Pollen assemblage richness does not reflect regional plant species richness: a cautionary tale


Correspondence author. E-mail: goring@wisc.edu


  1. Palaeoecological records of species richness spanning time intervals over which climate variables have shifted relative to one another can help reduce issues of colinearity that might affect our understanding of patterns of species richness.

  2. Fossil pollen assemblages have the potential to serve as a proxy for past plant richness because they record the presence of plant taxa. However, pollen assemblages are typically limited by low taxonomic resolution and taphonomic processes (pollen production, transport, deposition and preservation), which may degrade the degree to which pollen accurately represents vegetation communities.

  3. We combined pollen assemblages from modern lake sediments (n = 546; n = 167 in British Columbia, Canada) in the Pacific Northwest, with a detailed data base (n = 16 071) of plant presence across the province and a published record of gamma richness to test the accuracy of pollen assemblages as a proxy for regional patterns of plant richness.

  4. A generalized linear model using plant richness resolved at multiple taxonomic levels suggests taxonomic differences between plant and pollen taxa may reduce the ability of pollen richness to predict plant richness at the site level, but that this relationship is still recoverable, albeit with broad confidence intervals.

  5. Spatially explicit analysis using a generalized additive model shows that predicted plant richness has no relationship with raw pollen assemblage richness at all taxonomic levels.

  6. The taxonomic composition of the region (i.e. the ratio of wind-pollinated to insect-pollinated species) and/or the morphological specificity of the dominant pollen types in the region may play a role in limiting the reconstruction of plant richness from pollen richness. Nonetheless, we believe this study is the first to empirically test the relationship between plant and pollen richness, and fails to find a significant relationship.

  7. Synthesis. Palynological richness in itself cannot be considered a universally reliable proxy for inferring plant richness; however, broad spatial and temporal patterns of change in richness have been reported in the literature. Our findings suggest that more work is needed to understand previously reported patterns of pollen assemblage richness through time and in space. We suggest the use of functional diversity or phylogenetically based analysis may help link pollen richness to plant community richness.


The preservation of biodiversity meets several objectives: it preserves the intrinsic value of species (Ghilarov 2000), maintains population and community stability (Isbell, Polley & Wilsey 2009) and serves to provide and maintain certain ecosystem services (Balvanera et al. 2006). Reich et al. (2012) show the explicit value of biodiversity in maintaining regional productivity. However, the challenge to researchers is that the underlying causes of biodiversity remain complex, and the long-term stability of biodiversity remains poorly understood (Turner 2004; Jablonski, Roy & Valentine 2006; Archibald et al. 2010). This may be partly as a result of the fact that it is difficult to describe patterns of biodiversity in a palaeoecological context (Rull 2012), although a number of studies have examined diversity or richness measures through time, and interest in understanding biodiversity using palaeobiological tools is evident in the literature (e.g. Colombaroli et al. 2012).

Palaeoecological records based on fossil pollen are confounded by differential pollen production across species and uneven preservation of specimens. Certain taxa produce pollen that rarely enters the sedimentary record, whether through differential pollen production, dispersal or preservation (Birks & Birks 1980). Taxa that are pollinated by insect or animal vectors (entomophilous or zoophilous, respectively), and species with limited dispersal ability are rarely recorded in fossil pollen records (Faegri & Iversen 1989; Jackson 1994). Other taxa such as Poaceae, representing 11 462 species world-wide (The Plant List 2010), produce pollen that cannot be differentiated at the species or genus level, with some exceptions (e.g. Zea and Cereale in Poaceae). Nonetheless, palynological richness is often used as an indicator of past species richness, either explicitly or implicitly (Birks & Line 1992; Jaramillo, Rueda & Mora 2006; Berglund et al. 2008; Birks & Birks 2008; Brown et al. 2008; van der Knaap 2009; Morales-Molino, García Antón & Morla 2011; Colombaroli et al. 2012), despite the fact that the relationship between plant richness and pollen richness has not been well studied. While pollen richness at a site may vary over time, temporal changes in richness may not be directly correlated with regional plant species richness. A study of nine sites in Estonia demonstrated that palynological richness may be largely controlled by variability in local pollen productivity (Meltsov et al. 2011). At coarse spatial scales, where multiple vegetation zones may be encountered, differences in catchment area may play a much larger role in determining richness (Seppä 1998).

