AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 125:292302 (2004)

Geographic Distribution of Environmental Factors

Inuencing Human Skin Coloration
George Chaplin*
Department of Anthropology, California Academy of Sciences, San Francisco, California 94118
KEY WORDS
integument; spectrometry; reectometry; pigmentation; sunlight; adaptation;
photobiology; race; GIS
ABSTRACT
Skin coloration in indigenous peoples is
strongly related to levels of ultraviolet radiation (UVR). In
this study, the relationships of skin reectance to seasonal
UVR levels and other environmental variables were investigated, with the aim of determining which variables contributed most signicantly to skin reectance. The UVR
data recorded by satellite were combined with environmental variables and data on human skin reectance in a
geographic information system (GIS). These were then
analyzed visually and statistically through exploratory
data analysis, correlation analysis, principal components
analysis, least-squares regression analysis, and nonlinear

This paper presents an investigation of the specic nature of environmental inuences on the evolution of human skin coloration. It builds on previous research (Jablonski and Chaplin, 2000) in which
a clear causal relationship between average annual
ultraviolet radiation (UVR) and skin reectance was
established.
PREVIOUS RESEARCH ON GLOBAL
DISTRIBUTION OF ENVIRONMENTAL FACTORS
Few workers have looked systematically at the
environmental factors identied as possible inuences on skin color. Most have used latitude as a
surrogate for UVR because it has a strong correlation with skin color, with an r-value of around 0.8
(Relethford, 1997; Roberts, 1977; Roberts and
Kahlon, 1976; Tasa et al., 1985). Latitude is a poor
surrogate for UVR because other environmental
variables are correlated with it. Further, the correlation between UVR and latitude will be different
between the hemispheres due to orbital effects on
solar intensity (Chaplin and Jablonski, 1998;
Relethford, 1997). The rst use of direct measurements of UVR collected using a satellite (Jablonski
and Chaplin, 2000) showed the highest correlation
to annual average UVR and skin reectance at rvalues above 0.95, but no other environmental factors were taken into account.
Walter (1958) was the rst to use an estimate of
UVR to demonstrate a correlation between UVR and
skin color. He obtained his estimate by calculating
UVR at the Equator and correcting for latitude,

2004 WILEY-LISS, INC.

techniques. The main nding of this study was that the

evolution of skin reectance could be almost fully modeled
as a linear effect of UVR in the autumn alone. This linear
model needs only minor modication, by the introduction
of terms for the maximum amount of UVR, and for summer precipitation and winter precipitation, to account for
almost all the variation in skin reectance. A further
signicant nding was that the effect of summer UVR
seems to reach a threshold beyond which further adaptation is difcult. Am J Phys Anthropol 125:292302, 2004.

2004 Wiley-Liss, Inc.

clouds, moisture, vegetation, and albedo. He then

used statistical and visual map comparisons to conclude that there was a correlation of UVR with skin
color, and that it was modied by selection for thermoregulation.
Roberts and Kahlon (1976) showed that, in addition to the dominance of latitude, there were also
correlations of skin color with minimum, mean, and
maximum temperature, with minimum and maximum humidity, and with altitude. They combined
these variables in a multiple additive regression.
However, they posited no direct effect of these variables on skin color other than as modiers for
ground UVR estimation and in thermoregulation.
Roberts (1977) drew similar conclusions to Walter
(1958) and expanded them in a review. The skin
measurements Roberts and Kahlon (1976) used
were assumed to be representative of the geographic
area where they were measured, and not of the
whole of the area that the subjects occupied. The
variables of Roberts and Kahlon (1976) were more of
a point-by-point estimate than an area-wide average
*Correspondence to: George Chaplin, Department of Anthropology,
California Academy of Sciences, Golden Gate Park, San Francisco, CA
94118. E-mail: gchaplin@calacademy.org
Received 10 April 2002; accepted 3 January 2003.
DOI 10.1002/ajpa.10263
Published online 4 May 2004 in Wiley InterScience (www.
interscience.wiley.com).

ENVIRONMENT AND HUMAN SKIN COLORATION

(the only practical solution prior to the use of a

geographic information system (GIS)).
This study sought to replicate the work of Roberts
and Kahlon (1976), using directly measured UVR
instead of latitude. UVR and other environmental
variables were taken for the whole of the area that a
particular human population occupied. This approach was impossible before the advent of GIS.
In this study, the relationship of environmental
effects variables to skin color reectance was examined. Prior to this study, important variations in
environmental factors, such as the relative importance of maximum UVR and minimum UVR, the
temporal evenness of UVR dose, and the seasonal
aspect to the distribution of UVR throughout the
year, had not been explored in relation to skin color.
HYPOTHESES TESTED
Several environmental factors, especially those
related to UVR dosage and intensity, have been
proposed as being causally related. First, are these
hypotheses concerned with direct UVR effects and to
disease or trauma processes and skin coloration?
Previous researchers proposed that maximum UVR,
average UVR, or temporal patterns of UVR dose are
related to incidents of sunburning. Radiation-induced damage is known to cause cancer (melanoma
and basal- and squamous-cell carcinoma; Buzzell,
1993; Kollias et al., 1991; Longstreth et al., 1998;
Potten et al., 1993; Zihlman and Cohn, 1988) and to
damage sweat glands (Daniels, 1964; Pandolf et al.,
1992). Both UVB and long-wave UVR (UVA) can
cause deleterious effects in the skin, with those from
UVA being mediated by reactive oxygen species
(Cleaver and Mitchell, 2000; Roberts, 1977) This led
to formulation of the hypothesis that reduced skin
reectance is mediated directly by exposure to UVR
through reduced reproductive tness. Another cause
of reduced skin reectance, particularly in the tissue
(blue) reectance range, is the thickening of the
stratum corneum caused directly by UVR (Kollias et
al., 1991).
Second are hypotheses that suggest that UVR mediates the bodys production and storage of vital
nutrients. For example, a highly seasonal distribution of UVR throughout the year (especially where
winter levels are greatly reduced or nonexistent)
causes problems for vitamin D production and storage (Holick, 1981; Loomis, 1967; Murray, 1934; Jablonski and Chaplin, 2000). Some vital nutrients
stored in the body are subject to photolysis. Folate
may be destroyed by strong summer levels of UVA,
and in areas with sunny winters that have moderate
to high levels of long-wave UVR, both folate and
vitamin D are destroyed (Jablonski, 1992, 1999; Jablonski and Chaplin, 2000; Jones et al., 1998; Webb
and Holick, 1988). These theories led to formulation
of the hypothesis that metabolic processes related to
direct solar radiation are responsible for variation of
tness, leading to a correlation between skin reectance and UVR.

