Professional Documents
Culture Documents
See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12
APPLICATION OF XRF TO
THE ANALYSIS OF
ENVIRONMENTAL SAMPLES
John N. Driscoll
PID Analyzers, LLC
Walpole, Massachusetts
12.1 Introduction
12.2 Basics of XRF Spectroscopy
12.2.1 Detection Geometry
12.2.2 X-Ray Sources
12.2.3 Detectors
12.3 Sample Preparation
12.3.1 Sampling
12.3.2 Soils
12.3.3 Liquid Samples
12.3.4 Filter Paper
12.4 XRF Equipment and Operational Considerations
12.4.1 XRF Equipment
12.4.2 Operational Considerations for Tube-Excited XRF
12.4.3 Minimum Detectable Concentration
12.5 Errors in EDXRF
12.5.1 Total Error
Environmental Instrumentation and Analysis Handbook, by Randy D. Down and Jay H. Lehr
ISBN 0-471-46354-X Copyright # 2005 John Wiley & Sons, Inc.
251
10.1002/0471473332.ch12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/0471473332.ch12 by Universidade De La Coruña, Wiley Online Library on [19/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
252 APPLICATION OF XRF TO THE ANALYSIS OF ENVIRONMENTAL SAMPLES
12.1 INTRODUCTION
Table 12.1 Comparison of EDXRF with Other Spectroscopic Methods for Metals
Another advantage of XRF is that this technique is ideally suited for field screening
work4 as a result of the minimization of sample preparation and the portability.
With an X-ray tube and filters, XRF can be used for a variety of environmental ana-
lyses, including particulate matter in air, as well as metals in soil, and water. We
will be discussing some environmental applications in the following sections.
resulting characteristic X-ray emission. Since we are dealing with atoms, absorption
and emission lines are at the same wavelength or frequency. A typical energy level
diagram for an atomic species is shown in Figure 12.1 and a typical X-ray spectrum
is shown in Figure 12.2.
sample
105°
x-ray x-ray
source detector
be used to compensate for the X-ray continuum and improve detection limits. Tube
lifetimes are generally in excess of 5000 h and low-power tubes (10 W or less) can
be quite stable. The clear disadvantage of the tube is its power consumption, which
makes the development of a battery-operated instrument more challenging. It
is possible to operate an instrument like the XR1000 off an inverter (converts
12 V DC to 120 V AC). Thus a 12-V battery can be the source of power for this
instrument. A second choice is a gasoline-powered generator.
12.2.3 Detectors
Two types of detectors have been used in XRF instrumentation: proportional
counters, which are the more traditional, and cooled [by liquid nitrogen or electro-
nically (Peltier)] high-resolution semiconductor detectors, which have only been
used in the field very recently. The results obtained with the two types of detectors
can be quite different, as will be shown below.
The relative standard deviation (sQ ) of the pulse amplitude distribution for a
detector is given by
sQ C 1=2
¼
Q E
If the standard deviation is multiplied by 2.35, the energy resolution (R) of the
proportional counter can be calculated. At 10 keV, the energy resolution of a pro-
portional counter is 12.5% or 1250 eV.6 The resolution is given by
FWHM
Energy resolution ðRÞ ¼
H0
The Si(Li) detectors have been used since the 1960s for detecting X-rays and
low-energy gammas (<100 keV). These semiconductor detectors are useful in
that dimensions can be kept smaller than proportional counters because solid den-
sities are 1000 times greater than for a gas. As ionizing radiation passes through the
semiconductor, many electron–hole pairs are produced along the track. The low
value of the ionizing energy required (3 eV) for a semiconductor detector (about
1
10 that of a proportional counter) creates 10 times the number of charge carriers
than a proportional counter and is one of the factors that leads to increased resolu-
tion for the semiconductor detectors. The resolution for a semiconductor detector
is about 1.5% at 10 keV or 150 eV. The effect of detector resolution is shown in
Figure 12.4, where the energy spectra of an 55Fe source are compared for a
Si(Li) detector (resolution 185 eV) and a proportional counter (resolution 1200 eV).
10.1002/0471473332.ch12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/0471473332.ch12 by Universidade De La Coruña, Wiley Online Library on [19/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
258 APPLICATION OF XRF TO THE ANALYSIS OF ENVIRONMENTAL SAMPLES
Figure 12.6 Comparison of a silicon Pin detector and a proportional counter spectrum for lead
paint.
For these tests, only the detector was changed and all the electronics, including
the MCA, were the same. The actual resolution of the proportional counter was
1250 eV, in agreement with the value calculated above. The Mn Ka and Kb peaks
at 5.9 and 6.4 keV are well resolved with the Si(Li) detector, which had a resolution
of 180 eV, also in close agreement with the predicted value. It is clear from the
spectra in Figure 12.6 that with the poor resolution (>1000 eV) of a proportional
counter, adjacent element peaks are difficult to detect. Typically adsorption edge
filters are used to compensate for this problem. However, these filters reduce the
count rate and greatly affect the precision and accuracy of the results. A compar-
ison of the detector characteristics is given in Table 12.3.