Pollen richness is probably controlled only in part by the dominant vegetation in a region. At any one site, pollen richness is more likely to reflect plant richness when the taxonomic specificity of the pollen recorded at a site is high, and when there is a strong relationship between entomophilous and anemophilous (wind-pollinated) taxa. However, the relationship between the numbers of taxa for each pollination syndrome and the pollen specificity of the pollen taxa represented at a site is likely to vary spatially and temporally (Regal 1982). If relative numbers of entomophilous and anemophilous plant taxa vary spatially and these taxa have variable rates of deposition and preservation within the sedimentary environments from which fossil pollen is extracted, then we expect that the fundamental relationship between plant and pollen richness should not be universal and is likely to be weaker than the existing literature.

Odgaard (2001) outlines two competing but overlapping hypotheses on the mechanisms controlling pollen richness in sediments, those being (i) pollen richness is proportional to plant species richness and (ii) pollen richness is a function of pollen population evenness. These hypotheses relate directly to estimates of pollen richness using rarefaction analysis (Hurlbert 1971); thus, implicitly assuming that pollen reflects regional plant species richness, except inasmuch as analysis might affect this relationship. Weng, Hooghiemstra & Duivenvoorden (2006) and Peros & Gajewski (2008) present mathematical methods for testing pollen richness as a function of pollen counts, sediment accumulation rates, pollen concentration and palynological evenness, but these are based on a theoretical framework that assumes pollen richness is related to vegetation richness in a fundamentally knowable way. This may then be a case of ‘unknown knowns’ (Jackson 2012),that is, a literature exists relating pollen diversity to plant diversity, but the roots of this assumption are unclear.

Given the possible problems inherent in reconstructing plant species richness from pollen assemblages and the need for testing this relationship, we present a direct test of the relationship between plant and pollen richness using large plant and pollen data sets from British Columbia, Canada. We build spatial and point-based models relating pollen and plant taxonomic richness at multiple taxonomic levels to determine whether a relationship between pollen and plant richness can be detected, given the taxonomic and taphonomic constraints on pollen production and preservation and discuss the regional, global and disciplinary implications of our findings.

Materials and methods

Pollen data were obtained from the British Columbia Modern Pollen Database (BCMPD, Goring et al. 2009, 2010). This data set includes 546 lacustrine pollen sample sites from British Columbia, Canada, and Washington, Oregon, Idaho and Montana in the USA, of which 167 points are located within Canada. Plant data were obtained from the BEC Plants Database (Biogeoclimatic Ecosystem Classification Program 2013; nsites = 48 706). The BEC Plants data set represents a systematic survey of plant communities in the province of British Columbia using a standard 400 m2 plot design, except in alpine, grassland or wetland habitats. The sites cover a broad diversity of ecosystem types, soil classes, disturbance classes and other features. We removed all sites with low site-quality flags in the data set (= 33 067) leaving a total of = 14 529 sites. The pollen and plant data sets have different spatial distributions (Fig. 1), explained largely by differential investigator effort and purpose. Limitations of the distribution of pollen data are discussed further in Goring et al. (2009).

Figure 1.

Point distribution for the plant (a) and pollen (b) data sets across British Columbia, Canada, and into adjacent Washington, Idaho and Montana in the USA (for pollen).

