Science and Justice 49 (2009) 170–181

Contents lists available at ScienceDirect

Science and Justice

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / s c i j u s

Development of a searchable major and trace element database for use in forensic
soil comparisons
Kenneth Pye ⁎, Simon J. Blott
Kenneth Pye Associates Ltd, Crowthorne Enterprise Centre, Crowthorne Business Estate, Old Wokingham Road, Crowthorne, Berkshire RG45 6AW, UK

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

Article history: Forensic soil comparisons are normally undertaken on the basis of several physical, chemical and biological
Received 20 November 2007 properties, but in all cases the interpretation of results is dependent on the availability of relevant contextual
Received in revised form 6 February 2009 information. This paper summarises the results of major and trace element analyses performed using
Accepted 28 February 2009
inductively coupled plasma atomic emission spectrometry and inductively coupled plasma-mass spectro-
metry on the b 150 µm size fraction of 1896 soil samples collected in connection with forensic investigations
Keywords:
Soil
in England and Wales between 1999 and 2007. A number of new methods are described which facilitate
Major element composition inter-sample comparison. Although the available data do not provide uniform geographical coverage they do
Trace element composition provide useful information which can assist police search investigations and they provide valuable contextual
ICP-AES information which aids the evidential assessment of soil evidence when used in court. There is considerable
ICP-MS scope to improve the database by increasing the geographical coverage and increasing the number of soil
attributes which are included.
© 2009 Forensic Science Society. Published by Elsevier Ireland Ltd. All rights reserved.

1. Introduction use in forensic investigations. For example, a number of soil and

The forensic investigation of soil and other forms of physical
trace evidence involves two stages: (1) comparison of a questioned
sample with one or more samples of known origin, and (2)
evaluation of the significance of observed similarities and differ-
ences in order to arrive at a conclusion regarding a possible
association. The latter depends heavily on the availability of
background information relating to the frequency of occurrence
and geographical distribution of samples which show similar levels
of similarity or difference to those being compared. Comparison of
soils and sediment samples normally involves consideration of
several different properties, including particle size distribution,
colour, particle typology, mineralogy, chemical composition, pollen
assemblages and other microfossils [1–3]. Many techniques have
been developed over the past 20 years which allow these properties
to be quantified, in many cases with good accuracy and precision.
However, there are currently no comprehensive databases to assist
interpretation of the data obtained. Useful contextual information at
the national and regional scale is available in the form of maps,
atlases and descriptive compendia of soil properties, but the
supporting data are often presented in a form which limits their
⁎ Corresponding author.
E-mail address: k.pye@kpal.co.uk (K. Pye).
sediment sampling programmes have been carried out in England
and Wales over the past 40 years, principally for purposes of
geochemical mapping, mineral prospecting and land-use assess-
ment purposes [4–10], and similar exercises have been undertaken
in continental Europe [11], but the data mostly refer to composite
samples representing relatively large areas (1 km2 or more). To date
there have been few published attempts to develop databases of soil
and sediment properties specifically with forensic applications in
mind (e.g. [12]).
All databases relating to physical evidence are subject to limita-
tions imposed by the number of samples they contain, the degree to
which those samples are representative of the populations from
which they are drawn, the type and quality of the information
included, the discriminatory capacity of the criteria which can be
applied to the data, and the fact that some types of information may be
rendered out of date very quickly. Nonetheless, databases can provide
useful sources of contextual information which aids interpretation of
potential evidential significance [13].
This paper summarises an ongoing attempt to develop a database
containing information about the major and trace element composi-
tions of surface and near-surface soil samples in England and Wales,
and describes a number of methods which have been developed to
compare data for questioned samples with new control samples and
existing samples in the database. Information about soil properties
other than chemical composition is also recorded, and in casework
investigations comparisons are almost always made on the basis of
several criteria. For purposes of illustration in the present paper,
reference is made only to major and trace element compositional

1355-0306/$ – see front matter © 2009 Forensic Science Society. Published by Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.scijus.2009.02.007
 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181
data acquired from the b150 µm size fraction of 1896 soil samples
collected from known locations in England and Wales between 1999
and 2007. Results for samples collected since that time, or which
relate to different size fractions, locations outside England and
Wales, and to soil recovered from suspect vehicles, footwear,
implements etc., are not considered here.
2. Materials and methods
The samples referred to in this paper were collected from
numerous different locations in England, and some in Wales (Fig. 1).
The sampling locations are not uniformly distributed and were not
selected according to a pre-defined sampling framework; rather,
they reflect locations of interest in casework investigations during
the period 1999–2007. In some of the areas shown, relatively large
numbers of samples (N50) were collected within an area of a few
square kilometres, while in other areas the number of collected
samples was much smaller, or samples were collected across a wider
area (10–100 km2). As such, the dataset is not considered to be
statistically representative of the land area of England and Wales as
whole. Rather, the data should be considered in a similar way to the
‘survey’ information often used in connection with other forms of
physical trace evidence, such as the frequencies with which
particular fibre types have been recorded during laboratory
examination of submitted clothing items.
The great majority of the samples analysed were taken from
the ground surface (0–5 cm depth), although approximately 3%
were collected from shallow pits and graves (b50 cm depth). Most
were collected using a stainless steel trowel or similar instrument
which was thoroughly cleaned with distilled deionised water
before and after each sample was taken. In the laboratory, each
Fig. 1. Geographical distribution of the 1896 soil samples from England and Wales
contained in the dataset. Note that the locations shown include groups of several samples.
171
sample was homogenized by mixing prior to sub-sampling for a
range of analysis types. The b150 µm size fraction was separated
from one of the sub-samples in each case by wet-sieving though
disposable nylon mesh using distilled de-ionized water [14,15].
The b150 µm size fraction was selected partly because it has been
used extensively in previous geochemical investigations and partly
because tests have demonstrated that this size fraction is
practically efficient to obtain, normally yields sufficient material
for analysis from small forensic samples, and frequently provides a
good ‘proxy’ for the bulk elemental composition of sediments and
soils [16].
The concentrations of up to 10 major and 49 trace elements were
determined using inductively coupled plasma atomic emission
spectrometry (ICP-AES) and inductively coupled plasma-mass
spectrometry (ICP-MS) at two laboratories: the Department of
Geology, Royal Holloway, University of London (RHUL), and the
School of Earth Sciences, University of Greenwich (UoG). Some
samples were analysed at both laboratories, although most were
analysed at only one. Approximately a third of the samples were
analyzed only by ICP-AES, but the majority were analyzed using both
ICP-AES and ICP-MS to provide data for a wider range of elements.
The RHUL data were generated using a Perkin-Elmer Optima 3300
RL ICP-AES instrument and a Perkin-Elmer Sciex Elan quadropole
ICP-MS instrument. The UoG data were generated using a TGA Iris
Advantage instrument (ICP-AES) and a Thermo-Elemental X7
instrument with an HPI interface (ICP-MS). The sample preparation,
data processing and quality control procedures used in the two
laboratories, which include regular analysis of international stan-
dard materials and laboratory reference materials, have been
described in previous publications [15,17,18]. Owing to improve-
ments in analytical methods over the period of study, the number of
trace elements determined has increased over time. As a result, the
number of samples for which elemental data are available ranges
from 337 (thulium) to 1896 (all of the major elements except
silicon). The concentration s of major elements are reported in terms
of weight percent oxide (wt.%) and the concentrations of trace
elements in µg g− 1.
The compositional data were collated in Microsoft® Excel and
analysed using Microsoft XLSTAT®. The data generated by each
laboratory can be interrogated separately or in combination, using
major elements only, trace elements only, elemental ratios only or
various combinations of these. For the purposes this paper, only a
limited number of comparative examples are presented, using data for
both laboratories and a combination of major and trace element
concentrations and ratios.
3. Results
3.1. Inter-laboratory comparison
Table 1 summarises the results for 10 samples which were
analysed at both laboratories (different sub-samples of the same
ground powder, prepared as separate solutions at each labora-
tory). The results show relatively small differences between the
two laboratories in terms of the mean concentrations of the ten
major elements and most of the trace elements. Only three
elements (Ta, Pb and Co) showed relative differences in the mean
concentrations recorded by the laboratories larger than 20%, and
eight others (Ti, P, Cs, Hf, U, Th, Ho, Er) showed relative
differences larger than 10%. The inter-laboratory differences in
major oxide concentrations and elemental ratios derived from the
primary data were also small, with only the U/Th ratio showing a
relative difference larger than 10% (Table 1). This level of variation
is no greater than that found during analysis of different sub-
samples taken from the same parent sample at a single laboratory
over time [15]. It is therefore considered justified to combine the
172 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181

