS) are commonly used to meet the above requirements (Butler et al. 2006, 2009; Yamasaki 2000), all of these methods require time-consuming and potentially hazardous acid treatments for sample dissolution. Furthermore, a well-equipped chemical laboratory and the assistance of skillful analysts are indispensable. In contrast, X-ray fluorescence spectrometry (XRF) has many advantage in terms of: (1) simultaneous multi-element capability, (2) wider dynamic range, from parts per million (mg kg⁻¹) to 100%, (3) simple sample preparation procedure, (4) avoidance of poor recovery for certain elements during acid digestion, (5) easier machine operation, (6) higher sample throughput, (7) good reproducibility, (8) low operating costs. Because of these inherent advantages, XRF has been widely adapted for the analysis of major and trace elements in foods and plants (Margui et al. 2005; Nielson et al. 1991), farmyard manure (Matsunami et al. 2009), and soil and environmental samples (Gunicheva et al. 1995; Jørgensen et al. 2005; Nakayama et al. 2007; Orescanin et al. 2008; Yamasaki et al. 1980; Yu et al. 2002).

Despite these advantages, energy dispersive X-ray fluorescence (EDXRF) spectrometry is used to a lower extent comparatively than wavelength dispersive X-ray fluorescence (WDXRF) spectrometry owing to its lower energy resolution, particularly for light elements. However, technical advances have been made in EDXRF instrumentation. Because the conventional and widely adopted EDXRF uses unpolarized X-rays as its primary excitation radiation, background levels are higher because the unpolarized X-rays contain scattering from the X-ray tube and sample matrices. As a consequence, the detection limits are degraded for trace elements in light matrices, such as plant materials and high silicate rocks and soils (Heckel et al. 1992; Hittipathirana 2004). Spectrum analysis programs have also been developed that allow multiple-peak X-ray profiles to be fitted to spectra; such a fitting compensates for problems attributable to peak overlapping.

Taking all of these factors into consideration, we decided in the present study to examine the applicability of the recently developed EDXRF spectrometer with 3-D optics, in the hope that it will realize the efficient analysis of a large variety of elements in numerous soil and environmental samples with acceptable speed and accuracy.

Materials and methods

Instrumentation

The EDXRF spectrometer (Epsilon 5; PANalytical B. V., Lelyweg, The Netherlands) used in the present study was equipped with a Sc/W anode X-ray tube with the goal of increased sensitivity for lighter as well as heavier elements. The instrument has a series of user selectable secondary targets and is equipped with a liquid-nitrogen-cooled Ge solid-state high-resolution detector with a Be (8 μm) window. A 3-D design (or Cartesian geometry) was adopted for the instrument to eliminate the X-ray tube spectrum by polarization. Consequently, the background can be an order of magnitude lower than the traditional 2-D optics, resulting in much lower detection limits. The overlapping of the characteristic L lines of heavier elements with the K lines of lighter elements (a typical example is the overlap of Cd L lines on K K lines) is a major drawback in determining heavy elements using the EDXRF method. However, because the instrument used in the present study is equipped with a 100 kV Sc/W tube, even heavy rare earth elements (from Eu to Lu) can be analyzed using the more sensitive K lines, and hence the instrument is less susceptible to spectral overlapping with the K lines of lighter elements.

Construction of the calibration curves

The instrument was calibrated for 10 major constituents and 21 selected trace elements using 26 geological and environmental standard reference materials. Although calibration can be done using “homemade” (secondary) standards, including the method of standard additions, the most widely used and well-established method of utilizing standard reference materials was adopted in the present study. The reference materials are listed in Table 1, together with their basic information and the producers. More detailed information has been presented elsewhere (Ando et al. 1989; Govindaraju 1994; Imai et al. 1995a,b, 1996; National Bureau of Standards and Technology 1993a,b, 2000, 2002; Terashima et al. 1998). The selection of trace elements was based on previous information concerning the background concentration of each element in soils (Egashira et al. 1996, 2004, 2007; Kimura et al. 2008; Nanzyo et al. 2002; Takeda et al. 2004; Yamasaki et al. 2001, 2009). Although the existence of Cl and S was occasionally identified in the X-ray spectrum, these elements were not included because their occurrence rates were not very high. In contrast, although Br and I were observed in almost all of the samples, a lack of appropriate reference materials made it impossible to construct calibration curves for these elements.

