Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454 – 470

www.elsevier.com/locate/palaeo

Comparative study of modern phytolith assemblages

from inter-tropical Africa
Doris Barboni ⁎, Laurent Bremond 1 , Raymonde Bonnefille
CEREGE, CNRS-UAMIII / UMR 6635, B.P. 80, F13545 Aix-en-Provence Cedex 04, France
Received 10 March 2006; received in revised form 8 September 2006; accepted 18 October 2006

Abstract

We present a synthesis of modern phytolith studies from Africa, to infer the potential and limitations of phytolith assemblages to
reconstruct vegetation and tree cover density. The modern dataset includes 149 phytolith assemblages of surface soil samples from
10 phytogeographical zones and sub-zones from East and West Africa, as well as 500 m-resolution satellite estimates of the percent
tree cover at the sampling sites. To test the potential of phytolith assemblages to discriminate vegetation types we used principal
component analysis. For each phytogeographical zone and sub-zone, we also provided the mean values, standard errors, and 95%
confidence intervals for the means obtained on the modern African dataset for the abundance of the 13 most common phytolith
types preserved in soil samples, and for four phytolith indices.
Results from the modern African dataset show that 1) the relative abundances of 11 (out of 13) phytolith types allow discrimination
of all vegetation zones but the Somalia–Masaï steppe region, which at elevation b 600 m asl, exhibits a high proportion of rondel and
trapeziform short cell phytoliths, like in the Afromontane region; 2) the co-occurrence of rondels and trapeziform polylobates
characterises zones above 1900 m asl and/or current annual temperatures b 19 °C; 3) the relative abundance of globular phytoliths
(granulate, smooth, and echinate) is better correlated to 500 m-resolution satellite estimates of the tree cover (R2 = 0.60 for n = 149, and
R2 = 0.57 for n = 85, p b 0.005) than is the abundance of arboreal pollen (R2 = 0.42, p b 0.005, only for n = 85). The tree cover, however,
is largely under-estimated in the Afromontane zone, where globular phytoliths do not trace high-elevation forests.
Limitations in our interpretation do exist, but could be overcome in the future through additional studies along an elevation/
temperature gradient in the Somalia–Masaï region of East Africa, and with more precise identifications of phytolith types and
sub-types.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Hominids; Palaeo-anthropological sites; Silica bodies; PCA; Tree density; MODIS; Africa

1. Introduction important focus of palaeoenvironmental studies because

In Africa, reconstructing past vegetation, especially
past tree cover density at early hominid sites, is an
⁎ Corresponding author. Fax: +33 4 42 97 15 40.
E-mail address: barboni@cerege.fr (D. Barboni).
1
Present address: Center of Bio-Archaeology and Ecology (UMR5059
CNRS/USTL/EPHE), Institut de Botanique, Université Montpellier 2,
163 Rue Broussonet, F-34090 Montpellier, France.
0031-0182/$ - see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.palaeo.2006.10.012
theories on evolution often imply the influence of the
environment on human evolution, especially the
emergence of bipedalism in early hominids. For some
sites, such as the Ardipithecus ramidus localities of
Aramis, Kuseralee Dora, and Sagantole (White et al.,
1994; Wolde Gabriel et al., 1994) in the Middle Awash
Valley, Ethiopia, fossil pollen are poorly preserved, and
fossilized wood does not show diagnostic anatomical
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470 455

features necessary for identification. Because of these
limitations, phytolith studies were undertaken in an
effort to clarify the environment inhabited by the
hominids and other mammals from these localities.
Being preserved in oxidizing environments, phytoliths
(opal-A particles that precipitate in plant cells) may
compensate for the lack of other plant remains, as shown
by a preliminary phytolith study at Middle Awash
(Barboni et al., 1999) and at other palaeoanthropological
sites in East Africa (Albert et al., 2006). The field of
phytolith studies still is in its infancy, and uncertainties
regarding the interpretations of phytolith data can still be
controversial (Pearsall et al., 2004). However, it is now
well established that phytoliths can seldom relate to
certain groups of plants. Phytolith assemblages obtained
from soil samples proved to be good indicators of the
grass cover in many tropical and temperate regions (see
Strömberg, 2004, for review). Their capability to
characterize grasslands is due to the fact that phytoliths
are particularly distinctive within the Poaceae sub-
families (grasses), whereas pollen is not. However, one
phytolith type cannot relate to any specific plant taxon,
because many basic phytolith types are produced by
many different taxa (redundancy) and because one plant
taxon may produce many different phytolith types
(multiplicity) (e.g. Mulholland, 1989; Fredlund and
Tieszen, 1994).
The capability of phytolith assemblages to charac-
terize forests and to detect changes in the tree cover
density has been investigated at the local scale, and