Table 1
Average major and trace element concentrations of 10 soil samples determined by ICP-AES and ICP-MS at two laboratories (Lab A = RHUL, Lab B = UoG).

(n = 10) Lab A Lab B Percentage difference between labs

Mean concentration Standard deviation Mean concentration Standard deviation Average Maximum Minimum
Major oxides (wt %)
SiO2 50.6 6.1 51.6 6.3 3.0 4.4 2.0
Al2O3 9.8 2.1 10.2 1.9 5.5 14.8 0.5
Fe2O3 5.5 1.0 5.5 0.9 4.7 12.4 0.3
MgO 0.9 0.1 0.9 0.1 4.5 12.6 0.3
CaO 8.6 3.1 8.7 3.2 2.1 5.6 0.2
Na2O 0.2 0.0 0.3 0.0 9.0 17.6 5.5
K2O 1.6 0.2 1.7 0.2 7.8 18.3 0.4
TiO2 0.5 0.1 0.5 0.1 17.2 25.3 6.9
P2O5 0.6 0.1 0.5 0.1 17.0 20.0 14.2
MnO 0.1 0.0 0.1 0.0 5.8 16.5 1.3

Trace elements (µg g− 1)

Ba 289.9 28.6 313.5 25.5 7.6 14.6 2.4
Co 12.7 1.8 10.0 1.2 21.0 28.5 12.8
Cr 90.2 8.9 82.4 12.3 8.8 29.6 0.2
Cu 52.5 9.8 51.4 10.5 7.3 15.4 1.6
Ni 36.2 10.0 39.4 9.3 8.7 18.5 1.4
Pb 64.1 23.3 76.4 18.6 28.6 47.3 0.9
Sc 9.8 2.1 9.6 1.8 4.3 11.8 0.9
Sr 154.6 35.2 153.8 31.2 2.7 9.2 0.1
V 89.0 20.2 83.3 17.4 6.5 15.2 2.4
Zn 225.0 32.2 226.5 31.4 2.5 6.7 0.3
Zr 239.8 45.3 248.3 52.4 5.9 23.3 0.0
Rb 93.5 20.0 102.4 16.5 9.5 16.6 1.2
Nb 13.9 2.5 13.2 2.0 5.7 10.0 1.4
Cs 6.4 1.3 7.4 1.2 14.1 22.0 5.4
Hf 6.2 1.0 6.2 1.0 10.7 22.4 4.2
Ta 0.5 0.1 0.9 0.1 51.3 64.5 36.4
Y 23.5 3.5 25.0 2.5 8.8 15.7 1.6
U 1.9 0.2 1.6 0.1 11.3 23.5 1.6
Th 7.6 1.3 8.6 1.1 12.7 24.3 0.0
La 29.0 4.6 27.4 2.6 8.7 22.3 1.6
Ce 57.9 9.2 57.4 7.0 5.4 11.7 0.7
Pr 7.2 1.1 6.6 0.8 8.1 17.3 0.5
Nd 28.2 4.4 27.9 3.2 5.2 9.9 0.0
Sm 5.4 0.8 5.2 0.6 6.1 14.4 1.2
Eu 1.1 0.2 1.0 0.1 8.7 15.5 0.5
Gd 4.8 0.7 4.3 0.4 9.0 22.0 0.2
Dy 3.8 0.6 3.7 0.4 5.7 11.7 1.5
Ho 0.7 0.1 0.8 0.1 13.1 21.7 0.1
Er 2.3 0.3 2.1 0.2 10.6 21.5 2.7
Yb 2.0 0.2 2.0 0.2 3.4 9.6 0.1
Lu 0.3 0.0 0.3 0.0 7.4 17.1 0.0

