Soil and plant phytoliths from the Acacia-Commiphora mosaics at Oldupai Gorge (Tanzania)

Vegetation survey

The greater Serengeti area falls within the Somalia-Masai floristic region in which Acacia-Commiphora deciduous bushland is the climax vegetation (Kindt et al., 2011). This ecoregion is transitional to miombo woodlands to the south (Frost, 1996) and the Zanzibar Inhambane coastal forest mosaic to the east (Lovett & Wasser, 1993). The published plant inventory for Oldupai Gorge totals 32 genera (Herlocker & Dirchl, 1972). We studied 29 species representing 20 of these genera and 15 families. Characteristic Somalia-Masai taxa analysed here comprise the dominant members of the canopy (Acacia mellifera, A. nilotica), emergents (Boscia angustifolia), bushes (Barleria eranthemoides, Maerua triphylla), the herbaceous layer (Hypoestes forskaolii, Ocimum spp.), succulents (Sansevieria robusta, Aloe secundiflora), and the grasses (Sporobolus consimilis, Cynodon dactylon -Chloridoideae, Aristida adoensis - Aristidoideae).

Vegetation in the greater Oldupai region has been monitored for decades (Niboye, 2010). The average perennial cover today is 22% (Niboye, 2010), which is comparable to that observed several decades ago (Anderson & Talbot, 1965), with bare terrain exposing >45% of the area. Herlocker & Dirchl (1972) noticed an increased plant variety in the eastern sector of the gorge, introduced by the topographic changes from the canyon walls, and thus a higher diversity of bushes and grasses. They also noticed commonalities with the communities from the western gorge.

Ecologically, the plant community that typifies the natural vegetation at Oldupai Gorge is short Acacia and Commiphora woodland. Several tree species from these two genera co-dominate along with bushes such as Barleria eranthemoides. Succulents abound in ravines and depressions where several species of Sansevieria, Euphorbia, and Sarcostemma are common. Herbaceous plants are variably scattered across the landscape but, overall, they are low in numbers. Importantly, this domain is characterized by a paucity of grass taxa, which are typically inconspicuous and occasional (Kindt et al., 2011).

At the regional level (Fig. 1B) the vegetation was surveyed using qualitative criteria, aimed at investigating the range of species present in the greater Oldupai region, learning about their phytolith production and morphotype characteristics, and comparing the botanical dataset with the soil group. We did not attempt to collect plant specimens in numbers proportionate to frequency in each of their respective ecological communities, nor to other physiognomic variables or phylogenetic affiliation. Instead, we collected representative plants from the western and eastern gorge at the end of the wet season into the dry season in 38 sampling locations to target 29 species along an E-W transect (25 km long) from Granite Falls to Olbabal. Field samples were identified in the National Herbarium of Tanzania at the Tropical Pesticides Research Institute. When species level identification was uncertain and could belong to several species within the genus, we listed it as spp. Grouping and nomenclature come from the index of accepted names for the flowering plants of Sub-Saharan Africa (Klopper et al., 2006).

At the local level, the vegetation was surveyed using quantitative criteria. First, an aerial image covering 1 km² (Fig. 1C) taken at the end of the wet season in the eastern half of Oldupai Gorge, was georeferenced via ArcGIS - ArcMAP 10.5.1. A grid consisting of 100 cells, each 100 m², was superimposed. The cover-abundance value was digitized in each cell as a polygonal shape file, and classified it in four classes or ranks:

1) Rank zero, or ‘No-cover’: when plant cover constituted <10% within the spatial polygon (45.64% of the total area).

1. Rank one, or ‘Sparse’: when plant growth was 10%–20% (20.32% of the total).

2. Rank two, or ‘Open’: when plant growth was 20%–60% (26.41% of the total).

3. Rank three, or ‘Closed’: when plant growth was >60% (7.63% of the total).

This rank system was subjected to groundtruthing in 300 spots (Mercader et al., 2017: Fig. 2), and then we proceeded to conduct a qualitative vegetation survey of the entire 1 km² plot (Fig. 1C). After this, we zoomed into a 62,500 m² quadrat (Fig. 1D) to conduct a full inventory, plotting, and collection of all plants present within this smaller plot. According to plant cover rank (Fig. 2), thirteen samples (37.14%) were recovered from rank zero; eight (22.86%) from rank one; eight (22.86%) from rank two; and six (17.14%) from rank three. Samples were each assigned to one of two topographical categories: promontory or ravine. Ravines are at a mean elevation of 1,422.66 m with a range between 1,412 m and 1,441 m (29 m). Promontories are at a mean elevation of 1,452.41 m with a range between 1,416 m and 1,469 m (53 m). Most samples (n = 29, 82.86%) were recovered from promontory locations (Fig. S1).