Clearly, the accuracy depends on the resolution of the detector used and the
presence or absence of neighboring elements which may interfere. For the lead
paint sample, the peaks of lead and a neighboring element such as zinc are sharply
resolved for the solid-state detector. Note the significant quantity of zinc in this
paint sample. The spectrum for the XRF instrument with the proportional counter
is shown in Figure 12.6. The high concentration of zinc and its overlap with the lead
La peak make it difficult to quantify the lead peak.
The Si(Li) detector will provide the highest resolution, count rate, and efficiency
of the three detectors. The disadvantage of this detector is that it requires liquid
nitrogen to operate and 30–40 min to cool down. The Si Pin detector operates
over the whole range of elements Na–U, has an acceptable resolution of 230 eV,
and only requires about 5–10 min to cool down. The count rate and efficiency
are lower than the Si(Li) detector. The proportional counter is limited to only the
simplest applications where elements are not close together; otherwise significant
corrections and/or filters are required. If the user wants to cover a significant range,
more than one (two or three) proportional detectors (filters) would be required. The
count rate is considerably lower than the solid-state detectors, as explained earlier
in this chapter.
12.3.1 Sampling
One of the first considerations in the field is to ensure that the samples taken are
representative of the site. This is particularly true for soil samples that are typically
nonhomogeneous.
Sometimes multiple levels of contaminants are present at the site, and it is not
unusual to find data such as that shown in Figure 12.7, which profiles the arsenic
12.3.2 Soils
This is a key element for accurate XRF analysis. Samples to be analyzed must be
similar to standards (e.g., particle size, composition). In general, samples may have
to be sieved to a uniform particle size (sandy soils), ground with a mortar and pestle
(soils with a high clay content, sludge) to a uniform particle size, placed in a ball
mill or shatter box (for geological sample such as rocks), and so on. Soil samples
are analyzed directly after preparation by placing in a special low-volume polyethy-
lene sample cup fitted with a 0.15 film Mylar window and counting for 300 s. It is
essential that the standard(s) and the sample have the same surface characteristics
and similar composition. Particle size, compactness, and state (liquid, solid, etc.)
can affect the intensity of XRF.
Soils contain trace heavy metals, with the most common elements being Pb, As,
Hg, Fe, Cu, Zn, V, Ti, Cr, Mn, Cd, and Ni. The reasons for sieving the sandy soil
samples noted above are twofold: The majority of the metals are concentrated in the
fine particles and large particles can cause scattering.
Metals in groundwater, lakes, and ponds should only contain background levels
of heavy metals that are generally low- or sub-ppb levels. These are usually ana-
lyzed by ICP or AA down to < 1 ppb, but XRF can generally go down to
< 1 ppm. What can be done about this? If we take 1 L of water and pour it through
1 g of ion exchange resin, then dry and analyze that resin, we can detect 1 ppb of
metals in water. A spectrum at low-ppb levels is shown in Figure 12.8. This process
takes only about 5 min and can be easily done in the field.
Figure 12.9 XRF Spectrum of metal particulates (on filter paper) in ambient air.
detector and preamplifier, the amplifier, the multichannel analyzer (MCA), and the
microprocessor or computer.
where e is the counting efficiency (i.e., counts per second per unit concentration),
as determined from a single measurement or the slope of the calibration curve, and
B is an estimate of the background interference error based on the background
counts or the blank counts obtained from the zero intercept of the calibration
curve.
10.1002/0471473332.ch12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/0471473332.ch12 by Universidade De La Coruña, Wiley Online Library on [19/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
264 APPLICATION OF XRF TO THE ANALYSIS OF ENVIRONMENTAL SAMPLES
The most common errors encountered can have two very different meanings which
are frequently confused with each other:
1. Accuracy—the difference between the measured value and the true value.
2. Precision—the degree of agreement among replicate determinations made
under identical conditions.
The statistical counting error can be minimized by counting for the longest
interval consistent with the requirements of the total analysis problem. However,
the error cannot be reduced by dividing this interval into segments and taking
the mean of the individual measurements. For example, the standard deviation of the
mean of eight 100-s measurements of the above sample would be
35 counts (100/8).
If the same sample had been counted continuously for 800 s, 8000 counts would
have been accumulated with a standard deviation of
285 counts (80,000). The per-
cent error, however, is identical in both cases (0.35%). The standard deviation of the
mean value can be improved by making more measurements, and the standard
deviation of a measurement can be improved by counting for a longer time.
The magnitude of the other types of random errors (instrumental and procedural)
must be evaluated experimentally by making a sufficient number of replicate mea-
surements to define their frequency distribution. One of the best methods of eval-
uating the total random error is to prepare and analyze a large number of replicate
samples of the same material. This will provide an accurate estimate of the preci-
sion of the total procedure but will not necessarily be an accurate estimate of the
instrumental precision because it will include sampling and sample preparation
errors, which may be significant. The instrumental precision or reproducibility
may be determined by calculating the standard deviation of a series of measure-
ments of the same sample as a function of time or sample position using long count-
ing intervals to minimize the counting error contribution. The procedural errors can
be estimated by subtracting the counting and instrumental errors from the total
errors.