Element and oxide ratios

SiO2/Al2O3 5.4 1.4 5.3 1.3 5.6 14.1 0.2
SiO2/Fe2O3 9.6 2.2 9.6 2.0 5.8 12.7 1.1
SiO2/CaO 6.6 2.5 6.6 2.4 2.3 8.6 0.1
Al2O3/Fe2O3 1.8 0.1 1.8 0.1 3.0 4.6 2.2
Al2O3/K2O 6.1 0.5 6.0 0.6 6.1 17.4 3.3
CaO/MgO 9.7 3.1 9.9 3.5 3.8 14.8 0.0
K2O/Na2O 6.5 1.4 6.4 1.4 3.5 18.6 0.2
Rb/Sr 0.6 0.2 0.7 0.2 9.4 17.6 0.6
Ce/La 2.0 0.0 2.1 0.1 6.5 12.0 2.1
U/Th 0.3 0.0 0.2 0.0 23.1 27.7 14.5
Nd/Sm 5.3 0.1 5.3 0.3 4.1 9.8 0.0

data from the two laboratories, and for the purposes of this paper
the data have been divided into two sub-sets, one (A) containing
1874 samples with concentration data for 27 elements determined
by ICP-AES, and a second (B) containing 989 samples with
concentration data for 40 elements determined using both ICP-
AES and ICP-MS.
3.2. Overall concentration ranges and means
A summary of the maximum, minimum, mean, standard
deviation and percentage coefficient of variation (%CV) for each
element is provided in Table 2. Silica (SiO2) is by far the most
abundant element, followed by Al2O3, Fe2O3 and CaO. MnO and CaO
show the greatest variability (expressed by the %CV). Amongst the
trace elements, Zr, Sr and Pb have the highest average concentra-
tions, whilst Pb, Cr, Mo, Cu and Sn show the greatest variability
(highest %CV).
The concentration values of Al2O3, Fe2O3 and K2O approximate a
near-normal distribution (S-shaped curve) when plotted as cumu-
lative frequency curves, while SiO2, TiO2, Na2O, P2O5, MnO, MgO and
CaO show progressively greater departure from normality (Fig. 2).
Some trace elements, such as Ce, Rb, U and Th, also show a near-
 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181 173

Table 2
Summary of major and trace element concentrations in up to 1896 soil samples from England and Wales.

Element Sample n Mean Max Min Std dev CV % Type 5 Precision (CV %) Sub-set A (n = 1874) Sub-set B (n = 989)
ICP-AES elements
SiO2 1011 59.35 96.11 2.56 16.00 26.9 2.23 ✓
Al2O3 1895 9.16 29.98 0.33 3.50 38.2 2.02 ✓ ✓
Fe2O3 1895 4.63 40.72 0.16 2.14 46.2 2.47 ✓ ✓
MgO 1895 0.91 9.24 0.01 0.70 77.0 16.29 ✓ ✓
CaO 1891 4.42 54.04 bLLQ 6.80 153.9 5.47 ✓ ✓
Na2O 1894 0.48 4.31 bLLQ 0.32 65.9 4.36 ✓ ✓
K2O 1895 1.68 4.58 0.02 0.63 37.7 3.22 ✓ ✓
TiO2 1895 0.56 2.14 0.02 0.20 36.6 3.93 ✓ ✓
P2O5 1895 0.34 2.44 0.01 0.24 69.8 9.80 ✓ ✓
MnO 1890 0.10 6.01 bLLQ 0.19 180.4 2.82 ✓ ✓
Be 399 1.75 4.46 bLLQ 0.80 45.7 4.34
Sc 1874 9.25 50.00 bLLQ 4.14 44.8 22.64 ✓ ✓
Ni 1860 37.25 381.00 bLLQ 23.82 63.9 23.09 ✓ ✓
Zn 1870 226.80 4040.00 11.00 308.85 136.2 5.50 ✓ ✓
Y 1887 24.97 172.00 bLLQ 11.49 46.0 2.91 ✓ ✓
Ba 1895 411.31 4049.00 bLLQ 276.06 67.1 8.54 ✓ ✓

ICP-MS elements
V 1875 82.70 744.00 bLLQ 40.14 48.5 22.46 ✓ ✓
Cr 1875 86.89 9492.00 1.00 298.62 343.7 10.08 ✓ ✓
Co 1872 11.53 52.00 bLLQ 5.70 49.4 12.64 ✓ ✓
Cu 1860 65.15 3002.00 bLLQ 107.84 165.5 14.93 ✓ ✓
Ga 400 11.45 27.59 1.35 5.27 46.0 4.63
Rb 1003 78.18 704.61 1.05 38.29 49.0 3.06 ✓
Sr 1895 122.83 985.50 7.00 88.22 71.8 14.13 ✓ ✓
Zr 1896 216.50 1188.43 1.42 153.14 70.7 10.00 ✓ ✓
Nb 1022 12.69 144.09 0.23 6.94 54.7 5.44 ✓
Mo 400 2.51 80.73 bLLQ 4.97 198.1 21.74
Sn 400 7.37 205.67 bLLQ 12.08 164.0 13.26
Cs 1003 4.98 89.09 0.11 4.74 95.1 7.72 ✓
La 1887 30.68 142.00 1.00 11.38 37.1 4.59 ✓ ✓
Ce 1887 60.50 378.00 5.00 22.25 36.8 4.67 ✓ ✓
Pr 1003 6.86 32.06 0.70 2.91 42.4 3.65 ✓
Nd 1887 29.45 204.26 0.84 12.68 43.0 3.47 ✓ ✓
Sm 1887 5.31 31.84 0.09 2.30 43.3 2.96 ✓ ✓
Eu 1886 1.09 6.40 0.13 0.52 47.6 3.23 ✓ ✓
Gd 1003 4.68 22.90 0.66 1.94 41.5 4.16 ✓
Tb 400 0.79 2.99 0.10 0.33 41.7 2.84
Dy 1887 3.91 28.50 0.31 1.72 44.1 3.30 ✓ ✓
Ho 1003 0.93 3.60 0.12 0.35 36.9 9.09 ✓
Er 1003 2.46 9.08 0.32 0.91 36.8 2.74 ✓
Tm 337 0.41 1.23 0.05 0.16 38.1 25.94
Yb 1887 2.24 14.10 bLLQ 0.88 39.3 3.81 ✓ ✓
Lu 1001 0.39 1.10 0.05 0.13 32.3 3.17 ✓
Hf 1003 7.98 30.39 bLLQ 3.57 44.7 6.57 ✓
Ta 1002 0.95 23.33 bLLQ 0.97 102.6 13.44 ✓
W 400 2.16 26.26 bLLQ 2.50 115.8 136.61
Tl 640 0.43 2.36 bLLQ 0.40 93.0 37.53
Pb 1269 184.30 13,533.26 bLLQ 734.86 398.7 23.57
Th 1003 8.17 26.83 0.69 2.88 35.2 15.86 ✓
U 1003 2.12 29.15 0.17 1.11 52.5 7.36 ✓
Li 1337 51.35 365.00 bLLQ 28.98 56.4 nd