Taxonomy for each data set was standardized to be internally consistent to the species level, excluding hybrid species in the BC Plant data set, to the pollen taxonomy of Whitmore et al. (2005). Plant taxonomy for the BC Plants data (family, genus and species) was verified using the accepted taxonomy in the ITIS data base (http://www.itis.gov/) using the r package ritis (ROpenSci project: http://ropensci.org). After taxonomic standardization, the pollen data set included 67 pollen taxa and the plant data set included 1729 species from 490 genera and 97 families.

We assessed richness for both plant and pollen sites as the sum of species or taxa present in a sample. While measures of evenness are important for understanding ecological processes, it is important to note that individual pollen grains are not equivalent to individual plants and that pollen production by species varies enormously. Additionally, the plant data base is structured in such a way that evenness is difficult to assess at any one plot since per cent coverage of taxa is divided into multiple vertical strata. Since the spatial distribution of plant and pollen sample sites is uneven across the region, and since pollen richness is a representation of a regional taxon pool and plant richness in this data set is an explicit measure of plot level diversity, we do not explicitly compare the two richness measures. A regional vascular plant richness data set does exist for British Columbia (Warman & Scudder 2007); however, this data set uses both plot level data and herbaria samples, aggregated to 50 km grid cells, without correcting for sampling effort, meaning that those cells with higher sampling effort are likely to show higher richness as a result of sampling effort, rather than underlying patterns of richness. Further, herbaria are likely to over-represent rare species and may not be suitable for estimates of species diversity at finer scales (ter Steege et al. 2011).

To assess whether or not pollen accurately reflects plant richness, we divided our analysis into units along the path from plant community to pollen assemblages (Fig. 2). Although the stages of our diagram should result in a cumulative loss of fidelity between plant community richness and pollen richness, or a progressive increase in the unexplained variance in our models, we partitioned our analysis in such a way that analysis at each step attempts to explicitly test only the target process or processes. All analyses were carried out in r (R Development Core Team 2011), and the supplement contains the raw code (as an R Markdown file) necessary to replicate this analysis.

Figure 2.

Flow chart illustrating the process by which plant communities are transformed to pollen assemblages in lake sediments. Plant species have pollen morphologies that are phylogenetically constrained and each taxon produces pollen in various amounts, dependent on the species traits. Pollen is transported from the plant and deposited in a depositional basin where it is preserved. The analyst processes and counts a certain number of pollen grains, and thus constructs the pollen assemblage.

The Effects of Taxonomic Smoothing

The first step in the process diagram (Fig. 2) involves testing whether pollen production, the transition from plant taxa to pollen taxa (as opposed to productivity), affects the fidelity of pollen richness to plant richness. Taxonomy in the plant data set is not equivalent to pollen taxonomy. While individuals in plant community plots are generally identified to species, pollen taxa are often only identifiable to genus, and sometimes to family. Pollen morphology is largely constrained phylogenetically, such that within certain groups (such as Poaceae), it is generally considered too difficult to differentiate between taxa. Other pollen taxa may be identified directly to species (e.g. Tsuga heterophylla). Finally, within some groups, it may be possible for analysts to differentiate between taxa (e.g. Alnus: May & Lacourse 2012), but the practice is not uniformly undertaken, such that records in large data bases contain a mixture of taxonomic levels that must, ultimately, be resolved to the lowest taxonomic equivalent. Effectively this coarsened taxonomic resolution can significantly affect the ecological interpretations possible with pollen data (Finkelstein, Gajewski & Viau 2006).

If the effects of differential taxonomic resolution between plant and pollen taxa are large enough it could impact upon the ability of pollen richness to represent plant richness, so we stratified both the plant and pollen taxonomies. We express the plant taxonomy at species, genus and family levels for all plant species. We treat pollen data as either ‘resolved’ (pollen taxa are defined as in Whitmore et al. 2005) or ‘pollen family’ (the pollen taxonomy extends only to the family level) (Table 1). We refer to the transition between higher and lower taxonomic resolutions as ‘taxonomic smoothing’. We also smooth the plant data to pollen equivalents, that is, plant species are assigned pollen taxa from the Whitmore et al. (2005) taxonomy. Pollen equivalents for the plant data are either ‘resolved’ or ‘pollen family’ level. In practice, 21% of species in the plant data set, mostly entomophilous taxa, have no equivalent pollen taxon. Plant taxa with no pollen equivalent were smoothed out of the data set when pollen equivalents were used.

