|Home | About | Journals | Submit | Contact Us | Français|
There are no descriptions of phytoliths produced by plants from the ‘Zambezian’ zone, where Miombo woodlands are the dominant element of the largest single phytochorion in sub-Saharan Africa. The preservation of phytoliths in fossil records of Africa makes phytoliths a tool to study early plant communities. Paleo-ethnobotanical interpretation of phytoliths relies on the comparison of ancient types with morphotypes extracted from living reference collections.
Phytoliths were extracted from plant samples representing 41 families, 77 genera and 90 species through sonic cleaning, dry ashing and acid treatment; and phytoliths thus extracted were quantified. For each species, an average of 216 phytoliths were counted. The percentage of each morphotype identified per species was calculated, and types were described according to the descriptors from the International Code for Phytolith Nomenclature. Phytolith assemblages were subject to discriminant analysis, cluster analysis and principal component analysis.
Phytoliths were grouped into 57 morphotypes (two were articulated forms and 55 were discrete shapes), and provide a reference collection of phytolith assemblages produced by Miombo woody species. Common and unique morphotypes are described and taxonomic and grouping variables are looked into from a statistical perspective.
The first quantitative taxonomy of phytoliths from Miombos is presented here, including new types and constituting the most extensive phytolith key for any African ecoregion. Evidence is presented that local woody species are hypervariable silica producers and their phytolith morphotypes are highly polymorphic. The taxonomic significance of these phytoliths is largely poor, but there are important exceptions that include the morphotypes produced by members from >10 families and orders. The typical phytolithic signal that would allow scientists to identify ancient woodlands of ‘Zambezian’ affiliation comprises only half of the original number of phytoliths originally produced and might favour the more resilient blocky, cylindroid, globular and tabular forms.
Monosilicic acid [Si(OH)4] exists in a solute, monomeric state over a large part of the Earth's surface (Iler, 1979). It abounds in soils, where plants take it up through the roots, and distribute it as one of several sap constituents to shoots and leaves by means of the transpiration stream (Lewin and Reimann, 1969). With increasing concentrations in the sap, silicic acid combines chemically with other monomers to produce a network molecule, a polymerized silica gel (SiO2), that in some families can reach concentrations of >10 % of the plant's dry weight (Epstein, 1994). At the nanometre level, the gel solidifies inside and between cells as both amorphous and discretely shaped silica bodies called phytoliths (Snyder et al., 2007). In many species, these non-crystalline, bio-mineralized silica precipitates are a major mineral plant constituent (Epstein, 1994), but not all plants produce them. Phytoliths have a specific gravity of 1·5–2·3, are highly polymorphic, have the same optical properties in all directions and a refractive index of 1·4 (Elbaum et al., 2003). Depending on the extent of carbon coating, phytoliths may be clear, brown-pigmented or even totally opaque (Prychid et al., 2004). Phytolith production varies greatly among different genotypes of the same plant species and plant part (Epstein, 1999). With some exceptions, lower phylogenetic orders produce more phytoliths, although it is known that many basal monocots are non-producers (Prychid et al., 2004) and, contrarily, that some of the most derived clades produce high numbers of phytoliths (Hodson et al., 2005).
After plants die and decompose, the silica phytoliths contained in their tissues revert back to the soil. With death, the biogenic silica assemblage thus released suffers the insults of time, decay and differential preservation. As a result, over the course of several decades and plant generations, soils will contain a time-averaged, polygenic and biased record of local vegetation (Strömberg, 2004), not a floristical composition snapshot (see Thorn, 2008, p. 30 for an example of disagreement between soil phytolith assemblages and the overlying vegetation that grows on the spot at the time of sampling). Yet, phytoliths are helpful indicators of general plant physiognomy and the ecosystem's structure (Piperno, 2006) both today (Barboni et al., 2007) and in the past (Strömberg et al., 2007). Although phytolith studies have been carried out in modern soils from Central Africa (Runge 1999), West Africa (Bremond et al., 2005) and East Africa (Shahack-Gross et al., 2003, 2004; Albert et al., 2006; Bremond et al., 2008), quantitative work on silica bodies from living African plants is scarce and focuses on grasses from two phytochoria north of the equator (White, 1983): the Somalia-Masai (Palmer and Tucker, 1983) and the Sudano-Sahelian (Fahmy, 2008) vegetation zones. An extremely small amount of qualitative information is also known about the phytoliths produced by some African trees and bushes from the Guineo-Congolian region (Runge and Runge, 1997).