The measurement conditions for the major constituents and trace elements are given in Tables 2 and 3, respectively. As for the trace elements, the choice of secondary targets was a compromise between sensitivity and analytical speed. For example, the emitted X-ray intensities from Rb and Sr are much stronger when Zr is used as a target rather than Mo. However, the use of a Zr target makes it impossible to analyze both Zr and Y. As a result, these elements must be analyzed separately from Rb and Sr.

Sample preparation for X-ray spectrometry

The samples were prepared in the form of pressed powder pellets, without adding binders. They were first pulverized for 10 min in a planetary ball mill (pulverisette 7; Fritsch GmbH, Idar-Oberstein, Germany) at 450 rpm (135 m s⁻² or 13.8 G) and then pressed into 32-mm internal diameter pellets using a hydraulic press operated at a pressure of 20 tonnes (Specac, Kent, UK). To reduce errors resulting from any inhomogeneity caused by sample preparation, the spinner system in the spectrometer constantly rotated the samples during the measurements.

Table 1 Standard reference materials used for calibration

Name	Basic information	Producer
JA-1	Andesite (1982), Hakone volcano, Quaterary, Kanagawa, Japan	GSJ
JA-2	Andesite (1985), Goshikidai sanukitoid, Sakaide, Kagawa, Japan	GSJ
JA-3	Andesite (1986), Asama volcano, Tsumagoi, Gunma, Japan	GSJ
JB-1a	Basalt (1984), Kitamatsuura basalt, Sasebo, Nagasaki, Japan	GSJ
JB-2	Basalt (1982), Oshima volcano erupted in 1950–1951, Oshima, Tokyo, Japan	GSJ
JB-3	Basalt (1983), Fuji volcano, Narusawa, Yamanashi, Japan	GSJ
JF-2	Feldspar (1986), Kurosaka feldspar, Kurosaka, Ibaraki, Japan	GSJ
JGb-1	Gabbro (1983), Utsushigatake, Funehiki, Fukushima, Japan	GSJ
JGb-2	Gabbro (1991), Tsukuba-san leucogabbro, Yasato, Ibaraki, Japan	GSJ
JR-1	Rhyolite (1982), Wada Toge obsidian, Wada, Nagano, Japan	GSJ
JR-2	Rhyolite (1983), Wada Toge obsidian, Shimosuwa, Nagano, Japan	GSJ
JR-3	Rhyolite (1990), Ashizuri peralkaline rhyolite, Tosashimizu, Kochi, Japan	GSJ
JLk-1	Lake sediment (1987), Lake Biwa, Shiga, Japan	GSJ
JSd-1	Stream sediment (1988), Northern region, Ibaraki, Japan	GSJ
JSd-2	Stream sediment (1989), Eastern region, Ibaraki, Japan	GSJ
JSd-3	Stream sediment (1989), Central region, Ibaraki, Japan	GSJ
JS1-1	Slate (1988), Toyoma clay slate, Toyama, Miyagi, Japan	GSJ
JS1-2	Slate (1989), Toyoma clay slate, Okatsu, Miyagi, Japan	GSJ
SO-1	Regosolic clay soil, Hull, Quebec, Canada	CCRMP
SO-2	Podzalic B horizon soil, Montmorency forest, Quebec City, Canada	CCRMP
SO-3	Calcareous C horizon soil, Guelph, Ontario, Canada	CCRMP
SO-4	Chemozemic A horizon soil, Saskatoon, Saskachewan, Canada	CCRMP
NIST 2704	Buffalo River Sediment (reference material 8704)	NIST
NIST 2709	San Joaquin soil (baseline trace element concentrations)	NIST
NIST 2710	Montana soil (highly elevated trace element concentrations)	NIST
NIST 2711	Montana soil (moderately elevated trace element concentrations)	NIST
CCRMP, Canadian Certified Reference Materials Project; GSJ, Geological Survey of Japan; NIST, National Institute of Standards and Technology, USA.

Table 2 Measurement conditions for the major constituents

Analytes	X-ray tube		Secondary target	Measurement time (min)^†	Matrix correction procedure
	Voltage (kV)	Current (mA)
Na2O, MgO	25	24	Al	40	FP
A12O3∼K2O	40	15	CaF2	10	FP
CaO, TiO2	75	8	Fe	5	FP
MnO, Fe2O3	75	8	Ge	5	FP
^†Real time. FP, fundamental parameter method.