Particle size analysis

We carried out particle size analysis of selected soil samples (n = 10) via laser diffraction (Malvern Mastersizer 2000: Sperazza, Moore & Hendrix, 2004). A soil sample (1.0 g) was deflocculated with 0.1% sodium hexametaphosphate. Ultrasonics dislodged clays for 5 min. Sample was left on orbital shaker overnight at 200 rpm. We cleansed the sample of carbonates, phosphate, and organics via hydrochloric acid (HCl 37%), nitric acid (HNO₃ 60%) and hydrogen peroxide (H₂O₂ 30%) successively, rinsing the sample three times via centrifuging in between each treatment (3,000 rpm, 5 min). When measuring particle size in the Malvern 2000, we used obscuration rates above 15% but below 30%. The measurement of pH in all soils was through a ThermoFisher Orion 320 PerpHecT LogR meter, calibrated daily. The Ag/AgCl electrode was rinsed after calibration and in between samples. Sediment color was estimated dry under constant light conditions by two observers relative to the Munsell color system. Water loss on drying is the balance between soil field moisture in three grams of soil and the mass of the sample after drying at 110 °C for 24 h.

Soil phytoliths

Sampling was never conducted on archaeological sediments, nor within any of the well-known and mapped lacustrine and fluviatile outcrops that constitute the ancient beds (Leakey et al., 1972; Hay, 1976); our work was restricted to purely surficial sandy loam (Anderson & Talbot, 1965). Prior to collecting soil samples, and to better understand local soils and visualize their horizons, several trenches were dug reaching the C-horizon: a characteristic underlying hard pan described by earlier soil analysts (Anderson & Talbot, 1965; Jager, 1982). In section, we observed a thin A-horizon, restricted to the rhizosphere and litter zone, while a B-horizon seems absent. We scraped 25 cm² off the surface (5 ml) to a maximum depth of one cm within the A-horizon (Fishkis, Ingwersen & Streck, 2009; Mercader et al., 2010), excluding litter and not mixing or generating composite samples. Thirty-five loci were selected for stratified sampling, so that a comparable number of samples were collected from each plant cover type. The median distance between all soil assemblages is 516 m (measured distances, n = 522) over a range of 13 m–1,252 m. Geographical coordinates were documented as point features. The full dataset of soil samples including provenance, elevation, coordinates, processing weights, and individual counts are available at the Federated Research Data Repository (DOI https://dx.doi.org/10.20383/101.0122).

Detailed phytolith extraction protocols are available with the Federated Research Data Repository (DOI https://dx.doi.org/10.20383/101.0123), which are modified from (Albert et al., 1999). In brief, extraction from sediment used a 3.0 g aliquot that received 10 ml of 0.1% sodium hexametaphosphate for clay dispersion, followed by sonication (15 min) and overnight orbital shaking plus decantation of clays. Samples were rinsed and dried over two days (>70 °C). Soil samples received 10 ml equal parts 3N hydrochloric and nitric acids; then, rinsed and centrifuged at 3,000 rpm. Drying followed for two days (70 °C). Samples were weighed and transferred to Petri dishes on a hot plate (70 °C) whereby 10 ml of hydrogen peroxide (30%) was added to remove organics. Rinsing, drying, and weighing followed. Phytolith separation was done by transferring this fraction to 15 ml centrifuge tubes, to which 5 ml sodium polytungstate (∼2.4 s.g.) was added. Samples were vortexed for five minutes at 3,000 rpm. The supernatant was removed. A gradual decrease in specific gravity over the course of three subsets (subfractions 1, 2, 3) concentrated phytoliths in the third extract. Samples were rinsed with de-ionized water before centrifuging twice more, followed by two cycles of high centrifugation (4,500 rpm). Samples dried at 70 °C another time and a final weight obtained. The aliquot extracted for phytolith analysis weighed approximately 0.001 g. It was mounted with Entellan New and inspected while fresh; allowing particles to be rotated in three dimensions under the microscope. Phytoliths from soil samples were counted on one slide over 600 mm² to the total amount, reaching average counts of 250. Diatoms were tallied concurrently. Carbonate content was calculated as the loss on acid treatment relative to the original mass. Extraction weights are the balance between the original mass and that of the extract after defloculation, carbonate removal, destruction of phosphate, elimination of organics, rinsing, and drying.