293

Third are hypotheses that involve non-UVR agencies, such as forest cover, altitude, temperature, and
precipitation. These variables have been considered
in theories of skin coloration and were included in
this study. It was suggested that the intensity of
melanin pigmentation is related to the need for camouage in forest environments (Cowles, 1959). A
number of theories revolve around thermoregulation (Baker, 1958; Hamilton and Heppner, 1967;
Razi et al., 1980) or the prevention of cold injury
(Candler and Ivey, 1997; Post et al., 1975; Steegmann, 1967; Taylor, 1992). Precipitation was included because Glogers rule states that warmblooded animals living in warm and humid places
are more heavily pigmented than those in cool, dry
areas. Precipitation may also be indicative of the
need for shelter-seeking and cultural modication of
the UVR regime, leading to pleiotropic decline in
skin coloration (Brace, 1963; Daniels, 1964; Deol,
1975). This led to formulation of the hypothesis that
variables other than direct UVR are responsible for
skin reectance and are only correlated to UVR or
latitude by solar-driven processes such as vegetation, temperature, and precipitation.
MATERIALS
Skin reflectance data
The skin reectance data set used in this study
was essentially the same as that of Jablonski and
Chaplin (2000), but with a few new data points
added (see Appendix). The database comprises samples for reectance at 425 nm (blue lter), 545 nm
(green lter), and 685 nm (red lter), at a site on the
upper inner arm.
Environmental data
The UVR data used in this study represent the
relative daily areal exposure to the ultraviolet minimum erythemal dose (UVMED), i.e., the amount of
UVR effective in causing a barely perceptible reddening of light skin. The values for UVMED used in
this study were derived from readings taken by the
NASA total ozone mapping spectrometer (TOMS)
that was own aboard the Nimbus-7 satellite between 1978 1993 (Herman and Celarier, 1996). In
measuring UVR strength, all attenuating inuences
(humidity, rainfall, temperature, or elevation) were
already discounted, so that any effect that these
variables have, as modeled in this study, was the
result of their own action rather than how they were
affecting UVR transmittance. The treatment of
TOMS data was described in Jablonski and Chaplin
(2000). From the NASA TOMS data set, the following UVMED readings were derived: annual mean
for maximum, minimum, and range, and seasonal
average for winter, spring, summer, and autumn.
Data for global environmental variables were
taken from the unied data set, ArcAtlas: Our Earth
(ESRI, 1996). These data were provided at scales of
25 million to 1, and 10 million to 1, for selected

294

G. CHAPLIN

places. The following environmental data were used:

1) total solar radiation measured in millijoules (mJ)
per square mile; 2) air temperature readings for
yearly average, and for January and July measured
in degrees Celsius; 3) average length of the frost-free
period, dened as number of days per year; 4) total
number of days with snow on the ground (in days
per year); 5) average amount of all precipitation
reaching the ground (yearly average, and for months
of January and July, all measured in millimeters); 6)
reconstructed ground-cover polygons in the ArcWorld data, for the cover type as it existed before
human disturbance (Ground cover was classied as
desert, forest, or other. Where an area had more
than one ground-cover class type, it was assigned
the majority type, or where it was judged that there
was not a clear majority, it was designated as other.);
and 7) a classication of land surface relief in the
following terms: glaciers, high mountains, high
plains, low mountains, low plains, middle mountains, plateau, and tablelands.
In order that environmental variables be explored
for both hemispheres at the same time, it was decided to join them by seasons, which are offset by 6
months. To do this, each corresponding month was
joined together, e.g., April, the rst month of Southern Hemisphere autumn, was joined to the Northern
Hemispheres October. These were then averaged
over 3-month periods and renamed for their corresponding season.
METHODS
Statistical treatment
For each skin reectance sampled area, the cells
corresponding to it were extracted from the database by spatial join or overlay. These data were
placed into the S-Plus statistical data package for
further analysis (MathSoft, 1999).
Exploratory data analysis
The variables were studied for normality, kurtosis, and skew. Exploratory data analysis was conducted to detect and remove outliers and data anomalies.
Correlation
The skin reectance data set was used to estimate
the new variables, named Melanin Reectance, Tissue Reectance, and Blood Reectance. They were
so named because these three main light-absorbing
pigments have peaks in the red, blue, and green
wavelengths, respectively. These variables were derived, respectively, from the readings of either the
Evans Electroselenium Company Ltd. (EEL) or Photovolt reectometers for the following pairs of reectance lters: the E685.609-P670.Red red pair, the
E425.601-P420.Blue blue pair, and the E545.605P525.Green green pair.
The correlations within data were checked. The
correlation table included the raw values and the

trimmed and grouped data to show that this data

selection method had little effect on the overall pattern of variation within the data.
Principal components analysis
It was rst hoped to use stepwise regression to
identify the variables that inuenced skin coloration
in the manner of Roberts and Kahlon (1976). However, the existence of a very high amount of collinearity in the variables made stepwise regression
inadvisable (Johnston, 1978). Therefore, it was necessary to use principal components analysis (PCA)
as a rst investigation, because this has the advantage of rewriting the data into uncorrelated components (Johnston, 1978). A PCA was run using the
correlation matrix, because PCA is sensitive to different scales within the data if the covariance matrix is used.
Data reduction
PCA and the raw correlation readings were used
to identify a subset of variables that could explain a
large majority of the variation, so that the relationship could be further analyzed using linear and nonlinear regression techniques. This subset of terms
had equal explanatory power, but was much less
collinear.
Regression analysis
First, the identied subset of terms was used in a
stepwise regression, using a fully specied model to
see if there were any signicant interactions between variables.
Using this method, UVR range was eliminated
from further investigation. Winter precipitation was
only a minor component in two analyses, but was
retained as it still raised the coefcient of determination (r2) slightly.
Least squares regression was run, using the four
variables identied by data-reduction techniques
(UVR Autumn, UVR Winter, Winter Precipitation,
and Summer Precipitation).
Spatial bias and autocorrelation
To test if there was signicant error introduced by
treating geographic variables as independent, in the
light of possible spatial autocorrelation in data connected to geography (Cressie, 1993), the residuals of
the regression were inspected using correlation and
regression against both latitude and longitude separately, and latitude and longitude together. The
previous test was useful for checking that there was
no difference between hemispheres in skin reectance response, as suggested by Relethford (1997).
To test this premise further, t-tests were performed
on the residuals of the regression and on the data of
the two hemispheres, but with the span of latitudes
restricted to be the same (25), so that the samples
were not different in the two hemispheres (Chaplin
and Jablonski, 1998).

