J Forensic Sci, 2018

doi: 10.1111/1556-4029.13858
TECHNICAL NOTE Available online at: onlinelibrary.wiley.com

ANTHROPOLOGY

Sebastian K.T.S. W€
arml€ander,1,2,3 Ph.D.; Liivi Varul,4,5 M.A.; Juuso Koskinen,6 M.A.; Ragnar Saage,4
M.A.; and Stefan Schlager,7 Ph.D.

Estimating the Temperature of Heat-exposed

Bone via Machine Learning Analysis of SCI
Color Values: A Pilot Study†

ABSTRACT: Determining maximum heating temperatures of burnt bones is a long-standing problem in forensic science and archaeology. In
this pilot study, controlled experiments were used to heat 14 fleshed and defleshed pig vertebrae (wet bones) and archaeological human verte-
brae (dry bones) to temperatures of 400, 600, 800, and 1000°C. Specular component included (SCI) color values were recorded from the bone
surfaces with a Konica-Minolta cm-2600d spectrophotometer. These color values were regressed onto heating temperature, using both a tradi-
tional linear model and the k-nearest neighbor (k-NN) machine-learning algorithm. Mean absolute errors (MAE) were computed for 1000
rounds of temperature prediction. With the k-NN approach, the median MAE prediction errors were 41.6°C for the entire sample, and 20.9°C
for the subsample of wet bones. These results indicate that spectrophotometric color measurements combined with machine learning methods
can be a viable tool for estimating bone heating temperature.

KEYWORDS: forensic science, forensic anthropology, cremains, burned bone, color measurement, regression analysis

Estimating the temperatures to which cremated or otherwise
heated bones have been exposed is a long-standing problem in
forensic science and in archaeology (1–10). Heat exposure
induces alterations in numerous bone properties including size,
shape, hardness, chemical composition, crystallinity, and color
(1, 7, 8, 10–16). Such alterations interfere with forensic analysis
of, e.g., trace elements, isotopes, DNA, and morphological evi-
dence of blunt and sharp force trauma including cut marks and
bite marks (3, 1z0, 17, 18).
1 sent during heating (22–24), chemical interactions with soil dur-
Department of Biochemistry and Biophysics, Stockholm University,
106 91 Stockholm, Sweden.
2
Division of Commercial and Business Law, Link€oping University, 581 83
Link€oping, Sweden.
3
UCLA/Getty Conservation Programme, Cotsen Institute of Archaeology,
UCLA, Los Angeles, CA 90095.
4 muffler furnace was used to heat fleshed and defleshed pig ver-
Institute of History and Archaeology, University of Tartu, 50090 Tartu,
Estonia.
5
School of Humanities, Tallinn University, 10120 Tallinn, Estonia.
6
Department of Philosophy, History, Culture and Art Studies, University
of Helsinki, 00014 Helsinki, Finland.
7
Department of Anthropology, Medizinische Fakult€at der Albert Ludwigs,
University of Freiburg, 79085 Freiburg, Germany.
Corresponding authors: Sebastian K.T.S. W€arml€ander, Ph.D. E-mail:
seb@dbb.su.se and Stefan Schlager Ph.D. E-mail: stefan.schlager@anthro-
pologie.uni-freiburg.de
†
The bone experiments were carried out in line with the Estonian legisla-
tion and ethical rules for the handling of human and animal remains.
Prior presentation: The material in this manuscript has never been publicly
presented.
Grants/Funding: None.
Received 6 May 2018; and in revised form 18 June 2018; accepted 18
June 2018.
Many of the heat-altered bone properties have been tried as
predictors for maximum temperature exposure, with variable suc-
cess (4, 7, 9, 10, 12). So far, the best temperature estimates have
been provided by the bone crystallinity index (CI), measured
either with X-ray diffraction or with infrared spectroscopy (7–9,
12, 19–21). Even though it is generally accepted that bone color
changes systematically with the heating temperature, a large
number of researchers have concluded that bone color measure-
ments are not useful for accurate temperature estimations (3, 4,
7, 12, 22). The main argument has been that the bone color after
heat exposure is influenced by a multitude of factors, including
the burning time (22), the amount of oxygen and soft tissue pre-
ing or after the fire (2, 22, 25), and the possible influence of
coloring substances (contaminants) such as metals (10).