Element and oxide ratios

SiO2/Al2O3 1010 9.00 285.45 1.88 10.96 121.8 1.74
SiO2/Fe2O3 1010 19.22 336.36 0.31 27.67 143.9 2.64
SiO2/CaO 1005 109.63 9611.00 0.05 478.60 436.5 5.64
Al2O3/Fe2O3 1895 2.14 31.89 0.13 1.07 50.2 2.84
Al2O3/K2O 1895 5.84 161.50 1.82 5.51 94.3 2.81
CaO/MgO 1890 6.30 675.71 0.01 20.67 328.1 15.18
K2O/Na2O 1892 4.28 26.50 0.08 2.30 53.8 3.88
Rb/Sr 1002 0.80 4.43 0.01 0.46 57.8 13.08
Ce/La 1887 2.00 9.00 0.38 0.34 17.2 3.68
U/Th 1003 0.28 5.95 0.12 0.20 72.5 12.26
Nd/Sm 1887 5.86 129.17 1.67 4.00 68.3 1.98

Values for the 10 major elements are expressed at wt.% oxide and values for trace elements are expressed in µg/g− 1. b LLQ indicates values below the lower limit of quantitation. Type
5 precision values from [15]. Ticks indicate which elements are included in data sub-set A (27 elements in common) and which are included in sub-set B (40 elements in common).

normal distribution, but the majority (especially Cu, Pb, Zn and Sr) CaO and CaO/MgO, are also strongly skewed, but others, such as
have markedly skewed distributions (Fig. 3). Some of the major Al2O3/Fe2O3, Al2O3/K2O and Ce/La, are near-normally distributed
oxide and elemental ratios, including SiO2/Al2O3, SiO2/Fe2O3, SiO2/ (Fig. 4). These distributions obviously reflect the available dataset,
174 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181

Fig. 2. Cumulative frequency plots of major oxide concentrations recorded in the dataset. The number of samples for each oxide is given in Table 2.

and may not be representative of areas which have not been uranium (U) higher than 4 µg/g have been encountered relatively
sampled. However, a number of potentially useful inferences can be rarely (b1% of observations, Fig. 3), as have Al2O3/Fe2O3 ratios lower
drawn; for example, based on the data available, concentrations of than 1.0 (b3% of observations, Fig. 4).
 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181 175

Fig. 3. Cumulative frequency plots of selected trace element concentrations recorded in the dataset. The number of samples for each element is shown in Table 2.
176 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181

Fig. 4. Cumulative frequency plots of selected oxide and elemental ratios for the soil samples recorded in the dataset. The number of samples for each element is given in Table 2.
 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181 177

3.3. Inter-sample comparison Table 4

Comparison of 989 samples in sub-set B, showing the number of samples considered to
be analytically indistinguishable and distinguishable in terms of different numbers of
An important question concerns the degree to which any sample of elements.
interest is similar to, or different from, other samples of investigative
interest, or to other samples in the background dataset. Several Number Number of samples Number of samples Number of samples
of which are which are which can be
graphical and simple statistical methods are routinely used in the
elements indistinguishable from indistinguishable from distinguished from at
initial comparison of samples, a process which is frequently referred at least one other at least one other least one other sample
to as exploratory data analysis [1,19,20]. However, three new methods sample in terms of sample in terms of this in terms of this number
are described below which have been developed in our laboratory exactly this number of number of elements or of elements or more
elements more
specifically to facilitate comparison of soil samples on the basis of
elemental composition (although they are equally applicable other 40 14 14 975
39 15 29 960
criteria). 38 27 56 933
37 29 85 904
3.3.1. Method 1 36 29 114 875
Each sample can be compared with every other sample in the 35 34 148 841
34 35 183 806
dataset in terms of the concentrations of each element, allowing a pre-
33 43 226 763
specified degree of variation which takes into account differences 32 64 290 699
which inevitably arise due to sub-sampling, sample preparation and 31 66 356 633
instrumental measurement ‘error’ over periods of time (i.e. the ‘Type 30 74 430 559
5’ precision defined in ref. [15] and shown in Table 2). Samples may be 29 64 494 495
28 70 564 425
considered to be completely indistinguishable (in terms of elemental
27 61 625 364
composition) if the measured concentrations of all elements overlap 26 72 697 292
within the limits of the Type 5 precision error. In the example shown 25 44 741 248
in Table 3, 19 of the 1874 samples in Sub-set A can be considered as 24 48 789 200
23 47 836 153
completely indistinguishable from each other on the basis of the 27
22 35 871 118
elemental concentrations determined. In the case of Sub-set B, 14 of 21 29 900 89
the 989 samples can be considered to be completely indistinguishable 20 21 921 68
19 12 933 56
18 18 951 38
17 8 959 30
16 7 966 23
15 6 972 17
Table 3 14 3 975 14
Comparison of 1874 samples in data sub-set A, showing the number samples which are 13 1 976 13
analytically indistinguishable and distinguishable in terms of different numbers of 12 5 981 8
elements. 11 2 983 6
10 3 986 3
Number Number of samples Number of samples Number of samples 9 0 986 3
of which are which are which can be 8 2 988 1
elements indistinguishable from indistinguishable from distinguished from at 7 0 988 1
at least one other at least one other least one other sample 6 1 989 0
sample in terms of sample in terms of this in terms of this number 5 0 989 0
exactly this number of number of elements or of elements or more 4 0 989 0
elements more 3 0 989 0
27 19 19 1855 2 0 989 0
26 41 60 1814 1 0 989 0
25 45 105 1769 0 0 989 0
24 77 182 1692
Samples are considered to be analytically indistinguishable if values overlap within the
23 112 294 1580
Type 5 precision limits shown in Table 2.
22 110 404 1470
21 178 582 1292
20 205 787 1087
19 162 949 925
18 215 1164 710 in terms of 40 elemental concentrations (Table 4). A considerable
17 192 1356 518 amount of other information can be obtained from this table; for
16 138 1494 380 example, 975 of the 989 samples can be distinguished from at least
15 116 1610 264
one other sample on the basis of 40 elements, and 68 samples are
14 97 1707 167
13 59 1766 108 indistinguishable from at least one other sample in terms of 27
12 41 1807 67 elements.
11 19 1826 48
10 20 1846 28 3.3.2. Method 2
9 12 1858 16
Another method of comparing samples is to adopt an exclu-
8 4 1862 12
7 8 1870 4 sionary approach using selected elemental ratios or concentrations,
6 2 1872 2 taking into account the Type 5 analytical precision (or other
5 2 1874 0 prescribed ‘error’ range). The Al2O3/Fe2O3 and Al2O3/K2O ratios
4 0 1874 0
have previously been found to provide useful screening criteria in
3 0 1874 0
2 0 1874 0 many situations [1]. These major oxides often provide a good proxy
1 0 1874 0 for the relative proportions of feldspars and/ or different clay
0 0 1874 0 minerals in soil samples. Other ratios which have been found to be
Samples are considered to be analytically indistinguishable if values overlap within the useful discriminators include SiO2/Al2O3, SiO2/Fe2O3, SiO2/CaO, CaO/
Type 5 precision limits shown in Table 2. MgO, K2O/Na2O, Cu/Pb, Rb/Sr, Ce/La, Nd/Sm, and U/Th [1,14].
178 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181