Fig. 1. Phytogeographical zones from White (1983) and modern phytolith studies in Africa. a: Congo — 2 samples, b: Senegal — 4 samples
(Alexandre et al., 1997b), c: Central Africa — 17 samples (Runge, 1999), d: Ethiopia — 4 samples (Barboni et al., 1999, and one unpublished ),
e: Cameroon — 26 samples (Bremond et al., 2005a), f: Senegal and Mauritania — 61 samples (Bremond et al., 2005b), g: Tanzania, Mt Rungwe —
10 samples, and h: Mt Kenya — 13 samples (Bremond, 2003; Bremond et al., submitted for publication), i: Tanzania Serengeti Plains — 6 samples
(Bremond, unpublished ), and j: Massoko Lake — 6 samples (Bremond, 2003; Bremond et al., submitted for publication).
456 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
provided different results for different regions. For tem-
perate regions in the northern Hemisphere (see Ström-
berg, 2004, for review) and in the tropics at elevations
N1900 m asl (Bremond, 2003; Bremond et al., submitted
for publication), phytolith assemblages failed to reflect
the abundance of trees and shrubs. For tropical regions
but at low and medium elevation, however, phytolith
assemblages proved to reflect accurately local variations
in the tree and shrub cover between forests and savannas,
occurring along less than 1 km long transects (Runge,
1999; Runge and Fimbel, 1999; Bremond et al., 2005a).
At palaeo-hominid sites, it is important to be able to trace
the presence of trees, and to reconstruct the density of the
tree cover in order to determine the openness of the
environment associated with hominids. It is therefore
crucial to identify how well this feature of vegetation is
recorded in the phytolith assemblage, and what are the
potential biases that could lead to misinterpretations of
the fossil data. Another important point is to determine
how appropriate the modern dataset is against which the
fossil data would be compared, and/or what modern
analogues may be missing.
In this study, we aim to investigate the potential and
limitations of phytolith assemblages to reconstruct
vegetation and tree cover density at palaeoanthropolo-
gical sites. We use all modern phytolith data available
for Africa and compare them with tree cover estimates
from satellite data to assess the potential for phytolith
assemblages to estimate the tree cover. With a principal
component analysis (PCA), we also check how modern
vegetation types from different phytogeographical zones
and sub-zones in Africa can be discriminated by
phytolith assemblages.
2. Modern phytolith dataset and phytolith types
We compiled phytolith data (published and unpub-
lished) obtained from the analysis of surface soil sam-
Table 1
Number of surface soil samples in each phytogeographical zones (codes a to j refer to Fig. 1)
Phytogeographic zones (White, 1983)
AFMF Afromontane forest
AFMS Afromontane shrubland/grassland
EVF Evergreen forest (Guineo–Congolian)
MOSA Mosaic semi-deciduous forest and grassland
(Guineo–Congolian)
WOZA Woodland (Zambezian)
WOSU Woodland (Sudanian)
BUSA Deciduous bushland (Sahel)
GRSA Semi-desert grassland (Sahelian–Saharan)
EDAP Edaphic grassland and woodland (Somalia–Masaï)
GRSO Semi-desert grassland (Somalia–Masaï)
ples at 149 different locations in East and West Africa,
collected from low to high elevation and from sub-
desertic steppes to equatorial forests (Fig. 1, Table 1).
This dataset mostly contains phytolith counts or, where
not available, phytolith percentages. It also includes
geographical coordinates and description of the vegeta-
tion at the sampling sites with the names of the phyto-
geographical zones (White, 1983) where the sites are
located (Appendix 1). The phytolith names were
homogenized following the International Code for
Phytolith Nomenclature (ICPN Working group et al.,
2005).
Among all the phytolith types identified and counted
by the different analysts, 13 found in almost all
publications were considered for this study. The small
number of phytolith types selected for this study was
imposed by the disparity of the data available in the
literature. Despite their potential taxonomical/ecological
resolution, phytoliths recognized or counted by only one
analyst had to be excluded to avoid biases in numerical
analyses of all the available phytolith data together.
Bulliform and acicular hair cells were not counted in six
samples from three phytogeographical zones of West
Africa (Alexandre et al., 1997b). As we tested that the
lack of count for these two phytolith types did not
significantly modify our results because other numerous
samples from the same phytogeographical zones were
available, the six corresponding samples have been
integrated in all the statistical analyses after assigning
them a zero value for the non-counted bulliform and
acicular hair cells.
Among the 13 types of phytoliths considered here
(Fig. 2), nine are produced by grasses (Poaceae family).
Phytoliths rondel, trapeziform short cell, and polylobate
are typical for the Pooideae grass subfamily. Bilobates
and crosses are typical for the Panicoideae, and saddles
for the Chloridoideae. Rondels and saddles are also
largely produced by Bambusoideae. These six phytoliths
Samples References
10 [g, h] Bremond, 2003; Bremond et al., submitted for publication
13 [g, h] Bremond, 2003; Bremond et al., submitted for publication
17 [a, f] Alexandre et al., 1997b; Runge, 1999
29 [c, e, f] Runge, 1999; Bremond et al., 2005a,b
6 [j] Bremond, 2003; Bremond et al., submitted for publication
30 [f, b] Alexandre et al., 1997b; Bremond et al., 2005b
27 [f, b] Alexandre et al., 1997b; Bremond et al., 2005b
7 [f] Bremond et al., 2005b
6 [i] Unpublished
4 [d] Barboni et al., 1999, (one unpublished)
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
Fig. 2. Micrographs of the phytolith types considered for this study. Black bar represents 10 μm. 1: Rondel short cell, 2: Trapeziform short cell,
3: Trapeziform polylobate, 4: Bilobate, 5: Cross, 6: saddle. 7: Elongate, 8: Cuneiform bulliform cell, 9: acicular hair cell, 10: Cone-shaped,
11: Globular echinate, 12: Globular granulate, 13: Globular smooth.
are silicified short cells from grass epidermis. The
elongates (from epidermis long cells), cuneiforms (fan-
shaped) and parallepipedal bulliform cells, and acicular
hair cell phytoliths are typical for the grass family in
general; all subfamilies produce them. Palms, sedges and
some other plants also produce the acicular hair cells, and
a wide range of plants produce elongates (Piperno, 1988;
Strömberg, 2004). Assignment of phytoliths to grass
subfamilies is due to several authors and was made using
a variety of sources from the literature on phytolith
morphology (Metcalfe, 1960; Twiss et al., 1969;
Mulholland, 1989; Fredlund and Tieszen, 1994; Kondo
et al., 1994).
Plant taxa other than grasses produce distinct
phytoliths that are included in this study. The cone-
shaped type is produced in the epidermis of Cyperaceae
(sedge) (e.g. Kondo et al., 1994; Wallis, 2003). The
globular echinate type is produced in the wood and
leaves of Palmae (palms) (e.g. Kondo et al., 1994). The
globular granulate type is produced in the wood and
leaves of tropical dicotyledons (ligneous and herba-
ceous) (e.g. Scurfield et al., 1974), as well as in the
leaves of several tropical non-grass monocotyledons,
namely Juncaceae (Kondo et al., 1994), Cannaceae,
Marantaceae, and Strelitziaceae (Piperno, 1985). The
globular smooth type, which is abundant in modern
samples from some East African forests, may have
several origins. It was found in small quantities in some
tropical herbaceous monocots, in moderate quantities in
457
the leaves and seeds of some tropical arboreal dicoty-
ledons (Piperno, 1988), in large quantities in the roots of
some tropical grasses (Alexandre et al., 2000), and in the
leaves, twigs, and wood of some dicotyledons (Kondo
et al., 1994).
Relevant to the relations between the abundance of
grass species and climate, the ongoing work by Watson
and Dallwitz (1992 onwards) provides some informa-
tion on the climatic affinity of world grass species. Other
studies focused on the distribution of C3 and C4 grasses
along altitudinal/temperature and moisture gradients
(Vogel et al., 1978; Tieszen et al., 1979; Cabido et al.,
1997; Wooller et al., 2001; Winslow et al., 2003), but
Twiss (1992) provided the first synthesis on the world
distribution of grasses and on the potential for phytolith
assemblages to be interpreted in terms of climatic and
environmental conditions. In brief, Pooideae are com-
mon in temperate regions and at high elevation in the
tropics (Vogel et al., 1978; Tieszen et al., 1979;
Livingstone and Clayton, 1980). They exclusively use
the C3 photosynthetic pathway (GPWG, 2001). Pani-
coideae (including C3 and C4 grasses) are tall tropical
grasses adapted to warm climate and relatively high
available soil moisture (mesic environments). Chlor-
idoideae (mainly C4) are short tropical grasses but
adapted to dry climates and/or low available soil
moisture (xeric environments) (Vogel et al., 1978;
Tieszen et al., 1979; Livingstone and Clayton, 1980;
Cabido et al., 1997). C4 grasses also thrive in lower
458 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470

atmospheric CO2 than C3 grasses (Ehleringer et al.,

1997).

3. Phytogeographical zones and percent tree cover

at the modern sites

The 149 modern sites sample 10 different phytogeo-

graphical zones and sub-zones (White, 1983), but their
distribution in each zone is uneven (Fig. 1, Table 1). At
high elevation, in the Afromontane zone of Eastern
Africa, there are ten samples collected from the forests
and 13 from the shrublands of Mt Rungwe and Mt
Kenya. At low elevation, more samples are available.
There are 27 samples in the Sahelian bushlands, 30 in
the Sudanian woodlands, and 29 in the forest and
grassland mosaic at the transition between the Sudanian
and Guineo–Congolian zones. There are 17 samples in
the evergreen forests of the Guineo–Congolian zone,
and less than ten in all the other phytogeographical
zones (Table 1). Within each phytogeographical zone,
vegetation physiognomy exhibits substantial variations
(White, 1983). Variations in the tree cover density at the
sampling sites are usually described by the collectors
and taken into consideration, but rarely measured. Only
in Cameroon, a proxy for canopy density (the leaf area
index, Cournac et al., 2002) was measured and
correlated to phytolith (Bremond et al., 2005a) and
pollen abundances (Vincens et al., 2000) in corre-
sponding surface soil samples.
In order to compensate for the lack of measurements
of the tree density at other phytolith sampling sites, we
obtained percent tree cover values produced with
Moderate Resolution Imaging Spectroradiometer
(MODIS) data. Data derived from the MODIS visible
bands and NDVI (Normalized Difference Vegetation
Index) allow the proportion of tree cover to be estimated Fig. 3. Descriptive statistics on the relative abundance of phytoliths
attributed to woody plants and of arboreal pollen, in parallel to percent
for any given 500 × 500 m sample of Earth's surface.
tree cover (averaged) obtained from MODIS satellite data. Mean
MODIS percent tree cover corresponds to the amount of values for phytolith relative abundance in modern soil samples,
skylight obstructed by tree canopies equal to or greater standard error of the mean (SE), 95% confidence interval (1.96 ⁎ SE),
than 5 m in height (Hansen et al., 2003). This product is and number of samples are given for each phytogeographical zone
an annual representation of percent tree cover generated (Table 1, White, 1983).
from monthly composite data for the period from Nov.
2000 to Nov. 2001, somewhat significantly delayed
from the soil sampling dating 1980 in West Africa. Most surface soil samples were collected over areas of a
Multi-temporal metrics averaged over one year (as few hundred meters in size, therefore, this spatial
opposed to individual months) capture variations in the resolution of the satellite data seemed adapted for
tree cover due to phenological cycles (deciduousness), comparing phytolith or pollen content of soil samples
and avoid cloud contamination (Hansen et al., 2002). with tree cover values in the broad comparison under-
We used values of the percent tree cover obtained for taken here.
each 500 m-size grid cell including a sampling site, as At the phytolith sampling sites, MODIS percent tree
well as averages obtained over the nine 500 × 500 m grid cover happens to be in good agreement with vegetation
cells surrounding each sampling site (Appendix 1). description. It was found to be underestimated, however,
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470 459

Fig. 4. Ordination diagram of the surface soil samples and of the phytolith types based on the principal component analysis of the modern African
phytolith assemblages. Axes 1, 2, 3 and 4 explain 25%, 20%, 16%, and 10% of the variance, respectively.

