Comparison of Narrow-Band Reflectance Spectroscopy and Tristimulus Colorimetry for Measurements of Skin and Hair Color in Persons of Different Biological Ancestry

AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 112:17–27 (2000)
Comparison of Narrow-Band Re?ectance Spectroscopy and
Tristimulus Colorimetry for Measurements of Skin and Hair
Color in Persons of Different Biological Ancestry
MARK D. SHRIVER* AND ESTEBAN J. PARRA
Department of Anthropology, Pennsylvania State University, University
Park, Pennsylvania 16802
KEY WORDS
narrow-band spectrometer; tristimulus colorimeter;
skin and hair pigmentation
ABSTRACT
We have used two modern computerized handheld re?ectome-
ters, the Photovolt ColorWalk colorimeter (a tristimulus colorimeter; Photovolt,
UMM Electronics, Indianapolis, IN) and the DermaSpectrometer (a specialized
narrow-band re?ectometer; Cortex Technology, Hadsund, Denmark), to compare
two methods for the objective determination of skin and hair color. These in-
struments both determine color by measuring the intensity of re?ected light of
particular wavelengths. The Photovolt ColorWalk instrument does so by shining
a white light and sensing the intensity of the re?ected light with a linear
photodiode array. The ColorWalk results can then be expressed in terms of
several standard color systems, most importantly, the Commission International
d’Eclairage (CIE) Lab system, in which any color can be described by three
values: L*, the lightness; a*, the amount of green or red; and b*, the amount of
yellow or blue. Instead of a white light and photodiodes, the DermaSpectrometer
uses two light-emitting diodes (LEDs), one green and one red, to illuminate a
surface, and then it records the intensity of the re?ected light. The results of
these readings are expressed in terms of erythema (E) and melanin (M) indices.
We measured the unexposed skin of the inner upper arm, the exposed skin of the
forehead, and the hair, of 80 persons using these two instruments. Since it is
important for the application of these measures in anthropology that we under-
stand their relationship across a number of different pigmentation levels, we
sampled persons from several different groups, namely, European Americans
(n
55), African Americans (n
9), South Asians (n
7), and East Asians (n
9). In these subjects, there is a very high correlation between L* and the M index
for the inner arm (R2
0.928, P
0.001), the forehead (R2
0.822, P
0.001),
and the hair (R2
0.827, P
0.001). The relationship between a* and the E
index is complex and dependent on the pigmentation level. We conclude that
while both types of instruments provide accurate estimates of pigment level in
skin and hair, measurements using narrow-band instruments may be less af-
fected by the greater redness of certain body sites due to increased vasculariza-
tion. Am J Phys Anthropol 112:17–27, 2000.
© 2000 Wiley-Liss, Inc.
Over the past 50 years, skin pigmentation
levels have been objectively studied using
re?ectance spectroscopy. The two instru-
Grant sponsor: NSF; Grant number: 9610332.
ments that have been most widely used in
*Correspondence to: Mark D. Shriver, Department of Anthro-
anthropological studies are the E.E.L. in-
pology, Pennsylvania State University, 409 Carpenter Building,
University Park, PA 16802. E-mail: mds17@psu.edu
strument (Evans Electroselenium Co., Hal-
Received 16 December 1998; accepted 13 January 2000.
© 2000 WILEY-LISS, INC.

---

18
M.D. SHRIVER AND E.J. PARRA
Fig. 1.
A: Photovolt ColorWalk instrument being used to measure the inner arm of a subject (typically
the subject would be seated, as outlined in Subjects and Methods). B: DermaSpectometer, in a woman’s
hand.
stead, Essex, UK) and the Photovolt line of
?ectance technologies that have both been
instruments (UMM Electronics, Indianapo-
widely used in the ?eld of dermatology
lis, IN). Both of these instruments use col-
(Takiwaki et al., 1994, Fullerton et al.,
ored ?lters to measure the percent re?ec-
1996). The ColorWalk is a handheld tris-
tance of light of various wavelengths. The
timulus colorimeter, which uses photodiode
E.E.L. instruments have a series of nine
arrays in lieu of colored ?lters to measure
colored ?lters, while the original Photovolt
the intensity of particular wavelengths of
systems have six available ?lters. Although
light (see Fig. 1A). Tristimulus colorimetry
there is much data in the literature on pig-
was developed as a means of objectively rep-
mentation
levels
reported
using
these
resenting color in a manner analogous to
E.E.L. and Photovolt ?lter-based re?ecto-
the way the eye perceives color (Hunter,
meters (reviewed in Robins, 1991), newer
1942). The re?ectance level of light through
technologies have led to smaller and more
three particular broad wavelength ?lters
accurate devices.
(photodiode arrays on newer instruments) is
We compared two of these newer instru-
determined. Color parameters are then de-
ments for measurements of skin and hair
?ned by the levels of and differences
pigmentation, the Photovolt ColorWalk col-
among the re?ectance levels of these three
orimeter (Photovolt, UMM Electronics, In-
?lters. The most commonly used color pa-
dianapolis, IN) and the DermaSpectrometer
rameters are the Commission International
(Cortex Technology, Hadsund, Denmark).
d’Eclairage (CIE) L*a*b* system estab-
These instruments apply two different re-
lished in 1976. In the CIELab color system,