Plant phytoliths

Extraction followed Albert & Weiner (2001), with modifications. Botanical specimens were sorted into anatomical parts, and each was individually weighed and measured. Full datasets with species, plant part studied, provenance, elevation, coordinates, processing weights, and individual counts are available at the Federated Research Data Repository (DOI https://dx.doi.org/10.20383/101.0122). All samples were cleansed by ultrasonication, then rinsed and dried overnight >80 °C. Dry weight was obtained, and samples placed in a muffle furnace (24 h, 400 °C). The resulting ashes were weighed and transferred to Petri dishes to which 10 ml of 3N hydrochloric and nitric acids were added to a hot plate at 70 °C for 30 min. Samples were transferred to 15 ml tubes and rinsed of remaining acids in three wash and centrifugation cycles at 3,000 rpm, with excess water being boiled off. Hydrogen peroxide (10 ml, 30%) was added to the Petri dishes heated on hot plates (70 °C), transferred to 15 ml centrifuge tubes, rinsed, and centrifuged (3,000 rpm) for three cycles. Samples were dried >70 °C overnight, and then weighed. An aliquot of approximately 0.001 g was analysed after mounting the sample for microscopy with Entellan New. Inspection captured three-dimensional characteristics during rotation conducted within 48 h of mounting and before the resin dried. Frequent polarization of the field of view under the microscope ensured that no anisotropic particles and crystals were confused for phytoliths. Neither fresh nor weather volcanic glass could be observed. We counted one full slide and only discrete phytoliths (average = 106 phytoliths), while articulated shapes were counted as one. As for very rich extracts, we tallied a minimum of 200 phytoliths and/or short cells for the grasses.

Phytolith classification

There are two approaches to environmental reconstruction in phytolith research: the indices (Barboni, Bremond & Bonnefille, 2007; Bremond et al., 2008) and the general approach (Strömberg, 2004). Different researchers adopt one approach or another depending on their work’s practice and scope. We followed the general approach here to detect plant variations at the spatial level. Quantitative analysis involved exploration aimed at discovering relationships amongst the soil and plant phytoliths, which, when combined and filtered, isolated groups, classes, and morphotypes unique to each set and those in common. Further investigation involved comparison of phytoliths from soils and plants to explore their relative distributions through Chi–square independence tests, establishing the statistical significance of the relationship between categorical variables within the two populations using a 95% confidence value (p = 0.05).

Two system, polarizing microscopes (Olympus BX51, Motic BA410E) allowed us to confirm the opal silica’s isotropy and tally phytoliths at 400x, following the International Code for Phytolith Nomenclature 1.0 to name and describe specimens (Madella, Alexandré & Ball, 2005), with exceptions duly noted. Geographically pertinent reference collections and published soil phytolith assemblages were consulted (Albert, Bamford & Cabanes, 2006; Albert, Bamford & Esteban, 2015; Bamford, Albert & Cabanes, 2006; Fahmy, 2008; Mercader et al., 2009; Mercader et al., 2010; Mercader et al., 2011; Barboni & Bremond, 2009; Eichhorn, Neumann & Garnier, 2010; Novello et al., 2012; Collura & Neumann, 2017; Neumann et al., 2017). Full datasets with individual counts for each morphotype are available at the Federated Research Data Repository (DOI https://dx.doi.org/10.20383/101.0122). We classified discrete phytolith shapes in four large groups (woody dicot, grass short cells, generic Poaceae, and rare) and 17 classes that encompass 76 types. The woody dicot group totalled the blocky class (Mercader et al., 2009; Novello et al., 2012; Collura & Neumann, 2017), cylindrical types from bark (Mercader et al., 2009), varied globular phytoliths (Runge, 1999; Mercader et al., 2009), tabular types from woody dicots (Mercader et al., 2009; Collura & Neumann, 2017), clavates (Mercader et al., 2009), and sclereid phytoliths (Collura & Neumann, 2017). Phytoliths from Poaceae short cells were subdivided in three classes: lobate (Fahmy, 2008; Neumann et al., 2017), rondels (Mercader et al., 2010; Novello et al., 2012; Neumann et al., 2017) and chloridoid saddles (Twiss, Suess & Smith, 1969; Mercader et al., 2010). Two additional groups comprised generic grass phytoliths and rare morphotypes.