Table 5
Example of an inter-sample comparison exercise using 'target' soil samples 1A from Simon's Wood and 1B from Arborfield Bridge, Berkshire, UK.

Soil samples from the centre of each grid Number of non-excluded samples, in terms of
Al2O3/ Al2O3/Fe2O3 Al2O3/Fe2O3 Al2O3/Fe2O3
Fe2O3 Al2O3/K2O Al2O3/K2O Al2O3/K2O
Nd/Sm Nd/Sm
CaO
1A Total number 2 1 1 1
(Simon's Wood) Number from Simon's Wood 1 1 1 1
1A ± 5% Total number 5 3 3 3
(Simon's Wood) Number from Simon's Wood 3 3 3 3
1A ± 10% Total number 6 3 3 3
(Simon's Wood) Number from Simon's Wood 3 3 3 3
1A ± 20% Total number 12 5 5 4
(Simon's Wood) Number from Simon's Wood 5 5 5 4
1B Total number 105 13 6 3
(Arborfield Bridge) Number from Arborfield Bridge 7 3 2 2
1B ± 5% Total number 221 37 29 6
(Arborfield Bridge) Number from Arborfield Bridge 8 6 6 4
1B ± 10% Total number 325 81 66 12
(Arborfield Bridge) Number from Arborfield Bridge 8 8 8 4
1B ± 20% Total number 576 228 196 37
(Arborfield Bridge) Number from Arborfield Bridge 8 8 8 5

A computer search routine was used to identify those samples which could not be excluded as ‘different’ after applying four successive filters (Al2O3/Fe2O3 ratio, Al2O3/K2O ratio, Nd/
Sm ratio, and %CaO) and successive ‘error’ limits around the determined elemental concentration values.

A simple example of the use of elemental ratios and concentra-
tions to exclude certain samples from a list, and to identify others
which require more detailed examination, is presented in Table 5.
In this instance one sample from each of two locations in Berkshire,
UK (sample 1A from Simon's Wood, and sample 1B from Arborfield
Bridge) was compared with 1895 other samples in the background
dataset in terms of Al2O3/Fe2O3 ratio, Al2O3/K2O ratio, Nd/Sm ratio
and % CaO content. The background dataset contained 8 other
samples from Simon's Wood and 8 from Arborfield Bridge. In effect,
the different ratios are used as successive ‘filters’ to exclude
progressively more samples as possible ‘matches’ for the selected
samples. Samples 1A and 1B were in turn initially compared with
all other 1895 samples by setting ‘error’ limits defined by the Type
5 precision values reported in Table 2. The total number of non-
excluded samples in the database as a whole, and from the same
location (Simon's Wood or Arborfield Bridge, respectively), was

Table 6
Results of an exercise to identify the least dissimilar samples to target sample 1A from Simon's Wood, based on Euclidean distance indicated by PCA performed on data sub-set A
(1874 samples with 27 elements) and sub-set B (989 samples with 40 elements.

Rank Sub-set A (1874 samples with 27 elemental analyses) Sub-set B (989 samples with 40 elemental analyses)
Case name Sample Euclidean distance Number of element concentrations Case name Sample Euclidean distance Number of element concentrations
ID from 1A indistinguishable from 1A ID from 1A indistinguishable from 1A
1 Simon's Wood 1A 0.000 27 Simon's Wood 1A 0.000 40
2 Simon's Wood 6A 0.403 18 Simon's Wood 6A 0.488 29
3 Simon's Wood 3A 0.405 11 Simon's Wood 4A 0.743 25
4 Simon's Wood 4A 0.423 17 Simon's Wood 2A 0.849 19
5 Simon's Wood 2A 0.594 12 Simon's Wood 5A 0.996 18
6 Simon's Wood 5A 0.872 13 Simon's Wood 7A 1.119 16
7 Simon's Wood 7A 0.975 9 Simon's Wood 3A 1.125 19
8 Spinaker/Jones 19S 1.219 7 Huntley KP/128A 1.702 10
9 Simon's Wood 8A 1.285 6 Spinaker/Jones 19S 1.718 11
10 Simon's Wood 9A 1.358 9 Huntley KP/153A 1.799 8
11 Huntley KP/128A 1.482 4 Huntley KP/20 A 1.990 8
12 Spinaker/Jones 17S 1.547 5 Simon's Wood 8A 2.108 8
13 Huntley KP/153A 1.550 4 Spinaker/Jones 17S 2.175 8
14 Maple KP512 1.669 3 Simon's Wood 9A 2.295 12
15 Spinaker/Jones 10S 1.699 5 Huntley KP/4 A 2.408 5
16 Huntley KP/20 A 1.774 4 Huntley KP/156A 2.440 8
17 Huntley KP/6 A 1.784 7 Huntley KP/6 A 2.457 10
18 Maple KP507 1.801 4 Maple KP509 2.497 6
19 Maple KP505 1.825 4 Maple KP512 2.538 4
20 Maple KP610 1.891 3 Maple KP713 2.582 5
21 Spinaker/Jones 22N 1.903 3 Maple KP511 2.629 6
22 Maple KP509 1.908 4 Maple KP606 2.633 7
23 Huntley KP/156A 1.922 4 Spinaker/Jones 22N 2.639 7
24 Huntley KP/4 A 1.933 3 Spinaker/Jones 10S 2.647 7
25 Maple KP713 1.942 4 Maple KP507 2.673 5
26 Maple KP500 1.987 5 Huntley KP/150A 2.679 9
27 Maple KP365 2.049 4 Maple KP505 2.682 5
28 Maple KP712 2.068 3 Maple KP712 2.717 4
29 Maple KP511 2.080 4 Maple KP365 2.724 5
30 Maple KP601 2.124 5 Maple KP605 2.733 7