The preservation of phytoliths in terrestrial and lacustrine sediments from the African Miocene (Retallack, 1992), Plio-Pleistocene (Barboni et al., 1999; Albert et al., 2006) and Late Quaternary (Alexandre et al., 1997; Mercader et al., 2000) makes them a promising tool to study ancient plant communities. However, paleo-ethnobotanical interpretation relies on actualistic comparisons with silica morphotypes from modern plants (e.g. Gallego and Distel, 2004; Wallis, 2003; Carnelli et al., 2004; Honaine et al., 2006; Tsartsidou et al., 2007) and, therefore, a good understanding of the phytolith production of modern plants is a prerequisite for such studies. In this article, we estimate biogenic silica production by African woody species from the Miombo woodlands of the Niassa province of northwestern Mozambique, and present a reference collection for a wide range of woody plants growing today along the Niassa Rift. In particular, we determine the taxonomic significance of phytolith types produced by indigenous trees and shrubs from this part of Africa, and hypothesize the phytolithic signal that ‘Zambezian’ plants would leave behind in the geological and archaeological record.
Virtually no modern phytoliths have been identified in the richest and most diversified flora and vegetation on the continent (White, 1983, p. 89), the Zambezian zone, where Miombo woodlands are the dominant element of the largest single phytochorion in sub-Saharan Africa. The total number of taxa in the Flora Zambesiaca is estimated at approx. 11 400, of which Mozambique possesses half (Timberlake, et al., 2006, pp. 751, 754). Miombo woodlands typically display a single storey stand with a discontinuous tree canopy and an underlying sparse layer of small trees, shrubs, sedges and helyophytic grasses (Campbell et al., 1996). In mature Miombos, 95–98 % of the above-ground biomass is comprised of woody plants (Frost, 2006, p. 22). When these woodlands suffer alteration, the tree to grass ratio changes drastically. Bloesch and Mbago (2006) estimate that the canopy cover in moderately modified settings is 40 %, and others claim that in disturbed Miombos the grass biomass surpasses that from the woody component (Ribeiro et al., 2008). Most of the vegetation is briefly deciduous, but, depending on precipitation, some areas support totally deciduous and evergreen plants. White (1983, p. 93) classified Miombo woodlands into ‘wetter’ (>1000 mm of rainfall) and ‘drier’ Miombos (<1000 mm), and the study area supports both. Niassa's woodlands intersperse with other formations; namely (a) semi-evergreen forest; (b) savannas; and (c) hydromorphic grasslands known as ‘dambos’ or ‘mbungas’. This complex scenario of divergent influences is homogenized by the conspicuous dominance of trees from one genus only: Brachystegia (Bloesch and Mbago, 2006, p. 8), although, to a much a lesser extent, it may be co-dominant with Julbernardia. It is this dominance by the subfamily Caesalpinioideae (Fabaceae) that gives both floristic coherence and uniform physiognomy to the Miombo ecoregion (Exell et al., 1960; White, 1983; Frost, 1996).
The study area (Fig. 1) comprises the escarpment flanking the Lake Niassa Rift at an altitude of 465 m a.s.l. and the adjacent highlands (1841 m a.s.l.), which represents the third highest elevation in Mozambique. The districts where our work was conducted (Lago and Sanga) belong to the ‘Lichinga-Cobue’ geological unit (Lächelt, 2004) comprising two stocks: the crystalline basement and the Proterozoic cover with granodiorite, granosyenite, granites, and diverse gneisses and paragneisses (Lächelt, 2004). In the lowlands, red soils are shallow, neutral and they have a silty sand texture (classified as Ferric Lixisols by FAO, 1998; Instituto Nacional de Investigação Agronómica, 1995). In the highlands, red clayey soils have alkaline to acidic pH, and are deep and oxic (also known as Rhodic Ferralsols; FAO, 1998). The mean annual rainfall ranges from 700 mm in the lake-bordering lowlands by Lake Niassa to 1400 mm in the adjacent highlands (Gama, 1990, pp. 31–33). The precipitation system is unimodal, with a copious rainy season that lasts 5 months (January–May) and a severe, drought-prone season from June to December.
From the available sources, we have compiled a list of the 65 most common genera of vascular plants for the province of Niassa (Table 1; Gama, 1990; da Silva et al., 2004; Timberlake et al., 2004; Ribeiro et al., 2008; Bloesch and Mbago, 2006) so that, by comparison with our list of specimens processed (see Supplementary Data, available online), the reader has a measure of the overall ecological coverage and representativeness achieved in this study. The majority of our samples were field-collected by our crew around two loci: Metangula (Lago district, lowlands) and Njawala (Sanga district, highlands; Fig. 1). We targeted all trees and shrubs that our indigenous collectors were able to sample within a radius of 5 km. Botanical specimens were taken at the end of their annual growing cycle, during the dry season, to ensure better phytolith build up in their tissues. For the sake of completeness, when a species was unavailable to us, but it was estimated to be important for our phytolith work, specimens taken from past collections of ours carried out in neighbouring regions and countries since 1993 were used instead. In other cases, samples were obtained from the largest herbarium in sub-Saharan Africa (Pretoria). To identify woody taxa we relied on (a) the expertise of several local collectors; (b) the plant names given to our samples by these collectors; (c) the direct experience with the local flora acquired by our group over the course of five field seasons; (d) the comparison with keys from the standard field guides (Palgrave, 2002; Burrows and Willis, 2005); (e) the taxonomic dictionaries of local plant names mentioned above; and (f) the identifications provided by the Pretoria Herbarium. Our botanical nomenclature is that of the index of accepted names and synonyms from the ‘Checklist of flowering plants of Sub-Saharan Africa’ (Klöpper et al., 2006).