at three sites from the transitional (samples 68 and 69)
and Guineo–Congolian (sample 24) zones. For sites in
the Afromontane zone (1 to 23, Wooller et al., 2001),
shrublands happen to have higher percent tree cover
averages than forests (Fig. 3, Appendix 1). MODIS data
underestimate the tree cover at the six sites in the
Zambezian woodlands (70 to 75). In this case, samples
were collected close to a lake, and the pixel might have
included part of the water surface. Despite such
discrepancies, for each phytogeographical zone the
percent tree cover values evaluated at the sampling sites
by MODIS are relatively well centred on the mean and
confidence intervals are narrow. For example, for the
Sudanian woodland (WOSU, 30 samples) standard error
460 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
Fig. 5. Descriptive statistics on the relative abundance of the most
common grass (Poaceae) phytoliths in the surface soil phytolith
assemblages. Mean values for phytolith relative abundance in modern
soil samples, standard error of the mean (SE), 95% confidence interval
(1.96 ⁎ SE), and number of samples are given for each phytogeo-
graphical zone (Table 1, White, 1983).
of the mean tree cover is ± 1.8% and the 95% confidence
interval includes values between 9% and 16%. Mean
values for MODIS tree cover differ more between than
within the phytogeographical zones, such as for
example between the Guineo–Congolian evergreen
forests (EVF, 76%) and Sudanian woodlands (WOSU,
12%) (Fig. 3, Appendix 1).
4. Method
Details on the extraction methods applied to the soil
samples are given by the authors, and will not be
repeated here. Only noted is the fact that sieving of the
material at 60 μm in most studies (except in Runge,
1999) restricts the analysis to phytoliths of the silt
fraction (2–60 μm).
We undertook several principal component analyses
(PCAs) using CANOCO (ter Braak, 1986), including or
excluding some of the phytolith types in order to
identify those clearly discriminating the vegetation
types, and those responsible for just bringing noise
into the analysis. The final PCA (Fig. 4) was carried out
with the 149 modern phytolith assemblages. We focused
the scaling on the inter-sample distances to obtain an
Fig. 6. Descriptive statistics on the relative abundance of the most
common phytoliths of ligneous plants (trees, shrubs, as well as palms
and forbs with woody basis) in the surface soil phytolith assemblages.
Mean values for phytolith relative abundance in modern soil samples,
standard error of the mean (SE), 95% confidence interval (1.96 ⁎ SE),
and number of samples are given for each phytogeographical zone
(Table 1, White, 1983).
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470 461

Fig. 7. Correlation between satellite percent tree cover and relative abundance of globular phytoliths from ligneous plants at the sampling sites in
Africa. The correlation is highest if restricted to sites at low and mid-elevation (circles) (r = 0.87, n = 126, see Table 2 for R2), because the relative
abundance of globular phytoliths is unexpectedly low for forest and shrubland sites of the Afromontane zone (squares in the shaded area).

ordination diagram that displays more accurately the
dissimilarities between the phytolith compositions of
different samples. We also centred and standardized by
species, so that the PCA was done on the correlation
matrix of the phytolith relative abundances.
Among the 13 types, the elongate type (ubiquitous)
and the cone-shape phytolith typical for Cyperaceae
(sedges), which brought very little discrimination
between the samples (PCA results not shown), were
excluded from the final PCA and percentages of the 11
other phytolith types were recalculated. The phytolith
sums used for the final PCA (Fig. 4) and for the statistics
(Figs. 3,5–8) consequently included eight grass phyto-
lith types (bilobate, cross, saddle, rondel, trapeziform
short cell, polylobate, cuneiform bulliform cells, and
acicular hair cells), and three phytoliths of woody plants
and palms (globular smooth, granulate, and echinate).
As descriptive statistics, we provide for each phytogeo-
graphical zone and type of phytolith, the mean for
phytolith relative abundance, the standard error interval
of the mean (SE), and the 95% confidence interval
(1.96 ⁎ SE). We chose to give the standard error of the
mean for phytolith abundances (which takes into
account the sample size) rather than the standard
deviation, because the number of samples in each
phytogeographical zone is statistically low (b 30).
We also calculated the mean values for four phytolith
indices. The tree cover density index D/P is the ratio of
the ligneous dicotyledons phytoliths (globular granu-
late) versus the Poaceae phytoliths (i.e. sum of short
cells plus bulliform and hair cells). It has been used to
estimate the density of woody plants in different types of
grass savannas between 15° N and 4° S in West Africa
(Alexandre et al., 1997b). In modern samples, a value of
7 was found for the semi-evergreen forest in Congo,
whereas values b 1 characterized savannas with abun-
dant grass cover in Senegal. The aridity index Iph, first
used to appreciate humidity–aridity tendencies in
marine cores (Diester-Haass et al., 1973), is the ratio
of chloridoid phytoliths (saddles) versus total chloridoid
plus panicoid phytoliths (i.e. saddles plus bilobates and
crosses). It expresses the percentage of Chloridoideae
among the C4 grasses. High Iph values suggest grass-
lands dominated by Chloridoideae, i.e. xerophitic
grasses, and hence prevalence of dry climatic condi-
tions. Conversely, low Iph values indicate associations
in which Panicoideae, i.e. mesophytic grasses dominate,
suggesting humid climatic and/or high available soil
moisture. The climatic index Ic was defined to estimate
the relative proportion of C3 grasses present in the
American Great Plains (Twiss, 1987, 1992). It is the
ratio of pooid (i.e. rondel plus trapeziform short cell plus
polylobate phytoliths) versus the sum of short cell
phytoliths. High Ic values suggest an abundance of
Pooideae, and hence an abundance of C3 grasses.
Abundance of C3-Pooideae characterizes high latitudes
(Twiss, 1992) and high altitudes in the tropics (Tieszen
et al., 1979; Livingstone and Clayton, 1980). Conse-
quently, high Ic values suggest cool climate. The water
stress index Fs, which is the percentage of cuneiform
bulliform cells (fan-shaped phytoliths) to the sum of
grass phytoliths minus the elongate phytolith type, was
462 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
Fig. 8. Descriptive statistics on the phytolith indices D/P (Alexandre
et al., 1997b), D/P with a zoom on values b1, Iph (Diester-Haass et al.,
1973), Ic (Twiss, 1987), and Fs (Bremond et al., 2005b). Mean values
for indices, standard error of the mean (SE), 95% confidence interval
(1.96 ⁎ SE), and number of samples are given for each phytogeo-
graphical zone (Table 1, White, 1983).
defined to measure water stress and transpiration
suffered by the grass cover along a latitudinal gradient
in West Africa (Bremond et al., 2005b). High Fs values
suggest low water availability.
5. Results
The PCA indicates that 70% of the variance included
in the modern phytolith dataset is explained by the first
four axes, and that these axes relate to environmental
gradients. PCA-axis 1 (which explains 25% of the total
variance) distinguishes non-forested from forested
environments (Fig. 4). It is defined by saddle, bilobate,
cross, bulliform cells, and hair cells phytoliths (i.e. grass
phytoliths) at its negative end, and by globular
phytoliths (woody plants and palm phytoliths) at its
positive end. PCA-Axis 2 (20% of the variance) is
defined by rondel, polylobate, and trapeziform short cell
phytoliths (at its positive end) and by all the other types
(at its negative end). Axis 2 distinguishes high-elevation
from low-elevation vegetation types; however, sites
from the Middle Awash subdesertic steppe in the
Somalia–Masaï zone (elevation b 600 m) fall in the
diagram with those from high elevation. PCA-axis 3
(16% of the variance) is defined by bilobate and cross
phytoliths (at its negative end), and by bulliform and
hair cells phytoliths (at its positive end) (Fig. 3). The
dispersion of sites indicates that PCA-axis 3 distin-
guishes mesic from xeric sites. Finally, PCA-axis 4
(10% of the variance) is defined essentially by globular
smooth phytoliths (at its negative end) and globular
echinate phytoliths (at its positive end). The dispersion
of sites indicates that PCA-axis 4 represents a gradient
in the abundance of palms at the sampling sites; its
positive end being defined by site 41, a Raphia swamp
in Cameroon (Appendix 1).
Without considering the phytolith assemblages as a
whole, such as for example with a PCA, it can prove
hard to discriminate vegetation types because the
occurrence of phytolith types, at least the 11 considered
here, is redundant in many phytogeographical zones
(Figs. 5 and 6). Apart from polylobates, which are only
present (but never exceed 10%) in samples from the
Afromontane zone, none of the other phytoliths seems
to occur in only one place. Rondel and trapeziform short
cells for example, increase in abundance with increasing
elevation (up to 60% above 1900 m), but are not typical
phytoliths for the Afromontane zone because they may
still represent as much as 20% in samples from the
Somalia–Masaï phytogeographical zone despite the low
elevation (GRSO,b 600 m). For this reason, the semi-
desert steppes of East Africa (GRSO) are hard to
discriminate from the Afromontane vegetation, but are
well discriminated from the West African vegetation
types, whose phytolith assemblages have no rondels.
Although they occur in samples from all vegetation
zones, bilobate and cross phytoliths are most abundant
in soils from the sub-Saharan zones of West Africa, such
as the Sudanian and Sahelian phytogeographical zones
(WOSU and BUSA, abundance of bilobate is N 33%,
Fig. 5). Saddles are found in most samples from the sub-
Saharan zones and the Somalia–Masaï zone (BUSA,
EDAP, and GRSO, between 5% and 20% of saddles),
and found in samples from the Afromontane zone
(AFMF and AFMS, up to 12%, but less than 4% in
samples from the bamboo forest). Saddles (like rondel
and trapeziform short cells) occur in significant
abundance in soils from both high and low elevation
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470 463