Table 3 Measurement conditions for the trace elements

Analytes	X-ray tube		Secondary target	Measurement time (min)^†	Matrix correction procedure^‡
	Voltage (kV)	Current (mA)
V	75	8	Fe	5	Fe-C
Cr∼Zn	75	8	Ge	10	Ge-C
Rb∼Y	100	6	Mo	10	Mo-C
Zr∼Nd	100	6	A12O3 ^§	10	Ce-C
Pb, Th	100	6	Mo	10	Mo-C
^†Real time. ^‡Compton scattering method. ^§Barkla scatter.

Correction of matrix effects

The X-ray spectra were analyzed using AXIL software (Analysis of X-ray Spectra by Interactive Least Squares Fitting [Van Espen et al. 1986; Vekemans et al. 1994]). The use of powder pellets pressed without a binder eliminates sample dilution and, therefore, increases the sensitivity and lowers the detection limits for trace elements. Additional advantages of the above technique are its simplicity and the fact that no specialized skills are required for sample preparation. Disadvantages include the difficulty of obtaining pellets with sufficient physical strength from soil samples rich in siliceous sand. Particle size effects and matrix effects are other negative factors with this approach. Therefore, to cope with these problems Compton scatter radiation was used as an internal standard to compensate for variations in the sample matrix, particle size, packing density and operating characteristics of the instrument (Table 3). The fundamental parameter method (FP) was also used for matrix correction for major constituents (Table 2). This method aims to establish a proper correction for the inter-elemental effect caused by other matrix elements in the standard reference materials by using several physical constants (fundamental parameters). Accordingly, information concerning all of the major constituents (including loss on ignition) in the reference materials should be registered in the data file, even when there is only one analyte of interest. As various specific parameters related to the instrument are also included in the software provided by the manufacturer, the details of the correction procedures are kept closed to the users. However, the effectiveness of this approach was shown by the fact that correlation coefficients obtained using FP were higher than those without FP.

Sample dissolution

The EDXRF spectrometry results were compared with the results obtained by chemical methods and the borate fusion-WDXRF method (SiO₂ only), using three groups of samples. Group 1 contained the oldest samples collected nationwide from 1980 to 1982 as cooperative research work on “Ando Soils in Japan” (Wada 1986). The sampling sites were selected in such a way as to cover typical volcanic ash soils common to Japan. Group 2 consisted of part of the samples systematically collected from agricultural lands from 1979 to 1997, where the concentrations of more than 40 trace elements have already been determined (Kimura et al. 2008). Group 3 contained the newly sampled specimens and chemical analyses were carried out in parallel with the EDXRF measurements.

The sample dissolution techniques were based on an attack with mixed acids containing HF and were, therefore, essentially similar. However, the details of the procedures were considerably different between the three groups of samples because the experiments were carried out in different laboratories and at different times.

The most primitive technique was used for acid digestion for Group 1 samples. One gram of a finely ground sample was treated with 10 mL of HClO₃ + HNO₃ (1:1 mixture) to decompose the organic matter in the soils, followed by a double treatment with 15 mL of HClO₃ + HF (1:2 mixture) in a polytetrafluoroethylene (PTFE) beaker on a hot plate manufactured for domestic use. The residue was heated with 5 mL of HNO₃ and dissolved by the addition of 30–50 mL of H₂O with gentle boiling and finally made up to a volume of 100 mL. The final solutions were stored in plastic bottles until the measurements were conducted (Yamasaki 1978; Yoshida et al. 1992).

For Group 2, a more efficient and less sample-consuming and acid-consuming microwave digestion technique was adopted. In brief, 0.1 g of finely ground sample was first treated with 2 mL of HNO₃ in a tightly sealed PTFE container in a domestic microwave oven for 15 min, then with 0.5 mL of HF for another 15 min, and finally with 0.5 mL of HClO₄ for 15 min. The digest was dried up on a hot plate after removing the sealed cup. The residue was heated with 2 mL of HNO₃ and dissolved by adding 10–20 mL of H₂O with gentle boiling and finally made up to a volume of 40 mL (Kimura et al. 2008; Mizushima et al. 1996).

For Group 3, the sample amounts were increased to avoid possible errors resulting from sample inhomogeneity. With this procedure, a specially designed heat block (DigiPREP MS; SCP SCIENCE, Bale-d' Urfé, Quebec, Canada) and digestion vessels made of PTFE were used. In this apparatus, it was possible to treat up to 48 samples at one time under a more precisely controlled temperature and heating period. The finely ground sample (0.500 g) was first treated with 4 mL of HClO₄–HNO₃ (1:1 mixture) for 45 min at 120°C and then with 4 mL of HClO₄–HF (1:1 mixture) twice for 45 min at 170°C. The residue was heated with 5 mL of HNO₃ and dissolved by adding 20 mL of H₂O at 100°C for 3–4 h and finally made up to a volume of 100 mL (Yamasaki et al. 2009).