The aim of this pilot study is to challenge the notion that bone
color provides inferior information about heating temperature. A
tebrae, as well as dry human vertebrae recovered from an
archaeological excavation, to temperatures of 400, 600, 800, and
1000°C. This temperature range represents a number of burning
scenarios, such as 400–600°C for open camp fires, 600–900°C
for fire pits or improvised ovens, no more than 700°C for prairie
fires but 800°C or more for a forest fire front, 600–1100°C for
cremation furnaces, and up to 1000°C in burning cars and
houses (7, 26–30). As previous work has suggested differences
in thermal alteration between bones from bird, fish, and mam-
mals (4), but little difference between bones from different mam-
mal species (7), having both pig and human bones in the sample
was considered non-problematic. The chemical integrity of the
bones after heating was investigated using a portable X-ray

© 2018 American Academy of Forensic Sciences 1

2 JOURNAL OF FORENSIC SCIENCES
fluorescence (pXRF) instrument, and their color values were
recorded using a digital spectrophotometer. The usefulness of
color data for temperature estimation was evaluated by regres-
sion analysis, employing a linear model as well as a machine
learning algorithm.
Materials and Methods
Sample
Loin cuts from pigs (Sus scrofa domesticus) that included ver-
tebrae together with ilium parts (Fig. 1A) were purchased at a
local market in Tartu, Estonia. The pigs were young adults with
unfused epiphyses. Vertebrae were cut out and prepared either
fleshed (n = 4) or defleshed (n = 5)-these are the “wet” samples.
Defleshing was carefully conducted with a scalpel, leaving only
minute fragments of soft tissue attached to the bones. Vertebrae
from adult humans (n = 5) were also investigated (Fig. 1C),
obtained as dry bones from commingled archaeological material
excavated from the St Jakobi church cemetery in Tartu, Estonia.
This cemetery, which dates to the mediaeval to early modern
period, was rescue-excavated in 2014 as part of a construction
project.
Heating Procedure
A KL-22 muffle furnace (Kerako, Estonia) equipped with a
programmable controller was used to heat the bone samples to
four different temperatures, i.e., 400, 600, 800, and 1000°C. The
furnace was well ventilated during heating, thereby providing an
FIG. 1––(A) Pig vertebrae in the stainless steel bowl used for the heat treatment. (B) Pig vertebrae after heating to 600°C for 1 h (sample 2). (C) Human
archaeological vertebrae before heating. (D) Defleshed pig vertebrae after heating to 400°C for 1 h (sample 5). Note the blue discoloration of sample 5
induced by residue from the chromium-containing stainless steel bowl. Photos by LV.
oxidizing environment. The bone samples were placed in a bowl
of stainless steel and then put in the furnace. It took between 10
and 45 min for the furnace to reach the desired temperature
(longer time for higher temperatures; Table S1). The temperature
was then kept stable for 60 min. Finally, it took between 4 and
8 h for the furnace to cool down with the sample inside. All
samples were weighed and photographed before and after heat-
ing (Table S1 and Figs. S1–S13).
Color Analysis
Two approaches were used to characterize the colors of the
heat-exposed bones. First, the bone colors were manually
matched to Munsell soil color charts (Munsell Color Company
Inc., USA). Second, color values were recorded with a portable
Konica-Minolta cm-2600d spectrophotometer (Konica-Minolta,
Osaka, Japan), equipped with a xenon pulse light as light source
and operating with firmware version 1.23. Color data were
recorded from 360 to 740 nm with an 8 mm opening window, a
wavelength pitch of 10 nm, and a half bandwidth of approxi-
mately 10 nm. All measurements were conducted in a dark room
where the only light source besides the instrument itself was a
small fluorescent light located approximately 10 m away, direc-
ted away from the instrument. The spectrophotometer was zero-
and white-calibrated immediately before the measurements,
which were carried out with the following settings: 100% UV,
standard D65 illuminant, di:8°, de:8° geometry, 10° observer
angle, L*a*b* (CIE1976) color space, and reflectance as mea-
surement type. No materials cover or other medium was placed
between the 8 mm measuring window and the bone surface
 €
WARML €
ANDER ET AL.
analyzed. For each bone sample, three color measurements from
trabecular bone areas and three from cortical bone areas were
recorded. The color data were collected as specular-component
included (SCI) L*a*b* values in areas of the dominant color:
regions displaying stains or color anomalies were avoided.