Difference ¼ 0.253
Ka—1:487 KeV
Ka—1:740 KeV
Ka—6:93 KeV
Energy (KeV)
Ka—6:4 KeV
Figure 12.10 HNU model XR1000 table-top XRF.
Element
Co
Fe
Al
Si
Metals and Light Elements
Atomic Number
26
27
13
14
10.1002/0471473332.ch12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/0471473332.ch12 by Universidade De La Coruña, Wiley Online Library on [19/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
268 APPLICATION OF XRF TO THE ANALYSIS OF ENVIRONMENTAL SAMPLES
S1
0.28
MgO
Concentration [%]
0
0 Mg Ka Intensity [cpa] 25.41
There are two different software programs that can be used: empirical and FPT.
The empirical method requires the use of multiple standards (minimum of three
plus one for each element used for correction) to bracket the unknowns. This soft-
ware uses a linear regression fit with a Simplex variation to minimize the error in x
and y coordinates. Standards (use the AA or ICP data as the actual concentration)
can be prepared by using the AA or ICP data from the same site. Note: If this pro-
cedure is used, a representative aliquot of the soil samples should be available. Now
the soil data by XRF should correlate with the AA or ICP data for the site. Some
S1
0.28
MgO
Concentration [%]
0
0 37.11
Mg Ka Intensity [cps]
site data in Figure 12.13 compares AA with XRF data. The correlation coefficient is
0.995. The empirical method is useful from ppm to percentage levels, but multiple
calibration curves would be required to cover this range. An accuracy of 98þ%
may be obtained, but this depends upon the accuracy of the standards, the count
rate, matrix effects, and other factors.
The importance of interelement corrections in environmental analysis can be seen
with the analysis of lead and arsenic. The Ka line of arsenic and the La line of lead
are both at 10.5 keV. If both are present, there are several options: (a) use interele-
ment corrections (will need standards with known arsenic and lead concentrations)
or (b) use the Kb line of arsenic and the Lb line of lead. The lower limit of detection
will not be as low with the b lines, but the interference will be eliminated.
The FPT standardized program can be used for high ppm to 100%. All of the
elements in the sample must be known and must add up to 100%. In this program,
corrections are made for detector efficiency, tube output, other elements (absorption
and enhancement), and so on. At least one standard is needed. Typical accuracy
ranges from 90 to 95%.
Driscoll and Johnson9 have shown the neural networks can be successfully
applied to qualitative and quantitative identification of elements such as lead (Pb)
and arsenic (As). the Pb La and As ka X-ray spectra overlap at 10.6 keV. Using a
single neuron and 15 inputs per element, they were able to predict the shape of the
As and Pb spectra (qualitative). For quantitative analysis 45 inputs were required
with three processing neurons to determine As and Pb concentrations in complex
soil mixtures. This technique has significant promise for improving algorithms used
in spectroscopic software.
In this chapter we have described a number of basic aspects of XRF in order to
provide the reader with a basic understanding of both XRF theory and sample pre-
paration. Should the reader be interested in learning more about XRF applications
or detectors and electronics, Ref. 1 and 6 are recommended.
10.1002/0471473332.ch12, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/0471473332.ch12 by Universidade De La Coruña, Wiley Online Library on [19/01/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
270 APPLICATION OF XRF TO THE ANALYSIS OF ENVIRONMENTAL SAMPLES
REFERENCES
1. V. Valkovic, X-Ray Spectroscopy in the Environmental Sciences, CRC Press, Boca Raton,
FL, 1989.
2. Anonymous, Data Quality Objectives for Remedial Response Activities, RI/FS
Activities at a Site with Contaminated Soils & Ground Water, PB 88-131388, EPA,
Washington, DC, March 1987.
3. Anonymous, Data Quality Objectives for Remedial Response Activities, Development
Processes, PB 88-131370, EPA, Washington, DC, March 1987.
4. R. K. Stevens and W. F. Herget, Analytical Methods Applied to Air Pollution, Ann Arbor
Science Publishers, Ann Arbor, MI, 1974.
5. J. N. Driscoll, J. K. Marshall, E. S. Atwood, C. Wood, and T. Spittler, A New Instrument for
the Determination of Heavy Metals & Radioactivity at Mixed Waste Sites, Am. Lab., July
1991, p. 131.
6. G. F. Knoll, Radiation Detection and Measurement, Wiley, New York, 1989.
7. EPA, Superfund Innovative Technology Evaluation: Characterization and Monitoring
Program, http://www.clu-in.com/site/charmon.htm.
8. J. N. Driscoll, Field Analysis of Metals in the Environment by XRF, Int. Env. Tech.,
September/October 1998.
9. J. N. Driscoll and W. Johnson, Use of Neural Networks in XRF, Pittsburgh on Anal.
Chem. & Applied Spectroscopy, Paper 1968P New Orleans, LA, March 2002.