---

MEASUREMENT OF SKIN AND HAIR PIGMENTATION
19
any color can be represented by three vari-
SUBJECTS AND METHODS
ables: L*, the lightness-darkness axis; a*,
Subjects were seated for 5 min with their
the red-green axis; and b*, the blue-yellow
arms at their sides prior to being measured.
axis, which can be plotted in three-dimen-
While waiting to be measured, they were
sional space. Tristimulus colorimeters like
asked a number of questions, namely, their
the ColorWalk and the commonly used Mi-
name, date and place of birth, biological an-
nolta Chroma Meter 200 and 300 series ma-
cestry and ethnicity, and whether they had
chines (Minolta Co., Osaka, Japan) are usu-
recently colored their hair. Eighty persons
ally able to report color values for a number
participated in the study: 55 of European
of other color systems as well.
ancestry (Europeans or European Ameri-
The DermaSpectrometer is a different
cans), 9 of African ancestry (African or Af-
type of instrument that was developed spe-
rican Americans), 7 South Asians (India and
ci?cally for measurements of skin pigments,
Pakistan), and 9 East Asians (China, Tai-
namely hemoglobin and melanin. The Der-
wan, Philippines, and Korea). All persons
maSpectrometer (see Fig. 1B) and related
were current residents of Pittsburgh, PA or
instruments, namely the Erythema/Mela-
the surrounding areas, and all were mea-
nin Meter (DiaStron, DiaStron Ltd., Hamp-
sured during the second and third weeks of
shire, UK) and the Mexameter (Courage
August 1998. Before use, both instruments
Khazaka), are based on the work of Diffey et
were calibrated using the speci?c white and
al. (1984). Hemoglobin and melanin are the
black calibration standards supplied by the
principal pigments visible in the skin: he-
manufacturers. Measurements were ?rst
moglobin in the blood of the capilaries in the
taken with the DermaSpectrometer of the
dermis, and melanin in the keratinocytes
following sites in this order: inner upper
and melanocytes of the epidermis. Hemoglo-
right arm, inner upper left arm, forehead,
bin and melanin both absorb much light at
and hair. Three measurements were taken
the lower wavelengths, with hemoglobin
of each site, moving the measurement head
showing a large peak in the green wave-
a few centimeters between measurements.
lengths and then a sharp drop-off, absorbing
As has been suggested, care was taken not
very little light in the red wavelengths,
to apply too much pressure on the measure-
which is why blood is red. Melanin, both in
ment head of the DermaSpectrometer, since
vivo and in vitro, shows absorbance of light
doing so could occlude blood from the region
of all wavelengths, essentially a ?at line
being measured (Fullerton et al., 1996).
sloping down from the lower wavelengths to
Measurements of the hair were only
the higher wavelengths (Kollias and Baqer,
taken on those persons who said they did
1985). Based on these differences in the
not color or bleach their hair. For these mea-
spectral curves of hemoglobin and melanin,
surements, we carefully pressed down the
Diffey et al. (1984) suggested that the re?ec-
hair throughout the parietal region, making
tance of narrow-band light in the red spec-
sure that the scalp was not visible, and then
trum would yield reasonable estimates of
applied the measurement head to this area.
the melanin content of a persons skin, fol-
As with the skin, three measurements were
lowing the equation
taken in different areas around the parietal,
to get a better average of the total level of
M
log10 (1/% red reflectance).
hair pigmentation. After measuring with
the DermaSpectrometer, we measured the
The degree of skin redness or erythema can
same body sites in the same manner with
be calculated by subtracting the absorbance
the Photovolt ColorWalk instrument (see
due to melanin from the absorbance of the
Fig. 1A for how a measurement is taken).
green ?lter and is calculated as
The ColorWalk allows the user to select
among the two commonly used reference il-
E
log
luminants and observation angles in tris-
10 (1/% green reflectance)
timulus colorimetry. As previously sug-
log10 (1/% red reflectance).
gested by Weatherall and Coombs (1992),