Statistical Analysis of Color Data
Temperature predictions based on the SCI/L*a*b* color mea-
surements of the bones were computed for the entire sample as
well as for sample subsets, using two different regression models:
traditional linear regression and the non-parametric k-nearest
neighbor (k-NN) algorithm. The k-NN method is one of the sim-
plest machine learning algorithms, and was considered suitable
given the somewhat small sample size. The limited sample size
furthermore prompted us to use a k-fold cross-validation approach
to estimate the model accuracies and optimize the regression
parameters. In this approach, 1000 random partitions were cre-
ated, each consisting of a training sample (80% of data) and a test
sample (20% of data). For each of these 1,000 rounds, the mean
absolute error (MAE) of prediction was calculated based on the
temperature values predicted for the test sample. Rather than a
single error value, error distributions were obtained for both the
k-NN and the linear model predictions. As these distributions do
not necessarily follow a Gaussian distribution, especially for the
k-NN predictions, the medians of the errors are reported instead
of their means. The final parameters for the wet and the combined
samples consisted of a cosine kernel, a Minkovski distance of 1,
and a neighborhood of 4. For the dry data, a triweight kernel was
used, as it showed the best performance. As an additional statisti-
cal investigation, standard linear discriminant analysis (LDA)
with leave-one-out cross-validation was used to separate heated
wet bones from heated dry bones based on color data. All statisti-
cal analyses were performed using the R platform (31), and more
specifically the R packages caret (32), kknn (33), Morpho, Rvcg
(34), and rgl (35).
Elemental Analysis
A Tracer III-SD portable X-ray fluorescence (pXRF) unit
(Bruker Ltd., USA) operating at 40 kV and 10.7 lA with a
305 lm Al + 25 lm Ti filter was used to investigate the ele-
mental composition of the surfaces of some of the bone samples
after heating, in order to detect possible contamination from the
metal bowl used during the heating procedure. Under the set-
tings employed, the pXRF instrument detects the elements Ca to
Sb (K lines) and W to Bi (L lines).
Results
Heat Treatment
Photos of all the samples, before and after heat treatment, are
shown in Figs. S1–S13. Photos of four select bone samples are
shown in Fig. 1. Heat treatment of the wet pig bones clearly
changed their colors, with similar results obtained for the fleshed
and defleshed samples. The first round of heating was done with
sample 5, a defleshed pig sample heated to 400°C. During this
heating step, a large amount of surface residue flaked off the
stainless steel bowl. The interaction of these flakes with the pig
bone produced a metallic blue color, present on top of the other-
wise black/charred bone sample (Fig. 1D). Subsequent heating
rounds produced much less flaking, and the fleshed pig sample
. ESTIMATING HEATING TEMPERATURE FROM BONE COLOR 3
heated to 400°C charred the soft tissue and turned the bones uni-
formly black (i.e., no contamination/discoloration). Heating the
wet bones to 600°C (samples 2 and 6) destroyed almost all the
soft tissue and turned the bones into various shades of gray,
ranging from dark gray to almost white (Fig. 1B). Heating the
wet bones to 800°C and 1000°C (samples 3, 4, 7, and 8) com-
pletely disintegrated the soft tissue and fragmented the bones,
which turned white. For the dry human bones, heating to 400
and 600°C (samples 9 and 10) produced almost no visible color
change, although the part of sample 9 that was in contact with
the steel bowl turned black. The dry human bones heated to 800
and 1000°C (samples 11 and 12) became somewhat lighter.
Munsell color values for all samples are shown in Table S1,
together with the details of the heating procedures. All heated
samples displayed some weight loss, in particular the fleshed
samples (as expected). Weight loss and shrinking are well-docu-
mented effects of bone heating (7, 10, 36), and will not be fur-
ther discussed here.