We secured vernacular names through collaboration with five bush doctors over the course of five field seasons, and, to the best of our knowledge, we collected samples from >90 % of the flora most frequently employed by the Yao and Nyanja collectives from Niassa for building, cooking, clothing, medicinal and ritual purposes. Nine specimens bear indigenous names which do not appear in the existing ethnobotanies by Watt et al. (1962), Binns (1972), Jansen and Mendes (1990), Gama (1990), De Koning (1993), Morris (1996), Williamson (2005) and Fowler (2007), and therefore are presented here for the first time. Whenever taxonomic identification was not possible, we have used instead the plant name in Chinyanja/Yao languages. Otherwise, a specimen appears listed as ‘unknown’.
Phytolith extraction from botanical samples followed the dry ashing methodology outlined by Albert and Weiner (2001; cf. Parr et al., 2001), with minor modifications. All specimens were cleaned by immersion in 5 % lab grade soap solution (Micro-90) and sonication (Fisher Scientific, FS 60) for 30 min. In situations where samples were dirt-coated the specimen was soaked in a 5 % soap–sodium hexametaphosphate water-based solution overnight in order to defloculate contaminants, followed by successive 30 min sonication cycles. Specimens were then dried at >100 °C overnight. After cooling, mass was measured on a high precision balance. The samples then went into a muffle furnace for combustion over the course of 36 h at 500 °C. The mass of the resulting ash was noted, and it then received a 10 ml 50: 50 solution made of hydrochloric and nitric acids at 3 n and was boiled. Successive washing cycles removed acids from the sample by (5 min) centrifugation at 3000 r.p.m. When acid elimination was complete, and the remainder of a sample had been dried and weighed, phosphate and carbonate loss was estimated by calculating mass differential. After this, approx. 10 ml of hydrogen peroxide at 30 % was added to the sample to destroy organic matter (if the matrix is rich in carbonates, sodium hypochlorite at 6 % is used instead; Mikutta et al., 2005). The sample was dried at >100 °C overnight. The resulting biominerals formed the acid-insoluble fraction (AIF) and it was here where phytoliths, among other biogenic precipitates, were found. An aliquot of 0·001 g was taken for mounting, after proper mixing of this extract by vortexing. The mounting medium was made up of two droplets of resin solution ‘Entellan New’. The aliquot was well mixed, and the microscopic inspection and counting took place within 48 h of mounting before media were able to dry (3-D shifting of phytoliths was necessary to carry out identification: system microscope, Olympus BX51; ×400 magnifications). Inspection and counting were done under polarized light, regular light microscopy and differential interference contrast (DIC) which greatly enhances contrast and resolves fine structural details.
In total, 180 discrete samples representing >41 families, >77 genera and >90 species of vascular plants have been analysed. Two bryophytes were also included. Stem (n = 78) and leaf (n = 78) tissue were processed separately. Inflorescences, exocarps, nutshells, legume cases and seeds were also processed. The total number of phytoliths counted is 20 372. Average count per species, all tissues combined, is 216. Specimens that yielded very little silica were scanned at least ten times (the coverslip measures 22 mm in length) in adjacent but not overlapping lines. The percentage of each morphotype per species was calculated (Table 2), and types are described according to the descriptors from the International Code for Phytolith Nomenclature 1·0 (Madella et al., 2005). Phytoliths were grouped into 57 morphotypes (two were articulated forms and 55 were discrete shapes) (Table 2).