Table 2
Coefficient of determination (R2) between MODIS satellite estimates of the percent tree cover (averaged values over 1.5 km size window), and the
relative abundance (in %) of globular phytoliths, the D/P index (Alexandre et al., 1997b), and the relative abundance of arboreal pollen in the n
surface soil samples
MODIS % tree cover Σ Globular % Globular % Globular % Globular D/P % Arboreal
(averaged) phytoliths granulate smooth echinate pollen
All samples
n = 149 0.60 0.48 0.23 0.05 0.33 -
n = 85 0.57 0.62 ns 0.10 0.39 0.42
Samples at low and mid elevation (afromontane excluded)
n = 126 0.75 0.62 0.26 0.06 0.40 -
n = 77 0.66 0.72 ns 0.11 0.43 0.41
F-probability distribution and associated p-level were calculated to test the significance of R2, which are significant at p b 0.005 except where
indicated (ns: non-significant). The complete phytolith dataset includes 149 sites among which 85 have pollen counts. At low and mid elevation
(Afromontane sites excluded) there are 126 samples, among which 77 have pollen counts.

zones; only in soils from the forests and forest–
grassland mosaics of the Guineo–Congolian zone
(EVF and MOSA), the proportion of saddle is b4%.
Hair cells and bulliform cells phytoliths are found most
abundant (30% to 39%) in samples from the semi-desert
Sahelian–Saharan grasslands (GRSA) (for data on
bulliforms see Fig. 8).
As expressed by PCA-axis 1, tracing the tree cover at
the scale of inter-tropical Africa with phytoliths should
be possible, because the strongest phytolith signal in
modern assemblages is the gradient between forested
and non-forested environments that is linked to the
abundance of globular phytoliths. The abundance of
globular phytoliths is indeed very high (65% to 90%) for
evergreen forests and forest–grassland mosaics of the
Guineo–Congolian zone (EVF, MOSA), high for the
Zambezian woodlands (WOZA, 50%), and low (b10%)
for grasslands and bushlands (GRSA, BUSA, EDAP)
(Figs. 3 and 6). Unexpectedly, the mean sum of globular
phytoliths for forests and shrublands of the Afromon-
tane zone (AFMF, AFMS) is b 10%, a very low value
despite the abundance of Ericaceae and other ligneous
dicotyledons at all sampling sites (but two from a
bamboo forest). The abundance of globular phytoliths is
found high for the semi-desert grasslands of the
Somalia–Masaï zone for which the sum of globular
phytoliths averages 22% (GRSO). This high value is
accounted for by few samples mostly collected close to
local riparian forest, that are not representative of the
tree distribution in subdesertic steppe (Fig. 3). Satellite
data provide an objective estimate of the tree cover, but
the 500-m-resolution of MODIS data does not allow
tracing local features of the landscape, such as the
presence of riparian forests. Despite this, we found
positive and significant correlations (at p b 0.005)
between the percent tree cover estimated by satellite
images, and the abundance of globular granulate
(R2 = 0.48), globular smooth (R2 = 0.23), and globular
echinate (R2 = 0.05) phytoliths (Table 2, n = 149). The
correlation is highest between the percentage tree cover
estimated by MODIS satellite images and the sum of
globular granulate plus smooth plus echinate types
(R2 = 0.60). Also, at 85 sites where both phytolith and
pollen data were available (n = 85) (Lézine, 1988;
Lézine and Edorh, 1991; Vincens et al., 2000, 2006),
we found that the abundance of globular phytoliths
correlates better to the satellite estimates of the tree
cover (R2 = 0.57) than does the abundance of arboreal
pollen (R2 = 0.42). We also found that correlations are
always higher with the percent tree cover averaged
(1.5 km window) than with the percent tree cover value
(500 m window, not shown), because averaged values of
the tree cover for the Afromontane shrublands are
lowered by 26% compared to the non-averaged values.
Besides, the strength of the correlation between the sum
of globular phytoliths and satellite tree cover is
improved when only applied to low and mid-elevation
data (R2 = 0.75, n = 126) (Fig. 7, Table 2).
Mean values, standard errors, and confidence inter-
vals for the phytolith indices D/P, Iph, Ic, and Fs are
given in Fig. 8. The D/P index is successful in marking
the difference of tree density existing between the phy-
togeographical zones. In fact, the difference between
EVF, MOSA, WOZA, and WOSU is better marked by
the D/P index (+ 683% from WOSU to WOZA; +465%
from WOZA to MOSA, and + 189% from MOSA to
EVF) than by the sum of globular phytoliths (+ 505%
from WOSU to WOZA; + 35% from WOZA to MOSA,
and +40% from MOSA to EVF). However, because the
D/P index is proportional to the abundance of Poaceae
phytoliths (P), the correlation with MODIS tree cover
data is stronger with the sum of globular phytoliths
(R 2 = 0.60) than with the D/P index (R 2 = 0.33)
(Table 2). If D/P is a reliable index, however, Iph and
464 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
Ic indices may not provide taxonomical/ecological
resolution in this dataset, which considers distribution
of the vegetation at the scale of inter-tropical Africa.
This is because many rondel, trapeziform short cell,
polylobate, and saddle phytoliths occur in soil samples
at both high and low elevation in East Africa. Those two
indices, but mostly Ic, exhibit the same range of values
for Somalia–Masaï lowland samples (Ic, 30% to 59%)
and for Afromontane highland samples (Ic, 43% to
72%) (Fig. 8).
6. Discussion
6.1. Co-occurrence of rondel and polylobate phytoliths
trace current cool temperatures
Phytolith analysis of modern soils indicates that it is
hard to discriminate highland from lowland formations
of East Africa, because short cell phytoliths rondel and
trapeziform, characteristic for temperate Pooideae
grasses, are abundant (up to 22%) also in surface soils
from the Somalia–Masaï grasslands and woodlands, at
low elevation (b 600 m) under much hotter and drier
conditions than in highlands. In the surface soil samples
of the Middle Awash Valley, rondel and trapeziform
short cell phytoliths are associated with less than 1%
polylobates, which indicates that extra-local input of
phytoliths from vegetation above 1900 m asl is most
unlikely. High percentages for rondels and trapeziform
short cells in surface soil samples could be attributed to
some non-conventional phytolith production by local
grasses, such as those of the Chloridoideae sub-family.
Among the Chloridoideae, Sporobolus, Eleusine,
Chloris and Eragrostis produce rondel and trapeziform
short-cell phytoliths instead of or in addition to saddles
(Watson and Dallwitz, 1992 onwards). In the Serengeti
Plains, at Olduvai Gorge, and Olbalbal (Tanzania),
where a quantitative phytolith study was conducted, the
C4 grasses Sporobolus consumilis, S. spicatus, and Cy-
nodon dactylon produce 35% to 80% of C3-type rondel
phytoliths (“tower type short cells”, Bamford et al.,
2006). The presence of Chloridoideae species at the
Middle Awash sampling sites is most likely given the
prevailing arid conditions, but their abundance is
unknown. In an early botanical survey of East Africa,
the presence of Chloridoideae species such as Dacty-
loctenium robecchii, Sporobolus variegatus and Tetra-
pogon villosus was recorded in the xerophitic
woodlands of the arid and semi-arid parts of the
Awash Valley (Pichi-Sermolli, 1957). More recently, a
survey conducted in and near the Awash National Park,
but at higher elevation than the Middle Awash sampling
sites, indicates that other Chloridoideae species, such as
Sporobolus natalensis, Eleusine indica, Eragrostis
cilianensis, and Chloris pychnothrix occur in abundance
between 750–1200 m asl, in heavily as well as lightly
grazed open grasslands (Abule et al., 2005). Further
north, in the Arabian Peninsula, Sporobolus spp and
Urochondra setulosa are also abundant, and were
found to produce rondel and trapeziform short cell
phytoliths rather than saddles (Ball, 2002). More (soil
and plant) samples from the Middle Awash Valley and
East Africa are needed to determine the phytolith signal
of the Somalia–Masaï subdesertic steppes, and potential
biases due to the production of rondel phytoliths by
Chloridoideae species. In addition, future studies could
test if Chloridoideae-rondels can be discriminated from
Pooideae-rondels. According to Metcalfe (1960), some
Chloridoideae species produce rondel phytoliths with
slightly crescentic outlines and tops.
The data synthesis presented here shows that the co-
occurrence of polylobate and rondel phytoliths is re-
stricted to surface soil samples collected above 1900 m
asl. Polylobates occur mainly in Afromontane soil