Table 1. An example of several taxonomic equivalences. Pinus contorta, Tsuga heterophylla and Poa glauca all have equivalent pollen taxa; however, Pinus pollen can only be resolved to the genus level, while the fully resolved taxonomy for T. heterophylla is to the species level and P. glauca has a fully resolved pollen equivalence at the family level. When the pollen equivalence is smoothed to the family level P. contorta and T. heterophylla are both defined as Pinaceae, so the pollen equivalent richness for a site with both tree species in the plant data set would have a pollen family equivalence of only 1. Gentiana calycosa has no pollen equivalence, and so any plant plot with G. calycosa presence would have a pollen equivalence that excludes that taxon. The full table is available in Table S1 in the Supporting Information
SpeciesGenusFamilyFully resolvedPollen family
P. contorta Pinus Pinaceae Pinus Pinaceae
T. heterophylla Tsuga Pinaceae T. heterophylla Pinaceae
P. glauca Poa PoaceaePoaceaePoaceae
G. calycosa Gentiana Gentianaceaen.a.n.a.

The relationship between all plant taxonomic levels was examined using generalized linear models (GLM; ‘glm’ in r package stats; R Development Core Team 2011) relating the lower taxonomic level to the higher taxonomic level. Thus, for example, raw species richness is predicted from family richness. Predicting species level richness from family richness more accurately reflects the process used to estimate plant richness through time using fossil pollen data. In all cases, the GLM is fit using a Poisson's distribution.

The Effects of Taphonomy

The process of linking plant communities to pollen assemblages preserved in sediment occurs through a set of linked taphonomic processes (‘Effects of Taphonomy’, Fig. 2). Unfortunately, it is difficult to directly measure the individual taphonomic processes. As such, we used differences in the spatial patterns of richness between plant communities and pollen assemblages as a surrogate for the sum of the taphonomic effects.

The relationship between pollen and plant richness was tested with both the BC plant data set and the plant richness estimates generated by Warman & Scudder (2007). The relationship between predicted richness from the 50 km gridded richness estimates provided by Warman & Scudder (2007) and the observed pollen richness from the BCMPD was estimated using a GLM with a Poisson's distribution in R.

To ensure that we were using comparable measures of richness for pollen and plant data, we produced generalized additive models (‘GAM’ in r package mgcv; Wood 2011) that related alpha richness at various taxonomic levels to geographic position for both plant data and pollen data (northing, easting and elevation, all in metres with a BC Albers projection). Elevation is excluded from the pollen models due to the smaller sample size. Using these spatial models, we predicted plant richness at the locations of the pollen sample sites and tested for significant relationships. We did the same for the plant data, using predictions from the pollen models. We also tested for relationships between the plant models and the equivalent lower taxonomic resolutions within the plant data set, thus accounting for the effects of taxonomic smoothing discussed earlier.

The Effects of Rarefaction

Finally, we were interested in understanding how well the pollen counts reported in the pollen data set reflect the actual pollen richness at the sites (‘Effects of Rarefaction’, Fig. 2). We could not implement the methods of Weng, Hooghiemstra & Duivenvoorden (2006), since we were evaluating modern sediments and accumulation rates were unavailable for many of the modern assemblages. Accordingly, we rarefied all sites using ‘rarefy’ in r package vegan (Oksanen et al. 2012), with raw counts reported to estimate richness at a sample count of 300 to investigate the relationship between rarefied and raw counts for the data set since this is a common baseline value in palynological studies.