The elemental composition of some of the samples was deter-
mined with a hand-held portable XRF unit. Sample number 5,
which displayed the metallic blue discoloration (Fig. 1D),
showed elevated levels of Cu, Cr, Fe, and Mn (Fig. 2). These
are precisely the elements that make up the stainless steel alloy
of the bowl used in the heating procedure (it was found to con-
sist of 84.1% Fe, 10.6% Mn, 4.9% Cr, and 0.4% Cu). A pXRF
analysis of the non-contaminated sample 2 showed that small
amounts of Cu, Fe, and Zn are naturally present as trace ele-
ments in the bones, in addition to the dominating amount of cal-
cium originating from the bone apatite (Fig. 2).
Statistical Color Analysis
The SCI/L*a*b* color values measured for the different bone
samples were plotted in a 3D graph that turned out to be quite
informative (Figs. 3 and S14). The color values for the dry bone
and the wet bone samples are very well separated, and a simple
linear discriminant function was able to classify the heated bone
samples into wet or dry with 100% accuracy (data not shown).
The wet samples heated to, respectively, 400, 600, 800, and
1000°C form relatively well-separated clusters, with only minor
color differences for fleshed/defleshed and trabecular/cortical
bone samples. For the dry bones, it is more difficult to distin-
guish well-defined clusters (Figs. 3 and S14).
The sample groups that form well-defined clusters in Fig. 3
also produce the most accurate temperature predictions with the
regression analysis. These temperature predictions were carried
out using two approaches: traditional linear regression and a
slightly more sophisticated k-NN machine learning approach.
For the wet sample, the median for the mean absolute error
(MAE) of prediction is 47.5°C with linear regression, but merely
20.9°C with the k-NN model (Table 1 and Fig. 4). The predic-
tion errors are larger for the dry sample, with MAE medians of
92.3°C with linear regression and 80.0°C with the k-NN model.
As expected, the prediction errors for the combined sample are
in between those of the wet and the dry samples, i.e., MAE
medians of 72.2°C with linear regression and 41.6°C with k-NN
(Table 1 and Fig. 4). The samples heated to 1000°C are the easi-
est to predict, with median MAE errors of 21.9°C with the k-NN
method and 48.7°C with the linear method (Table 2). The bones
heated to 600°C and 800°C display the largest prediction errors
for both regression methods (Table 2), which is in line with the
color values for samples heated to these two temperatures show-
ing less well-defined clusters in Fig. 3.
4 JOURNAL OF FORENSIC SCIENCES
FIG. 2––Overlaid pXRF spectra recorded from the surfaces of pig bone samples 2 (red) and 5 (blue), normalized to the Ca peak at 3.7 keV. In addition to
Ca, sample 2 shows trace levels of Fe, Cu, and Zn—normal trace elements in bone. Sample 5 shows much higher levels of Cu, Fe, Mn, and Cr, which appear
to be contamination from the stainless steel bowl shown in Fig. 1A. Spectra by RS.
FIG. 3––3D plot of the SCI/L*a*b* color values recorded from the sur-
faces of the heat-treated bone samples. Squares – defleshed wet pig bone;
Triangles – wet pig bone with flesh; Circles – dry human bone; Small sym-
bols – cortical bone; Large symbols – trabecular bone; Black – not heated;
Green – 400°C; Turquoise – 600°C; Gold – 800°C; Purple – 1000°C. Image
by SS.
Discussion
Using the k-NN regression model, temperature predictions with
relatively small median MAE prediction errors (Table 1, Fig. 4)
were obtained for both the entire sample (41.6°C) and the sub-
sample of wet bones (20.9°C). These are very promising results,
strongly suggesting that color measurements can be used for reli-
able estimation of maximum bone heating temperature. Because
the approach presented here allows the prediction errors to be cal-
culated (Table 1, Fig. 4), it would likely pass the Daubert stan-
dard (37, 38) and could thus be useful in applied forensics work,
which most likely will continue to involve identification and anal-
ysis of burnt bone or contested cremains (13, 28, 30, 39–41).
Future work using a larger sample should be able to better estab-
lish the accuracy of color-based temperature estimates, which
TABLE 1––Median values for the mean absolute error (MAE) after 1000
rounds of temperature prediction, computed for the wet bones, dry bones,
and their combination using two different regression models.