samples (2% to 10%). They do not occur elsewhere in
East Africa, not even in the edaphic grasslands of the
Serengeti Plains (1250 m to 1725 m asl). Polylobates
may be present in some modern samples from a tree
savanna in Cameroon because of abundant Panicoideae
grasses (Bremond et al., 2005a), but their proportion is
b3% and they do not occur in association with rondels.
Higher taxonomical resolution within this dataset would
have shown that trapeziform polylobates typical for
Pooideae occur at high elevation, whereas cylindrical
polylobates, typical for grasses of the PACCAD clade
(Panicoideae, Arundinoideae, Chloridoideae, Centho-
thecoideae, Aristidoideae, and Danthonioideae) occur
at low elevation (Strömberg, 2004, pp 256–257). In
tropical grasses of Central and South America, trapezi-
form polylobates were found to have a “one-to-one
correspondence” with the Pooideae, a subfamily
geographically restricted to the highlands (Piperno and
Pearsall, 1998). Polylobates (trapeziform?) do not seem
to occur at low elevation in the tropics, even outside
Africa (e.g. in northeastern Australia — Thorn, 2004).
In soil samples from warm temperate, temperate, and
cool temperate regions such as the Great Plains of North
America (Fredlund and Tieszen, 1994), the Pacific
North West (Blinnikov, 2005), the Mediterranean region
(Bremond et al., 2004) and the eastern North Island in
New Zealand (Carter, 2002), trapeziform polylobates
are always found in abundance, and in association with
rondels. In the Great Plains, trapeziform polylobates and
rondels increase in abundance with decreasing summer
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
temperature, and the abundance N10% of polylobates is
essentially found at sites where mean summer temper-
ature is b20 °C (Fredlund and Tieszen, 1997). In East
Africa, proportions of Pooideae and Arundinoideae (C3-
grasses) versus Panicoideae (C3 and C4 grasses), which
increase with increasing altitude and are N50% above
2080–2600 m asl, are well reproduced in the surface soil
phytolith assemblages by the increasing abundance of
rondels, trapeziform short cells and trapeziform poly-
lobates, which become more abundant in samples
collected above 1900–2000 m asl (Bremond, 2003;
Bremond et al., submitted for publication), that is where
mean annual temperature is b19 °C (Leemans and
Cramer, 1991). The altitudinal/temperature distribution
of C3 and C4 grasses found for East Africa is in
agreement with modelling results showing that the
transition from domination by C3 grasses to domination
by C4 grasses should occur at about daytime growing
season temperatures of 21°–25 °C for current CO2
concentrations (ca 350 ppmV) (Ehleringer et al., 1997).
Therefore, the joint occurrence of both rondels and
trapeziform polylobates is a good indication of cool
temperature. However, more data from high elevation
sites in Africa are needed to evaluate if the temperature
threshold is 19 °C (or 21 °C for the daytime growing
season temperature). Even ascertained by more data, the
value of this threshold depends upon current CO2
concentration, and may not be valid for older geological
periods. Under higher CO2 levels such as predicted for
the Late Miocene (500 ppmV), the C3/C4 grass
transition is expected to have occurred at temperatures
of about 31 °C (Ehleringer et al., 1997). Therefore,
climatic interpretation of fossil phytolith assemblages
would need more profound investigation.
6.2. Globular phytoliths trace the current tree cover
density at mid- and low-elevation
The good correlation coefficient obtained between
the sum of globular phytoliths and the satellite percent
tree cover (R2 = 0.60, p b 0.005, n = 149) demonstrates
that the current tree cover is traceable with globular
phytoliths (granulate, smooth, and echinate) at the scale
of inter-tropical Africa. Despite the production of
globular smooth phytoliths by both woody (e.g.
Piperno, 1988) and non-woody (e.g. Alexandre et al.,
2000) plants, the PCA assigns globular smooth
phytoliths to woody environments because R2 = 0.23
(p b 0.005) between the abundance of globular smooth
phytoliths and the percent tree cover. It is unclear why
the smooth type is most abundant in soils from the
Guineo–Congolian evergreen forests (Runge, 1999) and
465
the Zambezian woodland (miombo forests from Tanza-
nia) (Bremond, 2003), and is rare in soils from the semi-
deciduous forest–grassland mosaics where the globular
granulate type is most abundant (Alexandre et al.,
1997b; Bremond et al., 2005a,b). Investigating the
phytolith production of the plants growing in these
habitats could add some taxonomical/ecological dis-
crimination to the globular smooth and granulate
phytolith types. However, summing globular smooth,
globular granulate, and globular echinate (palms) gives
the best estimation of the satellite tree cover. Yet, the
correlation between satellite tree cover and globular
phytoliths is not perfect. There are problems of under-
and of over-estimation of the tree cover especially for
the Zambezian woodlands and the Afromontane vege-
tation. For the Zambezian woodlands, the mean
abundance of globular phytoliths is 45%, while mean
satellite tree cover is 16%. This apparent over-esti-
mation of the miombo tree cover by spherical phytoliths
is artificial, and linked to the fact that MODIS data
systematically, and for various reasons, underestimate
the tree cover of miombo deciduous woodlands by 10–
20% (Hansen et al., 2005). In addition, the six sampling
sites are located close to Lake Massoko, which probably
affects the satellite tree cover averaged value by reduc-
ing it if one or more of the nine windows of 500 × 500-m
surrounding the sites include some water area. Field
measurements in miombo woodlands from Zambia
indicate a large range of values from 10% to 60% for
tree crown cover (Frost, 2000). Therefore, the abun-
dance of globular phytoliths at Massoko sampling site
remains realistic within this range of tree cover values.
For Afromontane vegetation, the satellite tree cover
(averaging 30% and 45%) is under-estimated by
globular phytoliths, which account for less than 20%
in the Afromontane soil samples. The low abundance of
globular phytoliths in soils from the Afromontane zone
(Bremond et al., submitted for publication), cannot be
explained by the occurrence of bamboos in the
Afromontane forests. Only two samples out of the 23
from the Afromontane zone were collected in bamboo
forests (Appendix 1, samples 4 and 5). It could be
explained, however, by the fact that some woody taxa of
the Afromontane zone such as Hagenia, Vaccinium,
Schefflera, Hypericum, Juniperus produce few or no
phytoliths at all (Runge, 1996; Kealhofer and Piperno,
1998; MU Phytolith database, 2004; Blinnikov, 2005),
therefore leading to an over-representation of grasses
versus woody trees and shrubs phytoliths. For compar-
ison, in the Central Alps the annual production of
biogenic silica from subalpine Ericaceae could be one
order of magnitude smaller than that from subalpine and
466 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
alpine meadows (contributions of the above ground
biomass, Carnelli et al., 2001). The low abundance of
globular phytoliths in Afromontane soils could likely be
explained by the fact that the Afromontane taxa produce
phytolith types other than the globular type used here to
trace the tree cover. Silicified epidermal jigsaw cells,
stomata complexes, and vessels are abundant in
Ericaceae, while globular types are not found at all
(Carnelli et al., 2004). Given that not all woody
dicotyledons produce the globular type, reconstructing
tree cover density merely from the globular phytoliths
may lead towards under-estimated tree cover values. It
was suggested (Strömberg, 2004) that the phytolith
signature of forests might be improved by considering
all size fractions (rather than fraction b 60 μm), because
phytolith indicators of forest range in size from 2 to
800 μm (Hansen et al., 1998; Runge, 1999). In this
study, the data compiled deal with fractions limited to
size b 50–63 μm, which does not allow large silicified
cells from woody taxa to be taken into account.
However, silicified cells produced by woody dicotyle-
dons are most sensitive to dissolution because of their
large specific surface area (Bartoli and Wilding, 1980;
Bartoli, 1985). They are more prone to disappearing in
fossil sediments or to loosing their characteristic shape
and surface features than compact globular phytolith
types (Alexandre et al., 1997a). Future studies should
consider phytoliths of all sizes, and then test if including