Once quantified (Albert et al., 1999, 2003; Tsartsidou et al., 2007), the phytoliths were subjected to statistical analyses. Cluster analysis (CA) and principal component analysis (PCA) were employed to reduce the large number of variables in the collection to a smaller set of principal components, reveal internal structure, group type variables and detect covariance (Kim and Mueller, 1978; Basilevsky, 1994; Jollife, 2002). We did not normalize the data for the PCA and ran it on the original correlation matrix. For the CA, we used Ward's method and the squared Euclidean distance with the variables rescaled to 0–1. PCA and CA had matching results. Due to this triangulation, we decided against conducting additional robustness tests. Discriminant analysis (DA) has long been used to identify variables that may classify unknown specimens into known categories because of its ability to detect group-indicators (or predictor phytoliths) and quantify the score on which such grouping variables indicate membership of a given group (Huberty, 1994). Beginning with Fisher's (1936) pioneering work, such studies have proven useful in plant classification (e.g. Ball et al., 1999; Lu and Liu, 2005). For the present study, the quantified phytolith morphotype data were subjected to a series of exploratory DAs using the SYSTAT® 10 software package (SPSS Inc. 2000) that were aimed at identifying morphotype associations that might identify appropriate groups accurately. The large number of null values was problematic and action had to be taken to reduce their numbers, and therefore all cases with fewer than 50 phytoliths and unique morphotypes (see below) were excluded. Exploratory data analysis revealed strong differences in both phytolith quantity and morphotype presence between stem and leaf tissues from the same taxon. It was reasoned that combining morphotype samples from stem and leaf tissues would tend to obscure possible patterning in taxon-specific morphotype associations, and the data set was split into two groups containing phytoliths from either stem or leaf tissue only.
In Table 3 we present the reader with AIF percentages expressed in two separate columns as functions of both the plant's dry weight and its ash. Note that the plant's dry weight contains an arbitrary content of volatile organic components, which disappears by reduction of the sample to ashes. Ash and biomineral contents are expressed in Table 3. Silica production averages 1·42% of the plant's dry weight (range: 0·005–21·79 %). We provide a silica-accumulator rank as well, in which higher units denote greater silica production. The mean content of biogenic silica per ash sample is 13·52 %. Leaf tissue produces an average of 18·60 % AIF (range: 0·43–81·71 %), while stems generate a mean AIF of 9·52 % (range: 0·053–64–69 %). High biomineral content (i.e. twice the average) in leaves has been recorded among members of the Malphigiales, Malvales, Arecales, Rosales and Fabales, while stem biominerals peak among the Gentianales, Liliales, Lamiales, Malvales, Arecales, Asterales and Fabales. The Fabales display the widest production range and plot throughout the spectrum. Yet, it is worth noting that four members are among the top ten producers, above well-known silica accumulators such as the Arecales, Asterales and, to a lesser extent, the Ericales.
From a morphotype diversity point of view, the average number of morphotypes produced by a given family (Table 4) is 11 (range: 1–35), 21 families created ≤10 types, 12 families generate between 11 and 20 morphotypes, and only four yielded >20 types (Fabaceae, 35; Rubiaceae, 25; Asteraceae/Euphorbiaceae, 23). Within the Fabaceae, at the subfamily level, the morphotype variability rank is highest among the Caesalpiniodeae (seven species, 30 types) followed by the Papilionoideae (eight species, 23 types), the Mimosoideae (>10 species, 15 types) and the Faboideae (two species, 15 types).
Eighty-five per cent of the variability documented here is accounted for by a sub-set of phytoliths comprising ten morphotypes. In order of frequency these are as follows.
This morphotype represents leaf epidermal tissue in which the cell shape is isodiametric; it is very abundant (mean value per species is 36 % of the phytoliths produced; range: 1–100 %). A total of five families are above-average producers (at least twice more than average) including the Anacardiaceae, Euphorbiaceae, Fabaceae, Moraceae and the Verbenaceae. Among these top producers, the maximum length of the cell ranges from 18 to 33 µm. Their cell wall thickness varies from 1 µm (Lonchocarpus capassa, Pseudolachnostylis maprouneifolia) to 1·8 µm (Sclerocarya birrea) and 3·7 µm (Uapaca kirkiana) The cell surface can be psilate, granulate or reticulate. The richest producer of this type is Phyllanthus spp. (Euphorbiaceae), in which 100 % of the phytoliths found belong to this category.
This comprises the leaf stomatal complex, trichome bases and hairs. This morphotype reaches a mean frequency of 25 % of the types produced by a given species (range: 1–82 %). A very diverse group of unrelated families produce it in numbers that exceed twice the average, including the Asteraceae, some bryophytes, the Ebenaceae, Euphorbiaceae, Fabaceae and the Urticaceae. The highest value is documented in Pouzolzia mixta, of the Urticaceae, in which 82 % of the phytoliths produced by the species belong to this group. The long trichome is slender and measures between 50 and >100 µm (e.g. Fabaceae and Urticaceae), while the small type is short and thick with a maximum length <50 µm. Both types can be psilate or papilate (e.g. Aspilia mosambicensis) and the two types may appear in the same specimen. Some species do possess hairs with both very square and acute tips (e.g. Albizia anthelmintica).).
This morphotype derives from the xylem's tracheary elements. The average frequency per species is 19 % (range: 1–91 %). Four families produce numbers twice as high as the average, and these are the Clusiaceae, Apocynaceae, Combretaceae and the Fabaceae. Importantly, the Fabaceae is the topmost producer, with values >50 % in Mundulea sericea (91 %).