Number of Color k-NN Linear
Sample Readings (n) Model (°C) Model (°C)
Wet bones (fleshed and 48 20.9 47.5
defleshed)
Dry bones 30 80.0 92.3
Wet and dry bones 78 41.6 72.2
appear to perform better than, e.g., bone crystallinity measure-
ments, where the major changes occur around 500–700°C and
where no changes to the crystal structure appear beyond 800°C
(20, 42). The upper temperature limit where bone color changes
no longer are induced still remains to be determined.
To facilitate future work on color analysis of heat-exposed
bones, a number of comparisons were carried out in this small-
sample pilot study. The k-NN approach clearly performed better
than the traditional linear regression model for every sample and
subsample, although the difference is smaller for the smallest
sample sizes (Table 1, Fig. 4). A more advanced machine learn-
ing algorithm than k-NN might perform even better, especially if
a large sample database is available.
The L*a*b* color values recorded for the heated wet pig bones
and the heated dry archaeological human bones were very differ-
ent (Fig. 3), and discriminant analysis was able to predict with
100% accuracy which group a given bone belonged to. As previ-
ous work has shown that bones from different mammal species
respond similarly to thermal alteration (7), the primary reason for
the observed color differences is likely the wet/dry starting condi-
tion, even though some small soil-induced discoloration to the
archaeological bones cannot be ruled out. The main conclusion
from this comparison is arguably that color data for heated wet
and heated dry bones should not be combined. Because dry bones
are not commonly involved in fire scenarios, forensic research
should focus on investigating heat-exposed wet bones.
Comparisons were made also between fleshed and defleshed
pig bones, and between color readings from cortical and trabecu-
lar pig bone. Relatively small differences in color values were
 €
WARML €
ANDER ET AL.
FIG. 4––Distributions of the mean absolute error (MAE) after 1000 rounds of temperature prediction, computed for the wet bones (A), dry bones (B), and
their combination (C) using two different regression models. Images by SS.
TABLE 2––Median values for the mean absolute error (MAE) after 1000
rounds of temperature prediction, listed by sample temperature, and com-
puted using color readings from both wet and dry bones (n = 78) using two
different regression models.
Sample Temperature k-NN Model (°C) Linear Model (°C)
400°C 33.4 72.4
600°C 47.2 78.4
800°C 47.3 76.7
1000°C 21.9 48.7
Overall 41.6 72.2
observed (Fig. 3), which could not be statistically investigated
here due to the small sample size. Future studies with a larger
sample will likely be able to clarify the importance of these
parameters. The amount of soft tissue present has been reported
to influence the bone crystallinity after burning (24), and a simi-
lar effect on bone color is not impossible. Further topics to
investigate include color analysis of burnt teeth and heat-expo-
sure scenarios outside laboratory conditions (43–47).
Some limitations to the proposed method should be mentioned.
Because of the numerous factors that may affect the color of
burnt bone, earlier researchers have refrained from using color
for temperature estimations. Factors such as burning time (22)
and the amount of soft tissue present (24) may affect the color,
but this can likely be controlled for by creating reference samples
with different burning times and soft tissue amounts. The amount
of available oxygen is known to influence the color (22, 23), but
most forensic burning scenarios arguably involve reasonably
good ventilation, i.e., oxidizing conditions. Reference samples of
bone burnt under reducing conditions should, however, be cre-
ated and studied. Chemical analysis might be able to reveal if
bones have been heat-exposed under oxidizing or reducing condi-
tions. Metals and other contaminants (including soil) have the
capacity to modify bone color (2, 10, 22, 25), but such contami-
nation is much more likely to happen after rather than during
burning. Thus, color contamination will certainly be a problem
for color measurements from buried human remains, but perhaps
not for fresh forensic cases. In those cases where freshly burnt
bones have been in contact with possible contaminants—for
example personal jewelry—chemical analysis such as the pXRF
measurements carried out here (Fig. 2) should be able to reveal if
foreign elements are present. Recent studies have demonstrated
that pXRF analysis can be used to relatively accurately quantify
concentrations of trace elements including heavy metals such as
lead in human and animal bones (48, 49). As portable XRF units
are easy to bring to the field, it can be expected that various
. ESTIMATING HEATING TEMPERATURE FROM BONE COLOR 5
forensic uses will be developed for this non-destructive technique
in the near future (48). Screening for chemical contamination
may be one such use.