large silicified cells adds to the resolution in tree cover
estimates (Strömberg, 2004).
Trees and shrubs of north temperate regions in
Europe lack globular phytoliths in their wood (Ström-
berg, 2004). Such observation may suggest that
production of globular phytoliths in the wood tissues
might be temperature-dependant. Although this relation
should be measured by analysing the globular phytolith
content in wood of particular taxa of trees along gra-
dients of altitude and/or temperature, we observe with
our dataset of surface soil samples, that the abundance of
globular phytoliths seems to decrease with increasing
altitude (decreasing temperature), such as the abundance
of globular phytoliths is close to 0% at 3500 m asl. At
such altitude, mean annual temperature (MAT) is ca
10 °C (Leemans and Cramer, 1991). Interestingly, in
temperate and boreal regions of North America and
Europe (e.g. Bozarth, 1993; Delhon et al., 2003;
Bremond et al., 2004; Blinnikov, 2005), where MAT
is b12 °C (MAT given by the authors or obtained from
FAO, 2002), globular phytoliths are absent from forest
soils. Yet, in warm temperate and temperate forests of
New Zealand, globular phytoliths can be very abundant
(75% to 95%, except in Pine forests; Kondo et al., 1994;
Carter, 2002) even at sites where MAT is 10° to 12 °C
(NIWA, 2005). As suggested, the lesser amount of
phytoliths in plants at high elevation could be explained
by a decreasing turnover of biogenic silica as a conse-
quence of reduced biological activity due to increasing
altitude (Carnelli et al., 2001). The altitude effect should
however affect all plants (grasses as well as woody
plants). Environmental factors such as temperature,
water availability, and silicon content in the soil in-
fluence the production of phytoliths (e.g. Rosen and
Weiner, 1994, and references therein). But also, an in-
creasing number of studies indicate that phytolith pro-
duction is primarily controlled by the phylogenetic
position of a plant (e.g. Hodson et al., 2005). Hence,
plants have a greater chance to be big phytolith
producers if they are angiosperms rather than gymnos-
perms, and if they belong to the clades Poales (that
includes grasses) and Arecales (palms) rather than Eri-
cales (Hodson et al., 2005). Future work is necessary to
test what factor is most important, phylogenetic position
or environmentally controlled variation in silica output.
6.3. Interpretation of the phytolith indices at the scale
of inter-tropical Africa
Fs index is the relative abundance in the phytolith
assemblages of cuneiform bulliform phytoliths, which
precipitate in the grass epidermis when the grass
experiences a strong drought stress. It is possible to
estimate environmental aridity from those phytoliths,
such as shown along the latitude gradient in West Africa
(Fs index, Bremond et al., 2005b), and in Brazil (Borba
Roschel et al., 2006). Indeed, in the study presented here
(Fig. 8), greater values of Fs are observed in xeric
GRSO to GRSA than in mesic (EVF and MOSA) forests
and woodlands.
Ic index measures the proportion of phytoliths
rondel, trapeziform short cell, and trapeziform poly-
lobate over all grass short-cell phytoliths. It was defined
to estimate the relative abundance of C3-temperate
grasses over C4-warm grasses in North America (Twiss,
1992). A recent study including 24 samples from Lake
Massoko, Mt Rungwe (Tanzania), and Mt Kenya
(Kenya) in East Africa, showed that the correlation
between Ic and elevation is strong (R2 = 0.92) (Bremond,
2003; Bremond et al., submitted for publication).
Calculated in relation to the whole African dataset, the
correlation remains strong (R2 = 0.81, n = 143), which
indicates the potential for Ic index to be applicable at the
scale of inter-tropical Africa. However, our results show
that if Ic values N 60% are typical for high-elevation C3-
Pooideae-dominated vegetation and Ic values b 40% are
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
typical for low elevation C4-Chloridoideae-dominated
vegetation, the distinction is not so clear between the
two vegetation types for intermediate Ic values between
40% and 60%. This overlap of Ic values is due to the fact
that phytoliths for typical C3-temperate grasses (rondel
and trapeziform short cells) may also be produced in
abundance by some C4-warm grasses such as Sporo-
bolus, Urochondra, and Cynodon in the Chloridoideae
sub-family (Ball, 2002; Bamford et al., 2006), and that
typical phytoliths for C4-warm grasses (saddles) are
produced by C3-temperate grasses such as bamboos
(Metcalfe, 1960). In Africa, the only phytoliths to be
present almost exclusively in soils from high-elevation
vegetation zones are the trapeziform polylobates, but
their relative abundance never exceeds 10%, which is
too low to reduce the overlap of Ic values obtained for
the Afromontane zone (high elevation) and the Soma-
lia–Masaï zone (low elevation). In the Americas, where
the collapsed saddles were found characteristic for some
Bambusoid grasses (subtribes Guadineae and Chusqui-
neae) (Piperno and Pearsall, 1998; Strömberg, 2004),
the Ic index could be redefined taking this precision into
account. More data from East African lowlands and
along an elevation gradient are needed to fully establish
the significance of lower Ic values within arid or sub-
arid region/vegetation, such as obtained here in
preliminary results from the Serengeti area.
Iph index measures the proportion of saddle phyto-
liths over saddle, bilobate, and cross short cell phytoliths
to approximate the relative abundance of Chloridoideae
(C4-short grasses in the most xeric environments, except
Eragrostis walteri and Merxmuellera rangei, C3,
GPWG, 2001) over Panicoideae (C3 and C4-tall grasses
in mesic environments) (Diester-Haass et al., 1973).
However, Chloridoideae (Sporobolus, Urochondra,
Cynodon) produce phytoliths possibly confused with
C3-type phytoliths, and Bambusoideae, which also
produce saddle type phytoliths, have not been distin-
guished. These discrepancies may explain why our
study does not indicate any clear difference between
vegetation types regarding distribution of the Iph index.
Highest values of the Iph index were found for semi-
desert vegetation types where xeric conditions prevail,
but high Iph values were also found under locally and
periodically humid environments near temporary
flooded rivers or swamps where the studied samples
were collected at Olduvai and Middle Awash, and where
the grass phytolith signal may have been modified
versus more mesic C4. While interpreting fossil
phytolith data, special attention will need to be made
in separating ecological information corresponding to
local environment from more regional conditions that
467
prevailed in the wider region and corresponding to
global climate.
6.4. Further discrimination of phytolith types should
improve discrimination of vegetation types
Identification of all phytolith types and sub-types
present in a sample could largely improve the inter-
pretation of phytolith assemblages (Strömberg, 2004).
The African phytolith dataset in this work contains 13
phytolith types securely identified by all authors. Sub-
types of saddle, polylobate, and rondel phytoliths are
also identified by other authors (e.g. Mulholland, 1989;
Piperno and Pearsall, 1998; Strömberg, 2004), and need
to be individually counted in future studies from Africa.
Among the phytolith sub-types that may contribute to the
discrimination of Afromontane vegetations are the
collapsed saddles typical for American Bambusoid and
the trapeziform polylobates typical for Pooideae. Among
phytolith sub-types discriminating of subdesertic steppes
are the crescent-shaped rondels that may be typical for
some species of Chloridoideae (Metcalfe, 1960).
For future work, we recommend to 1) study phytoliths
of all sizes (2–250 μm) to check for forest indicators,
2) observe and count individually all phytolith types and
sub-types, and 3) compare the relative frequencies of
all phytoliths types and sub-types, even of the ones that
are not currently diagnostic (Strömberg, 2004). Such as
routinely done in African palynology, it seems crucial to
observe phytoliths at ×1000 magnification, and to mount
slides with a liquid medium (e.g. glycerine) to allow
identification of phytoliths complex types and sub-types
in three dimensions. Also, taxonomical and ecological
interpretation of phytoliths would be helped by analys-
ing the phytolith assemblages of surface soil samples in
connection to dominant plant taxa at the site.
7. Conclusions
This study shows that, in the pool of phytoliths that