Reference methods used for comparison

Inductively coupled plasma-mass spectrometry

After a 10–100-fold dilution, the acid digests were analyzed by ICP-MS for all of the trace and ultra-trace elements. Indium (In) was used as an internal standard element. The validity of the proposed method was examined by analyzing several reference materials, as shown in Table 1. The working standards were prepared from a series of SPEX Multi-Element Plasma Standards (XSTC-1, XSTC-7, XSTC-8 and XSTC-13) supplied by SPEX Industries (Edison, NJ, USA). It was possible to obtain working standard solutions containing up to 70 elements in this way.

For Group 2 samples, a quadrupole type ICP-MS, HP-4500 (Hewlett Packard [now Agilent Technologies], Palo Alto, CA, USA), was used for most of the trace elements (>0.1 mg kg⁻¹ soil). However, for several ultra-trace elements (<0.1 mg kg⁻¹) and/or trace elements susceptible to interferences resulting from spectral overlaps, it was necessary to use a high-resolution ICP-MS, ELEMENT (Finnigan MAT [now Thermo Scientific], Bremen, Germany). This type of ICP-MS is equipped with a double focusing mass spectrometer to separate the analyte ions from the overlapping polyatomic ions originating from gaseous components and/or matrix elements (Yamasaki 2000). The commonly used quadrupole type ICP-MS instruments are incapable of separating chemically different ions at the same nominal mass values and are, therefore, more prone to inaccurate results, particularly when the concentrations of the analytes become low. For the latter instrument, a special attachment (capacitive decoupling device) was installed to enhance the sensitivity.

For Group 3, an Elan DRC II ICP-MS (Perkin Elmer SCIEX, Concord, Ontario, Canada) was used. This instrument is equipped with a dynamic reaction cell (DRC) to suppress any Ar-originated interferences. However, it was not necessary to activate this device because the concentration levels of the elements susceptible to the above interferences were sufficient enough in all samples analyzed. The detection limits attained by this new type of instrument were approximately three orders of magnitude lower than those obtained using HP-4500. Therefore, it was possible to obtain acceptable results without using a high-resolution ICP-MS. The operational conditions and analytical settings for all of the instruments described above were basically those recommended by the manufacturers.

Inductively coupled plasma-atomic emission spectrometry

The instrument used for this part of the study was Vista-PRO (Varian, Palo Alto, CA, USA). Preliminary experiments showed that matrix effects were not negligible when acid digests were aspirated without dilution. Accordingly, all of the major constituents (Na₂O, MgO, Al₂O₃, K₂O, CaO, TiO₂, MnO and Fe₂O₃) and trace elements (V, Sr and Ba) were analyzed after a 10-fold dilution, although the concentration levels of Cu and Zn became too low to obtain reliable results. First, the emission intensities from each element were measured at the positions of 4–8 different prominent lines using the simultaneous multi-wavelength capability of the instrument. Then, the most suitable analytical lines for each element, shown in Table 4, were selected after taking into consideration such factors as signal intensities, background levels and the reproducibility of the check standard solutions analyzed after every 10–15 real samples.

The mixed working standards solutions were prepared to cover the concentration ranges of most of the samples, while at the same time keeping the total concentrations of each solution approximately equal. The analytical results for V, Sr and Ba by ICP-AES were within ±5% of those obtained by ICP-MS in most cases.

Atomic absorption spectrometry

Manganese, Fe, Cu and Zn were also determined by AAS using a Varian AA240FS Fast Sequential AAS (Varian Australia, Mulgrave, Victoria, Australia). All of the elements except Fe were determined by direct aspiration of the acid digests. For Fe, however, samples were analyzed after a 10-fold dilution, using the less sensitive line of 372.0 nm rather than the commonly used 248.3 nm. The values obtained by AAS were in good agreement with those determined by ICP-AES and ICP-MS (Cu and Zn).

Table 4 Analytes and analytical lines for inductively coupled plasma-atomic emission spectrometry

Analyte	Ionization state	Analytical line (nm)
Na2O	I	588.995
MgO	I	285.213
A12O3	I	396.152
K2O	I	766.491
CaO	I	422.694
TiO2	II	334.941
V	II	292.401
MnO	II	257.610
Fe2O3	II	238.204
Sr	II	407.701
Ba	II	455.403
I, neutral line; II, singly ionized line.

