We searched the literature and constructed a database of published SEM images of grass pollen surface ornamentation. This literature included SEM surveys of taxa from several different subfamilies within Poaceae (Watson & Bell 1975; Page 1978; Köhler & Lange 1979; Perveen & Qaiser 2012; Mander et al. 2013), focused studies of a single species (Datta & Chaturvedi 2004) or subfamily (Andersen & Bertelsen 1972; Grant 1972; Peltre et al. 1987; Liu et al. 2004) and isolated SEM images published as part of other morphological work (Skvarla et al. 2003; Sivaguru et al. 2012). We also include images from the PalDat pollen image database (http://paldat.org), with the understanding that the broader palynological community contributed these images and that their taxonomic provenance is subject to greater uncertainty. Each SEM image in these publications was visually inspected and only those images in which the surface ornamentation was clear and not masked by folding of the exine, poor illumination or deposits on the surface of the pollen grain were included in our database. Each grass taxon in the database was assigned to a Poaceae subfamily using A World-Wide Phylogenetic Classification of Poaceae (available via http://www.tropicos.org; published as Soreng (2000, 2001) and Soreng et al. (2015)).

Each SEM image in this database was grouped into one of the six morphotypes shown in Figure 1. Four of these, the Hordeum-type, Triticum-type, Avena-type and Setaria-type, are morphotypes that have been established during previous work (Köhler & Lange 1979). The Hordeum-type ‘is characterised by single detached spinules generally not grouped in [areolae]’ (Köhler & Lange 1979, p. 134) (Fig. 1a). The Triticum-type ‘is characterised by small [areolae] bearing only one to three, rarely more, spinules’ (Köhler & Lange 1979, p. 137) (Fig. 1b). The Avena-type ‘has larger irregularly polygonal or elongate [areolae] studded with several, [1–10] spinules’ (Köhler & Lange 1979, p. 137) (Fig. 1c). The Setaria-type ‘is characterised by extensive field-like [areolae] of irregularly polygonal outlines. Their bulging surface is studded with very small pointed spinules’ (Köhler & Lange 1979, p. 139) (Fig. 1d). In these descriptions, the term ‘areolae’ replaces ‘insulae’, following Punt et al. (2007).

To these four established morphotypes we add the Pariana-type and the Stipa-type. The Pariana-type is characterized by ‘fused and spinulose areolate elements grouped into irregular islands interspersed with globular/granular exine particles approximately 0.25 µm in diameter’ (Skvarla et al. 2003, p. 925) (Fig. 1e), while the Stipa-type is characterized by simple scabrate ornamentation that lacks areolae and has sparsely distributed granula (Fig. 1f). Our database is summarized in Table 1.

In our database, there are SEM data on the surface ornamentation of the pollen of 99 species of grass. This represents 0.89% of the 11 157 grass species that are currently recognized (GPWG 2012) and there are SEM data on pollen surface ornamentation for just six of the 13 subfamilies within Poaceae (Table 1). There is a positive relationship between the species richness of each subfamily and the number of species with an SEM record of their surface ornamentation (Fig. 2). Larger subfamilies have the most species with an SEM record of their pollen surface ornamentation, whereas smaller subfamilies have received considerably less attention (Fig. 2).

The Poaceae subfamilies are traditionally grouped into three clades (Table 1). There are no SEM data on the pollen of the early diverging APP clade of grasses, which consists of three small subfamilies and contains a total of 28 species (Fig. 3). In the BEP clade, which contains 5423 species within three subfamilies that each exclusively use the C3 photosynthetic pathway (Christin et al. 2013), there are SEM data on the surface ornamentation of one species within the Ehrhartoideae subfamily, six species within the Bambusoideae subfamily and 53 species within the Pooideae subamily (Fig. 3). In the PACMAD clade, which contains a total of 5706 species and includes those that have evolved C4 photosynthesis (Christin et al. 2013), there are SEM data on the surface ornamentation of 22 species within the Panicoideae subfamily, one species within the Arundinoideae subfamily and 16 species within the Chloridoideae subfamily (Fig. 3). However, there are no SEM data on the pollen of the six species contained in the Centropodia clade or the other three subfamilies within the PACMAD clade (Fig. 3).