Mean production reaches 20 % per species (range: 1–100 %). A total of six families can be considered very high producers of globular psilates and they include the Arecaceae, Chrysobalanaceae, Euphorbiaceae, Fabaceae, Kirkiaceae and Proteaceae. The highest abundance is seen in Kirkia acuminata (Kirkiaceae: 100 %). Size is bimodal, but the mean maximum length always is <10 µm. Protea angolensis produces globular psilates that measure 3·4 ± 0·73 µm (n = 107). The mean size of these globular types in Pericopsis angolensis is 5·1 ± 1·7 µm (n = 200), while it increases to 7·8 ± 2·4 µm (n = 171) for Kirkia acuminata and to 8·5 ± 3 mm (n = 200) in Parinari spp. This morphotype may occur indiscriminately in both stem and leaf tissue of the same species (e.g. in the Fabaceae); however, there are families in which globular psilate shapes are exclusive to the leaf (Apocynaceae, Bignoniaceae, Clusiaceae, Myrsinaceae, Myrtaceae, Pandanaceae, Podocarpaceae, Proteaceae, Solanaceae, Sterculiaceae and Verbenaceae) or the stem (Annonaceae, Arecaceae, Bombacaceae, Chrysobalanaceae, Dipterocarpaceae, Euphorbiaceae, Flacourtiaceae, Kirkiaceae, Polygalaceae and Rubiaceae).
The average quantity per species is 20 % of the total (range: 1–100 %). Top producers are members of the Cucurbitaceae, Fabaceae, Annonaceae, Apocynaceae and the Musaceae. The highest rates have been documented among the Apocynaceae and Musaceae, in which this morphotype represents most or all of the phytoliths present. In these cases, the size ranges from an average of 12 µm (e.g. Musaceae) to 20 µm (e.g. Apocynaceae, Annonaceae). Sometimes globular granulates appear in both stem and leaf tissue. Yet, in general, this morphotype reaches higher absolute and relative frequencies in stem tissue compared with other plant parts such as leaf, exocarp and inflorescence. In five families the globular granulate is restricted to leaf tissue: Apiaceae, Arecaceae, Bignoniaceae, Myrsinaceae and Solanaceae. Among the Fabaceae, we note that this type is often present, but generally in low proportions (<14 % of the total per species, range: 2–53 %). In fact, we rarely see any species with truly high production rates (exception: Dalbergiella nyasae).
This is is produced exclusively by the Arecaceae. The average percentage in a given species is 33 (range: 11–64 %). This phytolith is produced in stem, leaf and inflorescence tissues. The highest yield observed by us is by Hyphaene spp. Size varies across species, from about 7 µm for Borassus aethiopum to 14 µm for Hyphaene spp.
This type is a body with a psilate to ridged texture that may appear articulated or individually. Its size ranges from 25 to >50 µm. Few families produce this morphotype in a significant proportion; namely, the Apocynaceae (Ectadiopsis oblongifolia), Asteraceae (Brachylaena spp.) and Fabaceae (Albizia anthelmintica). For the most part, this morphotype is generated in stems.
This type represents leaf epidermal tissue in which the cell shape is irregular. Mean frequency is 22 % of the phytoliths produced by a given species (range: 1–74 %). Only one family is a high producer (twice or more than average), the Thelypteridaceae; although four families produce just above average: Euphorbiaceae, Asteraceae, Combretaceae and Fabaceae (in Afzelia quanzensis this morphotype represents 40 % of the total assemblage produced by the species). Among these top producers, the maximum length of the cell ranges from 28 to 65 µm. Their cell wall thickness is around 1·2 µm. The cell wall in some species is pitted (e.g. Afzelia quanzensis). The cell surface is mostly psilate, but some specimens display sulcate textures (e.g. Cyclosorus spp.).
The mean frequency of this type is 9 % (range: 1–73 %). Only two families produce this type at a level twice the average, and these are the Fabaceae (Piliostigma thonningii and Pterocarpus angolensis) and the Arecaceae (Phoenix reclinata). This type appears in small and medium sizes, but in some instances it can be >50 µm long, even >100 µm.
The average representation per species is 7 % (range: 1–29 %). Five families produce twice the average: Acanthaceae, Rhamnaceae, Fabaceae, Asteraceae and Rubiaceae. The highest production rate (29 %) is documented in the stem of Pavetta crassipes (Rubiaceae). This type appears in stem tissue twice as frequently as it is documented in leaf, exocarp or the legume case. Only rarely are blocky shapes found in both stem and leaf tissue of the same species (Mundulea sericea, Pterocarpus angolensis; Fabaceae/Sterculia quinqueloba, Sterculiaceae). It is exclusive to the stem tissue of the Annonaceae, Apocynaceae, Asteraceae, Amaranthaceae, Clusiaceae, Combretaceae, Euphorbiaceae, some members of the Fabaceae, Polygalaceae, Rhamnaceae, Rubiaceae and Verbenaceae. Conversely, the blocky type is exclusive to the leaf tissue of the Apiaceae, Arecaceae, Clusiaceae, Flacourtiaceae, Myrsinaceae, Podocarpaceae, Proteaceae and Solanaceae.