---

20
M.D. SHRIVER AND E.J. PARRA
TABLE 1. Summary of the results for melanin content (M and L*), and hemoglobin content (E and a*), in the
groups included in the present study1
African Americans
East Asians
European Americans
South Asians
(n
9)
(n
9)
(n
55)
(n
7)
Mean
s.d.
CV
Mean
s.d.
CV
Mean
s.d.
CV
Mean
s.d.
CV
M
56.62
14.78
26%
31.79
2.39
8%
30.50
2.82
9%
37.13
4.19
11%
L*
47.50
9.95
21%
67.30
1.62
2%
69.86
3.26
5%
61.90
3.76
6%
E
2.68
5.08
190%
6.92
0.83
12%
6.64
1.20
18%
6.81
0.75
11%
a*
11.17
2.25
20%
12.46
1.21
10%
12.72
1.89
15%
13.25
0.77
6%
1 s.d., standard deviation; CV, coef?cient of variation (s.d./mean)*100.
we used the most recently established stan-
ropean ancestry have the lightest skin, high
dards, namely an observation angle of 10°
L*, and low M. Persons of East Asian ances-
and the D65 light source.
try have pigment levels which cluster at the
Both the ColorWalk and the DermaSpec-
lower end of the European distribution. Per-
trometer perform internal calculations to
sons of South Asian ancestry (Indian and
convert the raw re?ectance readings to the
Pakistani) are the next darkest group and
output variables, namely the E and M indi-
overlap with some of the African-American
ces for the DermaSpectrometer, and L*, a*,
subjects, who have the darkest skin as well
and b*, in addition to other color systems for
as the widest variance in pigmentation
the ColorWalk. Linear and nonlinear re-
level.
gression lines were calculated using stan-
Figure 3 shows the relationship between
dard statistical software packages.
L* and M for measurements of the forehead
RESULTS
of the 80 subjects studied. As with the mea-
surements of the inner arm, there is a clear
Skin
correlation between L* and the M index
In Table 1, we summarize the results ob-
(R2
0.870, exponential regression, Table
tained using the ColorWalk and the Der-
2). It is also clear that the relationship be-
maSpectrometer for measuring the melanin
tween L* and M for the forehead is not as
content (M and L*) and hemoglobin content
strong as for measurements of the inner
(E and a*) in a sample of 80 individuals
upper arm.
from different ethnic groups. Figure 2 shows
Both a* and the E index have been used
the relationship between the L* level, the
by dermatologists as indicators of the de-
lightness in the CIELab color system, and
gree of skin redness or erythema (Diffey et
M, the melanin index, as measured at the
al., 1984; Seitz and Whitmore, 1988; Serup
inner upper arm, for the 80 persons studied.
and Agner, 1990; Westerhof et al., 1990;
There is a clear relationship between these
Takiwaki et al., 1994). Our data (not shown)
two values: as L* decreases, indicating less
indicate that the relationship between a*
lightness and less re?ectance, the M index
and E is complex, and dependent on the
increases, indicating higher melanin con-
level of pigmentation. There is a clear posi-
tent in the skin. The linear equation for this
tive correlation between a* and E in persons
line is
with low melanin content (M
40). How-
L*
94.15
8132M
ever, heavily pigmented persons (M
40)
show a much lower correlation and a much
(R2
0.928, P
0.001). However, it is ap-
less steep relationship between a* and E
parent from the ?gure that the relationship
than lightly pigmented persons.
between L* and M is not strictly linear,
Given this complex relationship between
especially at high melanin concentrations,
a* and E, it is important to understand both
and a slightly better ?t is obtained using an
how E varies with respect to M, and how a*
exponential equation (R2
0.962, Table 2).
varies with respect to L*. Figures 4 and 5
Population differences in pigmentation level
show the results of these comparisons for
are evident in Figure 2. The subjects of Eu-
measurements of the inner upper arm.