It should be noted that the tabular class, with 12 shape variants, contains a confuser morphotype called here tabular sensu stricto, and defined as a narrow, thin parallelepiped shape with straight edges that may occur in both monocots and dicots. This confuser morphotype differs from the tabular psilate observed in woody dicots (this study) in that the latter is a wide and thick tabular shape with a very slight sinuosity to the outline.

Soil phytoliths

Soil characteristics and phytolith abundance

Local topsoils (Table 1) have a mean particle size of 344 µm, with pH of 8.7 and 5% water content. Unimodal particle size distribution is apparent in the poorly sorted soil matrices of most samples, while the coarse sand component in samples from the eastern half of our plot generates a bimodal population in which the poorly sorted fines represent the soil, and the well-sorted, coarse sands originate from easterly aeolian import (Fig. 3). Silt and very fine sands average 34% per sample, while carbonates are 69% by mass. The extracted phytolith-bearing fraction, which includes both biogenic and geogenic silica, is 54.53% (M = 53.85%; range = 45.54%, 25.23% –70.77%). Dissolution and mechanical breakage were rarely observed, and caution must be used not to confuse the effect of dissolution causing holes randomly through a phytolith versus the cavate textures that heavily decorate some blocky and tabular morphotypes (Figs. 4Q, 4W) from woodland taxa (cf. Mercader et al., 2009: Fig. 6g,k,l,m).

Table 1:

Particle size analysis of selected topsoils, textural classification, phytolith-bearing fraction, modality, pH, color, and water content.

Sample	Mean particle size, µm		Particle class	Silt & Very Fine Sand, %		Modality	Sediment pH		Sediment colour		Sediment H2O content, %
43	294.2		Medium sand	39.62		Unimodal	8.7		Dark yellowish brown		3.9
63	513.8		Coarse sand	22.73		Bimodal	8.3		Moderate yellowish brown		5.4
85	216.9		Fine sand	49.49		Unimodal	8.8		Dark yellowish brown		6.5
138	94.7		Very fine sand	72.86		Unimodal	8.7		Dark yellowish brown		3.6
146	335.8		Medium sand	38.33		Unimodal	9.1		Pale yellowish brown		3.7
192	469.6		Medium sand	37.59		Bimodal	8.6		Moderate yellowish brown		2.7
229	322.1		Medium sand	26.31		Unimodal	8.7		Dark yellowish brown		2.8
236	670.3		Coarse sand	17.75		Bimodal	8.4		Dark yellowish brown		3.7
279	736.8		Coarse sand	6.54		Unimodal	9.2		Dark yellowish brown		2.9
286	300.1		Medium sand	35.52		Bimodal	8.6		Dark yellowish brown		4.0
Mean	344.3		Medium sand	29.55		Unimodal	8.7		Yellowish brown		3.8

DOI: 10.7717/peerj.8211/table-1

No soil sample was void of phytoliths. In total, 10,745 phytoliths were tallied ($\bar{X}=$307, M = 285 per sample; range = 839, 16–855) containing 64 morphotypes that can be grouped into 15 major classes (Fig. 4, Table 2). Of these, six are prominent classes (Figs. 5, 6) and account for 83.78% of all soil phytoliths: tabular: 41.74%, blocky: 13.11%, globular: 8.52%, rondel: 6.98%, cylindric: 6.96%, and lobate: 6.47%. Within the tabular class, the three most prominent contributors represent 82.16% of the subtotal: scrobiculate: 45.40%; sinuate: 20.47%; and thick lacunate: 16.30%. Two types of blocky phytoliths represent 91.12% of the class; blocky: 62.38% and ridged: 28.74%. One single type of globular phytolith represents 85% of all globular types: a globular to hemispherical, facetate body with a centric cavity; possibly from mesophyll cells found in the leaves from several taxa in the Lauraceae (Figs. 4A and 4D; Thorn, 2001; Murungi, 2017). In contrast, the classic indicator of arboreal cover in the African tropics, the globular granulate, amounts to only 9% of the globular class. Two globular morphotypes were recovered from soils only: globulose bisected and segmented. Within the cylindric class, two phytoliths reach 91.44%: scrobiculate, 76% and sinuate, 15.37%.