Also shown are the numbers of elements for which concentrations are analytically indistinguishable from those for sample 1A when allowance is made for the Type 5 precision errors
shown in Table 2.
 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181 179

recorded. This exercise was then repeated by increasing the ‘error’
margins by ± 5%, ± 10% and ± 20%. As the ‘error’ margins is
widened, so the number of non-excluded samples increases. This
process can be regarded simply as a sensitivity testing exercise
which provides information helpful in the interpretation of the
data.
Table 5 shows that, when sample 1A from Simon's Wood was
compared with the other 1895 samples in the dataset, using the
Al2O3/Fe2O3 ratio and Type 5 precision limits only, all but two of
the samples could be excluded as dissimilar, one of which was also
from Simon's Wood. When the second, Al2O3/K2O ratio filter was
applied, all but the one sample from Simon's Wood could be
excluded as dissimilar. The effect of increasing the error margins
around the Type 5 precision limits was to progressively increase
the number of samples which could not be excluded as dissimilar
(some from Simon's wood and some from elsewhere). However,
even allowing for an additional 20% ‘error’ margin, only 12 samples
could not be excluded as dissimilar on the basis of Al2O3/Fe2O3
ratio only, with 5 of the 12 originating from the same location
(Simon's Wood).
In the case of sample 1B from Arborfield Bridge, 105 of the
other samples in the dataset (including 7 from Arborfield Bridge)
could not be excluded as dissimilar based on the Al2O3/Fe2O3 ratio
and Type 5 precision limits alone. When the Al2O3/K2O filter was
also applied, the number of samples which could not be excluded
was reduced to 13, 3 of which originated from Arborfield Bridge.
Successive application of the Nd/Sm and %Ca0 filters further
reduced the number of samples which could not be excluded to 3,
2 of which originated from the same location. However, applica-
tion of the Type 2 precision error limits also led to false exclu-
sion of 6 of the 8 other samples from Arborfield Bridge. When
additional 5%, 10% and 20% ‘error’ margins were added to the
Type 5 precision limits, the number of false exclusions was
progressively reduced to 3, but the number of false inclusions also
increased to 32. However, consideration of the nature and geo-
graphical origins of the comparatively small number of non-
excluded samples is able to provide useful information about
possible alternative geographical sources, and/or the type of soil
which requires consideration.
3.3.3. Method 3
A third method of inter-sample comparison is to consider a large
number of elements simultaneously and to calculate the Euclidean
distance between a questioned sample and all other samples in the
dataset using principal component analysis (PCA). Table 6 illustrates
the results of such an exercise using sample 1A from Simon's Wood
and the data sub-sets A and B. The Euclidean distance increases as
samples become more dissimilar from sample 1A in terms of the
elements considered. Table 6 shows that, using Sub-set A with data for
27 elements, 9 of the top 10 least dissimilar samples originate from the
same location at Simon's Wood. Using Sub-set B, which includes data
for 40 elements, the top 7 least dissimilar samples originate from
Simon's Wood, but the other two samples from this location occupy
twelfth and fourteenth positions in the dissimilarity ranking. The
reason for this is that these two samples differ slightly from the others
taken from the same area, notably in terms of the concentrations
certain trace elements [21].
It is also informative that Table 6 shows that even the least
dissimilar samples from the Simons Wood Site (1A and 6A) only 18
out of 27 elements in Sub-set A and 29 out of 40 elements in sub-set
B could be described as analytically indistinguishable (i.e. overlap
occurs when allowance is made for Type 5 precision errors).
Table 7 shows the results of a similar exercise using the data for
samples from Arborfield Bridge. Considering data sub-set A, the

Table 7
Results of an exercise to identify the least dissimilar samples to target sample 1B from Arborfield Bridge, based on Euclidean distance indicated by PCA performed on data sub-set A
(1874 samples with 27 elements) and sub-set B (989 samples with 40 elements.

Rank Sub-set A (1874 samples with 27 elemental analyses) Sub-set B (989 samples with 40 elemental analyses)
Case name Sample Euclidean Number of element concentrations Case name Sample Euclidean Number of element concentrations
ID distance from 1B indistinguishable from 1B ID distance from 1B indistinguishable from 1B
1 Arborfield Br. 1B 0.000 27 Arborfield Br. 1B 0.000 40
2 Arborfield Br. 4B 0.736 20 Arborfield Br. 4B 0.873 30
3 Arborfield Br. 2B 0.863 20 Arborfield Br. 2B 0.881 30
4 Arborfield Br. 3B 0.998 18 Arborfield Br. 3B 1.037 29
5 Arborfield Br. 9B 1.195 13 Arborfield Br. 9B 1.510 17
6 Arborfield Br. 7B 1.357 14 Arborfield Br. 7B 1.931 15
7 Arborfield Br. 6B 1.380 14 Honeycombe KP03a 1.954 15
8 Arborfield Br. 8B 1.476 16 Arborfield Br. 6B 1.958 15
9 Arborfield Br. 5B 1.493 14 Honeycombe KP01a 1.963 16
10 Atkinson KP 1A 1.756 9 Arborfield Br. 8B 2.068 17
11 Avon KP1006 1.761 9 Avon KP1006 2.154 14
12 Honeycombe KP03a 1.867 8 Afzal NAJ11A 2.223 19
13 Atkinson KP 4A 1.891 9 Afzal NAJ10A 2.223 20
14 Afzal NAJ11A 1.940 13 Arborfield Br. 5B 2.345 14
15 Honeycombe KP01a 1.944 8 Honeycombe KP02a 2.353 15
16 Afzal NAJ10A 1.987 14 Avon KP1030 2.397 15
17 Xmas Trees KP6 2.060 10 Honeycombe KP06a 2.481 18
18 Xmas Trees KP7 2.071 10 Avon KP1026 2.533 12
19 Avon KP1030 2.076 11 Atkinson KP 4A 2.588 15
20 Goldfinch KP113 2.133 8 Atkinson KP 1A 2.646 12
21 Goldfinch KP110 2.142 8 Maple KP506 2.660 19
22 Quy KP100 2.151 13 Comet DC 16 2.686 14
23 Gamekeeper KA19 2.163 9 Honeycombe KP04a 2.732 17
24 Spinaker/Jones 77S(b) 2.169 5 Huntley KP/110A 2.815 15
25 Goldfinch KP45 2.187 8 Spinaker/Jones 73S 2.833 11
26 Honeycombe KP02a 2.202 9 Honeycombe KP22a 2.884 9
27 Spinaker/Jones 6N 2.208 9 Comet DC 03 2.902 15
28 Avon KP1026 2.208 7 Cherry KP03A 2.907 8
29 Avon KP1020 2.217 6 Comet KP56A 2.933 18
30 Maple KP506 2.218 10 Honeycombe KP19a 2.945 7