---

MEASUREMENT OF SKIN AND HAIR PIGMENTATION
21
Fig. 2.
Relationship between L* and the M index for the inner arm average of all persons measured.
L* was measured using the ColorWalk, and the M index using the Dermaspectrometer, as described in
Subjects and Methods. Also indicated is the biological ancestry of the persons measured: Europeans and
European Americans (open circle), East Asians (solid diamond), South Asians ( ), and Africans and
African Americans (open square).
TABLE 2. Relationship between the parameters used
but in persons with M values higher than 40
for estimating melanin content (M and L*), as
(high pigmentation levels), a signi?cant
measured in the inner arm, forehead and hair
negative correlation is observed (R2
0.976,
Best ?t
R2
P
0.001). On the contrary, the plot of L*
Inner arm
vs. a* shows a clear negative correlation in
L* vs. M
L*
110.1e 0.0151M
0.9617
the range corresponding to L* values higher
M vs. L*
M
300.53
63.646Ln(L*)
0.9617
Forehead
than 60 (low pigmentation levels, R2
L* vs. M
L*
94.032e 0.0127M
0.8705
0.565, P
0.001), and this correlation is not
M vs. L*
M
317.06
68.647Ln(L*)
0.8705
Hair
observed in the region of L* values lower
L* vs. M
L*
167.45
30.494Ln(L*)
0.8852
than 60 (high pigmentation levels, R2
M vs. L*
M
222.8e 0.029L*
0.8852
0.0000, n.s.).
Figure 6 shows a histogram of the popu-
lation distribution of the L* inner upper
Again, the relationship between these pa-
arm measures for persons of European an-
rameters differs, depending on the level of
cestry. We constructed this histogram using
pigmentation, and there are important dif-
a bin width of two L* units, as suggested by
ferences between the E vs. M plot (Fig. 4)
Weatherall and Coombs (1992). The mini-
and the plot of a* and L* (Fig. 5). There is no
mum and maximum L* values were 58.7
signi?cant correlation between E and M in
and 75.3, respectively, and the mode was
the groups showing M values lower than 40
observed in the range of 71–73 (average L*,
(low pigmentation levels, R2
0.0515, n.s.),
69.9). The L* distribution is highly skewed

---

22
M.D. SHRIVER AND E.J. PARRA
Fig. 3.
Relationship between L* and the M index for the forehead average of all persons measured.
L* was measured using the ColorWalk and the M index using the Dermaspectrometer as described in
Subjects and Methods. Also indicated is the biological ancestry of the persons measured: Europeans and
European Americans (open circle), East Asians (solid diamond), South Asians ( ), and Africans and
African Americans (open square).
and very similar to another report from the
rithmic regression, R2
0.885, P
0.001).
literature on the CIE color system in Euro-
Notable is the limited variability in the hair
peans (Weatherall and Coombs, 1992). The
pigment level of non-European persons. Eu-
M index showed a very similar distribution,
ropeans demonstrate hair re?ectance levels
but it was skewed to the right instead of the
that span the range of variation observed.
left (data not shown). Both of these distri-
DISCUSSION
butions are similar in shape to the log-nor-
mal distribution.
In the present study, we used two hand-
held re?ectometers to compare two methods
Hair
for the determination of skin and hair color:
In addition to studies of the skin, re?ec-
a narrow-band spectrometer (DermaSpec-
tometers have been used to objectively
trometer) and a tristimulus colorimeter
quantify the color and degree of pigmenta-
(ColorWalk). We sampled persons of differ-
tion of the hair (Sunderland, 1956; Little
ent biological ancestry in an effort to span
and Wolf, 1981). Figure 7 shows our results
the range of variability in human pigmen-
for the measurement of the hair of the 64
tation levels. Previous studies were per-
persons in this survey who did not color or
formed by dermatologists in efforts to ex-
bleach their hair. Figure 7 shows the rela-
plore the relationships between the two
tionship between L* and the M index for
types of instruments that we consider in
these persons. There is a clear correlation
this study (e.g., Takiwaki et al., 1994; Ful-
between the two measures L* and M (loga-
lerton et al., 1996). One important consider-