ENVIRONMENT AND HUMAN SKIN COLORATION

Nonlinear investigation
Further, graphical techniques were run to investigate the relationships as identied in the nal
regression model. The nonlinear techniques consisted of local regression, (loess) regression, and
spline tting. Smoothing splines that minimize a
penalized residual sum of squares was done using
four degrees of freedom (d.f.), because this will not
force oversmoothing (S-Plus, 2000).
Analysis of categorical variables
To investigate categorical variables (relief and
land-cover type), boxplot techniques were used, and
an analysis of variance was performed. The latter
looked for variation that could be further explained
by inclusion of these categorical variables.
Prediction from final regression model
The nal regression model was used to predict
skin reectance for each color lter. These predicted
skin colors were then mapped. The colors of the map
were produced from the multiple regression formulae for each color lter. This map was, in effect, an
intersection of all the input maps. The legend was
created by assigning a color to each unique combination derived from the values of the red, green, and
blue (RGB) lters, to derive a RGB color value using
the GIS programming language. These colors are
realistic approximations of the true color of skin,
within the limits of the color space of the particular
medium in which it is viewed.
The large number of features and color combinations meant that this map needed generalization for
visualization of any trends in predicted skin color. It
was generalized as follows: cells were rst dissolved
until they had unique combinations of red, green,
and blue values. Then the lowest and highest values
in the distribution tail, that had only a few cells
each, were unioned into a single shape. Finally,
shapes that were almost identical in color were
merged with their neighbors to further reduce the
number of shapes. By this method, the number of
shapes was reduced to just 17 multipart shapes. The
generalization sought to reach a balance between
equal area representation and equal interval representation. The colors in the legend are also the actual skin colors as dened above. This map was
suitable for visualization techniques (Cleveland,
1993), but not for statistical analysis.
RESULTS
Exploratory data analysis
Outliers that were detected were examined, and
where their explanation could be understood in
terms of migration, genetic admixture, or incorrect
placement on the map, they were removed from
further analysis. These consisted of very few cases.
The rst were the Ammassalimiut (Inuit from
Greenland) and Yemeni Jews; these were removed

295

because these populations had the most extreme

outlier values. The Yemeni Jews are known not to
have a long history at their location (Lourie, 1973).
This may be true for the Inuit too, or it could be that
they are a special case related to their diet (Jablonski and Chaplin, 2000). The next to be removed were
two anomalous cells erroneously assigned as inhabited by Nunoans that were on the very summit of
Andean mountains at extreme UVR; the other cells
for Nunoans (on the Altiplano) were left. Lastly, the
Bolivians and mestizos of Peru were removed, who
were both reported to have some mixed blood and
whose residuals were large enough to conrm that
the admixture was probably great enough to warrant their removal.
Correlation results
The correlation results were very high when performed on the by cell data (Table 1). These correlations rose as outliers were removed (column 2,
Table 1). For the variables containing estimated values derived from either the EEL or the Photovolt
readings (column 3, Table 1), all of the correlations
between environmental variables and skin reectance measures dropped slightly to about the same
level as the raw data in column 1. However, it
should be noted that the number of cells rose considerably using these data, and this rise probably
accounts for much of the small drop in the correlations. These measures were then grouped by population, as explained in Methods, and the correlations
for these data are given in the right three columns of
Table 1.
Most of the variables displayed high collinearity,
as would be expected from solar processes or processes driven by solar energy such as precipitation.
There were very high correlations of skin reectance
to UVR, temperature measures, and absolute angle
of latitude. UVR Autumn was more highly correlated to Melanin (red) Reectance than was Latitude but not for Tissue (blue) Reectance, where
Latitude was more highly correlated. The highest
correlation of direct UVR was UVR Autumn, with
the absorption peak of Blood (green) Reectance (r
0.927, d.f. 70, P 0.0001), which was extremely
high considering the heterogeneity of the skin color
reectance sample. UVR Autumn, UVR Winter, and
UVR Minimum all had r-values greater than the
UVR Annual Average for Blood (green) Reectance
and Melanin (red) Reectance, but not for Tissue
(blue) Reectance, where the UVR Average was
highest. It is unlikely that some component of sunlight other than UVR is having an effect, as the
variable Total Solar Average Radiation was less correlated with skin reectance than was UVR Average
or in the UVR Winter and UVR Autumn. The number of days with snow was correlated with lighter
colors of skin; another variable with the same pattern was Winter Precipitation, which has snow as
one of its components.

296

G. CHAPLIN
TABLE 1. Pearson correlation coefficients, for skin reflectance measures to environmental variables1

d.f.
UVR Average
UVR Autumn
UVR Winter
UVR Minimum
UVR Spring
UVR Summer
UVR Maximum
UVR Range
Latitude
Annual precipitation
Summer precipitation
Winter precipitation
Average solar radiation
Snow
Frost-free days
Average temperature,
annual
Average temperature,
summer
Average temperature,
winter

Raw by cell
for Red Filter
E685.609

Trimmed
by cell for
Red Filter
E685.609

Melanin
reectance by
cell, including
estimates

Melanin
reectance
grouped by
population

Tissue
reectance
grouped by
population

Blood
reectance
grouped by
population

3,706.0
0.66***
0.76***
0.73***
0.76***
0.18***
0.31***
0.16***
0.54***
0.75***
0.16***
0.05***
0.26***
0.27***
0.35***
0.36***
0.42***

3,674.0
0.69***
0.78***
0.75***
0.78***
0.19***
0.33***
0.16***
0.56***
0.76***
0.16***
0.05**
0.26***
0.30***
0.35***
0.38***
0.43***

4,355.0
0.66***
0.76***
0.72***
0.76***
0.17***
0.33***
0.16***
0.56***
0.75***
0.15***
0.05**
0.24***
0.30***
0.31***
0.36***
0.41***