Effects of Taxonomic Smoothing

Plant species, genus and family richness are closely related for all plant taxonomic levels and their pollen equivalents (Fig. 3). Genus richness declines only slightly as the taxonomy is smoothed from species richness (Fig. 3a). Of the 490 genera in the plant data set, 27.6% appear as congenerics within plant community plots, and of those, only 7.3% have more than three species of the same genera in a plot at any time. The most frequent genera with congeneric presence are Vaccinium (16 sp. in BC, congeners in 27% of plots), Ranunculus (20 sp. in BC, congeners in 15% of plots) and Carex (113 sp. in BC, congeners in 10% of plots). Each of these genera has pollen equivalents at the family level (Ericaceae, Ranunculaceae and Cyperaceae, respectively) and so should be represented in the pollen record. There is a significant relationship between the number of taxa in a genus and the frequency of congeneric taxa from that genus in plots (glm, Poisson family, inline image = 19 700, < 0.001).

Figure 3.

The loss of plant species α-richness due to taxonomic smoothing to plant genus and family levels (a) and pollen equivalents (b) using the British Columbia plant data set. Convex hulls outline the point clouds for each relationship. The steeper the slope of the points, the more difficult it becomes to predict species richness from the taxonomically smoothed richness.

Family richness declines more sharply with increasing species richness, due to the greater prevalence of confamilial coexistence within plant communities (Fig. 3a). Asteraceae (211 sp. in BC, confamilials in 55% of plots), Poaceae (188 sp. in BC, confamilials in 49% of plots) and Ranunculaceae (62 sp. in BC, confamilials in 41% of plots) are the families with the most confamilial presence in plots, and all species in these families have family-level pollen equivalents. The relationship between the co-occurrence of confamilials in a plot and the number of species in a family is significant (glm, Poisson family, inline image = 96 200, < 0.001).

When species are directly transformed to their pollen equivalents, it becomes clear that, even though the relationship between species richness and pollen richness can still be recovered, it is even weaker than that between species richness and familial richness (Fig. 3a,b) with higher slopes resulting in broader confidence intervals. This is probably a result of both the weak taxonomic signature of the pollen types within British Columbia and the absence of many entomophilous taxa from pollen assemblages.

The Effects of Taphonomy

Spatial models (latitude, longitude and elevation) for the plant richness data (at species, genus and family levels and pollen equivalents; statistics reported in Table 2) show a complex pattern of diversity across the province that is highly correlated across all levels of taxonomic smoothing (Fig. 4, weakest correlation, between panels a and d; rpearson = 0.67, < 0.01). The smoothed surface does show edge effects where single points, near the margins of the spatial extent, can exert undue influence on the fit curve. These edge effects are most visible in the far northern corners of the province. This feature is limited to these edges and is far from the vast majority of pollen sites and so is unlikely to strongly affect our analysis.

Table 2. Results for spatially smoothed models of richness at various taxonomic levels, based on plant data
Taxonomic levelDeviance explained, %Est. d.f.χ2P-value
All plant species26.710717 300<0.001
Plant genera2710715 900<0.001
Plant families271079300<0.001
Plant species–pollen equivalents25.11066500<0.001
Plant species–pollen families23.91055200<0.001
Pollen genera46.636.0300<0.001
Pollen families39.326.9200<0.001
Figure 4.

Regional plant (a) and pollen (b) richness fit to the data using a three dimensional thin plate spline for latitude, longitude and elevation (only latitude and longitude for pollen). Only cells within 100 km of a sample site are shown. The data reveal differing patterns of richness for all plant species (a) and pollen richness (b). Other plant richness metrics are not shown because of the high correlation to plant species richness. High richness values in the extreme north-west (for pollen) and the north-east (for plants) are likely artefacts of smoothing, but have little effect on the overall correlation.