Also shown are the numbers of elements for which concentrations are analytically indistinguishable from those for sample 1B when allowance is made for the Type 5 precision errors
shown in Table 2.
180 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181

samples from Arborfield Bridge occupy the top 9 places, and in
the case of sub-set B they occupied the top 6 places. The other 3
samples from Arborfield Bridge occupy 8th, 10th and 14th places in
the dissimilarity ranking. Sample 4B, which is indicated by the
PCA analysis to be the least dissimilar to the questioned sample
1B using both data sub-sets, has only 20 out of 27 elements (sub-
set A), and 30 out of 40 elements (sub-set B) which can be classified
as analytically indistinguishable when the Type 5 precision limits
are applied. This demonstrates that even samples taken just a few
tens of centimetres apart are unlikely to be ‘completely indistingui-
shable’ in terms of all major and trace element concentrations.
Table 8 provides a casework example in which four questioned
soil samples from a car owned by a person suspected of child
abduction were compared (using PCA performed on elemental
concentration data) with samples from a suspected crime scene,
other control samples taken from the surrounding area, and
samples from numerous other locations. In this example, one of
the samples taken from the suspect vehicle (KP9A) was selected
as the ‘target’ sample with which all other samples were
compared. Using both data Sub-set A and Sub-set B, the three
least dissimilar samples to KP9A were also from the suspect
vehicle. In the case of Sub-set A, samples from the crime scene
occupied 9 of the next 11 least dissimilar positions. The highest
ranking (most dissimilar) sample from the local area outside the
crime scene had a rank of 84. In terms of Sub-set B, the samples
from the suspect vehicle and crime scene occupied the top 13
positions, while the highest ranking sample from the surrounding
area beyond the crime scene had a rank of 97. In the case of sub-
set A, the two least dissimilar samples identified by PCA had 22
out of 27 elements which are analytically indistinguishable; in the
case of sub-set B the two most similar samples had 33 out of 40
individual elements which were indistinguishable.
4. Conclusions
Major and trace elemental analysis of standardized soil size
fractions by ICP-AES and ICP-MS provides a precise and cost-
effective method for the comparison of soil samples. It can be
undertaken relatively quickly and can therefore provide a practical
tool which can be used to screen soil samples during ‘live’
investigations and to identify samples and areas of geographical
interest which may warrant more detailed examination. As with
all methods of trace evidence analysis, care must be taken to
minimise the risk of false exclusions as well as inclusions; the
adoption of a sensitivity testing approach, in which progressively
larger variance around the measured value is considered, can
assist in this process.
The fact that two samples are shown to be completely
indistinguishable, or very similar, in terms of elemental composition
does not in itself provide definitive evidence of a common source, or
other direct association. Elemental analysis of questioned soil
samples but should always be carried out in conjunction with
other methods which provide information about other soil attri-
butes, including mineralogy, particle size distribution, grain shape
characteristics, colour, and biological assemblages. There is con-
siderable scope to improve existing soil data base information by
increasing the number of samples and geographical coverage for
which major and trace element data are available, and by
incorporating data for further soil properties.
Acknowledgements
We thank Debra Croft, Samantha Witton, Nick Walsh, David Wray
and Lorna Dyer for assistance with sample collection and analysis, and
three anonymous referees for comments on earlier versions of the paper.

Table 8
Comparison of a target soil sample KP9A removed from a suspect motor vehicle with other soil samples taken from the vehicle (KP14A, KP16A, KP21A), a suspected crime scene, the
wider area around the crime scene, and samples from other areas, based on Euclidean distance indicated by PCA performed on dataset A (1874 samples for 27 elements) and data sub-
set B (989 samples with 40 elements).