96.0
0.82***
0.84***
0.83***
0.82***
0.65***
0.66***
0.63***
(0.13)
0.83***
0.23***
0.32***
0.35***
0.63***
0.46***
0.59***
0.67***

70.0
0.89***
0.86***
0.86***
0.82***
0.79***
0.78***
0.79***
(0.14)
0.92***
0.28***
0.38***
0.46***
0.78***
0.48***
0.77***
0.82***

70.0
0.90***
0.92***
0.93***
0.90***
0.75***
0.73***
0.74***
(0.02)
0.93***
0.29**
0.40***
0.44***
0.71***
0.54***
0.74***
0.84***

0.09***

0.09***

0.08***

0.38***

0.61***

0.55***

0.55***

0.55***

0.53***

0.71***

0.74***

0.82***

Note differences in degrees of freedom of different samples due to differing numbers of missing values. Three left-hand columns show
effect, on correlation in data, of having raw data trimmed by having ve outliers removed and for data converted from different
reectometers. Three right hand columns are for data summarized by population. All correlations are very highly signicant, except
UVR Range.
* P 0.05.
** P 0.01.
*** P 0.0001.

Latitude, UVR Autumn, UVR Winter, UVR Minimum, and UVR Average were all much more highly
correlated to skin reection measures than were
UVR Maximum, UVR Summer, UVR Spring, and all
other environmental factors.
Principal components analysis
Principal components analyses were performed on
both the trimmed by cells and the trimmed by
population data sets. The results were comparable.
The diagnostic metrics for the more signicant
components, for the grouped by population PCA,
are given in Table 2. The biplot of this PCA (Fig. 1)
represented both the original variables and the
transformed observations on the rst two principal
components axes. The original variables were represented as arrows that graphically indicated the proportion of the original variance explained by the rst
two principal components.
The rst component comprised the major UVR
measures and also was negatively correlated to skin
reectance variables, so that an increase in UVR
covaries with a decrease in reection, which means
a darker skin color. The direction of the arrow
showed the relative loadings of the variables on PCA
components 1 and 2 (S-Plus, 2000). In Figure 1, the
arrows span much of the data spread, showing that
the two rst components accounted for much of the
variation (cumulative proportion 0.821). The temperature measure arrows were all shorter than

those for other variables, indicating that their covariance was nested within the variation explained
by the UVR measures.
The variables UVR Winter (UV Win), UVR Autumn (UVAut), UVR Minimum (UV Min), and Winter Temperature (Win Tmp) were highly collinear.
These four variables showed the greatest negative
covariance with skin reectance measures in the
biplot graph (Fig. 1 and Table 2). These four were
offset by the number of days with snow on the
ground (Snow) and by Winter Precipitation
(Winavpcp), which was inclusive of snow. That is,
these two variables positively covaried with skin
reectance, with which they grouped.
Another group of variables consisted of Total Average Solar Radiation, UVR Maximum, UVR Summer, and Summer Temperature, that lay to the left
of the second component axis, and that thereby also
correlated with darker skin colors.
The variables Frost-Free Days, UVR Average, and
Annual Temperature, which were averaging measures of the other variables, lay between the groups
from maximum and minimum measures, as would
be expected.
The loadings showed that most of the variance
(65%) was accounted for by the interaction of skin
reectance and snow to solar processes. The next
component (17%) showed an axis whereby solar processes were offset by increasing precipitation, with
increases in UVR range being associated with

ENVIRONMENT AND HUMAN SKIN COLORATION

297

TABLE 2. Results of principal components analysis by

population1
Importance of components
(Comp.)
Comp. Comp. Comp. Comp.
1
2
3
4
Standard deviation
Proportion of variance
Cumulative proportion
Loadings
UVR Summer
UVR Autumn
UVR Winter
UVR Spring
UVR Maximum
UVR Minimum
UVR Range
UVR Average
Annual average precipitation
Summer precipitation
Winter precipitation
Average solar
Average snow days
Average frost-free days
Average annual temperature
Average summer temperature
Average winter temperature
Tissue (blue) Reectance
Blood (green) Reectance
Melanin (red) Reectance
1

3.61
0.65
0.65
0.24
0.26
0.26
0.25
0.25
0.24
0.47
0.27
0.45
0.39
0.15
0.25
0.18
0.25
0.26
0.21
0.24
0.26
0.26
0.24

1.84
0.17
0.82
0.17
0.15
0.19
0.20
0.21
0.39
0.45
0.44
0.22
0.21

0.22
0.17
0.14
0.16

1.08
0.06
0.88

0.91
0.04
0.92

0.19
0.12
0.13
0.15
0.14
0.20

0.10

0.32

0.44

0.34

0.63
0.24
0.23
0.22
0.29
0.20
0.20
0.25

0.23
0.16
0.11

Number of variables is 20; number of observations is 69.

lighter skin reectance in the longer wavelength

reectance but not for Tissue (blue) Reectance. Tissue (blue) Reectance was negatively correlated
(darker skin reectance) on the minor third axis
(6%) with snow and harsh winter weather. The last
axis, which was below 5%, was the contrast between
snow days and other Winter Precipitation, where a
decrease in snow and an increase in other forms of
winter precipitation were correlated with a lighter
skin reectance.
Regression analysis
It was possible to use the correlations (Table 1)
and the PCA analysis (Table 2) to pick variables
warranting greater investigation (Johnston, 1978).
UVR Autumn appeared to be the main single effector, and UVR Maximum with Summer Precipitation
were very minor effectors. These were offset by precipitation in the winter (which includes, in part, the
number of days with snow).
The PCA showed that the temperature measures
appeared to be so closely aligned to the UVR values
that they could be ignored. This was conrmed by
performing regressions with nested variables. For
example, Blood (green) Reectance alone to UVR
Autumn had a higher coefcient of determination
than that for Blood (green) Reectance and Average
Temperature together or Average Temperature
alone. The contribution of Average Temperature
was also nonsignicant (P 0.25, d.f. 69). This is
not to say that variation in temperature measures
lacked explanatory power, but rather they were so
collinear with UVR that they had no extra indepen-