Two types dominate the rondels (97.33%): Towers, 63.33% and rondels sensu stricto, 34%. Lastly, for the lobate class, 86.33% of the variability is represented by small bilobates with concave ends (61.58%), small bilobates with flat ends (16.04%), and small bilobates with convex ends (8.34%). Four lobate types were found exclusively in soils: small bilobate with notched ends, large bilobate with flat ends, large bilobate with notched ends, and cross. The saddle class is uncommon (5.87%) and the short saddle dominates it (74.32% of the subtotal), while squat saddles are rare. Generic tabulars (3.88% of the total), bulliforms and lanceolates (2.89% of the total), generic cylindric phytoliths (1.23% of the total), sclereids (1.1% of the total), fusiforms / gutiforms (0.74% of the total) (nomenclature not per ICPN 1.0), and clavates (0.25% of the total) are scarce. The remainder includes composite epidermal tissue, polygonal prisms with subrounded top (cf. Commelinaceae), and papillae; and they all are extremely uncommon.

Table 2:

Topsoil phytolith counts per group, class, and morphotype.

Major group	Phytolith class	Phytolith morphotype	Count
Poaceae short cell	Lobate	Bilobate, long concave	38
		Bilobate, long convex	8
		Bilobate, long flat	14
		Bilobate, long notched	4
		Bilobate, short concave	428
		Bilobate, short convex	58
		Bilobate, short flat	114
		Bilobate, short notched	28
		Cross	1
		Polylobate	2
	Rondel	Rondel, horned	8
		Rondel, truncated	12
		Rondel, tower	475
		Rondel	255
	Saddle	Saddle, ovate	24
		Saddle, short	469
		Saddle, squat	140
	Blocky	Blocky	879
		Blocky corniculate	2
		Blocky ridged	405
		Blocky scrobiculate	23
		Blocky sinuate	28
		Blocky thick lacunate	72
	Cylindrical	Cylinder corniculate	3
		Cylinder psilate	132
		Cylinder scrobiculate	569
		Cylinder sinuate	115
		Cylinder thick lacunate	61
	Fusiform	Fusiform	11
		Guttiform	68
	Sclereid/Clavate	Clavate	27
		Sclereid	119
	Spherical	Globular bisected	1
		Globular facetate	778
Woody		Globular granulate	83
		Globular granulate, large	3
		Globular granulate oblong	24
		Globular psilate	10
		Globulose	2
		Globulose segmented	1
		Oblong granulate	10
		Hemisphere psilate	3
Tabular	Tabular corniculate	11
		Tabular ellipsoidal	35
		Tabular ellipsoidal, large	21
		Tabular facetate	8
		Tabular laminate	186
		Tabular oblong	16
		Tabular pilate	3
		Tabular scrobiculate	2036
		Tabular sinuate	918
		Tabular strangulated	11
		Tabular sulcate	509
		Tabular thick lacunate	731
Undetermined grass	Grass, undetermined	Bulliform	47
		Trichome	264
Rare/Unknown	Composite	Epidermal	3
		Vessel	18
	Drum	Polygonal prism	2
	Sedge	Papillae	1
	Short cell, other	Trapeziform sinuate	1
	Tabular undetermined	Tabular	70
		Tabular crenate	36
		Tabular psilate	311

DOI: 10.7717/peerj.8211/table-2

Soils also produced 1,564 diatoms in addition to the phytolith complement summarized above. The mean number of diatoms per sample is 44.68 and the median recovery is 9.50 with a range of zero –301. The specimens are mostly from the Nitzschioid group (Figs. 4af–4ah), which grows in wet soils. The diatom count per plant cover class is inversely proportional to density of plant cover today, with the highest frequency in soil samples without vegetation and/or sparse growth probably because of pooling water in these areas (Albert, Bamford & Cabanes, 2006).

Extraction weight and plant cover

Conventionally, extraction weights are reported as a proxy of phytolith concentration. Like others before us (Blinnikov, Bagent & Reyerson, 2013), however, we did not see a discernible pattern between plant cover, phytolith count, and extraction weight (Fig. S1) with values randomly distributed. Thus, plant cover rank zero varied between 43.6% –69.5%, for rank one between 40.96% –70.77%, rank two between 25.23% –63.56%, and rank three between 41.1% –56.16%. The scores for plant cover ranks overlap, and so do those for extraction weight versus terrain elevation.