Wavelength dispersive X-ray fluorescence

For the samples in Group 1, SiO₂ was determined by WDXRF using an X-ray fluorescence spectrometer (model AFV 777; Tokyo Shibaura Electric Company, Tokyo, Japan). A finely powdered sample (0.500 g) and 4.5 g of Na₂B₄O₇ were fused using a high-frequency furnace at approximately 1100°C to make a glass bead_. Calibration curves were constructed using 12 rock standards provided by the US Geological Survey (AGV-1, BCR-1, G-2, GSP-1 and PCC-1 [Flanagan 1973]), the Geological Survey of Japan (JB-1 and JG-1 [Ando et al. 1974]) and the Centre de Rechereches Pétrographique et Geochimigues, France (BR, GA, GH, Mica Fe and Mica Mg [Roubault et al. 1970]). A mathematical approach was used to correct errors resulting from matrix effects. More details of these procedures have been presented elsewhere (Yamasaki et al. 1980).

Spectrophotometry

The phosphorus (P₂O₅) content of the samples in Group 1 was measured photometrically as molybdenum blue. A 5 mL aliquot of the acid digest containing less than 0.1 mg of P was transferred to a 50 mL measuring flask. After adding 20–30 mL of water, 8 mL of the mixed reagent (30 mL of ammonium molybdate solution [40 g L⁻¹], 60 mL of ascorbic acid solution [17.6 g L⁻¹], 10 mL of potassium antimony tartrate solution [2.7 g L⁻¹] and 100 mL of 2.5 mol L⁻¹ H₂SO₄) was added and finally made up to a volume of 50 mL. The absorbance of the developed blue color was measured at 880 nm (Murphy and Riley 1962).

Results and discussion

Calibration curves

Tables 5 and 6 list the correlation coefficients, the root mean square error (RMSE) and concentration ranges of the calibration curves for the major constituents and trace elements examined. The numbers of samples used for the construction of the calibration lines differed from analyte to analyte and were less than the total number of reference materials (26) because the chemical values of certain elements were not available. When the concentration level of a certain element was extremely high compared with the level of the other samples, such reference materials were also excluded from the statistical calculations.

Among the 31 examined analytes, the correlation coefficients of only eight analytes were less than 0.98. Lower values for correlation coefficients were observed in low atomic number major constituents, between Na₂O and P₂O₅. The lower correlation coefficients of these components apparently result from the lower total counts originating from the lower detection efficiency of the Ge detector at lower energy ranges (Table 7).

As the weight percentages of Na₂O and MgO in soil samples are generally lower than those of geological reference materials (Tables 5,9), it can be concluded that there appears to be little possibility of successfully using EDXRF for the determination of these two components.

The smaller value obtained for P₂O₅ can obviously be attributed to the combined effects of the lower weight percentage of this component in the reference materials used to construct the calibration lines and the lower detection efficiency, as can be observed from the values listed in Tables 5 and 7. The differences between the certified values and the calculated values using the regression equation were within the range of ±10% when the contents of P₂O₅ were more than 0.2%. This may suggest a higher possibility of using the EDXRF approach for analyzing surface soils obtained from intensive agricultural lands because, in general, such samples contain much higher amounts of P₂O₅.

Table 5 Analytical figures of merit for the major constituent calibrations

Analyte	N	Range (wt %)	R	RMSE (wt %)
Na2O	23	0.41–4.69	0.9660	0.30
MgO	22	0.01–8.42	0.9527	0.88
Al2O3	23	5.80–23.5	0.9395	1.28
SiO2	23	33.7–75.7	0.9738	2.49
P2O5	19	0.003–0.294	0.9475	0.03
K2O	23	0.06–12.9	0.9989	0.15
CaO	22	0.09–9.82	0.9996	0.12
TiO2	23	0.010–1.600	0.9914	0.06
MnO	23	0.001–0.266	0.9970	0.005
Fe2O3	23	0.06–15.1	0.9967	0.35
N, number of reference materials used for the calculation; R, correlation coefficient of the linear regression lines; RMSE, root mean square error. RMSE = (Σ (Xa − Xb)^2/(N − 2))^1/2, where Xa is the content in the reference material, Xb is the content calculated using the regression equation and N is the number of reference materials.