are easily preserved in surface soils, 11 out of 13
morphological types allow discrimination of most
vegetation types occurring in Africa today. High
elevation (Afromontane) vegetation, however, cannot
be discriminated from subdesertic grasslands wide-
spread in the Somalia–Masai region at mid- and low-
elevation, because production of rondel phytoliths is
redundant to many grass species in East Africa. Our
study also shows that phytolith assemblages allow
tracing the tree cover density, as the abundance of
globular phytoliths almost reproduces the percent tree
cover estimated from 500-m-ground resolution satellite
468 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
data (MODIS). However, phytolith assemblages under-
estimate tree cover for Afromontane vegetation.
Phytolith assemblages carry a temperature signal
through the co-occurrence of two phytolith types. How-
ever, associated rondels and polylobates, typical for
present-day Afromontane vegetations cannot be directly
applied to fossil sediments to extract climatic para-
meters, because C3–C4 grass distribution depends not
only on atmospheric temperature and aridity, but also on
CO2 levels. This needs consideration while addressing
environmental reconstruction during older geological
time such as the Miocene.
Improvement in phytolith studies will come from
more precise identification of phytolith sub-types, such as
the collapsed saddles of Bambusoideae, the trapeziform
polylobates of Pooideae, and the crescent-shaped rondels
of Chloridoideae. Additional studies along an elevation/
temperature gradient, notably in East Africa are needed.
Acknowledgements
We thank S. Alleaume for extracting the MODIS data,
C. Hély for extracting the temperature averages for
Africa, J-J. Motte for drawing and N. Guibert for
assistance in preparing some samples. We are very
grateful to T. White and A. Alexandre for fruitful dis-
cussions, and to two anonymous reviewers and P.
Kershaw for their useful comments on an earlier version
of the manuscript. DB thanks R. Byrne at UC Berkeley
for hosting her in his lab in 2005. LB's work was
supported by the French programs PNEDC (Programme
National d'Etudes Dynamique du Climat, INSU-CNRS,
projects ECHO and “Calibration des indices phytolithi-
ques pour l'intégration de la dynamique des biomes
herbacés dans la modélisation des végétations passées”),
ACI “Ecologie Quantitative” (RESOLVE), and by the 5th
EU PCRDT (EVK2-CT-2002-00153: MOTIF). Thanks
also to the NSF-HOMINID program Revealing Hominid
Origins Initiative (rhoi.berkeley.edu) for support.
Appendix A. Supplementary data
Supplementary data associated with this article
can be found, in the online version, at doi:10.1016/
j.palaeo.2006.10.012.
References
Abule, E., Smit, G.N., Snyman, H.A., 2005. The influence of woody
plants and livestock grazing on grass species composition, yield
and soil nutrients in the Middle Awash Valley of Ethiopia. Journal
of Arid Environments 60, 343–358.
Albert, R.M., Bamford, M.K., Cabanes, D., 2006. Taphonomy of
phytoliths and macroplants in different soils from Olduvai Gorge
(Tanzania) and the application to Plio–Pleistocene palaeoanthro-
pological samples. Quaternary International 148, 78–94.
Alexandre, A., Meunier, J.D., Colin, F., Koud, J.M., 1997a. Plant
impact on the biogeochemical cycle of silicon and related
weathering processes. Geochimica et Cosmochimica Acta 61,
677–682.
Alexandre, A., Meunier, J.-D., Lézine, A.-M., Vincens, A., Schwartz,
D., 1997b. Phytoliths indicators of grasslands dynamics during the
late Holocene in intertropical Africa. Palaeogeography, Palaeocli-
matology, Palaeoecology 136, 213–219.
Alexandre, A., Bouvet, M., Meunier, J.-D., 2000. Phytolith and the
biochemical cycle of silicon in a savanna ecosystem. Presented at
3rd I.M.P.R. Man and the (Palaeo)Environment. The Phytolith
Evidence, Tervuren, Belgium.
Ball, T.B., 2002. Phytoliths of Dhofar, Oman, Arabia. Brigham Young
University, Provo, UT. CD-ROM.
Bamford, M.K., Albert, R.M., Cabanes, D., 2006. Plio–Pleistocene
macroplant fossil remains and phytoliths from Lowermost Bed II
in the eastern palaeolake margin of Olduvai Gorge, Tanzania.
Quaternary International 148, 95–112.
Barboni, D., Bonnefille, R., Alexandre, A., Meunier, J.D., 1999.
Phytoliths as palaeoenvironmental indicators, West Side Middle
Awash Walley, Ethiopia. Palaeogeography, Palaeoclimatology,
Palaeoecology 152, 87–100.
Bartoli, F., 1985. Crystallochemistry and surface properties of biogenic
opal. Journal of Soil Science 36, 335–350.
Bartoli, F., Wilding, L.P., 1980. Dissolution of biogenic opal as a
function of its physical and chemical properties. Soil Science
Society of America Journal 44, 873–878.
Blinnikov, M.S., 2005. Phytoliths in plants and soils of the interior
Pacific Northwest, USA. Review of Palaeobotany and Palynology
135, 71–98.
Borba Roschel, M., Alexandre, A., Drummond Chicarino Varajão, A.F.,
Meunier, J.-D., Chicarino Varajão, C.A., Colin, F., 2006. Phytoliths
as indicators of pedogenesis and paleoenvironmental changes in the
Brazilian cerrado. Journal of Geochemical Exploration 88, 172–176.
Bozarth, S.R., 1993. Biosilicate assemblages of boreal forests and aspen
parklands. In Current research in phytolith analysis: Applications in
archaeology and paleoecology, eds. Pearsall, D.M., Piperno, D.R.,
University Museum of Archaeology and Anthropology, University
of Pennsylvania, Philadelphia, 95–105.
Bremond, L., 2003. Calibration des fonctions de transfert entre
assemblages phytolithiques, structure des végétations et variables
bioclimatiques actuelles, pour l'intégration de la dynamique des
biomes herbacés dans les modèles de végétation. PhD thesis.
University of Aix-Marseille III, France. 197 pp.
Bremond, L., Alexandre, A., Vela, E., Guiot, J., 2004. Advantages and
disadvantages of phytolith analysis for the reconstruction of
Mediterranean vegetation: an assessment based on modern
phytolith, pollen and botanical data (Luberon France). Review of
Palaeobotany and Palynology 129, 213–228.
Bremond, L., Alexandre, A., Hély, C., Guiot, J., 2005a. A phytolith
index as a proxy of tree cover density in tropical areas: calibration
with Leaf Area Index along a forest–savanna transect in
southeastern Cameroon. Global and Planetary Change 45, 277–293.
Bremond, L., Alexandre, A., Peyron, O., Guiot, J., 2005b. Grass water
stress estimated from phytoliths in West Africa. Journal of
Biogeography 32, 311–327.
Bremond, L., Alexandre, A., Wooler, M.J., Hély, C., Williamson, D.,
Schäfer, P.A., Majule, A., Guiot, J., submitted for publication.
 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
Phytolith indices as proxies of grass subfamilies on East African
tropical mountains. Global and Planetary Change.
Cabido, M., Ateca, N., Astegiano, M.E., Anton, A.M., 1997.
Distribution of C3 and C4 grasses along an altitudinal gradient in
Central Argentina. Journal of Biogeography 24, 197–204.
Carnelli, A.L., Madella, M., Theurillat, J.-P., 2001. Biogenic silica
production in selected alpine plant species and plant communities.
Annals of Botany 87, 425–434.
Carnelli, A.L., Theurillat, J.-P., Madella, M., 2004. Phytolith types and
type-frequencies in subalpine–alpine plant species of the European
Alps. Review of Palaeobotany and Palynology 129, 39–65.
Carter, J.A., 2002. Phytolith analysis and paleoenvironmental
reconstruction from Lake Poukawa Core, Hawkes Bay, New
Zealand. Global and Planetary Change 33, 257–267.
Cournac, L., Dubois, M.-A., Chave, J., Riéra, B., 2002. Fast
determination of light availability and leaf area index in tropical
forests. Journal of Tropical Ecology 18, 295–302.
Delhon, C., Alexandre, A., Berger, J.-F., Thiébault, S., Brochier, J.-L.,
Meunier, J.-D., 2003. Phytolith assemblages as a promising tool
for reconstructing Mediterranean Holocene vegetation. Quaternary
Research 59, 48–60.
Diester-Haass, L., Schrader, H.J., Thiede, J., 1973. Sedimentological
and paleoclimatological investigations of two pelagic ooze cores
off Cape Barbas, North-West Africa. Meteor-Forschungsergeb-
nisse. Reihe C, Geologie und Geophysik 16, 19–66.
Ehleringer, J.R., Cerling, T.E., Helliker, B.R., 1997. C4 photosynthe-
sis, atmospheric CO2, and climate. Oecologia 112, 285–299.
FAO, 2002. FAOCLIM, Version 2.0, Database of the Agrometeorol-
ogy Group. FAO Environment and Natural Resources Service
SDNR (Working Paper Serie n°9).
Fredlund, G., Tieszen, L.T., 1994. Modern phytolith assemblages from
the North American Great Plains. Journal of Biogeography 21,
321–335.
Fredlund, G., Tieszen, L.T., 1997. Calibrating grass phytoliths
assemblages in climatic terms: application to the late Pleistocene