We explored the relationships between plant cover rank and phytoliths from woody dicots versus grasses. The woody phytolith group dominates all terrain ranks, and appears to increase with denser plant cover, while grass phytoliths peak in sparsely vegetated terrain. The null hypothesis posits that if plant cover has no influence on the relative proportions of woody and grass phytoliths, there will be no significant difference between overall proportions and plant cover rank proportions. For plant cover rank zero, vegetation density has not significantly influenced the relative proportions of woody and grass phytoliths present in the topsoil (Chi–square = 2.382, not significant at p < 0.10). There are significant differences in all other plant landscapes: Rank one has significantly more grass phytoliths than expected from the null hypothesis (Chi–square = 9.0812, p < 0.01) while ranks two (Chi–square = 7.2288, p < 0.01) and three (Chi–square = 8.0625, p < 0.01) have many more woody phytoliths. Statistically, therefore, sparse, open, and closed terrain influenced the relative proportions of woody and grass phytoliths released into the soil (Table 3), with sparse land containing more grass phytoliths, and open woodland / closed terrain supporting more woody dicot types.

For short cell classes (Table 4), rank zero yields rondels followed by lobates and saddles; rank one: lobates, saddles, and rondels; rank two and three: lobates, rondels, and saddles. These class fluctuations across plant cover, however, are artifacts of differing sample numbers per plant zone (Figs. S2–S4). Chi-square testing revealed no significant changes in short cell class related to plant cover rank. By morphotype, the most frequent phytoliths are towers, short saddles, and small bilobates with concave ends. In ranks zero and one, this order is preserved, but it reverses in ranks two and three. That is, the tandem of towers and short saddles are more frequent in areas of sparse to no plant cover (rank zero, one), while short saddles and small bilobates with concave ends are more frequent in areas of increased plant cover (rank two, three). However, Chi-square testing shows no significant change in the relative frequencies of short cell morphotypes related to plant cover today. In fact, where one phytolith type increases in frequency, so do the other two: These morphotypes vary positively with each other.

Plant phytoliths

Morphotypes

The morphotypes from woody plants are led by the globular class through three phytoliths: Oblong shapes with a granulate texture, globulose, and globular granulate. The tabular class includes three morphotypes in abundance: scrobiculate, sinuate, and ellipsoidal; while the cylindrical class peaks with cylinder sinuates and psilates. Blocky morphotypes are relatively common, but four blocky morphotypes could not be found in the local plants we studied: corniculate, ridged, sinuate, and thick lacunate. The Poaceae produce various rondel types, dominated by ovates, towers, and horned towers. Lobate phytoliths comprise six variants but are mostly manifested through two types: small bilobates with concave ends (e.g., Panicoid type) and large bilobates with convex ends (e.g., Aristidoid type). The saddle class comprises two types largely dominated by short saddles both being very common in Cynodon dactylon. Several cylindrical phytoliths were documented in thirteen local dicots but were absent or extremely infrequent among the Poaceae: cylindrical crenate, psilate, scrobiculate, and sinuate.

The Acacia-Commiphora Analog

In East Africa, the transition to arid-adapted woodlands and grasslands over the last two million years has been documented in pedogenic carbonates (Cerling, Bowman & O’Neil, 1988; Levin et al., 2004; Levin et al., 2011; Quinn et al., 2007), mammal, bird, and fish fossils (Bibi & Kiessling, 2015; Bibi et al., 2018; Prassack et al., 2018), micro- and macrobotanical remains like pollen, phytoliths, and wood (Bonnefille, 1982; Bonnefille, 1995; Barboni et al., 1999), and stable isotopes of hydrogen and carbon from leaf wax lipid biomarkers (Magill, Ashley & Freeman, 2013a; Magill, Ashley & Freeman, 2013b; Uno, Polissar & Jackson, 2016; Lupien et al., 2018).

The phytoliths from present-day vegetation and soil surfaces are the baseline to interpret palaeolandscapes and ecological niches that might have been occupied by early hominins. To study the influence of plant community structure and cover on human ecology, soil phytolith assemblages under known vegetation density and composition can provide a referential to interpret key palaeoecological aspects of early Pleistocene floristic regions and their main vegetation types.