Table 6 Analytical figures of merit for the trace element calibrations

Analyte	N	Range (mg kg^−1)	R	RMSE (mg kg^−1)
V	25	3–635	0.9851	28
Cr	26	2.47–436	0.9929	13
Co	26	0.46–60.1	0.8940	7.2
Ni	26	1.38–139	0.9928	4.8
Cu	23	0.78–426	0.9987	5.1
Zn	24	1.4–408	0.9962	8.2
Rb	25	2.9–453	0.9985	6.2
Sr	24	8.11–442	0.9991	5.6
Y	22	2.67–51.1	0.9905	4.3
Zr	22	6.73–270	0.9977	21
Nb	17	0.7–26.9	0.9801	1.5
Sn	16	0.13–32.5	0.9966	0.7
Sb	21	0.04–38.4	0.9928	1.1
Cs	20	0.26–30.6	0.9958	0.8
Ba	26	36.5–1199	0.9970	25
La	24	0.63–54.0	0.9877	2.4
Ce	25	0.84–112	0.9944	3.2
Pr	21	0.083–14.3	0.8386	2.0
Nd	22	0.33–57.0	0 9621	3.8
Pb	23	1.5–150	0.9900	5.9
Th	16	0.19–31.4	0.9892	1.5
N, number of reference materials used for the calculation; R, correlation coefficient of the linear regression lines. RMSE, root mean square error. RMSE = (Σ (Xa − Xb)^2/(N − 2))^1/2, where Xa is the content in the reference material, Xb is the content calculated using the regression equation and N is the number of reference materials.

Table 7 Instrument stability of the major constituents†

Analyte	Content (weight %)	RSD of X-ray intensity (%)	X-ray intensity per unit wt% (CPS/wt %)
Na2O	1.05	2.18	2.71
MgO	1.74	0.51	13.5
Al2O3	16.73	0.23	33.0
SiO2	57.16	0.09	54.1
P2O5	0.208	4.66	45.7
K2O	2.81	0.07	1595
CaO	0.686	0.26	752
TiO2	0.668	0.21	1379
MnO	0.266	0.36	2061
Fe2O3	6.93	0.06	2778
^†Sample: JLk-1; number of repetitions: 24. CPS, counts per second; RSD, relative standard deviation.

Table 8 Instrument stability of the trace elements†

Analyte	Concentration (mg kg^−l)	RSD of X-ray intensity (%)
V	1254	0.265
Cr	1058	0.280
Co	1022	0.150
Ni	1002	0.237
Cu	1176	0.256
Zn	1084	0.163
As	1004	0.167
Sr	79	0.516
Y	1028	0.105
Zr	1061	0.181
Cd	104	0.301
Sn	101	0.417
Sb	99	0.363
Ba	327	0.277
W	100	0.480
Pb	1012	0.223
Bi	101	0.531
^†Sample: JSO-2; number of repetitions: 24. RSD, relative standard deviation.

In the case of Co, the overlapping of the Fe Kβ line is responsible for the lower correlation coefficient. Because the Fe content is three to four orders of magnitude higher than that of Co in most soil and environmental samples, it is common for the signal from the Co Kα line to be completely swallowed up by the huge peak in the Fe Kβ line. The use of the Co Kβ line in place of the Co Kα line was also impractical because the intensity of the Co Kβ line was strongly weakened owing to the absorption of Fe. Accordingly, it should be concluded that the EDXRF approach does not appear to be feasible for the analysis of Co in soils and related samples.

Although it was not possible to specify the causes of the lower correlation coefficients for Pr and Nd, particularly for Pr in the present study, it might be attributable to the difficulty of obtaining accurate values for these elements in the reference materials.

Instrument stability

Instrument stability was examined using two reference materials, JLk-1 for the major constituents and JSO-2 for the trace elements, under the conditions listed in Tables 2 and 3, respectively. The X-ray intensity was measured 24 times consecutively at intervals of 1 h. The X-ray intensity results for JLk-1 and JSO-2 are summarized in Tables 7 and 8, respectively.

As for the major constituents, the relative standard deviation (RSD) was less than 1%, with the exceptions of Na₂O and P₂O₅. The lower reproducibility of these two analytes can be attributed to the lower total counts, as mentioned in the section on the calibration curve.

