The Cyperus margaritaceus-niveus complex is here understood to comprise a group of ten species (C. karlschumannii C.B.Clarke, C. kibweanus Duvign., C. ledermannii (Kük.) S.S.Hooper, C. margaritaceus Vahl, C. niveus Retz., C. nduru Cherm., C. obtusiflorus Vahl, C. somaliensis C.B.Clarke, C. sphaerocephalus Vahl, and C. tisserantii Cherm.) distributed throughout sub-Saharan Africa and Madagascar. The group has been considered a species complex since Kükenthal (1936), who treated these taxa as varieties under C. margaritaceus, except for C. niveus and C. obtusiflorus, which he considered to be separate species, and C. kibweanus, which was first described in 1963 (Duvigneaud and Denaeyer-de Smet 1963). Other varieties of C. margaritaceus were also recognised by Kükenthal (1936), but these have since been reduced to synonymy of other taxa in this group. A recent molecular phylogenetic study about resolving the relationships in the C₄Cyperus clade has shown that some members of the Cyperus margaritaceus-niveus complex form a monophyletic group separate from and sister to most of the remaining C₄Cyperus species (Larridon et al. 2020). Although three taxa – C. kibweanus, C. sphaerocephalus, and C. somaliensis – were not included in that phylogenetic study, the results suggested that the morphological combination of a capitate inflorescence of multiflowered spikelets with white distichously arranged glumes, sometimes with streaks of green or yellow, with modified culm bases is unique to this species complex.

Current species delimitations for the taxa in this complex are based on classical morphotaxonomy. Despite a considerable overlap of the characters used to delimit the taxa across the regional African Floras (Table 1), Flora treatments have not reached a consensus on the species limits in this complex. This can be explained by the fact that different authors placed different taxonomic emphasis on a particular character, i.e. whether it should be used at specific, subspecific, or varietal rank. The characteristic white glumes are not used as a character per se in any of the Floras, although the taxa are placed in the ‘capitate species’ group of Cyperus within the key along with other species with cream, yellow, or brown glumes. As a result of this lack of consensus, the Flora treatments published over the past 50 years differ considerably in their circumscription of the taxa (Table 2). In Flora of West Tropical Africa, Hooper and Napper (1972) recognised five taxa as distinct species, excluding C. kibweanus, C. obtusiflorus, C. niveus, C. somaliensis, and C. sphaerocephalus as these occurred outside the region covered by the flora. Haines and Lye (1983) only recognised C. margaritaceus and C. niveus as separate species based on small differences in stem base size and degree of spikelet compression. The remaining taxa, excluding C. kibweanus, C. karlschumannii, and C. somaliensis, were reduced to varieties of these two species, while C. obtusiflorus was considered to be a synonym of C. niveus var. leucocephalus (Kunth) Fosberg. This followed Fosberg (1977) who commented on the few differences between African specimens of C. niveus and C. obtusiflorus, and, in his account for Flora of Aldabra, he recognized three varieties of C. niveus, sinking C. obtusiflorus into C. niveus var. leucocephalus; niveus being the older name. Later, both Flora of Tropical East Africa (Hoenselaar et al. 2010) and Flora Zambesiaca (Browning et al. 2020) broadly followed the concepts of Haines and Lye (1983) separating C. margaritaceus and C. niveus but elevating C. nduru to species level, transferring C. tisserantii to a variety of C. niveus – not C. margaritaceus cf. Haines and Lye (1983) – and reducing C. ledermannii to synonymy under C. niveus var. leucocephalus. In the notes accompanying the species account of C. niveus written for Flora of Tropical East Africa, Hoenselaar et al. (2010) mentioned that presence and absence of rhizomes was used as an additional character in their key to separate C. margaritaceus from C. niveus but noted it is possible that they could represent one species. Lye (1997), in his account for the Flora of Ethiopia and Eritrea, only includes C. niveus, but recognises two varieties: var. leucocephalus and var. tisserantii. Haines and Lye (1983) considered C. sphaerocephalus to be no more than a variety of C. niveus placing it under C. niveus var. flavissimus; a concept followed by Lye (1995). Hoenselaar et al. (2010) and Browning et al. (2020) elevated the variety to species on account of its distinctive yellow inflorescence. Cyperus somaliensis was first described by Clarke (1895) from among many collections brought to him from present-day Somalia. Due to its endemicity, it has not been treated in other African Floras and thus lacks a morphological comparison with the other taxa, though Lye (1995) compared it with C. niveus in Flora of Somalia, separating the two taxa only on glume length. Cyperus kibweanus was first described by Duvigneaud (Duvigneaud and Denaeyer-de Smet 1963) from a single collection in present day Democratic Republic of the Congo and it is not present in any of the aforementioned Floras.