There is no reliable relationship between the smoothed (or regional) richness and the pollen richness as recorded at the BCMPD sites (Fig. 5) when actual pollen richness is compared against the regionally smoothed richness levels. Using the Warman & Scudder (2007) richness data for each pollen site also provides no reliable relationship (GLM, Poisson family, = 0.07, 2.5% deviance explained). Differences between Warman & Scudder (2007) richness maps and the richness maps presented here are likely to be the result of the different methods employed in creating the maps. As noted above, Warman & Scudder (2007) do not correct for sampling effort, and aggregate all species counts to the 50 km cell level; thus, cells with heavy sampling are more likely to have higher richness.

Figure 5.

The relationship between (a) smoothed pollen richness and observed plant richness in British Columbia and (b) smoothed plant richness and observed pollen richness. A linear fit shows a significant relationship between predicted and observed for the full data set, but this relationship accounts for little deviance and is negative, indicating that higher predicted values are associated with lower observed values for both pollen and plant data.

The lack of a relationship between regional richness and pollen richness at each pollen sample site in the BCMPD is supported by GLM modelling. The plant species richness model for all taxa (Figs 4a and 5a) produces a negative relationship between pollen richness and species richness across the province for the fully-resolved and family-level pollen classes, that is, higher pollen richness is associated with lower regional plant richness (inline image = 50.6, < 0.001). However, this model accounts for only 7% of the total model deviance. The regionally smoothed pollen equivalent richness (Fig. 4b) does not show a significant fit between pollen equivalent richness for all plant species and the actual pollen richness (Fig. 5b; inline image = 1.9, = 0.11). These results suggest that even when a fit between regional richness and local pollen richness can be found at the provincial scale, the amount of deviance left unexplained indicates a very poor fit and the likelihood that taphonomic effects swamp any relationships between pollen and plant richness at a site level.

The Effects of Rarefaction

Pollen richness in the BCMPD ranges from 3 to 27 taxa with a mean richness of 12 taxa. The reported richness and the rarefied estimates (300 grains) are strongly related (Table 3). Since most samples in the data set have relatively high numbers of identified pollen, rarefying counts to a baseline of 300 grains reduces the apparent richness, but the effect is nearly linear in the region of richness we are examining. This near-linear fit indicates that rarefaction is unlikely to be a major driver of apparent richness in our pollen samples. The linearity of the relationship following rarefaction and the large number of samples with percentage data only, for which rarefaction cannot be applied, supports the use of raw (non-rarefied) pollen richness in our analysis.

Table 3. Correlations for rarefied and raw pollen richness at the genera and family levels using Pearson's product-moment correlation
 Rarefied count
Estimate300 grains
Correlation of genera richness (rpearson)0.97 ± 0.01
Correlation of family richness (rpearson)0.97 ± 0.02


There is a clear relationship between species, genus and family richness represented in the raw British Columbia plant data, regardless of taxonomic smoothing or data set reduction. Richness for both plant and pollen data is probably related to both climate and historical patterns of plant migration during the late-glacial and Holocene. In particular, the presence of very high plant richness in the interior may be a result of the recent colonization of the region by more temperate, mesophilous taxa (Gavin et al. 2009). However, we failed to obtain clear and meaningful relationships between measures of plant richness and pollen richness at any spatial scale and at any taxonomic level.

The lack of a relationship between pollen richness and any measure of plant richness points to the joint importance of the regional vegetation and taphonomic processes in assessing the relationship between pollen richness and plant species richness. The high correlation between pollen counts and rarefied counts (Table 3) indicates that the number of pollen grains counted, and the evenness of the pollen assemblages (even though we did not measure it explicitly here) is unlikely to play a role in degrading this relationship since loss of fidelity through rarefaction can be ascribed to highly uneven samples (e.g. Olszewski 2004). Significant relationships between parent plant taxa and the smoothed pollen equivalents suggest that the effect of taxonomic smoothing on interpreting species richness from pollen data can be quite large, but can be recoverable (Fig. 3). However, the spatial models and the direct comparison between pollen data and the site level plant community data suggest that the differential pollen production, dispersal and preservation that occurs between (and inclusive of) pollen production and deposition overwhelm any such relationship.