Fig. 1. Biplot of component 1 vs. component 2 of PCA of skin

reectance measures and environmental variables of grouped by
population data. Melanin (red), Blood (green), and Tissue (blue)
Reectances are tightly grouped to right on rst axis, and loosely
grouped with average snow days (Snow) and Winter Precipitation
(Winpcp). All other variables are on opposite side of central axis.
Directly opposite to skin reectance variables is a cluster of UVR
Winter (UV Win), UVR Autumn (UV Aut), UVR Minimum (UV
Min), and Average Winter Temperature (Wintmp). UVR Range
(UV Range) with Summer Precipitation (Sumpcp) and Annual
Precipitation (Anpcp) are on an orthogonal axis. AvRad, average
solar radiation; UvMax, UVR maximum; Sumtp, summer temperature; UvSum, UVR summer; UvSprg, UVR Spring; Ffree, no.
of frost-free days; UvAvg, UVR average; Avtmp, average temperature.

dent explanatory power over that of UVR; therefore,

they were omitted.
The weak relationship of Summer Precipitation to
skin reectance measures was highly signicant
(r2 0.403, d.f. 70, P 0.0001).
Correlation, PCA, and stepwise regression validated the four-variable linear model developed. The
results are given in Table 3.
All r2 values were impressively high (multiple r2:
Melanin (red) Reectance 0.7283, Tissue (blue)
Reectance 0.8327, and Blood (green) Reectance 0.8872; the nonsignicant coefcient (Winter Precipitation) added, at most, 0.017 to the r2
value). There was a greater spread in the tted
values than in the residuals; the quartile-quartile t
of the residuals was not signicantly different from
normal; and there were very few outliers. These
observations further validate the competency of the
linear model. In conclusion, the multiple-regression
model was an adequate straight-line approximation
of the relationship between variables.

298

G. CHAPLIN
TABLE 3. Table of diagnostic statistics of
multiple regression equations

Blue Tissue Reectance UVR Autumn UVR Maximum

Summer Precipitation Winter Precipitation
Coefcients

Value

Standard
error

t-value

Intercept
41.7231
2.4860
16.7834
0.0001
UVAUT
0.0457
0.0080
5.7156
0.0001
UVMAX
0.0365
0.0072
5.0912
0.0001
SUMAVPCP
0.0014
0.0004
3.2372
0.0019
WINAVPCP
0.0006
0.0011
0.5947
0.5540
Residual standard error: 4.396 on 67 degrees of freedom
Multiple r2: 0.8327
F-statistic: 83.39 on 4 and 67 degrees of freedom, P 0.0001
Outliers: Japan Hidakka, Eastern Nepal
Green Blood Reectance UVR Autumn UVR Maximum
Summer Precipitation Winter Precipitation
Coefcients

Value

Standard
error

t-value

Intercept
43.5006
2.2443
19.3825
0.0001
UVAUT
0.0758
0.0071
10.7087
0.0001
UVMAX
0.0156
0.0065
2.4072
0.0188
SUMAVPCP
0.0011
0.0004
2.9080
0.0049
WINAVPCP
0.0018
0.0009
1.8976
0.0621
Residual standard error: 3.977 on 67 degrees of freedom
Multiple r2: 0.8872
F-statistic: 131.7 on 4 and 67 degrees of freedom, p 0.0001
Outliers: Eastern Nepal, Yemen
Red Melanin Reectance UVR Autumn UVR Maximum
Summer Precipitation Winter Precipitation
Coefcients

Value

Standard
error

t-value

Intercept
63.0801
3.8472
16.3963
0.0001
UVAUT
0.1014
0.0114
8.9182
0.0001
UVMAX
0.0035
0.0111
0.3186
0.7507
SUMAVPCP
0.0001
0.0006
0.2310
0.8178
WINAVPCP
0.0028
0.0014
1.9205
0.0579
Residual standard error: 7.342 on 92 degrees of freedom
2
Multiple r : 0.7283
F-statistic: 61.66 on 4 and 92 degrees of freedom, P 0.0001
Outliers: Australia Darwin, Cambodia, Namibia, and Yemen

Spatial bias and autoregression

The correlation of the regression residuals with
either absolute angle of latitude or longitude or both
was zero, showing that there was an absence of
spatial bias in the residuals. The residuals were
normally distributed, and there was a very close
concordance of the model to the data. The residuals
of the multiple regression against both latitude and
longitude showed that there was no unaccounted-for
variation related to geography, i.e., latitude, longitude, or the latitude:longitude interaction.
The test for hemispheric differences did not nd a
noticeable difference in pattern, contra Relethford
(1997). There was a slight nonsignicant difference
in magnitude of response (as found by Relethford,
1997), which was considered the result of sampling.
Therefore, it is safe to assume that modern humans
are one population for statistical purposes.
Nonlinear graphing
Nonlinear methods do not provide good diagnostic
statistics. The nonlinear individual variable graphs

Fig. 2. Spline graph of relationship of green lter reectance,

which approximates that of blood absorption, to UVR Maximum
(UVMAX).

showed a few interesting ndings. UVR Autumn

had a linear relationship with skin reectance,
whereas UVR Maximums straight-line relationship
only extended up to the limit of that for UVR Autumn, above which the relationship failed. The distribution of populations in areas beyond this level of
UVMED showed a high level of variance. There was
a suggestion of a much lower correlation of UVR
Maximum to darker skin color above this value (Fig.
2). Precipitation in the summer seemed to follow the
pattern of UVR Maximum to which it was uncorrelated (r 0.044, P 0.67). Winter precipitation had
a weak opposite effect.
Categorical variables for forest and relief
The analysis of variance produced nonsignicant
results and will not be reported. The boxplots of the
data (not shown) indicated why these nonsignicant
results were obtained. All variables had ranges that
overlapped, except for UVR Maximum, in high and
middle mountains, and Melanin (red) Reectance,
where there does appear to be a small nonsignicant
inuence of topographic relief on skin reectance.
Mapping environmental data
The map of UVR Autumn was very similar to that
for Annual Average UVR (Jablonski and Chaplin,
2000). The predominant effect caused by the angle of
the sun to the ozone layer manifested itself as latitudinal bands of nearly equal UVR. Over the land,
these bands were modied by factors related to water vapor, monsoon effects related to the size of the
west-to-east extent of the land masses, and altitude.
Autumn was without extremes of UVR. In autumn,
there is a monsoon continental effect, i.e., an amelioration of UVR on the eastern side of land masses.
The effects of climate are more stable in autumnwinter than in summer, which brings monsoons and
summer rain, and droughts around the tropics.
Spring had higher levels of UVR than autumn due to