We provide a phytolith analog for Acacia-Commiphora mosaics that may represent the trend for more open landscapes recorded in East Africa since the early Pleistocene. In this analog, the cohort of phytolith morphologies that identifies semiarid woodland include tabular scrobiculate, tabular sinuate, blocky, globular facetate, tabular thick lacunate, cylinder scrobiculate, tabular sulcate, tower, short saddle, small bilobate with concave ends, and blocky ridged. We did not find unique identifiers or highly diagnostic shapes for the most typical species that today make up the Acacia-Commiphora ecosystem, but this is hardly surprising and confirms similar observations in adjacent regional centers of plant endemism, such as the Zambezian woodland, where woody dicot types have very little taxonomic power (Mercader et al., 2009) because of redundancy across orders (Bozarth, 1992). Also in common with phytolith spectra from plants and soils in Zambezian woodlands (Mercader et al., 2009; Mercader et al., 2010; Mercader et al., 2011), there is severe overrepresentation of robust morphotypes such as tabulars, blockies, and globulars in addition to lobate short cells.

Do soil phytoliths mirror physiognomy and composition of vegetation aboveground?

We address this question at two levels, regional (plot = 100 ha) and local (quadrat = 6.25 ha). When considering the larger regional plot, all soil samples are dominated by phytoliths from woody dicots (Table S2– S4) (Albert, Bamford & Cabanes, 2006):

The blocky morphotypes that could not be found in local plants (corniculate, ridged, sinuate, and thick lacunate) were partly tracked by geo-referencing: Blocky ridged and thick lacunate associate with areas of sparse plant growth (Figs. S5–S6). Large numbers of these forms in soils suggest that plants not represented in our inventory might have produced them or that the species producing these types have disappeared in past decades or not seasonally available when we collected our samples.

Cylindrical phytoliths were common in many of the local but were either absent or extremely infrequent among the grasses. Cylinder corniculates and thick lacunates occur only in the soils, while crenates occur only in plants. Cylinder crenates appear in Acacia mellifera and Ocimum spp. Similarly, the few corniculates recovered from soils may not be inferentially important. However, the relatively large number of cylinder thick lacunates in soils is noteworthy, as plants from the Miombo woodlands (Mercader et al., 2009) produce them. Fusiforms, guttiforms, clavates, and sclereids are found in plants and soils.

The globular class is common in both soil and plant samples. The globular facetate (mesophyll cells) is very abundant, and geo-referencing shows an association with areas covered by plants and absence from bare places (Fig. S7). Two globulose morphotypes were recovered from soils only: globulose bisected and segmented. The former is a unique phytolith that, in adjacent Zambezian woodlands, is restricted to the Solanaceae (Mercader et al., 2009: Fig. 3AE).

Tabular phytoliths are relatively uncommon in the plants but very common in the soils. Four morphotypes could not be found amongst the plants, but were in soil samples: corniculate, ellipsoidal large, facetate, and pilate. Their georeferencing shows no pattern for three of them, but one type (tabular ellipsoidal large) appears solely in locations with plants growing today.

Poaceae phytoliths are skewed in their soil expression (Table 8, Table S5): the local grasses are demonstrated overproducers (Twiss, Suess & Smith, 1969; Novello et al., 2012) but they are not overrepresented in these soils, rather the opposite. We infer that topsoils embody plant communities, past and present, in a time-averaged palimpsest dominated by woodlands (Table S3). It is likely that the heterogenous landscape and the changing grazing history, erosion, and fire regime by pastoralists over the last 60 years (Sinclair et al., 2008) have created a phytolith record of ecological succession underscored by a reversion towards a dominance of woody vegetation, as independently recorded by studies of land cover change (Sinclair et al., 2007; Niboye, 2010) that denote a low human impact on ecosystem dynamics since the 1960s.

Within the short cells, three major classes were identified in plants and soils: lobates, rondels, and saddles; albeit saddles are overrepresented in soils. Ten phytolith morphotypes were recovered within the lobate class. Of these, six were found in both plant and soil samples and four were found only in soils (small bilobate with notched ends; large bilobate with flat ends; large bilobate with notched ends; cross). The lead lobate type (Fahmy, 2008; Neumann et al., 2017), small bilobate with concave ends, is produced abundantly in one local and one regional taxon: Penninsetum mezianum and Sporobolus africanus respectively. Aristidoid, large bilobates with convex ends are documented in Aristida adoensis while small bilobates with flat ends appear in Pennisetum mezianum. Rondel class phytoliths are common in both plants (Sporobolus specimens) and soils. Rondel ovates are unique to the plant sample, while truncated rondels occur only in soils. Rondels and horned rondels appear in Sporobolus spp. The absence of rondel ovates in the soils is notable given their preponderance in the genus Sporobolus. Saddles were recovered from three plants. Saddle squats are present in Cynodon dactylon and the abundance of this morphotype in the soil samples may suggest correlation. Short saddles are found in plants and soils while saddle ovates are from soils but not plants. In brief, short cell phytoliths from soils do not always reflect the grass species growing in them today.