Table 9 Results of the linear regression analysis

Analyte	Linear regression equation^†		Concentration of the samples (mg kg^−1)
	R	Slope	Range	Median
Group 1 (R > 0.98)
Sr	0.9950	0.9855	12–669	155
Ni	0.9923	1.1053	2–145	19
MnO	0.9896	1.0266	0.004–0.55^‡	0.09^‡
TiO2	0.9877	0.9828	0.12–2.46^‡	0.75^‡
Cs	0.9866	0.9798	0.5–20.2	4.4
Rb	0.9864	1.0977	7–133	51
Ba	0.9861	0.9941	37–740	383
Fe2O3	0.9841	1.0429	1.1–13.9^‡	6.3^‡
CaO	0.9839	1.0017	0.03–6.10^‡	1.6^‡
Ce	0.9819	1.0659	5–129	44
Zn	0.9813	1.0531	21–361	97
K2O	0.9811	1.0638	0.25–2.75^‡	1.51^‡
Group 2 (0.98 R > 0.92)
Al2O3	0.9787	0.9976	5.1–33.6^‡	18.2^‡
Nd	0.9757	1.0521	3.8–77.1	19.5
P2O5	0.9754	1.0583	0.09–0.48^‡	0.20^‡
SiO2	0.9751	0.9626	23.2–84.7^‡	57.7^‡
Cr	0.9734	1.4046	2–200	40
Cu	0.9723	1.0700	3–114	23
La	0.9718	1.1691	2.8–63	19.8
Th	0.9700	0.9320	0.9–21.8	5.9
Y	0.9688	1.2613	8.1–161	27
V	0.9546	1.1481	16–291	100
Nb	0.9512	1.1309	1.1–30.6	8.0
Pb	0.9248	1.0274	4.1–60.8	20.1
Group 3 (R < 0.92)
MgO	0.9134	1.3678	0.26–2.41^‡	1.05^‡
Pr	0.9016	0.9713	0.9–17.6	4.9
Sn	0.8932	2.0676	0.37–6.38	1.81
Sb	0.8130	0.9573	0.21–6.41	0.96
Na2O	0.7915	0.7556	0.02–3.53§	1.35§
Co	0.6059	1.4598	0.6–34	11.2
Zr	0.3977	1.7093	7–249	87
^†Linear regression equation was calculated in such a way as to force the equation to cross the Y-axis (energy dispersive X-ray fluorescence value) at zero when the X values (chemical results) are zero; ^‡Weight (%). R, correlation coefficient.

In the case of the trace elements, the RSDs were much lower than 1% for all of the elements measured. JSO-2 is a unique reference material in that the concentrations of a considerable number of trace elements are adjusted to be approximately 1000 or 100 mg kg⁻¹ by spiking chemicals to the original volcanic ash soil (Terashima et al. 2000, 2002). Accordingly, it can be expected that the RSDs of real world samples would not be as low as those listed in Table 8. However, as the RSDs of the non-spiked elements (Sr) were approximately 0.5%, it appears reasonable to assume that the RSDs would be better than 1.0%, even with real samples.

Comparison of the results

The values obtained by the EDXRF method were compared with those obtained by the reference methods (Table 9). The number of samples examined was approximately 70 for SiO₂ and P₂O₅, 100 for all of the other major constituents, and more than 230 for the trace elements. The soil and environmental samples were selected in such a way as to cover a wide range of chemical as well as mineralogical compositions. The best-fit lines were calculated in such a way as to force the equation to cross the Y-axis (EDXRF value) at zero when the X values (chemical results) are zero. The examined analytes were conveniently classified into three groups based on the correlation coefficients of their linear regression lines.

Out of 31 analytes examined, the number of analytes classified into Group 1 (R > 0.98) was 12, Sr was the highest among them, with the decreasing order of Ni > MnO > TiO₂ > Cs > Ba > Rb > Fe₂O₃ > CaO > Ce > Zn > K₂O. Figure 1 shows the results for Sr as a typical example of the analytes in this group. The regression lines were practically superimposed on the 1:1 lines (the lines of perfect fitness). As for the other analytes, the X-coefficients of the regression equation were also very close to unity (1.0), with the exception of Ni and Rb. Therefore, it can be concluded for the analytes of Group 1 that the results obtained in the present study compared favorably with those determined through time-consuming wet chemical pre-treatments.