There are seven morphotypes that are exclusive to the species level. These types are (1) blocky facetate (morphotype 4, n = 1, Uapaca nitida, Fig. 5q); (2) epidermal laminate (morphotype 21, n = 1, Cassia spp., Fig. 3b); (3) globulose bisected (morphotype 37 n = 2, Solanum panduriforme, Fig. 3ae); (4) cylindroid reticulate (morphotype 18, n = 6, Solanum panduriforme, Fig. 2f); (5) tabular thin pilate (morphotype 56, n = 8, Cassytha spp., Fig. 6i); (6) blocky hairy (morphotype 5, n = 9, Podocarpus falcatus, Fig. 5r); and (7) blocky radiating (morphotype 8, n = 56, Aeschynomene spp., Fig. 5u). Altogether these types represent 0·0043 % of the total assemblage, with negligible intraspecies percentages ranging from 0·002352941 to 0·185430464 %. Obviously, these idiosyncratic morphotypes have very high taxonomic value, but their low numbers make their detection in small collections exceedingly unlikely. Note that for the proportions of these unique phytoliths to reach a mere 1 % of a given assemblage the investigator would need to study a very species-rich botanical collection and carry out extremely high total phytolith counts, in the order of 75 000–4 000 000.
Backward, stepwise DAs were run on the stem vs. leaf data set to find morphotype associations that could be useful in classifying unknown morphotype samples to the order level (Table 5). These data sets contain relatively few cases per order [e.g. for the stem tissue data set, n varies between 1 and 11, with eight of the ten orders present in the sample being represented by only one (n = 4) or two (n = 4) cases]. These are small numbers, but, given the pioneering nature of the present study, it was felt that our results may guide future research efforts and develop a better understanding of the available data. DA of the leaf tissue sample did not result in successful identification of morphotype associations that could be used reliably to assign unknown samples to the order level. An exception to this overall result is the order Rosales, which has quite distinctive leaf tissue phytolith morphotype associations, but the failure of classification for the remaining orders and the misclassification of non-Rosales specimens as Rosales makes the production of a useful classification function problematic. It may be that further work with larger numbers of cases per order might resolve this problem but, for the present, leaf phytolith morphotype associations cannot be used to classify unknown materials.
DA of the stem tissue data set, on the other hand, initially appeared to be very successful. Of the ten orders present, 100 % were correctly classified by the analysis. Using the same data both to calculate discriminant functions and then to estimate their accuracy, however, tends to overestimate the usefulness of the discriminant functions in classifying new data. Cross-validation was conducted using the SYSTAT® 10 ‘Jackknife’ procedure (SPSS Inc.: I:292). Jackknifing involves computing functions using all available data except the cases being classified, i.e. as each case is classified it is excluded from the discriminant function calculation that is used to classify it. This procedure almost always produces poorer but more realistic results than using the entire data set to predict group membership on a post hoc basis. The results of the jackknifed classification procedure are much poorer than the initial run (Table 5); only 22 % of cases were correctly assigned. Nevertheless, jackknifing correctly classified three orders at 100 % (Caryophyllales, Malvales and Solanales), and two more (Fabales and Malphigiales) were correctly classified at 18–25 %. Divergent results from the two above discriminant procedures highlight the fact that there are too many predictors (SPSS Inc.:I:292). However, the better-than-chance outcome is an indication of an underlying association pattern that, with further study, might be useful in classifying unknown samples. It is anticipated that increasing the number of specimens per order could substantially improve the robustness of the results, and would allow us to explore if there could be predictors that may be able to identify, for example, Fabaceae subfamilies (cf. Soladoye, 1982).
The CA and PCA were run on the full data set to determine the consistency of the morphotype signal from different families and orders, and all types that did not load per iteration (seven iterations for the stem data) were dropped, until the data reached a simple structure in which each remaining morphotype loaded to only one component (Table 6; Fig. 7). The first two axes of the PCA account for 26·8 % of the total variance for stem, and 42·9 % for leaf tissue (not presented here). For stem data alone, we found that 15 phytolith morphotypes created seven principal components (types 6, 7, 10, 11, 14, 17, 27, 29, 30, 32, 34, 43, 49, 50 and 55). Note that the type ‘cylindroid scrobiculate’ was dropped because of perfect correlation with the type ‘cylindroid large’, and the latter was kept because it displayed more variance. Leaf tissue yielded six principal components (phytolith morphotypes 7, 11, 15, 16, 19, 30, 32, 39, 43, 46, 47, 48, 50 and 57). For stem and leaf tissue combined, the identification of specific morphotypes as principal components overlaps across five morphotypes: 7, 11, 30, 32, 43 and 50.