Conclusion
The results of this small-sample pilot study indicate that
machine-learning algorithms can be used to estimate rather accu-
rate heating temperatures from digital color measurements of
bone surfaces, especially for fresh forensic cases where the
bones have not spent long time in the ground, and provided that
the color readings are combined with chemical analysis to verify
that the bone surfaces are free from color-altering contamination.
Acknowledgments
We thank Riikka R€ais€anen at The Department of Teacher
Education, Helsinki University, for allowing JK to use the porta-
ble spectrophotometer. We thank also Martin Malve at Tartu
University for providing access to the archeological skeletal
material used in the study. The comments of two anonymous
reviewers greatly improved the paper.
Conflict of interest
The authors declare they have no conflicts of interest.
Ethics
The bone experiments were carried out in line with the Esto-
nian legislation and ethical rules for the handling of human and
animal remains.
References
1. Absolonova K, Dobisikova M, Beran M, Zocova J, Veleminsky P. The
temperature of cremation and its effect on the microstructure of the
human rib compact bone. Anthropol Anz 2012;69(4):439–60.
2. Bennett JL. Thermal alteration of buried bone. J Archaeol Sci 1999;26
(1):1–8.
3. Harbeck M, Schleuder R, Schneider J, Wiechmann I, Schmahl WW,
Grupe G. Research potential and limitations of trace analyses of cremated
remains. Forensic Sci Int 2011;204(1–3):191–200.
4. Nicholson RA. A morphological investigation of burnt animal bone and an
evaluation of its utility in archaeology. J Archaeol Sci 1993;20(4):411–28.
5. Schmidt CW, Symes SA, editors. The analysis of burned human remains.
London, U.K.: Academic Press, 2008.
6. Schmidt CW, Symes SA, editors. The analysis of burned human remains,
2nd edn. London, U.K.: Academic Press, 2015.
6 JOURNAL OF FORENSIC SCIENCES
7. Shipman P, Foster G, Schoeninger M. Burnt bones and teeth: an experi-
mental study of color, morphology, crystal structure and shrinkage. J
Archaeol Sci 1984;11(4):307–25.
8. Stiner MC, Kuhn SL, Weiner S, Bar-Yosef O. Differential burning,
recrystallization, and fragmentation of archaeological bone. J Archaeol
Sci 1995;22(2):223–37.
9. Thompson TJU, Gauthier M, Islam M. The application of a new method
of Fourier Transform Infrared Spectroscopy to the analysis of burned
bone. J Archaeol Sci 2009;36(3):910–4.
10. Ubelaker DH. The forensic evaluation of burned skeletal remains: a syn-
thesis. Forensic Sci Int 2009;183(1–3):1–5.
11. Snoeck C, Lee-Thorp JA, Schulting RJ. From bone to ash: compositional
and structural changes in burned modern and archaeological bone.
Palaeogeogr Palaeoclimatol Palaeoecol 2014;416:55–68.
12. Ellingham STD, Thompson TJU, Islam M, Taylor G. Estimating temper-
ature exposure of burnt bone – a methodological review. Sci Justice
2015;55(3):181–8.
13. Ellingham STD, Thompson TJU, Islam M. Scanning electron micro-
scopy-energy-dispersive X-ray (SEM/EDX): a rapid diagnostic tool to
aid the identification of burnt bone and contested cremains. J Forensic
Sci 2018;63(2):504–10.
14. Snoeck C, Schulting RJ, Lee-Thorp JA, Lebon M, Zazzo A. Impact of
heating conditions on the carbon and oxygen isotope composition of cal-
cined bone. J Archaeol Sci 2016;65:32–43.
15. Castillo RF, Ubelaker DH, Acosta JA, de la Fuente GA. Effects of tem-
perature on bone tissue. Histological study of the changes in the bone
matrix. Forensic Sci Int 2013;3:33–7.