The analytes belonging to Group 2 were those having correlation coefficients between 0.98 and 0.92, and were in the order of Al₂O₃ > Nd > P₂O₅ > SiO₂ > Cr > Cu > La > Th > Y > V > Nb > Pb. Good agreements were observed for Al₂O₃, SiO₂, P₂O₅ and Nd, although the correlation coefficients of the calibration lines of these analytes were lower (Tables 5,6,9). For these analytes, considerably greater scattering around the regression lines was observed, as manifested by the lower correlation coefficients. However, it can be concluded that the proposed EDXRF method is also acceptable for the analysis of these analytes because the X coefficients remained between 1.0 and 1.3, except for Cr. However, in the case of Cr, the slope was much higher than the other analytes, indicating that the values obtained by the proposed technique were significantly higher than those analyzed by the spectrophotometric methods. A similar tendency was also observed for Sn, and particularly for Zr, which were categorized as Group 3. One of the authors has already reported that the Cr and Zr values in certain samples containing considerable amounts of chromite and zircon, respectively, resulted in an underestimation when analyzed by ICP-MS (Yamasaki 1996; Yamasaki et al. 1990), presumably because of incomplete dissolution of these minerals. However, it has become evident in this study that the above underestimation is a rather universal phenomenon that is not limited to chromite-rich and zircon-rich specific samples, but is generally observed in a wide variety of soil samples. In fact, even though the total number of samples examined was more than 230, it was not possible to find a sample whose value of Zr using the proposed EDXRF method was lower than the value recorded using ICP-MS, as can be seen in Fig. 1. In the case of Cr, volatilization of Cr compound during acid digestion would be an additional factor resulting in lower values in a chemical analysis. In contrast, the lower value for Sn could be attributed to volatilization losses. The above results clearly indicate the superiority of EDXRF over chemical methods based on an acid attack. In fact, the excellent calibration curve for Zr can be attributed to the fact that the contents of Zr in most of the reference materials were obtained by XRF, although the details of the analytical procedures differed from sample to sample.

Figure 1 Energy dispersive X-ray fluorescence concentration data plotted against the values obtained by inductively coupled plasma-mass spectrometry or inductively coupled plasma-atomic emission spectrometry. The superimposed line of slope = 1.0 does not represent a line of “best fit”.

Group 3 was the worst class (R < 0.92). Except for Sn and Zr, the scatter plots were roughly within the range of RMSE. The causes of the poor results for these analytes were undoubtedly inherited from the lower correlation coefficients of the calibration lines (Tables 5,6) and/or the lower contents of these components in the samples (Tables 5,6,9). Therefore, it can be concluded that chemical methods are more preferable for these five analytes.

Conclusions

We thoroughly examined the feasibility of using polarizing EDXRF spectrometry with pressed powder pellets to establish a rapid simultaneous multi-element analysis technique for the major constituents and trace elements in soil and environmental samples. The analytes examined were 10 major constituents (NaO, MgO, Al₂O₃, SiO₂, P₂O₅, K₂O, CaO, TiO₂, MnO and Fe₂O₃) and 21 trace elements (V, Cr, Co, Ni, Cu, Zn, Rb, Sr, Y, Zr, Nb, Sn, Sb, Cs, Ba, La, Ce, Pr, Nd, Pb and Th). Twenty-six reference materials encompassing a wide range of rock, sediment and soil compositions were used for the calibrations. Compton scatter radiation was used as an internal standard to compensate for variations in the sample matrix, particle size, packing density and operating characteristics of the instrument. The correlation coefficients of the linear regression lines were >0.98, with the exception of Na₂O, MgO, Al₂O₃, SiO₂, P₂O₅, Co, Pr and Nd.

The values obtained using the proposed EDXRF method were compared with values obtained using conventional chemical methods for all of the elements except SiO₂, the value for which was determined by borate fusion-WDXRF. The number of samples examined was approximately 70 for SiO₂ and P₂O₅, 100 for all the other major constituents, and more than 230 for the trace elements. The soils and geological samples were selected in such a way as to cover a wide range of chemical as well as mineralogical compositions.

Out of the 31 analytes examined, the results for CaO, K₂O, TiO₂, MnO, Fe₂O₃, Ni, Zn, Rb, Sr, Cs, Ba and Ce obtained by the proposed EDXRF method compared favorably with those determined through conventional wet chemical methods. In contrast, the results for eight analytes (Na₂O, MgO, Cr, Co, Zr, Sn, Sb and Pr) obtained by the proposed EDXRF method exhibited poor agreements with the results obtained using the chemical methods. Among these analytes, the poor agreements of Cr, Zr and Sn were attributable to incomplete dissolution and/or volatilization losses during the chemical treatments based on an acid attack. It can be concluded, therefore, that there appears to be a high possibility of the underestimation of the concentrations of these elements if wet chemical methods are used. Although the results of the other 11 analytes were not as good as those of the first group, they still appeared to be of practical use, considering the time consuming and potentially hazardous wet chemical pretreatment.

Notes

Present address: National Institute of Livestock and Grassland Science, National Agriculture and Food Research Organization, Tochigi 329-2793, Japan