Given the differing taxonomic opinions outlined above, in this study, we use the currently accepted species, as presented in the relevant Floras, as our baseline taxa, which function as our hypotheses. Morphometrics is then used as an algorithmic process to test the results of the traditional taxonomic procedure and thereby test the robustness of the authors’ original circumscriptions from the Flora accounts. As such, C. margaritaceus is the most widespread species, occurring in west tropical Africa, central Africa, east Africa, and southern Africa but not in Madagascar. Cyperus karlschumannii, C. ledermannii, and C. tisserantii are native to west tropical Africa, although C. tisserantii extends south into the Democratic Republic of the Congo. Cyperus niveus and C. nduru – the latter characterised by bottle-like plant bases and hardened basal leaf sheaths (Fig. 1E, F) – have overlapping distributions; the former mostly found in east Africa and the latter mainly in west tropical Africa and southern tropical Africa. Cyperus obtusiflorus occurs from Sudan and Ethiopia down to South Africa and is widespread in Madagascar. Cyperus somaliensis is endemic to Somalia, while C. sphaerocephalus extends from tropical east Africa to South Africa. Figure 2 shows the locations of the herbarium specimens used in this study.

Morphometrics has been shown to be a useful tool to investigate species limits in closely related taxa (Marhold 2011), either used alone (e.g. Atkinson and Codling 1986; Brysting and Elven 2000; Rivero-Guerra 2011) or in conjunction with other approaches such as molecular phylogenetics (e.g. Perný et al. 2005; Barrett and Freudenstein 2009; Kučera et al. 2010). Within Cyperaceae, a number of species complexes have been assessed using morphometrics (e.g. Rosen 2006; Naczi and Moyer 2016; Di Natale et al. 2020), with a particular focus on Carex L. (e.g. Naczi et al. 1998; Smith and Waterway 2008; Míguez Rios 2017), which is the most species-rich genus in the family, as well as one of the largest angiosperm genera. Comparatively, few morphometric studies have focussed on Cyperus L. (e.g. Carter and Bryson 2000; Lowe 2018) although such studies have resulted in species new to science being described (Gardner et al. 2014; Gray and Stott 2017).

In this paper, we use multivariate analyses as a tool to support hypotheses concerning the taxa involved in the Cyperus margaritaceus-niveus complex. In the analyses, we include most of the characters used in the relevant Flora treatments (Table 1) to test their utility for assessing species limits. Unlike the Flora accounts based on partial subsets of the taxa of the complex, all species were included in the present study and across the entire geographical range, except for C. kibweanus due to a lack of sufficient number of specimens. Three different types of analysis were carried out: Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Classification and Regression Tree (CART) analysis. PCA and LDA used only the quantitative characters present in the data set, whereas CART used the full data set, i.e. including the qualitative characters.

This study aims to provide testable circumscriptions of taxonomic species (see Mayo 2022) based primarily on morphological data. The data sets (Supplementary material 1) and analysis scripts are available for future testing and improvement. Our objective is to contribute a computable morphological framework for future integrative species taxonomy in Cyperaceae (Dayrat 2005; Will et al. 2005; Pante et al. 2015).