16. Keough N, L’Abbe EN, Steyn M, Pretorius S. Assessment of skeletal
changes after post-mortem exposure to fire as an indicator of decomposi-
tion stage. Forensic Sci Int 2015;246:17–24.
17. Waltenberger L, Schutkowski H. Effects of heat on cut mark characteris-
tics. Forensic Sci Int 2017;271:49–58.
18. Kooi RJ, Fairgrieve SI. SEM and stereomicroscopic analysis of cut
marks in fresh and burned bone. J Forensic Sci 2013;58(2):452–8.
19. Thompson TJU, Islam M, Bonniere M. A new statistical approach for
determining the crystallinity of heat-altered bone mineral from FTIR
spectra. J Archaeol Sci 2013;40(1):416–22.
20. Thompson TJU. The analysis of heat-induced crystallinity change in
bone. In: Schmidt CW, Symes SA, editors. The analysis of burned
human remains, 2nd edn. London, U.K.: Academic Press, 2015;323–37.
21. Thompson TJU, Islam M, Piduru K, Marcel A. An investigation into the
internal and external variables acting on crystallinity index using Fourier
Transform Infrared Spectroscopy on unaltered and burned bone. Palaeo-
geogr Palaeoclimatol Palaeoecol 2011;299(1–2):168–74.
22. Walker PL, Miller KWP, Richman R. Time, temperature, and oxygen
availability: an experimental study of the effects of environmental condi-
tions on the color and organic content of cremated bone. In: Schmidt
CW, Symes SA, editors. The analysis of burned human remains. Lon-
don, U.K.: Academic Press, 2008;129–35.
23. Reidsma FH, van Hoesel A, van Os BJH, Megens L, Braadbaart F.
Charred bone: physical and chemical changes during laboratory simu-
lated heating under reducing conditions and its relevance for the study of
fire use in archaeology. J Archaeol Sci Rep 2016;10:282–92.
24. Ellingham ST, Thompson TJ, Islam M. The effect of soft tissue on tem-
perature estimation from burnt bone using Fourier Transform Infrared
Spectroscopy. J Forensic Sci 2016;61(1):153–9.
25. Bradfield J. Some thoughts on bone artefact discolouration at archaeolog-
ical sites. J Archaeol Sci Rep 2018;17:500–9.
26. Bohnert M, Rost T, Pollak S. The degree of destruction of human bodies
in relation to the duration of the fire. Forensic Sci Int 1998;95(1):11–21.
27. Dennison PE, Charoensiri K, Roberts DA, Peterson SH, Green RO.
Wildfire temperature and land cover modeling using hyperspectral data.
Remote Sens Environ 2006;100(2):212–22.
28. Okamoto K, Otake T, Miyamoto H, Honma M, Watanabe N. Burning
behavior of minivan passenger cars. Fire Saf J 2013;62(Part C):272–80.
29. Mangs J, Keski-Rahkonen O. Characterization of the fire behaviour of a burn-
ing passenger car. Part I: car fire experiments. Fire Saf J 1994;23(1):17–35.
30. Yuen ACY, Yeoh GH, Alexander R, Cook M. Fire scene reconstruction
of a furnished compartment room in a house fire. Case Studies Fire
Safety 2014;1:29–35.
31. R_Core_Team. R: a language and environment for statistical computing,
2015; https://www.R-project.org (accessed January 5, 2018).
32. Kuhn M, Wing J, Weston S, Williams A, Keefer C, Engelhardt A, et al.
caret: classification and regression training. R package version 6.0-21.
Vienna, Austria: CRAN, 2015.
33. Schliep K, Hechenbichler K. kknn: weighted k-nearest neighbors, 2016;
https://CRAN.R-project.org/package=kknn (accessed January 5, 2018).
34. Schlager S. Morpho and Rvcg – shape analysis in R. In: Zheng G, Li S,
Szekely G, editors. Statistical shape and deformation analysis. London,
U.K.: Academic Press, 2017;217–56.
35. Adler D, Murdoch D, Nenadic O, Urbanek S, Chen M, Gebhardt A,
et al. rgl: 3D visualization using OpenGL, 2017; https://CRAN.R-project.
org/package=rgl (accessed January 5, 2018).