There are not enough SEM data to reliably assess the phylogenetic distribution of grass pollen surface ornamentation patterns. However, the available information indicates that the Avena-, Hordeum-, Setaria- and Triticum-types all occur in both the BEP and PACMAD clades (Fig. 3), and this shows that at present these four morphotypes cannot be used to reliably distinguish major groups of grasses (Watson & Bell 1975; Page 1978; Kellogg 2015). It is intriguing, however, that the Pariana- and Stipa-types are restricted to the BEP clade, and that the Pariana-type only occurs in the Bambusoideae subfamiliy (Fig. 3). Although the data are extremely limited (only 6 of the 1462 species contained within the Bambusoideae subfamily are present in our database, for example (Fig. 3)), this raises the possibility that a simple morphotyping approach based on SEM images of surface ornamentation might be used to distinguish the BEP clade and the Bambusoideae subfamily in the fossil record and in samples of modern material.

In general, very little SEM data on grass pollen surface ornamentation have been acquired (e.g. Fig. 2). The pollen of cereal grasses such as rye (Secale cereale), wheat (Triticum spp.) and barley (Hordeum vulgare) have received considerable morphological attention in the context of archaeobotanical studies of crop domestication (e.g. Köhler & Lange 1979), but the pollen morphology of smaller subfamilies, such as the Micrairoideae and the Danthonioideae, is unknown (see Fig. 3). This lack of data is problematic for several reasons. From an evolutionary perspective, it prevents phylogenetic analyses of the mechanisms that have generated the extant diversity of grass pollen form. For example, it is unclear whether the occurrence of the Triticum-type in the Panicoideae, Chloridoideae and Pooideae (Fig. 3) reflects multiple character state gains or the loss, reversal or regaining of characters (e.g. Cronk 2009). In this context, data on the pollen morphology of the early diverging APP clade of grasses would be useful in order to generate comparisons between more primitive and more derived subfamilies. As noted by Kellogg (2015), the lack of data also hinders efforts to examine whether aspects of grass pollen morphology, such as surface ornamentation, have any clear functional significance or whether they correlate with other aspects of biology, such as environment or mode of pollination (e.g. Soderstrom & Calderón 1971; Skvarla et al. 2003).

The lack of SEM data on grass pollen surface ornamentation also represents an obstacle to the classification of fossil grass pollen in terms of extant taxa (e.g. Lee et al. 2004). This is because workers are currently unable to compare fossil specimens to a database or monograph that displays the full range of natural morphological variation of extant taxa (e.g. Birks & Birks 1980; Mander & Punyasena 2014), and the morphological boundaries of grass pollen morphotypes are somewhat unclear as a result (Mander et al. 2013). Additionally, the commonness and rarity of each morphotype is unclear. For example, in the context of the entire Poaceae phylogeny, the Setaria-type occurs 38 times while the Pariana-type occurs only twice (Fig. 3). However, there are not enough data at present to reliably indicate whether certain morphotypes are common or rare in each subfamily and this information might help to constrain the affinities of fossil specimens.

We remain encouraged by the possibility that grass pollen can play a role in future palaeoecological studies of grasslands. We do note, though, that the spread of morphotypes across the Poaceae phylogeny suggests that this role is likely to be restricted to the generation of raw diversity counts, which could be used to track diversification history and compositional change through time. Such diversity counts could be calibrated by data from fossil phytoliths, which have high taxonomic resolution and allow workers to assess the presence or absence of important subfamilies (Piperno 2006; Strömberg 2011). Owing to the subtlety of grass pollen morphology and, in particular, the surface ornamentation patterns that we have focused on here, we feel that computational approaches (Mander et al. 2013) offer the best chance of generating consistently reproducible classifications (e.g. Mander & Punyasena 2014). However, the morphotyping approach to grass pollen classification that we have reviewed in this paper could be used to generate broad groupings of morphologically similar grass pollen grains in fossil samples, which could then be classified to finer taxonomic levels using computational methods. Morphotypes generated by human analysts could also be used as a yardstick against which to compare computational classifications (e.g. Mander et al. 2014) and, of course, also represent an archive of ancient biodiversity in their own right.