Catchment

We explore the notion of catchment by zooming into one local quadrat covering 6.25 ha to conduct a side-by-side comparison of phytolith assemblages from topsoils (n = 6) and plants (n = 18) (Table 8, Tables S4– S5). To do this, we tallied all phytoliths by class. Of the thirteen phytolith classes represented in this quadrat, eight classes are severely underrepresented in the soil, including all three Poaceae short cell classes, two woody classes (globular/fusiform), epidermal cells in anatomical connection, and vessels (Bamford, Albert & Cabanes, 2006). Specifically, a positive balance exists for cylindrical shapes from woody tissue, along with sclereids and clavates. Moreover, dicot tabular morphotypes are overrepresented in the soil by a factor of seven. Several blocky morphotypes from woody tissue are four times more common in soils than in the plants that produced them, and lack of correspondence was similarly noticed for bulliforms and lanceolates. We attribute this imbalance to the phytolith signal from grasses being dampened by an overwhelming accumulation of woody phytoliths.

Rich phytolith producers such as the Poaceae display differential representation in soils, especially at sampling loci where they were growing. For instance, Cynodon dactylon produces short and squat saddles. In the quadrat where it grows today the average frequency of short saddles is half (n = 7.5) of the local average (n = 13), and the same applies to squat saddles (this quadrat = 1.8 versus 3.5 in the rest of the collection). Sporobolus panicoides preferentially generates rondel ovates, but none could be found in any of the six soil samples surrounding it. Aristida adoensis supports a majority of large bilobates with convex edges, and the average representation in the plant (n = 0.2) is similar to that in the soils nearby (n = 0.3). Importantly, one of the highest ranked short cells from both collections, soils and plants, the small bilobate with concave ends, attains lower than average representation in the soils under scrutiny (n = 4.8/6) perhaps because the main producers of this morphotype (Pennisetum mezianum, Sporobolus africanus) were not close.

Considering that all phytolith classes found in the living tissues of the eighteen plant taxa growing within this quadrat are recorded in the underlying soil surface it can be suggested that the catchment area for these soil phytolith assemblages is smaller than five hectares (Blinnikov, 1994; Fredlund & Tieszen, 1994; Blinnikov, Busacca & Whitlock, 2002).

Ecological disturbance, time averaging, heterogeneity, and sampling methods

The soil and plant phytolith assemblages are clearly distinguishable from one another. The presence of rare group phytoliths influences ratios in plants (26.98% of the total assemblage) and is much less an aspect of the soils (4.11%). However, even if we were to remove rare and delicate morphotypes, and consider short cells and woody dicot phytoliths only, the differences between the two datasets are obvious. In the plants, the two major groups are near parity (ratio = 1.031; Poaceae n = 1, 556 / woody n = 1, 509) while in the soils the two groups are quite disparate in relative frequency (ratio = 0.263; Poaceae n = 1, 509 and woody n = 7, 914).

Soil samples are not homogenous relative to one another. We indicated statistically significant relationships between the soil phytolith population in samples collected under a given plant cover and the rank of that zone, although not all individual samples show this relationship. The ratio between Poaceae short cells and woody phytoliths varies widely from 1:1 to 0:1. Moreover, soil samples with high number of Poaceae short cells have the highest number of woody phytoliths. Even in the most open parts of the landscape phytoliths from woody tissue outnumber those from grasses (Gao et al., 2018a; Gao et al., 2018b) and some woody types abound in soils but are absent from the local plants sampled (Albert, Bamford & Esteban, 2015).

We are suggesting decadal inheritance of soil phytoliths, as evidenced by:

1. The highest phytolith recovery is from areas without plant cover or Rank Zero.

2. No significant change in the relative frequencies of short cell morphotypes related to plant cover today.

3. Multiplicity of soil phytolith types, woody and grassy, that cannot be traced back to local plants.

4. Several phytolith classes are overrepresented in the soils.

That is, although soil phytoliths might represent ‘decay-in-place’ (Piperno, 2006), as well as transport by wind, water, fire and herbivory (Fredlund, Egan & Howell, 2001), the two phytolith populations reported here, from plants and soils, will always differ from each other because soil assemblages are the sum of phytolith accumulations and losses over time (Fredlund & Tieszen, 1994; Kerns, Moore & Hart, 2001).