One issue driving the lack of relationship between the richness measures may be the lack of taxonomic resolution in the pollen data set. The taxonomic standard used here is the same as in the North American Modern Pollen Database (Whitmore et al. 2005), except that taxa with both differentiated and undifferentiated types (e.g. Pinus haploxylon and diploxylon and Pinus undiff.) cannot be reliably compared at the differentiated level across samples since we cannot know from which subtype the undifferentiated pollen type is derived. The Asteraceae are an interesting problem. Although several Asteraceae types are readily distinguished, the North American Pollen Database and the BCMPD contain a number of sites with both differentiated (e.g. Aster Cichoroides) and undifferentiated Asteraceae pollen. This presents a problem in defining a standard taxonomy: should these be treated as two taxa, or one? In this analysis, we treat all Asteraceae as a single taxon with the exception of the most common, genera-level classes such as Ambrosia and Artemisia. In practice, this is an issue that should be resolved; however, since the relationship between pollen and plant species richness is so poor, it is unlikely that resolving this single taxonomic unit would resolve the relationship.

If researchers work to a common, higher taxonomic standard, identifying evolutionarily significant units (sensu Rull 2011) rather than higher-level taxonomic groupings, then it may be possible to obtain a closer relationship between pollen and plant richness. However, issues of pollen preservation and processing, as well as analytical skill will continue to affect our ability to resolve pollen to a high taxonomic resolution in large data sets. Salonen et al. (2012) show the possibility of developing a high resolution regional data set, given sufficient coordination among pollen analysts.

Although the two metrics, plant richness from local plots and pollen richness from lake sediments, may not be directly comparable (plant richness reflects local richness, pollen richness is some measure of regional richness), the spatial GAM for plant species richness does capture a regional pattern of plant species richness for British Columbia; thus, the comparisons of predicted plant richness and observed pollen richness, and vice versa, should show patterns should they exist, however, it appears that no such pattern exists.

While a regional relationship may not exist between plant richness and pollen richness in British Columbia, we cannot assume that the relationship is absent at all spatial scales in all regions, although the role of taphonomic processes might ensure that pollen richness will fail to accurately represent plant species richness elsewhere. Our results suggest that the pollen–plant richness relationship does not exist in British Columbia, but there are broad biogeographic patterns of richness in both anemophilous and entomophilous species (Regal 1982), for the taxonomic resolution of pollen taxa and for the relationship between family and species level diversity, suggesting that the lack of a relationship may not necessarily be pervasive in modern settings.

The dominant vegetation in British Columbia (e.g. coniferous, temperate rain forest dominated by Thuja plicata and T. heterophylla on the coast and boreal type forests dominated by Pinus and Picea species in the interior) is represented by pollen taxa that are generally poorly resolved taxonomically (e.g. 62 tree taxa identified in the data base are represented by only nine pollen taxa), and a significant proportion (21%) of the plant taxa do not show up in the pollen record at all. The low taxonomic resolution in pollen assemblages, the large number of entomophilous taxa and the taphonomic processes that occur between pollen liberation and its eventual preservation in depositional basins result in the lack of a meaningful statistical relationship between measures of plant and pollen richness. This calls into question the use of fossil pollen assemblages as a measure of species richness through time and has implications for understanding relationship between changes in climate and richness over time (Birks & Birks 2008; Brown et al. 2008). Further research to investigate relationships between pollen and plant richness at broad spatial scales is needed in other regions in order to assess the extent to which our results indicate a universal shortcoming in the use of palynological richness as a proxy for plant richness.

It is also possible that at very local scales (forest hollows for example) this relationship may be more significant, where local pollen influx dominates the pollen assemblages. However, if it is the case that broader spatial scales result in a gradual coarsening of this relationship, and if differing vegetation classes support differing plant–pollen richness relationships (Seppä 1998), then perhaps even forest hollows may be of little use at longer time-scales since the likelihood of regional vegetation change at a site would disrupt the relationship between plant and pollen richness over time.