---

MEASUREMENT OF SKIN AND HAIR PIGMENTATION
23
Fig. 4.
Relationship between M index and E index for inner upper arm average. Biological ancestry
of the persons measured: Europeans and European Americans (open circle), East Asians (solid diamond),
Southwest Asians ( ), and Africans and African Americans (open square)
ation regarding these studies is that the
signi?cant negative correlation between M
focus in the dermatological literature has
and L* (R2
0.314, P
0.001). In our larger
been most often on the measurement of er-
sample of persons of European ancestry
ythema, the reddening of the skin in re-
(n
55), we observed a higher correlation
sponse to irritation from ultraviolet light or
(upper inner arm, R2
0.624, P
0.001),
other causes; researchers have not consid-
and this correlation is even higher when we
ered the whole range of human pigmenta-
include individuals representative of all
tion levels. In contrast, in the anthropolog-
ethnic groups (R2 0.928, P
0.001). Both
ical literature, the main focus has been an
L* and M seem to be highly correlated with
objective determination of the melanin con-
the melanin content of the skin, but the
tent of the skin (e.g., Korey, 1980; Releth-
melanin index (M), which has been speci?-
ford et al., 1983; Robins, 1991).
cally designed by taking into account the
Melanin and hemoglobin are the two dom-
absorbance spectrum of melanin and hemo-
inant chromophores of the skin. The mela-
globin, may likely be a better indicator of
nin index, M (measured by means of a nar-
the melanin content than L*. The value of
row-band spectrometer), and the L* value of
L* is highly dependent on the re?ected
the CIELab space (measured by means of a
green light, where in fact hemoglobin has its
tristimulus colorimeter) have been used by
peak absorption, so that the L* value is not
dermatologists as indicators of the melanin
just a function of the melanin concentration.
content of the skin. A previous study of 10
This is clearly indicated by the signi?cant
individuals of Caucasian ancestry (Taki-
correlation observed between L* and a* in
waki et al., 1994) reported a moderate but
low-pigmented persons (Fig. 5). The more

---

24
M.D. SHRIVER AND E.J. PARRA
Fig. 5.
Relationship between L* and a* for inner upper arm average. Biological ancestry of the
persons measured: Europeans and European Americans (open circle), East Asians (solid diamond),
Southwest Asians ( ), and Africans and African Americans (open square).
red the skin is, the lower the L* is for these
complex relationship between a* and E, and
persons. This same trend was observed in
the substantial differences in what both pa-
previous studies (Takiwaki et al., 1994, Ful-
rameters are measuring. This is not surpris-
lerton et al., 1996, Takiwaki, 1998), and
ing, if we take into account the different
may be responsible for the decreased corre-
methodologies upon which the DermaSpec-
lation of L* and M in some comparisons
trometer and the ColorWalk are based.
(e.g., forehead, where there is increased vas-
The comparisons of M vs. E and L* vs. a*
cularization and sometimes quite dramatic
further stress these differences. M and E
intraindividual variability).
are not correlated in low-pigmented groups
Both a* and the E index have been used
(Fig. 4). Thus, E is a good indicator of hemo-
by dermatologists as indicators of the de-
globin content in those groups, and behaves
gree of skin redness or erythema. A high
independently of M. However, in groups
positive correlation has been observed be-
characterized by high melanin content,
tween a* and E in a small sample of 10
there is a signi?cant negative correlation
Caucasian male volunteers, in whom mea-
between E and M (R2
0.990, P
0.001),
sures were taken at 23 different anatomical
indicating that E is no longer a linear func-
sites (R2
0.846, P
0.001). We observed
tion of hemoglobin content. This is intrinsi-
in our sample of 55 persons of European
cally due to the methodological principle
ancestry a lower, but still signi?cant corre-
upon which the calculation of the melanin
lation (R2
0.379, P
0.001). However,
and erythema indices is based. The melanin
this correlation is not signi?cant in highly
index is calculated based on the amount of
pigmented persons (M
40), indicating the
red light re?ected, given that hemoglobin