36. Ellingham STD, Thompson TJU, Islam M. Thermogravimetric analysis
of property changes and weight loss in incinerated bone. Palaeogeogr
Palaeoclimatol Palaeoecol 2015;438:239–44.
37. Christensen AM. The impact of Daubert: implications for testimony and
research in forensic anthropology (and the use of frontal sinuses in per-
sonal identification). J Forensic Sci 2004;49(3):427–30.
38. Dirkmaat DC, Cabo LL, Ousley SD, Symes SA. New perspectives in
forensic anthropology. Am J Phys Anthropol 2008;Suppl 47:33–52.
39. Bartelink EJ, Sholts SB, Milligan CF, Van Deest TL, W€arml€ander SK.
A case of contested cremains analyzed through metric and chemical com-
parison. J Forensic Sci 2015;60(4):1068–73.
40. Crewe RJ, Stec AA, Walker RG, Shaw JE, Hull TR, Rhodes J, et al.
Experimental results of a residential house fire test on tenability: temper-
ature, smoke, and gas analyses. J Forensic Sci 2014;59(1):139–54.
41. Grevin G, Bailet P, Quatrehomme G, Ollier A. Anatomical reconstruc-
tion of fragments of burned human bones: a necessary means for forensic
identification. Forensic Sci Int 1998;96(2–3):129–34.
42. Etok SE, Valsami-Jones E, Wess TJ, Hiller JC, Maxwell CA, Rogers
KD, et al. Structural and chemical changes of thermally treated bone
apatite. J Mater Sci 2007;42(23):9807–16.
43. Beach JJ, Passalacqua NV, Chapman EN. Heat-related changes in tooth
color. In: Schmidt CW, Symes SA, editors. The analysis of burned
human remains, 2nd edn. London, U.K.: Academic Press, 2015;139–47.
44. Dozic A, Zukanovic A, Bajsman A, Secic S, Petaros A. Spectrophoto-
metric evaluation of color alterations of teeth exposed to different condi-
tions in time. Acta Stomatol Croat 2011;45(4):247–57.
45. Carroll EL, Smith M. Burning questions: investigations using field exper-
imentation of different patterns of change to bone in accidental vs delib-
erate burning scenarios. J Archaeol Sci Rep. In press. http://doi.org/10.
1016/j.jasrep.2018.02.001.
46. Waterhouse K. The effect of weather conditions on burnt bone fragmen-
tation. J Forensic Leg Med 2013;20(5):489–95.
47. Waterhouse K. Post-burning fragmentation of calcined bone: implications
for remains recovery from fatal fire scenes. J Forensic Leg Med 2013;20
(8):1112–7.
48. Byrnes JF, Bush PJ. Practical considerations in trace element analysis of
bone by portable X-ray fluorescence. J Forensic Sci 2016;61(4):1041–5.
49. Specht AJ, Parish CN, Wallens EK, Watson RT, Nie LH, Weisskopf MG.
Feasibility of a portable X-ray fluorescence device for bone lead measure-
ments of condor bones. Sci Total Environ 2018 Feb;15(615):398–403.
Supporting Information
Additional Supporting Information may be found in the online
version of this article:
Table S1. Sample properties
Figure S1. Sample 1 (pig bone with flesh, 400°C).
Figure S2. Sample 2 (pig bone with flesh, 600°C).
Figure S3. Sample 3 (pig bone with flesh, 800°C).
Figure S4. Sample 4 (pig bone with flesh, 1000°C).
Figure S5. Sample 5 (defleshed pig bone, 400°C).
Figure S6. Sample 6 (defleshed pig bone, 600°C).
Figure S7. Sample 7 (defleshed pig bone, 800°C).
Figure S8. Sample 8 (defleshed pig bone, 1000°C).
Figure S9. Sample 9 (dry human bone, 400°C).
Figure S10. Sample 10 (dry human bone, 600°C).
Figure S11. Sample 11 (dry human bone, 800°C).
Figure S12. Sample 12 (dry human bone, 1000°C).
Figure S13. Samples 13 (pig, defleshed) and 14 (dry human
bone).
Figure S14. 3D plot of the SCI/L*a*b* color values.