Within the Cyperus margaritaceus-niveus complex, the analyses classify the individuals, originally assigned to nine species, into eight groups in the LDA and six groups in the CART analysis. The two results we wish to highlight are the failure of C. niveus to separate as a group in both analyses and the retention of C. margaritaceus as a distinct entity mixed together with a small number of individuals of other species. In the PCA, although the plots for C. niveus and C. sphaerocephalus are embedded within the overall scatter plot, C. sphaerocephalus is still separated by LDA and CART, while C. niveus is not separated by either analysis. In both the LDA and CART cross-validation, the 12 specimens of C. niveus were distributed among several taxa, although without consistency between the two analyses (Tables 4, 5). It is possible that characters not included in this study would help to separate this species as a cross-validated group, but our present results imply that C. niveus is not sufficiently distinct from the other taxa in the complex. This corroborates the view of Fosberg (1977) who did not consider C. niveus as a separate species when compared to C. obtusiflorus even though he focussed on material from Aldabra. The taxonomic differences between the species according to Kükenthal (1936) were somewhat dismissed by Fosberg as “a weak difference indeed”; hence, his reduction of C. obtusiflorus into C. niveus, which subsequent regional Floras have followed. In our analysis the type specimen of C. niveus from India was also placed in the C. obtusiflorus group although Fosberg recognised the typical variety only of C. niveus as being confined to southern Asia. No Asian material was examined in this study, but clearly the taxonomic status of C. niveus requires further elucidation.

Cyperus margaritaceus shows considerable overlap with the other taxa in the PCA (Fig. 3) with 41% of the total sample study (Table 4) and the widest geographical range of the species in the complex. However, despite this apparent “clouding” of the data in the two-dimensional ordinations, the high percentage of correctly assigned individuals in the cross-validations of the LDA and CART analysis shows that C. margaritaceus is a robust entity, supporting its circumscription in the regional African Floras as a distinct species. Its delimitation as a separate species is further strengthened by a character not used in our analyses but highlighted by Hooper and Napper (1972): the presence of a rhizome consisting of a series of connected culm bases. While this character is also found in other species in the complex, in C. margaritaceus it is found in combination with discrete spikelets (as opposed to confluent spikelets) that are relatively few in number and moderate in length. The CART analysis separated the species into two groups based on the number of spikelets in an inflorescence; individuals in the group marg1.crt had less than 7.2 spikelets per inflorescence, while those in group marg2.crt, a considerably smaller group, had more spikelets. This character alone may not be sufficient for splitting C. margaritaceus into two species and furthermore, the number of original C. margaritaceus individuals in this group is tiny with no geographical correlation among them.

Cyperus nduru was treated as a variety of C. margaritaceus by Haines and Lye (1983) but as a separate species by Hoenselaar et al. (2010) and Browning et al. (2020). The LDA supports the assertion that C. nduru is a distinct entity, though the CART analysis separates them into two groups by glume width. All the specimens from west tropical Africa were assigned to the nduru1.crt group, while the specimens from D.R.Congo, east tropical and south tropical Africa were assigned to both the nduru1.crt and nduru2.crt groups. Our results support the notion that C. nduru is a distinct entity within the complex but that glume width alone may not be sufficient to recognise two separate taxa especially since the individual specimen assignments show no clear geographical distinction between the nduru.crt groups.

The taxonomic status of C. tisserantii differs amongst the regional Floras (Table 2). Hooper and Napper (1972) separated it from C. nduru based on the number of involucral bracts and their length in relation to the spikelets. Our analyses separate this species as a distinct group although the characters used by Hooper and Napper (1972) do not necessarily have high influence according to the LDA axes. Number of involucral bracts scored relatively high on both axes but length of involucral bracts scored very low for the first axis (Supplementary material 5). Nevertheless, the CART analysis shows that C. tisserantii is recognised as a distinct entity based on glume width, number of spikelets, length of bracts, and spikelet length. The length of the bracts was used by CART to separate this species from C. tisserantii, and although the PCA shows considerable overlap with other taxa, mainly C. nduru, the cross-validation analyses show that these two species – C. tisserantii and C. nduru – are two separate groups.

The analyses show strong support for C. karlschumannii as a distinct entity. This species can be easily distinguished from the other taxa in the complex by the considerably longer ovate spikelets, ranging from white to pale-yellow and with broader glumes.

The analyses differ as to the identity of C. ledermannii. While the LDA treats it as a separate entity, the CART analysis places most of the C. ledermannii individuals in the obtus.crt group. Hooper (1972) originally raised the status of C. ledermannii from variety to species commenting that “this taxon seems as worthy as others in the C. obtusiflorus-margaritaceus group of specific recognition”, although there was no direct comparison made with C. obtusiflorus. Cyperus ledermannii is restricted to west tropical Africa though Hooper commented that its occurrence in east tropical Africa seemed probable. Haines and Lye (1983), who kept it as a variety of C. niveus following Kükenthal (1936), noted that, for east Africa, this taxon is only recorded from Tanzania. The type of C. niveus var. ledermannii, observed for this study on JSTOR, has a large of head of many discrete spikelets and in this respect is most similar to C. obtusiflorus. The CART analysis supports Hoenselaar et al. (2010) and Browning et al. (2020) who reduced C. ledermannii to synonymy under C. niveus var. leucocephalus. It is therefore possible that C. ledermannii does not represent a separate morphospecies.