ENVIRONMENT AND HUMAN SKIN COLORATION

299

Fig. 3. Map of skin color reectance predicted from multiple regression. Map is generalized to reduce number of polygons.

drying prior to the monsoon. The lowest mean and

coefcient of variation were in autumn.
The map for UVR Range showed some unexpected
results. The range was lowest in the large equatorial
rainforest areas. There the range was about the
same as in Greenland, where the range was compressed by the low potential maximum. Northern
Europe had higher ranges than equatorial regions,
and the Mediterranean area was considerably
higher. The highest values for range were in the
high-altitude deserts, which also have the highest
UVR values.
Snow followed latitude to a great degree except for
Tibet, the southern Andes, Iran, Turkey, the TransCaucasus, and Japan.
The pattern discernible in autumn UVR had
higher values in Africa than in other equatorial regions. Africa was the most heterogeneous continent
with respect to UVR regimes, due to its large latitudinal range, large west-to-east extent north of the
equator, a wide variety of humidity regimes, and a
large altitudinal range from west to east (and within
the east from north to south).
Map of skin color reflectance predicted from
multiple regression formulae
The map of predicted skin colors produced from
the multiple regression formulae had more than
26,000 features, and the legend had over 1,200
unique combinations of the three reectance values.
The maps of predicted values showed that skin color
reectance was predicted to be more heterogeneous
in Africa. There were similar small niches in all
areas of high UVR near the equator. Areas with the
lowest value for predicted skin color reectance were
in New Zealands South Island fjords, Greenland,
and Tierra del Fuego. The next lowest predictions
were areas in Scandinavia and the northwestern

British Isles. The area with the darkest predicted

values for skin reectance were on the mountain
peaks of the Andes and the Ethiopian highlands. It
was easy to see in these maps that skin reectance
would be predicted to be a fairly smooth gradation of
values with disjuncts only at the most major of geomorphologic barriers such as the Himalayas. In areas away from the Equator, the size of ecological
patches increases with increasing latitude. This effect was mostly lost in the Southern Hemisphere
due to the lack of land farther to the south at high
latitudes. At smaller scales, the colors almost completely blend into one another, making the clinal
nature of skin color obvious. Therefore, the generalized map is of more use for visualizing the data.
DISCUSSION
The rst and more popular hypothesis to be tested
(that reduced skin reectance is related to direct
exposure to UVR) is only weakly supported by this
study. There are lower correlations, and a much
smaller contribution in the regression analysis, for
measures of periods of high UVR, e.g., skin reectance and UVR Maximum, UVR Summer, and UVR
Spring. These measures explain at least 17% less of
the variation in the data than those for times of low
UVR. This observation argues strongly against sunburning or sun-induced cancer as major causative
agents in the evolution of skin coloration, because
these diseases are modulated by UVB (summer
UVR). This result was predicted because cancers
arise mostly after the human reproductive career
has ended (Blum, 1961; Jablonski and Chaplin,
2000).
The breakdown of the linear response of skin reectance to UVR Maximum at very high levels of
UVR, as seen in the nonlinear techniques, indicates
that these populations exhibited a large range of

300

G. CHAPLIN

values and much variation about the line. The darkest people seemed to be those who have inhabited
their homelands for very long periods of time, e.g.,
Australian Aborigines and Ethiopians. As these populations are those that seem to be resident the longest, it can be concluded that the nal increments of
skin darkening appear to have required longer to
evolve. The darkest populations have similar reectance values, so that there may be a point where it is
not biologically possible to darken further, and approaching that point will slow the rate of adaptation.
The prediction that there is reduced skin reectance in the Tissue (blue) Reectance range due to
thickening of the stratum corneum caused by UVR
is supported. Tissue (blue) Reectance was slightly
more correlated with UVR Maximum than were the
other reectance measures, a nding that helps to
explain the thickened stratum corneum found in
African populations, although a thickened stratum
corneum would provide little photo-protection, as
can be seen in people with pigment disorders (Robins, 1991).
The second set of hypotheses, concerning theories
of metabolic processes relating direct solar radiation
to skin reectance, nds the strongest support. The
main original nding of this study was that the
evolution of skin color reectance could be almost
fully modeled as a linear effect of UVR in the autumn alone (UVR Autumn with the absorption peak
of Blood (green) Reectance; r 0.927, d.f. 70, P
0.0001). This is also true for all measures of the
period of lower UVR (those with an increased proportion of longer-wavelength UVR): UVR Autumn,
UVR Minimum, and UVR Winter. These measures
were all more important at explaining skin reectance variation than were the peak UVR measures.
This very strong nding gives considerable support
to the nutrient hypothesis. Folate photolysis leading
to folate deciency was proposed as an agent decreasing tness and promoting skin darkening in
lower latitudes near the Equator out to beyond the
tropics (Jablonski, 1992, 1999; Jablonski and Chaplin, 2000). Hypovitaminosis D was found to be the
agent in higher latitudes where there are prolonged
periods without vitamin-D-promoting UVR (UVB)
(Holick, 1995; Jablonski and Chaplin, 2000; Loomis,
1967). High levels of wintertime UVA without UVB,
as found in middle latitudes, would destroy both
vitamin D and folate by photolysis.
This extremely well-tted linear model explains
why latitude was such a good surrogate for UVR,
because Latitude and UVR Autumn were extremely
highly correlated (r 0.967, d.f. 102, P 0.0001),
as was Latitude and UVR Average (r 0. 937, d.f.
102, P 0.0001), but not Latitude and UVR Summer (r 0.767, d.f. 102, P 0.0001). Further, this
indicates that latitude is still useful for human ecological modeling.
Hypotheses about variables that are only correlated to UVR by latitude or through solar-driven
processes such as vegetation, temperature, and pre-