Given our findings, what are we to make of studies that present changes in richness over time and space that are based on pollen assemblages (e.g. Seppä 1998; Willis et al. 2007; Birks & Birks 2008; Morales-Molino, García Antón & Morla 2011; Colombaroli et al. 2012). Jaramillo, Rueda & Mora (2006) use an extended record to show a strong relationship between pollen and spore morphotype richness through time and climate. Morales-Molino, García Antón & Morla (2011) show a strong significant relationship between the percentage of herbaceous pollen and palynological richness, and relate that to processes of forest clearance, suggesting that fire activity promotes species diversity. Our results suggest that these changes in pollen richness may not be explicitly related to changes in plant richness, potentially calling these results into question.

Spatial changes in pollen richness in BC are not necessarily related to changes in plant richness, and as such must be driven by other processes. Seppä (1998) indicates that increases in pollen richness are associated with rapid community change and Meltsov et al. (2011) indicate that increased local richness is related to pollen source area and productivity. If the plant community composition strongly affects pollen richness, but plant community richness does not, how might we begin to understand patterns of plant richness through time?

Investigating the differences in the climatic relationships between plant and pollen richness, both in time and in space may provide more information about the relationship between plant and pollen richness as well. Relating downscaled climate data products such as ClimateWNA (Wang et al. 2012) to local richness measures in a modern context, and relating palaeo-richness from sediment cores to transient general circulation model runs (e.g. Liu et al. 2009) may provide an indication of both the stability of pollen richness metrics with respect to modern climate, and the relationship between pollen and vegetation richness over time, if coupled with species distribution models for plant species assemblages predicted through the Holocene to produce estimates of palaeo-plant richness.

It may also be possible to use functional trait or phylogenetic information to unite the plant and pollen taxa, such that the richness values from pollen are not evaluated on their own, but in a multivariate form that provides information about the structure of the pollen assemblage in an evolutionary or functional manner. This information may be integrated in measures of functional richness (Mason et al. 2005), but the choice of functional characters may strongly affect our ability to detect a relationship. If anemophily is phylogenetically constrained, then the use of functional data may prove to be a useful tool in connecting modern plant communities to pollen in a much more powerful way. This approach, using both species and traits, and accounting for phylogenetic constraints has been advocated on a broader scale for macroecological studies by Beck et al. (2012) and will allow greater integration of palaeoecological data and analysis into macroecological research. The development of phylogenies for the BC plant data and for the BC pollen data will allow us to compare various phylogenetic metrics to richness measures at local and regional scales. However, the taxonomic structure of pollen data may prove challenging to recreate in phylogenetic form, given that plant species within a single family may be identifiable as pollen only to family, to genus or to species.

Ultimately, pollen richness is one metric of many provided by palaeoecological data. Broad interest in patterns of richness means that pollen richness may be viewed as a valuable tool, but our results show that the use of this tool has important caveats, and requires further investigation before a straightforward interpretation of pollen richness can exist. The fundamental inability to relate pollen richness to plant richness in this analysis does not invalidate other studies that show robust changes in pollen richness over time, but it calls into question whether these changes reflect underlying changes in plant richness, or some other change in plant community composition or structure.


The authors would like to thank support from NSF Macrosystems Grant EF-1065656 for S.J.G.; R.W.M. and T.L. are supported by grants from NSERC. Two anonymous referees provided excellent comments that improved the manuscript considerably. S.J.G. would like to thank discussions with the Williams Lab at the University of Wisconsin-Madison, Will MacKenzie of the British Columbia Ministry of Forests and Leanne Warman for helping to clarify ideas and data contained herein. The authors have no conflict of interest to declare. The r package ritis is part of the ROpenSci project, http://ropensci.org.