---

MEASUREMENT OF SKIN AND HAIR PIGMENTATION
25
Fig. 6.
Population distribution of inner upper arm L* levels for Europeans.
does not absorb in this spectrum, and con-
termination of skin and hair color. Both in-
sequently does not interfere in the calcula-
struments operate based on two different
tion of the melanin content. On the con-
principles. Our results indicate that both
trary, both melanin and hemoglobin absorb
types of instruments provide good and cor-
light in the green part of the spectrum, and
related estimates of pigment level in skin
when high concentrations of melanin are
and hair. However, we ?nd that measure-
present, its effect on the amount of green
ments
using
narrow-band
instruments
light re?ected is substantial, and the E
(DermaSpectrometer, in this study) appear
value is no longer linearly related to the
to be less affected by the increased redness
hemoglobin content (Takiwaki et al., 1994).
of certain body sites due to increased vascu-
When we consider the relationship between
larization. It is also evident that E and a*,
L* and a*, exactly the opposite trend is ob-
the parameters normally employed for eval-
served. These values are highly correlated
uating the degree of erythema, show a com-
when the amount of melanin is low, but
plex relationship, which is dependent on the
there is no signi?cant correlation when the
melanin content of the skin.
melanin level is high (Fig. 5).
It is necessary to point out that there are
a number of portable and handheld true
CONCLUSIONS
spectrophotometers
currently
available
With recent technical advances in the
(e.g., the spectrophotometer CM-500 and
?eld of colorimetry and photometry, new in-
CM-2000 series by Minolta, Japan, and the
struments have become available which of-
Micro?ash series by Datacolor Interna-
fer substantial advantages over previously
tional, Charlotte, NC). Although somewhat
used instruments, in terms of precision,
more expensive (2–3 times the cost of the
portability, and ease of use. In this paper,
instruments we used in this study; $4,995
we used two new handheld re?ectometers,
for the ColorWalk, and $4,500 for the Der-
the Photovolt ColorWalk (a tristimulus col-
maSpectrometer),
these
spectrophotom-
orimeter) and the DermaSpectrometer (a
eters are more versatile in that they mea-
specialized narrow-band re?ectometer), to
sure the re?ectance at regular intervals
compare two methods for the objective de-
across the spectrum of visible light (400 –

---

26
M.D. SHRIVER AND E.J. PARRA
Fig. 7.
Relationship between the M index and L* for hair. Biological ancestry of the persons mea-
sured: Europeans and European Americans (open circle), East Asians (solid diamond), South Asians ( ),
and Africans and African Americans (open square).
700 nm). Both have internal software,
primary aim is the determination of skin
which computes values for different color
pigmentation due to melanin, the Der-
systems, including CIELab. Since these in-
maSpectrometer would likely be the pre-
struments can display and record re?ec-
ferred instrument, as the M index obtained
tance levels at narrow intervals across the
with this apparatus is less confounded by
visual spectrum, one could also calculate
levels of hemoglobin and thus better re?ects
the E and M indices using these instru-
the amount of melanin present in the skin.
ments. Additionally, having re?ectance in-
This is especially true if comparisons of dif-
formation across the whole visible spectrum
ferent body sites will be made (see also
would make it possible to better distinguish
Lock-Andersen and Wulf, 1998). In lightly
the effect of diverse skin chromophores
pigmented persons, differences in degree of
(melanin, oxy- and deoxy-hemoglobin, bili-
vascularization of different body sites make
rubin), and even different forms of melanin
the use of L* troublesome for these types of
(high and low molecular weight melanin,
comparisons.
Kollias and Baqer, 1987, 1988; and poten-
ACKNOWLEDGMENTS
tially, eumelanin and pheomelanin).
In summary, both the DermaSpectrom-
We thank all of the participants in this
eter and the ColorWalk provide accurate
study for their willingness and cooperation.
and objective measurements of skin and
This study was funded in part by NSF grant
hair color, which are highly correlated (Ful-
9610332 to M.S. We also thank Dr. Gary
lerton et al., 1996; Takiwaki, 1998). In an-
Grove of cyberDERM, Inc. (Media, PA) for
thropological and genetic studies where the
the loan of the DermaSpectrometer, as well

---

Document Outline

• Fig.1.
• SUBJECTS AND METHODS
  • TABLE 1.
• RESULTS
  • Fig. 2.
  • TABLE 2.
  • Fig. 3.
• DISCUSSION
  • Fig. 4.
  • Fig. 5.
  • Fig. 6.
• CONCLUSIONS
  • Fig. 7.
• ACKNOWLEDGMENTS
• LITERATURE CITED

---