cipitation generally received little support, although

these variables may alter the response of skin to
UVR and so magnify or mitigate the other two
groups of hypotheses.
The inability of any of the temperature measures
to explain variation in the regression analysis indicated that the small differences in thermoregulatory
performance of different skin colors seemed to be
swamped by the nutrient requirements and protective roles of skin.
Frost and Winter Precipitation in the PCA and
regression results provide tenuous support to the
structural reduction hypothesis by Brace (1963),
where defects accumulate in the pigmentary system
in the absence of positive selection. However, the
cultural adaptations of shelter-seeking and clotheswearing would have exaggerated problems with hypovitaminosis D. This was documented for immigrants into Northern Europe (Henderson et al.,
1987; Hodgkin et al., 1973; Solanki et al., 1995).
The last two factor loadings of the PCA indicate
that there was a contrast in the effects of precipitation measures. Tissue (blue) Reectance was negatively correlated on the minor third axis (6%) with
harsher winter weather, i.e., harsh weather was
correlated with darker skin in the blue spectrum.
This could be because of the response to extremes of
climate that cause trauma that thickens the stratum corneum. This effect may be magnied by
higher altitude. This mechanism could also account
for some of the different correlation observed for
Tissue (blue) Reectance in the correlation and regression analysis compared to the other skin reectences.
All the analysis methods show that Summer Precipitation is correlated with darker skin color. This
effect is nonlinear and mirrors that seen in the response to UVR Maximum. It appears to magnify the
effect of UVR. This nding adds support to Glogers
rule. Summer rain, particularly in warmer regions,
promotes sweating, which causes blood to rise into
the dermis to supply the sweat glands. Under these
conditions, the blood and its nutrients will be more
susceptible to UVR damage. In addition, wet skin
allows the skin to become more transparent to UVR
by refraction, so that sweating effectively lowers the
minimal erythemal dose (Moehrle et al., 2000).
The maps of predicted values of skin coloration
were consistent with ndings by Relethford (2000)
that skin reectance was more variable in Africa,
and that in areas where the size of ecological patches
is larger with increased latitude (primarily Eurasia), there was less diversity in skin color. This is a
result explained here by diversity and size of climate
niches. The four-term multiple regression was very
successful and was able to model complex relationships across relatively small areas, such as Micronesia, which showed a great diversity of habitats
within an area of limited latitudinal range.
This study conrms the observation that skin coloration is adaptive and clinal in nature, although

ENVIRONMENT AND HUMAN SKIN COLORATION

the adaptation of skin color to Summer UVR seemed

to reach a threshold beyond which further adaptation through increasing melanization was difcult.
The conclusion of an adaptive cline was supported
by: 1) the very high level of correlation between skin
reectance and environmental variables; 2) a nearly
straight-line t of skin reectance to UVR Autumn,
which was highly suggestive of a one-to-one response; and 3) the very low number of variables that
were able to account for the majority of variation in
the data. Although this does not constitute proof of a
causal relationship between skin reectance and environmental factors, this combination of factors
greatly increases the likelihood that one exists. This
is a position that was supported for other geographically varying associations, e.g., disease (Foster,
1987).
ACKNOWLEDGMENTS
The author thanks Elizabeth Weatherhead for
providing access to the NASA TOMS UVMED data,
to Kenneth Beals for providing a digital copy of the
unpublished database collated at Oregon State University, and to Charles Convis of the Environmental
Systems Research Institute (ESRI) for donations of
GIS software and data. The author thanks C. Philip
Wheater and Tom Poiker for their timely help and
advice. The author also thanks Nina G. Jablonski for
advice and fruitful discussions.
LITERATURE CITED
Baker P. 1958. Racial differences in heat tolerance. Am J Phys
Anthropol 16:287305.
Blum HF. 1961. Does the melanin of human skin have adaptive
value? Q Rev Biol 36:50 63.
Brace CL. 1963. Structual reduction in evolution. Am Nat 97:39
49.
Buzzell RA. 1993. Effects of solar radiation on the skin. Otolaryngol Clin North Am 26:111.
Candler WH, Ivey H. 1997. Cold weather injuries among U.S.
soldiers in Alaska: a ve-year review. Milit Med 162:788
791.
Chaplin G. 2001. The geographic distribution of environmental
factors inuencing human skin coloration. Dissertation submitted in partial fulllment of requirements for Degree of Master
of Science in Geographical Information Systems, Manchester
Metropolitan University, Manchester.
Chaplin G, Jablonski NG. 1998. Hemispheric difference in human
skin color [comment]. Am J Phys Anthropol 107:221224.
Cleaver J, Mitchell D. 2000. Ultraviolet radiation carcinogenesis.
In: Bast RJ, Kufe D, Pollock R, Weichselbaum R, Holland J,
Frei E, editors. Cancer medicine. Hamilton: B.C. Decker. p
31 41.
Cleveland W. 1993. Visualizing data. Summit, NJ: Hobart.
Cowles RB. 1959. Some ecological factors bearing on the origin
and evolution of pigment in the human skin. Am Nat 93:283
293.
Cressie N. 1993. Statistics for spatial data. New York: John
Wiley.
Daniels F Jr. 1964. Man and radiant energy: solar radiation. In:
Dill DB, Adolph EF, editors. Adaptation to the environment.
Washington, DC: American Physiological Society. p 969 985.
Deol MS. 1975. Racial differences in pigmentation and natural
selection. Ann Hum Genet 38:501503.
Ebbinghaus J. 1966. Differences in skin pigmentation between
the sexes. Masters thesis, Pennsylvania State University.