Rank Sub-set A (1874 samples with 27 elemental analyses) Sub-set B (989 samples with 40 elemental analyses)
Case name Sample ID Euclidean distance Number of element concentrations Case name Sample Euclidean distance Number of element concentrations
from KP9A indistinguishable from KP9A ID from KP9A indistinguishable from KP9A
1 Damascus KP 9A 0.000 27 Damascus KP 9A 0.000 40
2 Damascus KP 16A 0.629 22 Damascus KP 16A 0.844 33
3 Damascus KP 14A 0.763 20 Damascus KP 21A 0.917 34
4 Damascus KP 21A 0.831 22 Damascus KP 14A 0.920 30
5 Damascus KP84A 1.595 10 Damascus KP84A 1.800 15
6 Damascus KP 27A 1.652 13 Damascus KP 24A 1.851 23
7 Damascus KP 24A 1.788 14 Damascus KP 27A 1.918 21
8 Damascus KP86A 1.842 8 Damascus KP 34A 2.092 20
9 Damascus KP85A 1.955 8 Damascus KP86A 2.171 13
10 Damascus KP 34A 2.045 12 Damascus KP85A 2.266 12
11 Damascus KP 38A 2.152 10 Damascus KP 38A 2.319 15
12 Beverley KP26 2.243 9 Damascus KP 26A 2.400 16
13 Beverley KP25 2.257 9 Damascus KP88A 2.634 16
14 Damascus KP87A 2.303 9 Docker KP303A 2.709 19
15 Damascus KP 26A 2.312 12 Damascus KP87A 2.733 14
16 Docker KP303A 2.324 13 Comet DC 15 2.766 16
17 Beverley KP33 2.384 11 Damascus KP 40A 2.782 19
18 Beverley KP21 2.399 8 Avon KP128 2.782 17
19 Beverley KP23 2.489 10 Bexhill KP 1A 2.800 18
20 Comet DC 15 2.495 10 Comet DC 17 2.851 12
21 Beverley KP27 2.506 7 Bexhill KP 3A 2.891 20
22 Comet DC 17 2.514 10 Foreman KP2A 2.903 13
23 Avon KP128 2.546 12 Avon KP1015 2.943 12
24 Damascus KP88A 2.583 9 Docker KP220A 2.947 11
25 Goldfinch KP1105 2.587 10 Docker KP 504 2.981 14
26 Beverley AMS/343/39 2.588 9 Avon KP2016 2.994 19
27 Foreman KP2A 2.633 8 Spinaker/Jones 27S 3.010 18
28 Docker KP110A 2.635 9 Comet DC 07 3.011 19
29 Bexhill KP 1A 2.640 11 Soil Grids 5B 3.038 13
30 Fallin KP4 2.645 6 Avon KP1029 3.071 19

Also shown are the numbers of elements for which concentrations are analytically indistinguishable when allowance is made for the Type 5 precision errors shown in Table 2.
 K. Pye, S.J. Blott / Science and Justice 49 (2009) 170–181
References
[1] K. Pye, Geological and Soil Evidence — Forensic Applications, CRC Press, Boca
Raton, 2007.
[2] D.J. Croft, K. Pye, Multi-technique comparison of source and primary transfer soil
samples: an experimental investigation, Science and Justice 44 (2003) 21–28.
[3] K. Pye, D.J. Croft, Multi-technique comparison of source and primary transfer soil
samples: an experimental investigation — a reply, Science and Justice 44 (2006)
176–178.
[4] Applied Geochemistry Research Group and Imperial College London, The Wolfson
Geochemical Atlas of England and Wales, Clarendon Press, Oxford, 1978.
[5] British Geological Survey, Regional Geochemistry of North-east England, British
Geological Survey, Keyworth, 1996.
[6] British Geological Survey, Regional Geochemistry of North-west England and
North Wales, British Geological Survey, Keyworth, 1997.
[7] British Geological Survey, Regional Geochemistry of North Wales and Part of West-
central England: streamwater, British Geological Survey, Keyworth, 1999.
[8] British Geological Survey, Regional Geochemistry of North Wales and Part of West-
central England: Stream Sediment and Soil, British Geological Survey, Keyworth, 2000.
[9] S.P. McGrath, P.J. Loveland, The Soil Geochemical Atlas of England and Wales,
Blackie Academic and Professional, Glasgow, 1992.
[10] M.A. Oliver, P.J. Loveland, Z.I. Frogbrook, R. Webster, S.P. McGrath, Statistical and
geostatistical analysis of the National Soil Inventory of England and Wales, The
Technical Report, Report on Project SP0124, UK Ministry of Agriculture, Fisheries and
Food (now the Department of Environment, Food and Rural Affairs), London, 2002.
[11] C. Reimann, U. Siewers, T. Tarvainen, L. Bityukova, J. Eriksson, A. Gilucis, V.
Gregorauskiene, V.K. Lukashev, N.N. Matinian, A. Pasieczna, Agricultural soils in
northern Europe: a geochemical atlas, Geologisches Jahrbuch Sonderhefte, Reihe D,
Mineralgie, Petrographie, Geochemie, Lagerstattenkunde, Heft SD5, Herausgegeben
181
von der Bundesanstalt fur Geowissenschaften und Rohstoffe un den Staatlichen
Geologischen Diensten in der Bundsrepublik Deutschland, Hannover, 2003.
[12] S.E. Saye, K. Pye, Development of a coastal dune sediment database for England and
Wales: forensic applications, in: K. Pye, D.J. Croft (Eds.), Forensic Geoscience: Principles,
Techniques and Applications, Geological Society, London, Special Publications, vol. 232,
The Geological Society, London, 2004, pp. 75–96.
[13] M.M. Houck, Statistics and trace evidence: the tyranny of numbers, Forensic
Science Communications 1 (1999) 1–8.
[14] K. Pye, S.J. Blott, Comparison of soils and sediments using major and trace element
data, in: K. Pye, D.J. Croft (Eds.), Forensic Geoscience: Principles, Techniques and
Applications, Geological Society, London, Special Publications, vol. 232, The
Geological Society, London, 2004, pp. 183–196.
[15] K. Pye, S.J. Blott, D.S. Wray, Elemental analysis of soil samples for forensic purposes
by inductively coupled plasma spectrometry — precision considerations, Forensic
Science International 160 (2006) 178–192.
[16] K. Pye, S.J. Blott, D.J. Croft, S.J. Witton, Discrimination between sediment and soil
samples for forensic purposes using elemental data: an investigation of particle
size effects, Forensic Science International 167 (2007) 30–42.
[17] J.N. Walsh, Inductively coupled plasma-atomic emission spectrometry (ICP-AES), in:
R. Gill (Ed.), Modern Analytical Geochemistry, Longman, Harlow, 1997, pp. 41–66.
[18] D.S. Wray, Geological analysis. In S.M. Nelms Editor, ICP Mass Spectrometry Handbook,
Blackwell Publishing, Oxford, and CRC Press, Boca Raton (2005) 432–451.
[19] J.W. Tukey, Exploratory Data Analysis, Addison Wesley, Reading, Massachusetts, 1977.
[20] J.N. Miller, J.C. Miller, Statistics and Chemometrics for Analytical Chemistry, 5th ed.
Pearson Education Limited, Harlow, 2005.
[21] K. Pye, S.J. Blott, D.J. Croft, J. Carter, Forensic comparison of soil samples: assessment of
small-scale spatial variability in elemental composition, carbon and nitrogen isotope
ratios, colour and particle size distribution, Forensic Science International 163 (2006)
59–80.