Cyperus sphaerocephalus, known as the golden headed sedge in South Africa, was long considered as C. obtusiflorus var. flavissimus, until Hilliard and Burtt (1986) made morphological comparisons between it and the typical variety of C. obtusiflorus. They found that several quantitative characters (number of spikelets, spikelet length and width, glume length and width, involucral bract length, diameter of head), rather than inflorescence colour alone, separated the two varieties and hence elevated C. obtusiflorus var. flavissimus to species level as C. sphaerocephalus. Within South Africa, these two taxa have also been treated as separate species based on differing distributional patterns. Gordon-Gray (1995) observed how C. sphaerocephalus was mainly confined to Natal, while the “white-headed form” was predominantly at “Coastland though Midlands, into Zululand and N Natal”. Gordon-Gray (1995) stressed that further investigation should cover both entities across the whole of Africa. Since our methods take the whole range of these species into account, we conclude that our results support the notion of both species representing distinct entities, although the CART analysis separates the taxon further based on glume width. Individuals of C. sphaerocephalus make up the majority of both sphaer.crt groups but the individual assignments of this taxon for each group reveal a geographical correlation. The majority of individuals in the sphaer1.crt group are from South Africa, whereas those from sphaer2.crt are mostly from outside of South Africa. Further work is needed to determine whether they represent subspecies or varieties.

The entity of C. somaliensis also differs between the analyses. While the LDA distinguishes this species as a distinct entity, the CART analysis shows little consistency with regards to assigning individuals. This disparity may be due to the comparative lack of specimens used that belong to this taxon; only six specimens of C. somaliensis were used in the analyses. When Clarke (1895) originally described the species, no diagnosis was given though Clarke commented it was “near C. leucocephalus” a species from India, which has morphological similarities. In Flora of Somalia, Lye (1995) compared it with C. niveus, although C. niveus is not a distinct entity in any of the analyses and further work is needed to establish its relationship to other species in the complex.

The assignment of type specimens to different cross-validated groups across the analyses makes their classification difficult. Types may not be typical of the specimens assigned to their particular species, however, missing character values may alter their true position in the multispace and hence affect their assignment in the cross-validation.

For the complex as a whole, characters with the highest loadings (glume width, number of bracts, glume length, and number of spikelets per head) were also used by Haines and Lye (1983), Hoenselaar et al. (2010), and Browning et al. (2020) for delimiting some of these taxa at varietal rank. Similarly, these characters were used by Hooper and Napper (1972) to separate them at species level. Despite differences in taxonomic opinion (see Table 2), the fact that the characters used in these regional Floras also represent the highest loadings in the PCA indicate their taxonomic informativeness for delineating the taxa.

In summary, of the nine putative species studied, six morphospecies can be recognised based on our results: C. karlschumannii, C. margaritaceus, C. nduru, C. obtusiflorus, C. sphaerocephalus, and C. tisserantii. Each of the six morphospecies is distinguished by a combination of morphological characters and geographical distribution. The picture is more complex for C. ledermannii and C. somaliensis, given the differences between the LDA and CART analysis. Further work is desirable to elucidate the taxonomic affinities of these two taxa. None of the analyses considered C. niveus to be a separate entity for Africa. We have also shown that the morphological characters used in classical taxonomic studies for this complex are of sufficient taxonomic value to delimit similar morphogroups as presented in the regional Floras. The analyses used in this study provide a transparent and repeatable methodology for justifying the recognition of morphology-based taxa within this complex, which also enables morphological characters, not included in this study (e.g. nutlet dimensions, anther length), to be added to our dataset to corroborate our results. We recommend the application of this methodology to resolve other known species complexes in the Cyperaceae.

The R analysis script is deposited at

https://doi.org/10.5061/dryad.2ngf1vht5