The cluster ordination of stem morphotypes relative to families and species was conducted on the basis of both linkage distance among phytolith morphotypes and their associations. The resulting dendrogram (Fig. 8) shows well-defined groups that have within themselves similar samples; also the cluster diagram by phytolith morphotype matches the PCA results closely. The number of clusters thus identified is ten (Table 6). Cluster number three separates out from the rest clearly, but the short rescaled linkage distance for all of its samples (<1) highlights the fact the (taxonomically unrelated) families contained therein have the least distinctiveness of all clusters. Contrarily, the wider dissimilarity coefficient seen in the remaining nine clusters speaks to the distinctiveness displayed by several members of the Amaranthaceae, Annonaceae, Apocynaceae, Arecaceae, Asteraceae, Dipterocarpaceae, Euphorbiaceae, Fabaceae and Urticaceae. In a separate dendrogram (not presented here) leaf morphotypes cluster similarly to stem types, and the overlap includes species from the Amaranthaceae, Annonaceae, Apocynaceae and the Asteraceae.
We have supplied the reader with the phytolith spectra produced by woody species from the most widespread ecosystem within the Zambezian regional centre of plant endemism, and now turn to the assessment of two questions raised by our results: how do the documented morphometric variables indicate taxonomic membership and what is the phytolith profile that hypothetically, if found in a geo-edaphic or archaeological context, would allow the scientist to infer plant community structure similar to the one seen today in Miombo woodlands.
Over the last 10 years, the application of quantitative methods to study phytolith assemblages produced by modern plants has focused on questions such as the ability of specific morphotypes to discriminate subfamilies (Honaine et al., 2006), genera (Zucol, 1998, 2000) and species (Ball et al., 1999), the signal produced by C3 vs. C4 plant communities (Gallego and Distel, 2004), the grouping of taxonomically related species (Carnelli et al., 2004) and the connections between morphotype diversity and anatomical leaf features (Marx et al., 2004). The variables employed by these authors are the presence and frequency of types (Carnelli et al., 2004), the frequency of morphotypes expressed as a percentage of the total (Zucol, 1998, 2000; Gallego and Distel, 2004; Honaine et al., 2006), the ordination of both types and abundances (Marx et al., 2004) or simply the descriptive definition of keys (Ball et al., 1999). Their techniques have included PCA (Zucol, 1998, 2000; Honaine et al., 2006), CA (Carnelli et al., 2004; Gallego and Distel, 2004; Marx et al., 2004) and DA (Ball et al., 1999).
In our case, PCA, CA and DA results support the hypothesis that a subset of orders and families can be clustered and identified on the basis of phytolith morphotype abundance and associations. To some extent, statistical analyses confirm what has been known for classic arboreal phytolith types for many years (Bozarth, 1992): they have very little taxonomic power because of redundancy across orders. In this respect, it is significant that none of the common types (85 % of the total) was flagged by PCA as a principal component. Yet, stating that arboreal phytoliths have no explanatory value would be a gross oversimplification. It is within the ‘uncommon’ and ‘unique’ categories (about 15 % of the total) that we see promise: blind, independent statistical testing and validation were able to identify up to 15 distinctive morphotypes produced by several members of derived clades among the Asterids (Asterales/Asteraceae; Gentianales/Apocynaceae; Ericales/Ebenaceae; Myrsinaceae; Solanales/Solanaceae), some core eudicots (Rosales/Urticaceae; Caryophyllales/Amaranthaceae; Malpighiales/Euphorbiaceae; Fabales/Fabaceae; Malvales/Dipterocarpaceae, Sterculiaceae; Myrtales/Myrtaceae), monocots (Arecales/Arecaceae) and primitive monocots within the Magnolids (Magnoliales/Annonaceae) as arboreal groups that produce phytolith morphotypes with acceptable discriminatory capabilities.