assemblages from Kansas and Nebraska. Palaeogeography,
Palaeoclimatology, Palaeoecology 136, 199–211.
Frost, P.G.H., 2000. Vegetation structure of the MODIS validation site,
Kataba Forest Reserve, Mongu, Zambia. Report Prepared for the
NASA Southern African Validation of EOS (SAVE) Project, IGBP
Kalahari Transect Programme, and SAFARI 2000. Institute of
Environmental Studies, University of Zimbabwe, Harare, Zim-
babwe. 13 pp.
GPWG (Grass Phylogeny Working Group), 2001. Phylogeny and
subfamilial classification of the grasses (Poaceae). Annals of the
Missouri Botanical Garden 88, 373–457.
Hansen, B.T., Plew, M.G., Schimpf, M., 1998. Elucidation of size
patterning in phytolith assemblages by field-flow fractionation.
Journal of Archaeological Science 25, 349–357.
Hansen, M.C., DeFries, R.S., Townshend, J.R.G., Sohlberg, R.,
Dimiceli, C., Carroll, M., 2002. Towards an operational MODIS
continuous field of percent tree cover algorithm: examples using
AVHRR and MODIS data. Remote Sensing of Environment 83,
303–319.
Hansen, M.C., DeFries, R., Townshend, J.R.G., Carroll, M., Dimiceli,
C., Sohlberg, R., 2003. Global percent tree cover at a spatial
resolution of 500 meters: first results of the MODIS vegetation
continuous fields algorithm. Earth Interactions 7, 1–15.
Hansen, M.C., Townshend, J.R.G., DeFries, R.S., Carroll, M., 2005.
Estimation of tree cover using MODIS data at global, continental
and regional/local scales. International Journal of Remote Sensing
26, 4359–4380.
469
Hodson, M.J., White, P.J., Mead, A., Broadley, M.R., 2005.
Phylogenetic variations in the silicon composition of plants.
Annals of Botany 96, 1027–1046.
ICPN Working group, Madella, M., Alexandre, A., Ball, T.B., 2005.
International code for phytolith nomenclature 1.0. Annals of
Botany 96, 253–260.
Kealhofer, L., Piperno, D.R., 1998. Opal phytoliths in Southeast
Asian and Thai flora. Smithsonian Contributions to Botany 88,
1–39.
Kondo, R., Childs, C., Atkinson, I., 1994. Opal Phytoliths of New
Zealand. Manaaki Whenua Press. 85 pp.
Leemans, R., Cramer, W., 1991. The IIASA database for mean
monthly values of temperature, precipitation and cloudiness of a
global terrestrial grid. International Institute for Applied Systems
Analysis (IIASA) RR-91-18, 61.
Lézine, A.M., 1988. New pollen data from the Sahel, Senegal. Review
of Palaeobotany and Palynology 55, 141–154.
Lézine, A.M., Edorh, T.M., 1991. Modern pollen deposition in West
African Sudanian environments. Review of Palaeobotany and
Palynology 67, 41–58.
Livingstone, D.A., Clayton, W.D., 1980. An altitudinal cline in
tropical African grass floras and its paleoecological significance.
Quaternary Research 13, 392–402.
Metcalfe, C.R., 1960. Anatomy of the Monocotyledons. I. Gramineae.
Clarendon Press, Oxford. 731 pp.
MU Phytolith database. 2004. University of Missouri, http://www.
missouri.edu/~phyto/index.shtml.
Mulholland, S.C., 1989. Phytoliths shape frequencies in North Dakota
grasses: a comparison to general patterns. Journal of Archaeolog-
ical Science 16, 489–511.
NIWA, S., 2005. Map of New Zealand Mean Annual Temperature
from 1971 to 2000. http://www.niwascience.co.nz/edu/resources/
climate/overview/.
Pearsall, D.M., Chandler-Ezell, K., Chandler-Ezell, A., 2004. Maize
can still be identified using phytoliths: response to Rovner. Journal
of Archaeological Science 31, 1029–1038.
Pichi-Sermolli, R.E.G., 1957. Une carta geobotanica dell'Africa
orientale (Eritrea Etiopia, Somalia). Webbia 13, 15–132.
Piperno, D.R., 1985. Phytolith analysis and tropical paleo-ecology:
production and taxonomic significance of siliceous forms in New
World plant domesticates and wild species. Review of Palaeobo-
tany and Palynology 45, 185–228.
Piperno, D.P., 1988. Phytolith Analysis — An Archaeological and
Geological Perspective, vol. 1. Academic Press/Harcourt Brace
Jovanovich, New York (US). 280 pp.
Piperno, D.R., Pearsall, D.M., 1998. The silica bodies of tropical
American grasses: morphology, taxonomy, and implications for
grass systematics and fossil phytolith identification. Smithsonian
Contributions to Botany 85, 1–40.
Rosen, A.M., Weiner, S., 1994. Identifying ancient irrigation: a new
method using opaline phytoliths from emmer wheat. Journal of
Archaeological Science 21, 125–132.
Runge, F., 1996. Opal Phytolithe in Pflanzen aus dem humiden und
semi-ariden Osten Afrikas und ihre Bedeutung für die Klima-und
Vegetationgeschichte. Botanische Jahrbücher für Systematik,
Pflanzengeschichte und Pflanzengeographie 118, 303–363.
Runge, F., 1999. The opal phytolith inventory of soils in central Africa —
quantities, shapes, classification, and spectra. Review of Palaeobo-
tany and Palynology 107, 23–53.
Runge, F., Fimbel, R., 1999. Opal phytoliths as evidence for the
formation of savanna islands in the rain forest of Southeast
Cameroon. In: Heine, K., Runge, J. (Eds.), Proceedings of the
470 D. Barboni et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 246 (2007) 454–470
VXth INQUA Conference, Durban, South Africa. International
Union for Quaternary Research, Tokyo, pp. 171–189.
Scurfield, G., Anderson, C.A., Segnit, E.R., 1974. Silica in woody
stems. Australian Journal of Botany 22, 211–229.
Strömberg, C.A.E., 2004. Using phytolith assemblages to reconstruct
the origin and spread of grass-dominated habitats in the great plains
of North America during the late Eocene to early Miocene.
Palaeogeography, Palaeoclimatology, Palaeoecology 207, 239–275.
ter Braak, C.J.F., 1986. CANOCO-A FORTRAN Program for
Canonical Community Ordination by Partial, Detrended, Canon-
ical Correspondence Analysis, Principal Components Analysis and
Redundancy Analysis (Version 3.1).
Thorn, V.C., 2004. Phytolith evidence for C4-dominated grassland
since the early Holocene at Long Pocket, northeast Queensland,
Australia. Quaternary Research 61, 168–180.
Tieszen, L.L., Senyimba, M.M., Imbamba, S.K., 1979. The distribution
of C3 and C4 grasses and carbon isotope discrimination along an
altitudinal and moisture gradient in Kenya. Oecologia 37, 337–350.
Twiss, P.C., 1987. Grass–opal phytoliths as climatic indicators of the
Great Plains Pleistocene. In: Johnson, W.C.E. (Ed.), Quaternary
Environments of Kansas. Guidebook. Kansas Geological Survey,
pp. 179–188.
Twiss, P.C., 1992. Predicted world distribution of C3 and C4 grass
phytoliths. In: Rapp, G.R., Mulholland, S.C. (Eds.), Phytoliths
Systematics Emerging Issues. Plenum Press, New York, pp. 113–128.
Twiss, P.C., Suess, E., Smith, R.M., 1969. Morphological classifica-
tion of grass phytoliths. Proceedings - Soil Science Society of
America 33, 109–115.
Vincens, A., Dubois, M.A., Guillet, B., Achoundong, G., Buchet, G.,
Kamgang Kabeyenne Beyala, V., de Namur, C., Riera, B., 2000.
Pollen–rain–vegetation relationships along a forest–savanna
transect in southeastern Cameroon. Review of Palaeobotany and
Palynology 110, 191–208.
Vincens, A., Bremond, L., Brewer, S., Buchet, G., Dussouillez, P.,
2006. Modern pollen-based biome reconstructions in East Africa
expanded to southern Tanzania. Review of Palaeobotany and
Palynology 140, 187–212.
Vogel, J.C., Fuls, A., Ellis, R.P., 1978. The geographical distribution of
Kranz grasses in South Africa. South African Journal of Science
74, 209–215.
Wallis, L., 2003. An overview of leaf phytolith production patterns in
selected northwest Australian flora. Review of Palaeobotany and
Palynology 125, 201–248.
Watson, L., Dallwitz, M.J., 1992 onwardss. The Grass Genera of the
World: Descriptions, Illustrations, Identification, and Information
Retrieval; Including Synonyms, Morphology, Anatomy, Physiol-
ogy, Phytochemistry, Cytology, Classification, Pathogens, World
and Local Distribution, and References. Version: 28th November
2005. http://delta-intkey.com.
White, F., 1983. The Vegetation Map of Africa. UNESCO, Paris.
356 pp.
White, T.D., Suwa, G., Asfaw, B., 1994. Australopithecus ramidus, a
new species of early hominid from Aramis, Ethiopia. Nature 371,
306–312.
Winslow, J.C., Hunt, E.R., Piper, S.C., 2003. The influence of seasonal
water availability on global C-3 versus C-4 grassland biomass and
its implications for climate change research. Ecological Modelling
163, 153–173.
Wolde Gabriel, G., White, T.D., Suwa, G., Renne, P., de Heinzelin, J.,
Hart, W.K., Heiken, G., 1994. Ecological and temporal placement
of early Pliocene hominids at Aramis, Ethiopia. Nature 371,
330–333.
Wooller, M.J., Swain, D.L., Street-Perrott, F.A., Mathai, S., Agnew,
A.D.Q., 2001. An altitudinal and stable carbon isotope survey of
C3 and C4 graminoids on mount Kenya. Journal of East African
Natural History 90, 69–85.