301

Eckes LK. 1980. Anthropological aspects of skin color. Hautarzt Z

Dermatol Venerol Verwandte Gebiete 31:76 81.
ESRI. 1996. ArcAtlas: our Earth: Environmental Systems Research Institute. Redlands, CA: ESRI.
Foster HD. 1987. Disease family trees: the possible roles of iodine
in goitre, cretinism, multiple sclerosis, amyotrophic lateral
sclerosis, Alzheimers and Parkinsons diseases and cancers of
the thyroid, nervous system and skin. Med Hypotheses 24:249
263.
Hamilton W III, Heppner F. 1967. Radiant solar energy and the
function of black homeotherm pigmentation: an hypothesis.
Science 155:196 197.
Henderson JB, Dunnigan MG, McIntosh WB, Abdul-Motaal
AA, Gettinby G, Glekin BM. 1987. The importance of limited
exposure to ultraviolet radiation and dietary factors in the
aetiology of Asian rickets: a risk factor model. Q J Med
63:413 425.
Herman J, Celarier E. 1996. TOMS version 7. UV-erythemal
exposure: 1978 1993. NASA Goddard Space Flight Center
Ozone Processing Team. Greenbelt, MD.
Hodgkin P, Kay G, Hine P, Lumb G. 1973. Vitamin D deciency
in Asians at home and in Britian. Lancet 1973:167172.
Holick MF. 1981. The cutaneous photosynthesis of previtamin D3: a
unique photoendocrine system. J Invest Dermatol 77:5158.
Holick MF. 1995. Environmental factors that inuence the
cutaneous production of vitamin D. Am J Clin Nutr 61:638
645.
Jablonski NG. 1992. Sun, skin colour and spina bida: an exploration of the relationship between ultraviolet light and neural
tube defects. Proc Australas Soc Hum Biol 5:455 462.
Jablonski NG. 1999. A possible link between neural tube defects
and ultraviolet light exposure. Med Hypotheses 52:581582.
Jablonski NG, Chaplin G. 2000. The evolution of human skin
coloration. J Hum Evol 39:57106.
Johnston R. 1978. Multivariate statistical analysis in geography.
New York: Longman.
Jones G, Strugnell SA, DeLuca HF. 1998. Current understanding
of the molecular actions of vitamin D. Physiol Rev 78:1193231.
Kollias N, Sayre RM, Zeise L, Chedekel MR. 1991. New trends in
photobiology (invited review). J Photochem Photobiol B Biol
9:135160.
Little M, Wolff M. 1981. Skin and hair reectance in women with
red hair. Ann Hum Biol 8:231241.
Longstreth J, de Gruijl FR, Kripke ML, Abseck S, Arnold F,
Slaper HI, Velders G, Takizawa Y, van der Leun JC. 1998.
Health risks. J Photochem Photobiol B Biol 46:20 39.
Loomis WF. 1967. Skin-pigment regulation of vitamin-D biosynthesis in man. Science 157:501506.
Lourie J. 1973. Physical characteristics of Yemenite and Kurdish
Jews in Israel. Philos Trans R Soc Lond [Biol] 266:101112.
MathSoft. 1999. S-Plus 2000 professional release 1. Seattle:
MathSoft.
Moehrle M, Koehle W, Dietz K, Lischka G. 2000. Reduction of
minimal erythema dose by sweating. Photodermatol Photoimmunol Photomed 16:260 262.
Murray F. 1934. Pigmentation, sunlight and nutritional disease.
Am Anthropol 36:438 445.
Pandolf K, Gange R, Latzka W, Blank I, Kraning K, Gonzalez R.
1992. Human thermoregulatory responses during heat exposure after articially induced sunburn. Am J Physiol 262:610
616.
Pawson IG, Petrakis NL. 1975. Comparisons of breast pigmentation among women of different racial groups. Hum Biol 47:441
450.
Pollitzer W, Namboodiri K, Elston R, Brown W, Leyshon W. 1970.
The Seminole Indians of Oklahoma: morphology and serology.
Am J Phys Anthropol 33:1530.
Post PW, Daniels F Jr, Binford RT. 1975. Cold injury and the
evolution of white skin. Hum Biol 47:65 80.
Potten CS, Chadwick CA, Cohen AJ, Nikaido O, Matsunaga T,
Schipper NW, Young AR. 1993. DNA damage in UV-irradiated
human skin in vivo: automated direct measurement by image
analysis (thymine dimers) compared with indirect measure-

302

G. CHAPLIN

ment (unscheduled DNA synthesis) and protection by 5-methoxypsoralen. Int J Radiat Biol 63:313324.
Razi D, Prevulotzky A, Shkolnik A. 1980. Is a black coat in the
desert a means of saving metabolic energy? Nature 283:761
762.
Relethford JH. 1997. Hemispheric difference in human skin color.
Am J Phys Anthropol 104:449 457.
Relethford JH. 2000. Human skin color diversity is highest in
sub-Saharan African populations. Hum Biol 72:773780.
Roberts DF. 1977. Human pigmentation: its geographical and
racial distribution and biological signicance. J Soc Cosmetic
Chem 28:329 342.
Roberts DF, Kahlon DPS. 1976. Environmental correlations of
skin colour. Ann Hum Biol 3:1122.
Robins AH. 1991. Biological perspectives on human pigmentation. Cambridge: Cambridge University Press.
Sefcakova A. 1986. Spectrophotometric study of pigmentation in
albinos, their parents and helathy persons. Bratisl Lek Listy
86:401 407.

Solanki T, Hyatt RH, Kemm JR, Hughes EA, Cowan RA. 1995.
Are elderly Asians in Britain at a high risk of vitamin D
deciency and osteomalacia? Age Ageing 24:103107.
S-Plus. 2000. Guide to statistics. Seattle: Data Analysis Products
Division, MathSoft.
Steegmann A Jr. 1967. Frostbite of the human face as a selective
force. Hum Biol 39:131145.
Tasa GL, Murray CJ, Boughton JM. 1985. Reectometer reports
on human pigmentation. Curr Anthropol 26:511512.
Taylor MS. 1992. Cold weather injuries during peacetime military training. Milit Med 157:602 604.
Walter H. 1958. Der Zusammenhang von Hautfarbenverteilung und Intensitat der ultravioletten Strahlung. Homo 9:1
13.
Webb AR, Holick MF. 1988. The role of sunlight in the cutaneous
production of vitamin D3. Annu Rev Nutr 8:375399.
Zihlman AL, Cohn BA. 1988. The adaptive response of human
skin to the savanna. Hum Biol 3:397 409.

APPENDIX I. Skin reective values used in this study.

Reectance
Name

Ref.

Various
Huachipaeri
Mexico
Shipibo
Slovakia
United Kingdom
United Kingdom & Northern Ireland
United States, Florida
United States, Oklahoma
United States, Western

1
2
2
2
3
4
4
5
5
6

Tissue

Blood

12.8
19.3
15.75
28.8
42.3
42.3

19.1
27.1
21.4
36.9
45.6
45.6

19.1

33.3

Melanin

How mapped

48.6
48.1
53.2
62.5
69
68.9
47.7
50.4

Various
Estimated from atlas
Estimated as sierra region
Estimated from atlas
Western region of Slovakia
Estimated from atlas
Estimated from atlas
Estimated from atlas
Estimated from atlas
Estimated from atlas

References (Ref.) for data: 1, Jablonski and Chaplin, 2000; 2, Ebbinghaus, 1966; 3, Sefcakova, 1986; 4, Little and Wolff, 1981; 5,
Pollitzer et al., 1970; 6, Pawson and Petrakis, 1975. Full references in Chaplin (2001).