The phytolith profile studied here demonstrates that silicon is a low prominence mineral plant constituent in Miombo trees and bushes. This finding is in agreement with reported low values of silica concentrations among dicotyledonous plants, especially among the Fabales (Epstein, 1994, p. 12; Hodson et al., 2005, p. 1040; cf. Carnelli et al., 2001; Marx et al., 2004). Woody species from Niassa produce large phytoliths. Over half of our morphotypes are considerably larger than 50 µm, and, in fact, smaller types (<20 µm) are in a minority, making up about one-fifth of the total number of morphotypes. This finding indicates that extraction techniques that sieve sediments or use gravity sedimentation with a cut-off, discard point between 50·238 and 63·246 µm should not be used as these methodologies surely cannot extract phytoliths >50 µm. Another defining characteristic of Miombo phytoliths is their heavily pitted (decorated) textures, which are a natural feature of this population and not weathering markers left by taphonomic processes such as dissolution. With regard to contamination, we have observed instances in which stem specimens contained phytoliths known to be produced exclusively by leaf tissue or the grasses; therefore, these types can only be interpreted as foreign, unrelated silica particles (Tsartsidou et al., 2007: Fig. 7). Yet, the instances in which blatant contamination was detected were rare. Moreover, we used heavily polluted specimens (n = 3) to test if an aggressive cleaning methodology could reduce impurity levels. We noticed that soaking and deflocculating overnight prior to cleaning and the use of extended, successive sonication cycles were able to cut down contamination rates by 60 (Ficus spp.) to 75 % (Dolichos kilimandscharicus), as deduced from both AIF differences for the same specimen and absolute counts of control targets in the same sample before and after cleaning (e.g. clusters of isodiametric cells exclusive to the leaf tissue). Yet, it should be borne in mind that, at least in our reference collection from Niassa, these are extreme situations, and that a specific stem phytolith sample has to be studied under the microscope to establish contamination on an individual basis before it can be concluded that contamination is an issue for that specific sample, or that contamination of all stem samples in a collection is prevalent.
Detailed taphonomic analysis of phytoliths is beyond the scope of this study. Only the retrieval, quantification and especially the comparison of opal silica from local soils and sediments with the data presented here will establish the aspects of quality and source of bias in the regional fossil record. Many filters govern the long-term preservation of phytoliths, including silica supply, the matrix geochemistry, the likelihood of immediate (and permanent) burial, reworking, the impact of silica recycling by living plants and the capacity of a phytolith morphotype to resist the changing physical and geochemical conditions that typify the highly dynamic and time-averaged nature of topsoils (Behrensmeyer et al., 2000, p. 107). Until a meticulous phytolith catalogue from modern soils is available, we can only suggest that the compositional fidelity in Miombo assemblages will probably be low. We foresee selective destruction of any silica body that is partially silicified or thin, which alone would bring about a loss of around 50 % of the total assemblage (Fig. 9). Morphotypes that may stand better chances at preservation include globular psilates, globular granulates and echinates, blocky types, cylinders and tabular shapes. To this segment that comprises about half the available silica produced by local arboreal plants, we would have to add grass silica for the Miombo phytolith spectrum to be comprehensive.
We have provided the first quantitative taxonomy of phytoliths for the largest phytochorion of sub-Saharan Africa. This taxonomy includes new types and is the most extensive phytolith key for any African ecoregion (41 families, 77 genera, 90 species). We contribute to filling the current information gap in woody phytoliths and explore their taxonomic value in paleoecological reconstruction through phytolith analysis. More specifically, we have presented evidence that Miombo woody species are hypervariable silica producers and their phytolith morphotypes are many and highly polymorphic. The taxonomic significance of phytolith types produced by indigenous trees and shrubs from this part of Africa is largely poor, but there are important exceptions that include several morphotypes produced by members of >10 families and orders. Lastly, this paper has put forward a model illustrative of the phytolithic signal that ‘Zambezian’ woody species are likely to leave behind in the fossil record: a skewed segment of the total phytolith spectrum favouring blocky, cylindroid, globular and tabular forms.
This work could have not been accomplished without Arianna Fogelman, Lourenço Thawe, Justin Sondergaard, Sofia Sondergaard, and the numerous workers, friends and authorities in Niassa. The authors thank the Department of Anthropology and Archaeology at Eduardo Mondlane University for the support, collegiality and encouragement, especially Professor Hilário Madiquida. Work in Niassa was conducted under two permits to J.M. from Eduardo Mondlane University and the Ministry of Education and Culture (03-2003 and 01-2007). Temporary export of materials was conducted under ‘Certificate of Origin no. 0134’ from the Mozambican Chamber of Commerce, as well as the ‘Export License no. 24399’ from the Mozambican Customs Service. We thank the Canada Research Chairs programme and the Canada Foundation for Innovation for making much of this research possible through generous grants and research endowments (CFI grant no. 201550) to the lead author, and the Tropical Archaeology Laboratory at the University of Calgary, the Faculty of Social Sciences, Department of Archaeology and various internal programmes at the University of Calgary made available financial and logistical support to the authors. The Social Sciences and Humanities Research Council of Canada (File no. 410-2007-0697; CID: 148244), as well as the American Embassy in Maputo (Ambassador's Fund for Cultural Preservation) assisted this project with essential monetary aid. The Department of Anthropology at the George Washington University and the Human Origins Program at the Smithsonian Institution provided institutional support. The South African National Biodiversity Institute, specifically the Pretoria Herbarium, provided us with invaluable plant reference material, books, advice, specimen identification services and a friendly and efficient work environment. We thank Rosa M. Albert for her guidance.